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 8192
92 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2048
93 #define SECT_PER_MB_SHIFT 11
94 #define MAX_SECTOR_SIZE 4096
95 #define MULTIPLE_PPL_AREA_SIZE_IMSM (1024 * 1024) /* Size of the whole
100 * Internal Write-intent bitmap is stored in the same area where PPL.
101 * Both features are mutually exclusive, so it is not an issue.
102 * The first 8KiB of the area are reserved and shall not be used.
104 #define IMSM_BITMAP_AREA_RESERVED_SIZE 8192
106 #define IMSM_BITMAP_HEADER_OFFSET (IMSM_BITMAP_AREA_RESERVED_SIZE)
107 #define IMSM_BITMAP_HEADER_SIZE MAX_SECTOR_SIZE
109 #define IMSM_BITMAP_START_OFFSET (IMSM_BITMAP_HEADER_OFFSET + IMSM_BITMAP_HEADER_SIZE)
110 #define IMSM_BITMAP_AREA_SIZE (MULTIPLE_PPL_AREA_SIZE_IMSM - IMSM_BITMAP_START_OFFSET)
111 #define IMSM_BITMAP_AND_HEADER_SIZE (IMSM_BITMAP_AREA_SIZE + IMSM_BITMAP_HEADER_SIZE)
113 #define IMSM_DEFAULT_BITMAP_CHUNKSIZE (64 * 1024 * 1024)
114 #define IMSM_DEFAULT_BITMAP_DAEMON_SLEEP 5
117 * This macro let's us ensure that no-one accidentally
118 * changes the size of a struct
120 #define ASSERT_SIZE(_struct, size) \
121 static inline void __assert_size_##_struct(void) \
125 case (sizeof(struct _struct) == size): break; \
129 /* Disk configuration info. */
130 #define IMSM_MAX_DEVICES 255
132 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
133 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
134 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
135 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
136 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
137 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
138 #define JOURNAL_DISK __cpu_to_le32(0x2000000) /* Device marked as Journaling Drive */
139 __u32 status
; /* 0xF0 - 0xF3 */
140 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
141 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
142 #define IMSM_DISK_FILLERS 3
143 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
145 ASSERT_SIZE(imsm_disk
, 48)
147 /* map selector for map managment
153 /* RAID map configuration infos. */
155 __u32 pba_of_lba0_lo
; /* start address of partition */
156 __u32 blocks_per_member_lo
;/* blocks per member */
157 __u32 num_data_stripes_lo
; /* number of data stripes */
158 __u16 blocks_per_strip
;
159 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
160 #define IMSM_T_STATE_NORMAL 0
161 #define IMSM_T_STATE_UNINITIALIZED 1
162 #define IMSM_T_STATE_DEGRADED 2
163 #define IMSM_T_STATE_FAILED 3
165 #define IMSM_T_RAID0 0
166 #define IMSM_T_RAID1 1
167 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
168 __u8 num_members
; /* number of member disks */
169 __u8 num_domains
; /* number of parity domains */
170 __u8 failed_disk_num
; /* valid only when state is degraded */
172 __u32 pba_of_lba0_hi
;
173 __u32 blocks_per_member_hi
;
174 __u32 num_data_stripes_hi
;
175 __u32 filler
[4]; /* expansion area */
176 #define IMSM_ORD_REBUILD (1 << 24)
177 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
178 * top byte contains some flags
181 ASSERT_SIZE(imsm_map
, 52)
184 __u32 curr_migr_unit_lo
;
185 __u32 checkpoint_id
; /* id to access curr_migr_unit */
186 __u8 migr_state
; /* Normal or Migrating */
188 #define MIGR_REBUILD 1
189 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
190 #define MIGR_GEN_MIGR 3
191 #define MIGR_STATE_CHANGE 4
192 #define MIGR_REPAIR 5
193 __u8 migr_type
; /* Initializing, Rebuilding, ... */
194 #define RAIDVOL_CLEAN 0
195 #define RAIDVOL_DIRTY 1
196 #define RAIDVOL_DSRECORD_VALID 2
198 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
199 __u16 verify_errors
; /* number of mismatches */
200 __u16 bad_blocks
; /* number of bad blocks during verify */
201 __u32 curr_migr_unit_hi
;
203 struct imsm_map map
[1];
204 /* here comes another one if migr_state */
206 ASSERT_SIZE(imsm_vol
, 84)
209 __u8 volume
[MAX_RAID_SERIAL_LEN
];
212 #define DEV_BOOTABLE __cpu_to_le32(0x01)
213 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
214 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
215 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
216 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
217 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
218 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
219 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
220 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
221 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
222 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
223 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
224 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
225 __u32 status
; /* Persistent RaidDev status */
226 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
230 __u8 cng_master_disk
;
234 __u16 my_vol_raid_dev_num
; /* Used in Unique volume Id for this RaidDev */
240 /* Unique Volume Id of the NvCache Volume associated with this volume */
241 __u32 nvc_vol_orig_family_num
;
242 __u16 nvc_vol_raid_dev_num
;
245 #define RWH_DISTRIBUTED 1
246 #define RWH_JOURNALING_DRIVE 2
247 #define RWH_MULTIPLE_DISTRIBUTED 3
248 #define RWH_MULTIPLE_PPLS_JOURNALING_DRIVE 4
249 #define RWH_MULTIPLE_OFF 5
251 __u8 rwh_policy
; /* Raid Write Hole Policy */
252 __u8 jd_serial
[MAX_RAID_SERIAL_LEN
]; /* Journal Drive serial number */
255 #define IMSM_DEV_FILLERS 3
256 __u32 filler
[IMSM_DEV_FILLERS
];
259 ASSERT_SIZE(imsm_dev
, 164)
262 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
263 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
264 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
265 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
266 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
267 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
268 __u32 attributes
; /* 0x34 - 0x37 */
269 __u8 num_disks
; /* 0x38 Number of configured disks */
270 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
271 __u8 error_log_pos
; /* 0x3A */
272 __u8 fill
[1]; /* 0x3B */
273 __u32 cache_size
; /* 0x3c - 0x40 in mb */
274 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
275 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
276 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
277 __u16 num_raid_devs_created
; /* 0x4C - 0x4D Used for generating unique
278 * volume IDs for raid_dev created in this array
281 __u16 filler1
; /* 0x4E - 0x4F */
282 __u64 creation_time
; /* 0x50 - 0x57 Array creation time */
283 #define IMSM_FILLERS 32
284 __u32 filler
[IMSM_FILLERS
]; /* 0x58 - 0xD7 RAID_MPB_FILLERS */
285 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
286 /* here comes imsm_dev[num_raid_devs] */
287 /* here comes BBM logs */
289 ASSERT_SIZE(imsm_super
, 264)
291 #define BBM_LOG_MAX_ENTRIES 254
292 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
293 #define BBM_LOG_SIGNATURE 0xabadb10c
295 struct bbm_log_block_addr
{
298 } __attribute__ ((__packed__
));
300 struct bbm_log_entry
{
301 __u8 marked_count
; /* Number of blocks marked - 1 */
302 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
303 struct bbm_log_block_addr defective_block_start
;
304 } __attribute__ ((__packed__
));
307 __u32 signature
; /* 0xABADB10C */
309 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
311 ASSERT_SIZE(bbm_log
, 2040)
313 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
315 #define BLOCKS_PER_KB (1024/512)
317 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
319 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
321 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
322 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
323 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
326 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
327 * be recovered using srcMap */
328 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
329 * already been migrated and must
330 * be recovered from checkpoint area */
332 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of single PPL, without the header */
335 __u32 rec_status
; /* Status used to determine how to restart
336 * migration in case it aborts
338 __u32 curr_migr_unit_lo
; /* 0..numMigrUnits-1 */
339 __u32 family_num
; /* Family number of MPB
340 * containing the RaidDev
341 * that is migrating */
342 __u32 ascending_migr
; /* True if migrating in increasing
344 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
345 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
347 * advances per unit-of-operation */
348 __u32 ckpt_area_pba_lo
; /* Pba of first block of ckpt copy area */
349 __u32 dest_1st_member_lba_lo
; /* First member lba on first
350 * stripe of destination */
351 __u32 num_migr_units_lo
; /* Total num migration units-of-op */
352 __u32 post_migr_vol_cap
; /* Size of volume after
353 * migration completes */
354 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
355 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
356 * migration ckpt record was read from
357 * (for recovered migrations) */
358 __u32 curr_migr_unit_hi
; /* 0..numMigrUnits-1 high order 32 bits */
359 __u32 ckpt_area_pba_hi
; /* Pba of first block of ckpt copy area
360 * high order 32 bits */
361 __u32 dest_1st_member_lba_hi
; /* First member lba on first stripe of
362 * destination - high order 32 bits */
363 __u32 num_migr_units_hi
; /* Total num migration units-of-op
364 * high order 32 bits */
367 ASSERT_SIZE(migr_record
, 128)
372 * 2: metadata does not match
380 struct md_list
*next
;
383 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
385 static __u8
migr_type(struct imsm_dev
*dev
)
387 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
388 dev
->status
& DEV_VERIFY_AND_FIX
)
391 return dev
->vol
.migr_type
;
394 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
396 /* for compatibility with older oroms convert MIGR_REPAIR, into
397 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
399 if (migr_type
== MIGR_REPAIR
) {
400 dev
->vol
.migr_type
= MIGR_VERIFY
;
401 dev
->status
|= DEV_VERIFY_AND_FIX
;
403 dev
->vol
.migr_type
= migr_type
;
404 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
408 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
410 return ROUND_UP(bytes
, sector_size
) / sector_size
;
413 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
414 unsigned int sector_size
)
416 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
420 struct imsm_dev
*dev
;
421 struct intel_dev
*next
;
426 enum sys_dev_type type
;
429 struct intel_hba
*next
;
436 /* internal representation of IMSM metadata */
439 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
440 struct imsm_super
*anchor
; /* immovable parameters */
443 void *migr_rec_buf
; /* buffer for I/O operations */
444 struct migr_record
*migr_rec
; /* migration record */
446 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
447 array, it indicates that mdmon is allowed to clean migration
449 size_t len
; /* size of the 'buf' allocation */
450 size_t extra_space
; /* extra space in 'buf' that is not used yet */
451 void *next_buf
; /* for realloc'ing buf from the manager */
453 int updates_pending
; /* count of pending updates for mdmon */
454 int current_vol
; /* index of raid device undergoing creation */
455 unsigned long long create_offset
; /* common start for 'current_vol' */
456 __u32 random
; /* random data for seeding new family numbers */
457 struct intel_dev
*devlist
;
458 unsigned int sector_size
; /* sector size of used member drives */
462 __u8 serial
[MAX_RAID_SERIAL_LEN
];
465 struct imsm_disk disk
;
468 struct extent
*e
; /* for determining freespace @ create */
469 int raiddisk
; /* slot to fill in autolayout */
471 } *disks
, *current_disk
;
472 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
474 struct dl
*missing
; /* disks removed while we weren't looking */
475 struct bbm_log
*bbm_log
;
476 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
477 const struct imsm_orom
*orom
; /* platform firmware support */
478 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
479 struct md_bb bb
; /* memory for get_bad_blocks call */
483 struct imsm_disk disk
;
484 #define IMSM_UNKNOWN_OWNER (-1)
486 struct intel_disk
*next
;
490 unsigned long long start
, size
;
493 /* definitions of reshape process types */
494 enum imsm_reshape_type
{
500 /* definition of messages passed to imsm_process_update */
501 enum imsm_update_type
{
502 update_activate_spare
,
506 update_add_remove_disk
,
507 update_reshape_container_disks
,
508 update_reshape_migration
,
510 update_general_migration_checkpoint
,
512 update_prealloc_badblocks_mem
,
516 struct imsm_update_activate_spare
{
517 enum imsm_update_type type
;
521 struct imsm_update_activate_spare
*next
;
527 unsigned long long size
;
534 enum takeover_direction
{
538 struct imsm_update_takeover
{
539 enum imsm_update_type type
;
541 enum takeover_direction direction
;
544 struct imsm_update_reshape
{
545 enum imsm_update_type type
;
549 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
552 struct imsm_update_reshape_migration
{
553 enum imsm_update_type type
;
556 /* fields for array migration changes
563 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
566 struct imsm_update_size_change
{
567 enum imsm_update_type type
;
572 struct imsm_update_general_migration_checkpoint
{
573 enum imsm_update_type type
;
574 __u64 curr_migr_unit
;
578 __u8 serial
[MAX_RAID_SERIAL_LEN
];
581 struct imsm_update_create_array
{
582 enum imsm_update_type type
;
587 struct imsm_update_kill_array
{
588 enum imsm_update_type type
;
592 struct imsm_update_rename_array
{
593 enum imsm_update_type type
;
594 __u8 name
[MAX_RAID_SERIAL_LEN
];
598 struct imsm_update_add_remove_disk
{
599 enum imsm_update_type type
;
602 struct imsm_update_prealloc_bb_mem
{
603 enum imsm_update_type type
;
606 struct imsm_update_rwh_policy
{
607 enum imsm_update_type type
;
612 static const char *_sys_dev_type
[] = {
613 [SYS_DEV_UNKNOWN
] = "Unknown",
614 [SYS_DEV_SAS
] = "SAS",
615 [SYS_DEV_SATA
] = "SATA",
616 [SYS_DEV_NVME
] = "NVMe",
617 [SYS_DEV_VMD
] = "VMD"
620 const char *get_sys_dev_type(enum sys_dev_type type
)
622 if (type
>= SYS_DEV_MAX
)
623 type
= SYS_DEV_UNKNOWN
;
625 return _sys_dev_type
[type
];
628 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
630 struct intel_hba
*result
= xmalloc(sizeof(*result
));
632 result
->type
= device
->type
;
633 result
->path
= xstrdup(device
->path
);
635 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
641 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
643 struct intel_hba
*result
;
645 for (result
= hba
; result
; result
= result
->next
) {
646 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
652 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
654 struct intel_hba
*hba
;
656 /* check if disk attached to Intel HBA */
657 hba
= find_intel_hba(super
->hba
, device
);
660 /* Check if HBA is already attached to super */
661 if (super
->hba
== NULL
) {
662 super
->hba
= alloc_intel_hba(device
);
667 /* Intel metadata allows for all disks attached to the same type HBA.
668 * Do not support HBA types mixing
670 if (device
->type
!= hba
->type
)
673 /* Multiple same type HBAs can be used if they share the same OROM */
674 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
676 if (device_orom
!= super
->orom
)
682 hba
->next
= alloc_intel_hba(device
);
686 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
688 struct sys_dev
*list
, *elem
;
691 if ((list
= find_intel_devices()) == NULL
)
695 disk_path
= (char *) devname
;
697 disk_path
= diskfd_to_devpath(fd
);
702 for (elem
= list
; elem
; elem
= elem
->next
)
703 if (path_attached_to_hba(disk_path
, elem
->path
))
706 if (disk_path
!= devname
)
712 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
715 static struct supertype
*match_metadata_desc_imsm(char *arg
)
717 struct supertype
*st
;
719 if (strcmp(arg
, "imsm") != 0 &&
720 strcmp(arg
, "default") != 0
724 st
= xcalloc(1, sizeof(*st
));
725 st
->ss
= &super_imsm
;
726 st
->max_devs
= IMSM_MAX_DEVICES
;
727 st
->minor_version
= 0;
732 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
734 return &mpb
->sig
[MPB_SIG_LEN
];
737 /* retrieve a disk directly from the anchor when the anchor is known to be
738 * up-to-date, currently only at load time
740 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
742 if (index
>= mpb
->num_disks
)
744 return &mpb
->disk
[index
];
747 /* retrieve the disk description based on a index of the disk
750 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
754 for (d
= super
->disks
; d
; d
= d
->next
)
755 if (d
->index
== index
)
760 /* retrieve a disk from the parsed metadata */
761 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
765 dl
= get_imsm_dl_disk(super
, index
);
772 /* generate a checksum directly from the anchor when the anchor is known to be
773 * up-to-date, currently only at load or write_super after coalescing
775 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
777 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
778 __u32
*p
= (__u32
*) mpb
;
782 sum
+= __le32_to_cpu(*p
);
786 return sum
- __le32_to_cpu(mpb
->check_sum
);
789 static size_t sizeof_imsm_map(struct imsm_map
*map
)
791 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
794 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
796 /* A device can have 2 maps if it is in the middle of a migration.
798 * MAP_0 - we return the first map
799 * MAP_1 - we return the second map if it exists, else NULL
800 * MAP_X - we return the second map if it exists, else the first
802 struct imsm_map
*map
= &dev
->vol
.map
[0];
803 struct imsm_map
*map2
= NULL
;
805 if (dev
->vol
.migr_state
)
806 map2
= (void *)map
+ sizeof_imsm_map(map
);
808 switch (second_map
) {
825 /* return the size of the device.
826 * migr_state increases the returned size if map[0] were to be duplicated
828 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
830 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
831 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
833 /* migrating means an additional map */
834 if (dev
->vol
.migr_state
)
835 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
837 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
842 /* retrieve disk serial number list from a metadata update */
843 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
846 struct disk_info
*inf
;
848 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
849 sizeof_imsm_dev(&update
->dev
, 0);
854 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
860 if (index
>= mpb
->num_raid_devs
)
863 /* devices start after all disks */
864 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
866 for (i
= 0; i
<= index
; i
++)
868 return _mpb
+ offset
;
870 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
875 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
877 struct intel_dev
*dv
;
879 if (index
>= super
->anchor
->num_raid_devs
)
881 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
882 if (dv
->index
== index
)
887 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
890 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
891 __le16_to_cpu(addr
->w1
));
894 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
896 struct bbm_log_block_addr addr
;
898 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
899 addr
.dw1
= __cpu_to_le32((__u32
)(sec
>> 16) & 0xffffffff);
903 /* get size of the bbm log */
904 static __u32
get_imsm_bbm_log_size(struct bbm_log
*log
)
906 if (!log
|| log
->entry_count
== 0)
909 return sizeof(log
->signature
) +
910 sizeof(log
->entry_count
) +
911 log
->entry_count
* sizeof(struct bbm_log_entry
);
914 /* check if bad block is not partially stored in bbm log */
915 static int is_stored_in_bbm(struct bbm_log
*log
, const __u8 idx
, const unsigned
916 long long sector
, const int length
, __u32
*pos
)
920 for (i
= *pos
; i
< log
->entry_count
; i
++) {
921 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
922 unsigned long long bb_start
;
923 unsigned long long bb_end
;
925 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
926 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
928 if ((entry
->disk_ordinal
== idx
) && (bb_start
>= sector
) &&
929 (bb_end
<= sector
+ length
)) {
937 /* record new bad block in bbm log */
938 static int record_new_badblock(struct bbm_log
*log
, const __u8 idx
, unsigned
939 long long sector
, int length
)
943 struct bbm_log_entry
*entry
= NULL
;
945 while (is_stored_in_bbm(log
, idx
, sector
, length
, &pos
)) {
946 struct bbm_log_entry
*e
= &log
->marked_block_entries
[pos
];
948 if ((e
->marked_count
+ 1 == BBM_LOG_MAX_LBA_ENTRY_VAL
) &&
949 (__le48_to_cpu(&e
->defective_block_start
) == sector
)) {
950 sector
+= BBM_LOG_MAX_LBA_ENTRY_VAL
;
951 length
-= BBM_LOG_MAX_LBA_ENTRY_VAL
;
960 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
961 BBM_LOG_MAX_LBA_ENTRY_VAL
;
962 entry
->defective_block_start
= __cpu_to_le48(sector
);
963 entry
->marked_count
= cnt
- 1;
970 new_bb
= ROUND_UP(length
, BBM_LOG_MAX_LBA_ENTRY_VAL
) /
971 BBM_LOG_MAX_LBA_ENTRY_VAL
;
972 if (log
->entry_count
+ new_bb
> BBM_LOG_MAX_ENTRIES
)
976 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
977 BBM_LOG_MAX_LBA_ENTRY_VAL
;
978 struct bbm_log_entry
*entry
=
979 &log
->marked_block_entries
[log
->entry_count
];
981 entry
->defective_block_start
= __cpu_to_le48(sector
);
982 entry
->marked_count
= cnt
- 1;
983 entry
->disk_ordinal
= idx
;
994 /* clear all bad blocks for given disk */
995 static void clear_disk_badblocks(struct bbm_log
*log
, const __u8 idx
)
999 while (i
< log
->entry_count
) {
1000 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
1002 if (entries
[i
].disk_ordinal
== idx
) {
1003 if (i
< log
->entry_count
- 1)
1004 entries
[i
] = entries
[log
->entry_count
- 1];
1012 /* clear given bad block */
1013 static int clear_badblock(struct bbm_log
*log
, const __u8 idx
, const unsigned
1014 long long sector
, const int length
) {
1017 while (i
< log
->entry_count
) {
1018 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
1020 if ((entries
[i
].disk_ordinal
== idx
) &&
1021 (__le48_to_cpu(&entries
[i
].defective_block_start
) ==
1022 sector
) && (entries
[i
].marked_count
+ 1 == length
)) {
1023 if (i
< log
->entry_count
- 1)
1024 entries
[i
] = entries
[log
->entry_count
- 1];
1034 /* allocate and load BBM log from metadata */
1035 static int load_bbm_log(struct intel_super
*super
)
1037 struct imsm_super
*mpb
= super
->anchor
;
1038 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1040 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
1041 if (!super
->bbm_log
)
1045 struct bbm_log
*log
= (void *)mpb
+
1046 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1050 if (bbm_log_size
< sizeof(log
->signature
) +
1051 sizeof(log
->entry_count
))
1054 entry_count
= __le32_to_cpu(log
->entry_count
);
1055 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
1056 (entry_count
> BBM_LOG_MAX_ENTRIES
))
1060 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
1061 entry_count
* sizeof(struct bbm_log_entry
))
1064 memcpy(super
->bbm_log
, log
, bbm_log_size
);
1066 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
1067 super
->bbm_log
->entry_count
= 0;
1073 /* checks if bad block is within volume boundaries */
1074 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
1075 const unsigned long long start_sector
,
1076 const unsigned long long size
)
1078 unsigned long long bb_start
;
1079 unsigned long long bb_end
;
1081 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1082 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1084 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1085 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1091 /* get list of bad blocks on a drive for a volume */
1092 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1093 const unsigned long long start_sector
,
1094 const unsigned long long size
,
1100 for (i
= 0; i
< log
->entry_count
; i
++) {
1101 const struct bbm_log_entry
*ent
=
1102 &log
->marked_block_entries
[i
];
1103 struct md_bb_entry
*bb
;
1105 if ((ent
->disk_ordinal
== idx
) &&
1106 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1108 if (!bbs
->entries
) {
1109 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1115 bb
= &bbs
->entries
[count
++];
1116 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1117 bb
->length
= ent
->marked_count
+ 1;
1125 * == MAP_0 get first map
1126 * == MAP_1 get second map
1127 * == MAP_X than get map according to the current migr_state
1129 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1133 struct imsm_map
*map
;
1135 map
= get_imsm_map(dev
, second_map
);
1137 /* top byte identifies disk under rebuild */
1138 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1141 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1142 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1144 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1146 return ord_to_idx(ord
);
1149 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1151 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1154 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
1159 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1160 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1161 if (ord_to_idx(ord
) == idx
)
1168 static int get_imsm_raid_level(struct imsm_map
*map
)
1170 if (map
->raid_level
== 1) {
1171 if (map
->num_members
== 2)
1177 return map
->raid_level
;
1180 static int cmp_extent(const void *av
, const void *bv
)
1182 const struct extent
*a
= av
;
1183 const struct extent
*b
= bv
;
1184 if (a
->start
< b
->start
)
1186 if (a
->start
> b
->start
)
1191 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1193 int memberships
= 0;
1196 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1197 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1198 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1200 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
1207 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1209 static int split_ull(unsigned long long n
, void *lo
, void *hi
)
1211 if (lo
== 0 || hi
== 0)
1213 __put_unaligned32(__cpu_to_le32((__u32
)n
), lo
);
1214 __put_unaligned32(__cpu_to_le32((n
>> 32)), hi
);
1218 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1220 return (unsigned long long)__le32_to_cpu(lo
) |
1221 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1224 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1228 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1231 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1235 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1238 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1242 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1245 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1249 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1252 static unsigned long long vol_curr_migr_unit(struct imsm_dev
*dev
)
1257 return join_u32(dev
->vol
.curr_migr_unit_lo
, dev
->vol
.curr_migr_unit_hi
);
1260 static unsigned long long imsm_dev_size(struct imsm_dev
*dev
)
1264 return join_u32(dev
->size_low
, dev
->size_high
);
1267 static unsigned long long migr_chkp_area_pba(struct migr_record
*migr_rec
)
1269 if (migr_rec
== NULL
)
1271 return join_u32(migr_rec
->ckpt_area_pba_lo
,
1272 migr_rec
->ckpt_area_pba_hi
);
1275 static unsigned long long current_migr_unit(struct migr_record
*migr_rec
)
1277 if (migr_rec
== NULL
)
1279 return join_u32(migr_rec
->curr_migr_unit_lo
,
1280 migr_rec
->curr_migr_unit_hi
);
1283 static unsigned long long migr_dest_1st_member_lba(struct migr_record
*migr_rec
)
1285 if (migr_rec
== NULL
)
1287 return join_u32(migr_rec
->dest_1st_member_lba_lo
,
1288 migr_rec
->dest_1st_member_lba_hi
);
1291 static unsigned long long get_num_migr_units(struct migr_record
*migr_rec
)
1293 if (migr_rec
== NULL
)
1295 return join_u32(migr_rec
->num_migr_units_lo
,
1296 migr_rec
->num_migr_units_hi
);
1299 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1301 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1304 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1306 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1309 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1311 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1314 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1316 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1319 static void set_vol_curr_migr_unit(struct imsm_dev
*dev
, unsigned long long n
)
1324 split_ull(n
, &dev
->vol
.curr_migr_unit_lo
, &dev
->vol
.curr_migr_unit_hi
);
1327 static void set_imsm_dev_size(struct imsm_dev
*dev
, unsigned long long n
)
1329 split_ull(n
, &dev
->size_low
, &dev
->size_high
);
1332 static void set_migr_chkp_area_pba(struct migr_record
*migr_rec
,
1333 unsigned long long n
)
1335 split_ull(n
, &migr_rec
->ckpt_area_pba_lo
, &migr_rec
->ckpt_area_pba_hi
);
1338 static void set_current_migr_unit(struct migr_record
*migr_rec
,
1339 unsigned long long n
)
1341 split_ull(n
, &migr_rec
->curr_migr_unit_lo
,
1342 &migr_rec
->curr_migr_unit_hi
);
1345 static void set_migr_dest_1st_member_lba(struct migr_record
*migr_rec
,
1346 unsigned long long n
)
1348 split_ull(n
, &migr_rec
->dest_1st_member_lba_lo
,
1349 &migr_rec
->dest_1st_member_lba_hi
);
1352 static void set_num_migr_units(struct migr_record
*migr_rec
,
1353 unsigned long long n
)
1355 split_ull(n
, &migr_rec
->num_migr_units_lo
,
1356 &migr_rec
->num_migr_units_hi
);
1359 static unsigned long long per_dev_array_size(struct imsm_map
*map
)
1361 unsigned long long array_size
= 0;
1366 array_size
= num_data_stripes(map
) * map
->blocks_per_strip
;
1367 if (get_imsm_raid_level(map
) == 1 || get_imsm_raid_level(map
) == 10)
1373 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
,
1374 int get_minimal_reservation
)
1376 /* find a list of used extents on the given physical device */
1377 struct extent
*rv
, *e
;
1379 int memberships
= count_memberships(dl
, super
);
1382 /* trim the reserved area for spares, so they can join any array
1383 * regardless of whether the OROM has assigned sectors from the
1384 * IMSM_RESERVED_SECTORS region
1386 if (dl
->index
== -1 || get_minimal_reservation
)
1387 reservation
= imsm_min_reserved_sectors(super
);
1389 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1391 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
1394 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1395 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1396 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1398 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1399 e
->start
= pba_of_lba0(map
);
1400 e
->size
= per_dev_array_size(map
);
1404 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1406 /* determine the start of the metadata
1407 * when no raid devices are defined use the default
1408 * ...otherwise allow the metadata to truncate the value
1409 * as is the case with older versions of imsm
1412 struct extent
*last
= &rv
[memberships
- 1];
1413 unsigned long long remainder
;
1415 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1416 /* round down to 1k block to satisfy precision of the kernel
1420 /* make sure remainder is still sane */
1421 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1422 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1423 if (reservation
> remainder
)
1424 reservation
= remainder
;
1426 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1431 /* try to determine how much space is reserved for metadata from
1432 * the last get_extents() entry, otherwise fallback to the
1435 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1441 /* for spares just return a minimal reservation which will grow
1442 * once the spare is picked up by an array
1444 if (dl
->index
== -1)
1445 return MPB_SECTOR_CNT
;
1447 e
= get_extents(super
, dl
, 0);
1449 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1451 /* scroll to last entry */
1452 for (i
= 0; e
[i
].size
; i
++)
1455 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1462 static int is_spare(struct imsm_disk
*disk
)
1464 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1467 static int is_configured(struct imsm_disk
*disk
)
1469 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1472 static int is_failed(struct imsm_disk
*disk
)
1474 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1477 static int is_journal(struct imsm_disk
*disk
)
1479 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1482 /* round array size down to closest MB and ensure it splits evenly
1485 static unsigned long long round_size_to_mb(unsigned long long size
, unsigned int
1489 size
= (size
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
1495 static int able_to_resync(int raid_level
, int missing_disks
)
1497 int max_missing_disks
= 0;
1499 switch (raid_level
) {
1501 max_missing_disks
= 1;
1504 max_missing_disks
= 0;
1506 return missing_disks
<= max_missing_disks
;
1509 /* try to determine how much space is reserved for metadata from
1510 * the last get_extents() entry on the smallest active disk,
1511 * otherwise fallback to the default
1513 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1517 unsigned long long min_active
;
1519 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1520 struct dl
*dl
, *dl_min
= NULL
;
1526 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1529 unsigned long long blocks
= total_blocks(&dl
->disk
);
1530 if (blocks
< min_active
|| min_active
== 0) {
1532 min_active
= blocks
;
1538 /* find last lba used by subarrays on the smallest active disk */
1539 e
= get_extents(super
, dl_min
, 0);
1542 for (i
= 0; e
[i
].size
; i
++)
1545 remainder
= min_active
- e
[i
].start
;
1548 /* to give priority to recovery we should not require full
1549 IMSM_RESERVED_SECTORS from the spare */
1550 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1552 /* if real reservation is smaller use that value */
1553 return (remainder
< rv
) ? remainder
: rv
;
1557 * Return minimum size of a spare and sector size
1558 * that can be used in this array
1560 int get_spare_criteria_imsm(struct supertype
*st
, struct spare_criteria
*c
)
1562 struct intel_super
*super
= st
->sb
;
1566 unsigned long long size
= 0;
1573 /* find first active disk in array */
1575 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1579 /* find last lba used by subarrays */
1580 e
= get_extents(super
, dl
, 0);
1583 for (i
= 0; e
[i
].size
; i
++)
1586 size
= e
[i
-1].start
+ e
[i
-1].size
;
1589 /* add the amount of space needed for metadata */
1590 size
+= imsm_min_reserved_sectors(super
);
1592 c
->min_size
= size
* 512;
1593 c
->sector_size
= super
->sector_size
;
1598 static int is_gen_migration(struct imsm_dev
*dev
);
1600 #define IMSM_4K_DIV 8
1602 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1603 struct imsm_dev
*dev
);
1605 static void print_imsm_dev(struct intel_super
*super
,
1606 struct imsm_dev
*dev
,
1612 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1613 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1617 printf("[%.16s]:\n", dev
->volume
);
1618 printf(" Subarray : %d\n", super
->current_vol
);
1619 printf(" UUID : %s\n", uuid
);
1620 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1622 printf(" <-- %d", get_imsm_raid_level(map2
));
1624 printf(" Members : %d", map
->num_members
);
1626 printf(" <-- %d", map2
->num_members
);
1628 printf(" Slots : [");
1629 for (i
= 0; i
< map
->num_members
; i
++) {
1630 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1631 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1636 for (i
= 0; i
< map2
->num_members
; i
++) {
1637 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1638 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1643 printf(" Failed disk : ");
1644 if (map
->failed_disk_num
== 0xff)
1647 printf("%i", map
->failed_disk_num
);
1649 slot
= get_imsm_disk_slot(map
, disk_idx
);
1651 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1652 printf(" This Slot : %d%s\n", slot
,
1653 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1655 printf(" This Slot : ?\n");
1656 printf(" Sector Size : %u\n", super
->sector_size
);
1657 sz
= imsm_dev_size(dev
);
1658 printf(" Array Size : %llu%s\n",
1659 (unsigned long long)sz
* 512 / super
->sector_size
,
1660 human_size(sz
* 512));
1661 sz
= blocks_per_member(map
);
1662 printf(" Per Dev Size : %llu%s\n",
1663 (unsigned long long)sz
* 512 / super
->sector_size
,
1664 human_size(sz
* 512));
1665 printf(" Sector Offset : %llu\n",
1667 printf(" Num Stripes : %llu\n",
1668 num_data_stripes(map
));
1669 printf(" Chunk Size : %u KiB",
1670 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1672 printf(" <-- %u KiB",
1673 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1675 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1676 printf(" Migrate State : ");
1677 if (dev
->vol
.migr_state
) {
1678 if (migr_type(dev
) == MIGR_INIT
)
1679 printf("initialize\n");
1680 else if (migr_type(dev
) == MIGR_REBUILD
)
1681 printf("rebuild\n");
1682 else if (migr_type(dev
) == MIGR_VERIFY
)
1684 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1685 printf("general migration\n");
1686 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1687 printf("state change\n");
1688 else if (migr_type(dev
) == MIGR_REPAIR
)
1691 printf("<unknown:%d>\n", migr_type(dev
));
1694 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1695 if (dev
->vol
.migr_state
) {
1696 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1698 printf(" <-- %s", map_state_str
[map
->map_state
]);
1699 printf("\n Checkpoint : %llu ", vol_curr_migr_unit(dev
));
1700 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1703 printf("(%llu)", (unsigned long long)
1704 blocks_per_migr_unit(super
, dev
));
1707 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1709 printf(" RWH Policy : ");
1710 if (dev
->rwh_policy
== RWH_OFF
|| dev
->rwh_policy
== RWH_MULTIPLE_OFF
)
1712 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1713 printf("PPL distributed\n");
1714 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1715 printf("PPL journaling drive\n");
1716 else if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
1717 printf("Multiple distributed PPLs\n");
1718 else if (dev
->rwh_policy
== RWH_MULTIPLE_PPLS_JOURNALING_DRIVE
)
1719 printf("Multiple PPLs on journaling drive\n");
1720 else if (dev
->rwh_policy
== RWH_BITMAP
)
1721 printf("Write-intent bitmap\n");
1723 printf("<unknown:%d>\n", dev
->rwh_policy
);
1725 printf(" Volume ID : %u\n", dev
->my_vol_raid_dev_num
);
1728 static void print_imsm_disk(struct imsm_disk
*disk
,
1731 unsigned int sector_size
) {
1732 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1735 if (index
< -1 || !disk
)
1739 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1741 printf(" Disk%02d Serial : %s\n", index
, str
);
1743 printf(" Disk Serial : %s\n", str
);
1744 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1745 is_configured(disk
) ? " active" : "",
1746 is_failed(disk
) ? " failed" : "",
1747 is_journal(disk
) ? " journal" : "");
1748 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1749 sz
= total_blocks(disk
) - reserved
;
1750 printf(" Usable Size : %llu%s\n",
1751 (unsigned long long)sz
* 512 / sector_size
,
1752 human_size(sz
* 512));
1755 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1757 struct migr_record
*migr_rec
= super
->migr_rec
;
1759 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1760 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1761 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1762 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1763 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1764 set_migr_chkp_area_pba(migr_rec
,
1765 migr_chkp_area_pba(migr_rec
) / IMSM_4K_DIV
);
1766 set_migr_dest_1st_member_lba(migr_rec
,
1767 migr_dest_1st_member_lba(migr_rec
) / IMSM_4K_DIV
);
1770 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1772 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1775 void convert_to_4k(struct intel_super
*super
)
1777 struct imsm_super
*mpb
= super
->anchor
;
1778 struct imsm_disk
*disk
;
1780 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1782 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1783 disk
= __get_imsm_disk(mpb
, i
);
1785 convert_to_4k_imsm_disk(disk
);
1787 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1788 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1789 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1791 set_imsm_dev_size(dev
, imsm_dev_size(dev
)/IMSM_4K_DIV
);
1792 set_vol_curr_migr_unit(dev
,
1793 vol_curr_migr_unit(dev
) / IMSM_4K_DIV
);
1796 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1797 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1798 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1800 if (dev
->vol
.migr_state
) {
1802 map
= get_imsm_map(dev
, MAP_1
);
1803 set_blocks_per_member(map
,
1804 blocks_per_member(map
)/IMSM_4K_DIV
);
1805 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1806 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1810 struct bbm_log
*log
= (void *)mpb
+
1811 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1814 for (i
= 0; i
< log
->entry_count
; i
++) {
1815 struct bbm_log_entry
*entry
=
1816 &log
->marked_block_entries
[i
];
1818 __u8 count
= entry
->marked_count
+ 1;
1819 unsigned long long sector
=
1820 __le48_to_cpu(&entry
->defective_block_start
);
1822 entry
->defective_block_start
=
1823 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1824 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1828 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1831 void examine_migr_rec_imsm(struct intel_super
*super
)
1833 struct migr_record
*migr_rec
= super
->migr_rec
;
1834 struct imsm_super
*mpb
= super
->anchor
;
1837 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1838 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1839 struct imsm_map
*map
;
1842 if (is_gen_migration(dev
) == 0)
1845 printf("\nMigration Record Information:");
1847 /* first map under migration */
1848 map
= get_imsm_map(dev
, MAP_0
);
1850 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1851 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1852 printf(" Empty\n ");
1853 printf("Examine one of first two disks in array\n");
1856 printf("\n Status : ");
1857 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1860 printf("Contains Data\n");
1861 printf(" Current Unit : %llu\n",
1862 current_migr_unit(migr_rec
));
1863 printf(" Family : %u\n",
1864 __le32_to_cpu(migr_rec
->family_num
));
1865 printf(" Ascending : %u\n",
1866 __le32_to_cpu(migr_rec
->ascending_migr
));
1867 printf(" Blocks Per Unit : %u\n",
1868 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1869 printf(" Dest. Depth Per Unit : %u\n",
1870 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1871 printf(" Checkpoint Area pba : %llu\n",
1872 migr_chkp_area_pba(migr_rec
));
1873 printf(" First member lba : %llu\n",
1874 migr_dest_1st_member_lba(migr_rec
));
1875 printf(" Total Number of Units : %llu\n",
1876 get_num_migr_units(migr_rec
));
1877 printf(" Size of volume : %llu\n",
1878 join_u32(migr_rec
->post_migr_vol_cap
,
1879 migr_rec
->post_migr_vol_cap_hi
));
1880 printf(" Record was read from : %u\n",
1881 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1887 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1889 struct migr_record
*migr_rec
= super
->migr_rec
;
1891 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1892 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1893 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1894 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1895 &migr_rec
->post_migr_vol_cap
,
1896 &migr_rec
->post_migr_vol_cap_hi
);
1897 set_migr_chkp_area_pba(migr_rec
,
1898 migr_chkp_area_pba(migr_rec
) * IMSM_4K_DIV
);
1899 set_migr_dest_1st_member_lba(migr_rec
,
1900 migr_dest_1st_member_lba(migr_rec
) * IMSM_4K_DIV
);
1903 void convert_from_4k(struct intel_super
*super
)
1905 struct imsm_super
*mpb
= super
->anchor
;
1906 struct imsm_disk
*disk
;
1908 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1910 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1911 disk
= __get_imsm_disk(mpb
, i
);
1913 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1916 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1917 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1918 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1920 set_imsm_dev_size(dev
, imsm_dev_size(dev
)*IMSM_4K_DIV
);
1921 set_vol_curr_migr_unit(dev
,
1922 vol_curr_migr_unit(dev
) * IMSM_4K_DIV
);
1925 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1926 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1927 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1929 if (dev
->vol
.migr_state
) {
1931 map
= get_imsm_map(dev
, MAP_1
);
1932 set_blocks_per_member(map
,
1933 blocks_per_member(map
)*IMSM_4K_DIV
);
1934 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1935 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1939 struct bbm_log
*log
= (void *)mpb
+
1940 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1943 for (i
= 0; i
< log
->entry_count
; i
++) {
1944 struct bbm_log_entry
*entry
=
1945 &log
->marked_block_entries
[i
];
1947 __u8 count
= entry
->marked_count
+ 1;
1948 unsigned long long sector
=
1949 __le48_to_cpu(&entry
->defective_block_start
);
1951 entry
->defective_block_start
=
1952 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1953 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1957 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1960 /*******************************************************************************
1961 * function: imsm_check_attributes
1962 * Description: Function checks if features represented by attributes flags
1963 * are supported by mdadm.
1965 * attributes - Attributes read from metadata
1967 * 0 - passed attributes contains unsupported features flags
1968 * 1 - all features are supported
1969 ******************************************************************************/
1970 static int imsm_check_attributes(__u32 attributes
)
1973 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1975 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1977 not_supported
&= attributes
;
1978 if (not_supported
) {
1979 pr_err("(IMSM): Unsupported attributes : %x\n",
1980 (unsigned)__le32_to_cpu(not_supported
));
1981 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1982 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1983 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1985 if (not_supported
& MPB_ATTRIB_2TB
) {
1986 dprintf("\t\tMPB_ATTRIB_2TB\n");
1987 not_supported
^= MPB_ATTRIB_2TB
;
1989 if (not_supported
& MPB_ATTRIB_RAID0
) {
1990 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1991 not_supported
^= MPB_ATTRIB_RAID0
;
1993 if (not_supported
& MPB_ATTRIB_RAID1
) {
1994 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1995 not_supported
^= MPB_ATTRIB_RAID1
;
1997 if (not_supported
& MPB_ATTRIB_RAID10
) {
1998 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1999 not_supported
^= MPB_ATTRIB_RAID10
;
2001 if (not_supported
& MPB_ATTRIB_RAID1E
) {
2002 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
2003 not_supported
^= MPB_ATTRIB_RAID1E
;
2005 if (not_supported
& MPB_ATTRIB_RAID5
) {
2006 dprintf("\t\tMPB_ATTRIB_RAID5\n");
2007 not_supported
^= MPB_ATTRIB_RAID5
;
2009 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
2010 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
2011 not_supported
^= MPB_ATTRIB_RAIDCNG
;
2013 if (not_supported
& MPB_ATTRIB_BBM
) {
2014 dprintf("\t\tMPB_ATTRIB_BBM\n");
2015 not_supported
^= MPB_ATTRIB_BBM
;
2017 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
2018 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
2019 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
2021 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
2022 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
2023 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
2025 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
2026 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
2027 not_supported
^= MPB_ATTRIB_2TB_DISK
;
2029 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
2030 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
2031 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
2033 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
2034 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
2035 not_supported
^= MPB_ATTRIB_NEVER_USE
;
2039 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
2047 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
2049 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
2051 struct intel_super
*super
= st
->sb
;
2052 struct imsm_super
*mpb
= super
->anchor
;
2053 char str
[MAX_SIGNATURE_LENGTH
];
2058 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2060 time_t creation_time
;
2062 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
2063 str
[MPB_SIG_LEN
-1] = '\0';
2064 printf(" Magic : %s\n", str
);
2065 printf(" Version : %s\n", get_imsm_version(mpb
));
2066 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
2067 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
2068 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
2069 creation_time
= __le64_to_cpu(mpb
->creation_time
);
2070 printf(" Creation Time : %.24s\n",
2071 creation_time
? ctime(&creation_time
) : "Unknown");
2072 printf(" Attributes : ");
2073 if (imsm_check_attributes(mpb
->attributes
))
2074 printf("All supported\n");
2076 printf("not supported\n");
2077 getinfo_super_imsm(st
, &info
, NULL
);
2078 fname_from_uuid(st
, &info
, nbuf
, ':');
2079 printf(" UUID : %s\n", nbuf
+ 5);
2080 sum
= __le32_to_cpu(mpb
->check_sum
);
2081 printf(" Checksum : %08x %s\n", sum
,
2082 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
2083 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
2084 printf(" Disks : %d\n", mpb
->num_disks
);
2085 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
2086 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
2087 super
->disks
->index
, reserved
, super
->sector_size
);
2088 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
2089 struct bbm_log
*log
= super
->bbm_log
;
2092 printf("Bad Block Management Log:\n");
2093 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
2094 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
2095 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
2097 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2099 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2101 super
->current_vol
= i
;
2102 getinfo_super_imsm(st
, &info
, NULL
);
2103 fname_from_uuid(st
, &info
, nbuf
, ':');
2104 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
2106 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2107 if (i
== super
->disks
->index
)
2109 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
2110 super
->sector_size
);
2113 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2114 if (dl
->index
== -1)
2115 print_imsm_disk(&dl
->disk
, -1, reserved
,
2116 super
->sector_size
);
2118 examine_migr_rec_imsm(super
);
2121 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
2123 /* We just write a generic IMSM ARRAY entry */
2126 struct intel_super
*super
= st
->sb
;
2128 if (!super
->anchor
->num_raid_devs
) {
2129 printf("ARRAY metadata=imsm\n");
2133 getinfo_super_imsm(st
, &info
, NULL
);
2134 fname_from_uuid(st
, &info
, nbuf
, ':');
2135 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
2138 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
2140 /* We just write a generic IMSM ARRAY entry */
2144 struct intel_super
*super
= st
->sb
;
2147 if (!super
->anchor
->num_raid_devs
)
2150 getinfo_super_imsm(st
, &info
, NULL
);
2151 fname_from_uuid(st
, &info
, nbuf
, ':');
2152 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2153 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2155 super
->current_vol
= i
;
2156 getinfo_super_imsm(st
, &info
, NULL
);
2157 fname_from_uuid(st
, &info
, nbuf1
, ':');
2158 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
2159 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
2163 static void export_examine_super_imsm(struct supertype
*st
)
2165 struct intel_super
*super
= st
->sb
;
2166 struct imsm_super
*mpb
= super
->anchor
;
2170 getinfo_super_imsm(st
, &info
, NULL
);
2171 fname_from_uuid(st
, &info
, nbuf
, ':');
2172 printf("MD_METADATA=imsm\n");
2173 printf("MD_LEVEL=container\n");
2174 printf("MD_UUID=%s\n", nbuf
+5);
2175 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
2176 printf("MD_CREATION_TIME=%llu\n", __le64_to_cpu(mpb
->creation_time
));
2179 static void detail_super_imsm(struct supertype
*st
, char *homehost
,
2184 struct intel_super
*super
= st
->sb
;
2185 int temp_vol
= super
->current_vol
;
2188 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2190 getinfo_super_imsm(st
, &info
, NULL
);
2191 fname_from_uuid(st
, &info
, nbuf
, ':');
2192 printf("\n UUID : %s\n", nbuf
+ 5);
2194 super
->current_vol
= temp_vol
;
2197 static void brief_detail_super_imsm(struct supertype
*st
, char *subarray
)
2201 struct intel_super
*super
= st
->sb
;
2202 int temp_vol
= super
->current_vol
;
2205 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2207 getinfo_super_imsm(st
, &info
, NULL
);
2208 fname_from_uuid(st
, &info
, nbuf
, ':');
2209 printf(" UUID=%s", nbuf
+ 5);
2211 super
->current_vol
= temp_vol
;
2214 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
,
2215 size_t serial_buf_len
);
2216 static void fd2devname(int fd
, char *name
);
2218 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2220 /* dump an unsorted list of devices attached to AHCI Intel storage
2221 * controller, as well as non-connected ports
2223 int hba_len
= strlen(hba_path
) + 1;
2228 unsigned long port_mask
= (1 << port_count
) - 1;
2230 if (port_count
> (int)sizeof(port_mask
) * 8) {
2232 pr_err("port_count %d out of range\n", port_count
);
2236 /* scroll through /sys/dev/block looking for devices attached to
2239 dir
= opendir("/sys/dev/block");
2243 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2254 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2256 path
= devt_to_devpath(makedev(major
, minor
));
2259 if (!path_attached_to_hba(path
, hba_path
)) {
2265 /* retrieve the scsi device type */
2266 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2268 pr_err("failed to allocate 'device'\n");
2272 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2273 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2275 pr_err("failed to read device type for %s\n",
2281 type
= strtoul(buf
, NULL
, 10);
2283 /* if it's not a disk print the vendor and model */
2284 if (!(type
== 0 || type
== 7 || type
== 14)) {
2287 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2288 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2289 strncpy(vendor
, buf
, sizeof(vendor
));
2290 vendor
[sizeof(vendor
) - 1] = '\0';
2291 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2292 while (isspace(*c
) || *c
== '\0')
2296 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2297 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2298 strncpy(model
, buf
, sizeof(model
));
2299 model
[sizeof(model
) - 1] = '\0';
2300 c
= (char *) &model
[sizeof(model
) - 1];
2301 while (isspace(*c
) || *c
== '\0')
2305 if (vendor
[0] && model
[0])
2306 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2308 switch (type
) { /* numbers from hald/linux/device.c */
2309 case 1: sprintf(buf
, "tape"); break;
2310 case 2: sprintf(buf
, "printer"); break;
2311 case 3: sprintf(buf
, "processor"); break;
2313 case 5: sprintf(buf
, "cdrom"); break;
2314 case 6: sprintf(buf
, "scanner"); break;
2315 case 8: sprintf(buf
, "media_changer"); break;
2316 case 9: sprintf(buf
, "comm"); break;
2317 case 12: sprintf(buf
, "raid"); break;
2318 default: sprintf(buf
, "unknown");
2324 /* chop device path to 'host%d' and calculate the port number */
2325 c
= strchr(&path
[hba_len
], '/');
2328 pr_err("%s - invalid path name\n", path
+ hba_len
);
2333 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2334 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2338 *c
= '/'; /* repair the full string */
2339 pr_err("failed to determine port number for %s\n",
2346 /* mark this port as used */
2347 port_mask
&= ~(1 << port
);
2349 /* print out the device information */
2351 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2355 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2357 printf(" Port%d : - disk info unavailable -\n", port
);
2359 fd2devname(fd
, buf
);
2360 printf(" Port%d : %s", port
, buf
);
2361 if (imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2363 printf(" (%s)\n", buf
);
2378 for (i
= 0; i
< port_count
; i
++)
2379 if (port_mask
& (1 << i
))
2380 printf(" Port%d : - no device attached -\n", i
);
2386 static int print_nvme_info(struct sys_dev
*hba
)
2394 dir
= opendir("/sys/block/");
2398 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2399 if (strstr(ent
->d_name
, "nvme")) {
2400 sprintf(buf
, "/sys/block/%s", ent
->d_name
);
2401 rp
= realpath(buf
, NULL
);
2404 if (path_attached_to_hba(rp
, hba
->path
)) {
2405 fd
= open_dev(ent
->d_name
);
2406 if (!imsm_is_nvme_supported(fd
, 0)) {
2413 fd2devname(fd
, buf
);
2414 if (hba
->type
== SYS_DEV_VMD
)
2415 printf(" NVMe under VMD : %s", buf
);
2416 else if (hba
->type
== SYS_DEV_NVME
)
2417 printf(" NVMe Device : %s", buf
);
2418 if (!imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2420 printf(" (%s)\n", buf
);
2433 static void print_found_intel_controllers(struct sys_dev
*elem
)
2435 for (; elem
; elem
= elem
->next
) {
2436 pr_err("found Intel(R) ");
2437 if (elem
->type
== SYS_DEV_SATA
)
2438 fprintf(stderr
, "SATA ");
2439 else if (elem
->type
== SYS_DEV_SAS
)
2440 fprintf(stderr
, "SAS ");
2441 else if (elem
->type
== SYS_DEV_NVME
)
2442 fprintf(stderr
, "NVMe ");
2444 if (elem
->type
== SYS_DEV_VMD
)
2445 fprintf(stderr
, "VMD domain");
2447 fprintf(stderr
, "RAID controller");
2450 fprintf(stderr
, " at %s", elem
->pci_id
);
2451 fprintf(stderr
, ".\n");
2456 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2463 if ((dir
= opendir(hba_path
)) == NULL
)
2466 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2469 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2470 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2472 if (*port_count
== 0)
2474 else if (host
< host_base
)
2477 if (host
+ 1 > *port_count
+ host_base
)
2478 *port_count
= host
+ 1 - host_base
;
2484 static void print_imsm_capability(const struct imsm_orom
*orom
)
2486 printf(" Platform : Intel(R) ");
2487 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2488 printf("Matrix Storage Manager\n");
2489 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2490 printf("Virtual RAID on CPU\n");
2492 printf("Rapid Storage Technology%s\n",
2493 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2494 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2495 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2496 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2497 printf(" RAID Levels :%s%s%s%s%s\n",
2498 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2499 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2500 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2501 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2502 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2503 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2504 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2505 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2506 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2507 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2508 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2509 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2510 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2511 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2512 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2513 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2514 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2515 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2516 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2517 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2518 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2519 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2520 printf(" 2TB volumes :%s supported\n",
2521 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2522 printf(" 2TB disks :%s supported\n",
2523 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2524 printf(" Max Disks : %d\n", orom
->tds
);
2525 printf(" Max Volumes : %d per array, %d per %s\n",
2526 orom
->vpa
, orom
->vphba
,
2527 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2531 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2533 printf("MD_FIRMWARE_TYPE=imsm\n");
2534 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2535 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2536 orom
->hotfix_ver
, orom
->build
);
2537 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2538 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2539 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2540 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2541 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2542 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2543 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2544 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2545 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2546 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2547 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2548 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2549 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2550 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2551 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2552 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2553 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2554 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2555 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2556 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2557 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2558 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2559 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2560 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2561 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2562 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2563 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2564 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2567 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2569 /* There are two components to imsm platform support, the ahci SATA
2570 * controller and the option-rom. To find the SATA controller we
2571 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2572 * controller with the Intel vendor id is present. This approach
2573 * allows mdadm to leverage the kernel's ahci detection logic, with the
2574 * caveat that if ahci.ko is not loaded mdadm will not be able to
2575 * detect platform raid capabilities. The option-rom resides in a
2576 * platform "Adapter ROM". We scan for its signature to retrieve the
2577 * platform capabilities. If raid support is disabled in the BIOS the
2578 * option-rom capability structure will not be available.
2580 struct sys_dev
*list
, *hba
;
2585 if (enumerate_only
) {
2586 if (check_env("IMSM_NO_PLATFORM"))
2588 list
= find_intel_devices();
2591 for (hba
= list
; hba
; hba
= hba
->next
) {
2592 if (find_imsm_capability(hba
)) {
2602 list
= find_intel_devices();
2605 pr_err("no active Intel(R) RAID controller found.\n");
2607 } else if (verbose
> 0)
2608 print_found_intel_controllers(list
);
2610 for (hba
= list
; hba
; hba
= hba
->next
) {
2611 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2613 if (!find_imsm_capability(hba
)) {
2615 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2616 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2617 get_sys_dev_type(hba
->type
));
2623 if (controller_path
&& result
== 1) {
2624 pr_err("no active Intel(R) RAID controller found under %s\n",
2629 const struct orom_entry
*entry
;
2631 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2632 if (entry
->type
== SYS_DEV_VMD
) {
2633 print_imsm_capability(&entry
->orom
);
2634 printf(" 3rd party NVMe :%s supported\n",
2635 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2636 for (hba
= list
; hba
; hba
= hba
->next
) {
2637 if (hba
->type
== SYS_DEV_VMD
) {
2639 printf(" I/O Controller : %s (%s)\n",
2640 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2641 if (print_nvme_info(hba
)) {
2643 pr_err("failed to get devices attached to VMD domain.\n");
2652 print_imsm_capability(&entry
->orom
);
2653 if (entry
->type
== SYS_DEV_NVME
) {
2654 for (hba
= list
; hba
; hba
= hba
->next
) {
2655 if (hba
->type
== SYS_DEV_NVME
)
2656 print_nvme_info(hba
);
2662 struct devid_list
*devid
;
2663 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2664 hba
= device_by_id(devid
->devid
);
2668 printf(" I/O Controller : %s (%s)\n",
2669 hba
->path
, get_sys_dev_type(hba
->type
));
2670 if (hba
->type
== SYS_DEV_SATA
) {
2671 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2672 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2674 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2685 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2687 struct sys_dev
*list
, *hba
;
2690 list
= find_intel_devices();
2693 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2698 for (hba
= list
; hba
; hba
= hba
->next
) {
2699 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2701 if (!find_imsm_capability(hba
) && verbose
> 0) {
2703 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2704 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2710 const struct orom_entry
*entry
;
2712 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2713 if (entry
->type
== SYS_DEV_VMD
) {
2714 for (hba
= list
; hba
; hba
= hba
->next
)
2715 print_imsm_capability_export(&entry
->orom
);
2718 print_imsm_capability_export(&entry
->orom
);
2724 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2726 /* the imsm metadata format does not specify any host
2727 * identification information. We return -1 since we can never
2728 * confirm nor deny whether a given array is "meant" for this
2729 * host. We rely on compare_super and the 'family_num' fields to
2730 * exclude member disks that do not belong, and we rely on
2731 * mdadm.conf to specify the arrays that should be assembled.
2732 * Auto-assembly may still pick up "foreign" arrays.
2738 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2740 /* The uuid returned here is used for:
2741 * uuid to put into bitmap file (Create, Grow)
2742 * uuid for backup header when saving critical section (Grow)
2743 * comparing uuids when re-adding a device into an array
2744 * In these cases the uuid required is that of the data-array,
2745 * not the device-set.
2746 * uuid to recognise same set when adding a missing device back
2747 * to an array. This is a uuid for the device-set.
2749 * For each of these we can make do with a truncated
2750 * or hashed uuid rather than the original, as long as
2752 * In each case the uuid required is that of the data-array,
2753 * not the device-set.
2755 /* imsm does not track uuid's so we synthesis one using sha1 on
2756 * - The signature (Which is constant for all imsm array, but no matter)
2757 * - the orig_family_num of the container
2758 * - the index number of the volume
2759 * - the 'serial' number of the volume.
2760 * Hopefully these are all constant.
2762 struct intel_super
*super
= st
->sb
;
2765 struct sha1_ctx ctx
;
2766 struct imsm_dev
*dev
= NULL
;
2769 /* some mdadm versions failed to set ->orig_family_num, in which
2770 * case fall back to ->family_num. orig_family_num will be
2771 * fixed up with the first metadata update.
2773 family_num
= super
->anchor
->orig_family_num
;
2774 if (family_num
== 0)
2775 family_num
= super
->anchor
->family_num
;
2776 sha1_init_ctx(&ctx
);
2777 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2778 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2779 if (super
->current_vol
>= 0)
2780 dev
= get_imsm_dev(super
, super
->current_vol
);
2782 __u32 vol
= super
->current_vol
;
2783 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2784 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2786 sha1_finish_ctx(&ctx
, buf
);
2787 memcpy(uuid
, buf
, 4*4);
2792 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2794 __u8
*v
= get_imsm_version(mpb
);
2795 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2796 char major
[] = { 0, 0, 0 };
2797 char minor
[] = { 0 ,0, 0 };
2798 char patch
[] = { 0, 0, 0 };
2799 char *ver_parse
[] = { major
, minor
, patch
};
2803 while (*v
!= '\0' && v
< end
) {
2804 if (*v
!= '.' && j
< 2)
2805 ver_parse
[i
][j
++] = *v
;
2813 *m
= strtol(minor
, NULL
, 0);
2814 *p
= strtol(patch
, NULL
, 0);
2818 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2820 /* migr_strip_size when repairing or initializing parity */
2821 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2822 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2824 switch (get_imsm_raid_level(map
)) {
2829 return 128*1024 >> 9;
2833 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2835 /* migr_strip_size when rebuilding a degraded disk, no idea why
2836 * this is different than migr_strip_size_resync(), but it's good
2839 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2840 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2842 switch (get_imsm_raid_level(map
)) {
2845 if (map
->num_members
% map
->num_domains
== 0)
2846 return 128*1024 >> 9;
2850 return max((__u32
) 64*1024 >> 9, chunk
);
2852 return 128*1024 >> 9;
2856 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2858 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2859 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2860 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2861 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2863 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2866 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2868 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2869 int level
= get_imsm_raid_level(lo
);
2871 if (level
== 1 || level
== 10) {
2872 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2874 return hi
->num_domains
;
2876 return num_stripes_per_unit_resync(dev
);
2879 static __u8
imsm_num_data_members(struct imsm_map
*map
)
2881 /* named 'imsm_' because raid0, raid1 and raid10
2882 * counter-intuitively have the same number of data disks
2884 switch (get_imsm_raid_level(map
)) {
2886 return map
->num_members
;
2890 return map
->num_members
/2;
2892 return map
->num_members
- 1;
2894 dprintf("unsupported raid level\n");
2899 static unsigned long long calc_component_size(struct imsm_map
*map
,
2900 struct imsm_dev
*dev
)
2902 unsigned long long component_size
;
2903 unsigned long long dev_size
= imsm_dev_size(dev
);
2904 long long calc_dev_size
= 0;
2905 unsigned int member_disks
= imsm_num_data_members(map
);
2907 if (member_disks
== 0)
2910 component_size
= per_dev_array_size(map
);
2911 calc_dev_size
= component_size
* member_disks
;
2913 /* Component size is rounded to 1MB so difference between size from
2914 * metadata and size calculated from num_data_stripes equals up to
2915 * 2048 blocks per each device. If the difference is higher it means
2916 * that array size was expanded and num_data_stripes was not updated.
2918 if (llabs(calc_dev_size
- (long long)dev_size
) >
2919 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
2920 component_size
= dev_size
/ member_disks
;
2921 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
2922 component_size
/ map
->blocks_per_strip
,
2923 num_data_stripes(map
));
2926 return component_size
;
2929 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2931 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2932 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2934 switch(get_imsm_raid_level(map
)) {
2937 return chunk
* map
->num_domains
;
2939 return chunk
* map
->num_members
;
2945 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2947 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2948 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2949 __u32 strip
= block
/ chunk
;
2951 switch (get_imsm_raid_level(map
)) {
2954 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2955 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2957 return vol_stripe
* chunk
+ block
% chunk
;
2959 __u32 stripe
= strip
/ (map
->num_members
- 1);
2961 return stripe
* chunk
+ block
% chunk
;
2968 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2969 struct imsm_dev
*dev
)
2971 /* calculate the conversion factor between per member 'blocks'
2972 * (md/{resync,rebuild}_start) and imsm migration units, return
2973 * 0 for the 'not migrating' and 'unsupported migration' cases
2975 if (!dev
->vol
.migr_state
)
2978 switch (migr_type(dev
)) {
2979 case MIGR_GEN_MIGR
: {
2980 struct migr_record
*migr_rec
= super
->migr_rec
;
2981 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2986 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2987 __u32 stripes_per_unit
;
2988 __u32 blocks_per_unit
;
2997 /* yes, this is really the translation of migr_units to
2998 * per-member blocks in the 'resync' case
3000 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
3001 migr_chunk
= migr_strip_blocks_resync(dev
);
3002 disks
= imsm_num_data_members(map
);
3003 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
3004 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
3005 segment
= blocks_per_unit
/ stripe
;
3006 block_rel
= blocks_per_unit
- segment
* stripe
;
3007 parity_depth
= parity_segment_depth(dev
);
3008 block_map
= map_migr_block(dev
, block_rel
);
3009 return block_map
+ parity_depth
* segment
;
3011 case MIGR_REBUILD
: {
3012 __u32 stripes_per_unit
;
3015 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
3016 migr_chunk
= migr_strip_blocks_rebuild(dev
);
3017 return migr_chunk
* stripes_per_unit
;
3019 case MIGR_STATE_CHANGE
:
3025 static int imsm_level_to_layout(int level
)
3033 return ALGORITHM_LEFT_ASYMMETRIC
;
3040 /*******************************************************************************
3041 * Function: read_imsm_migr_rec
3042 * Description: Function reads imsm migration record from last sector of disk
3044 * fd : disk descriptor
3045 * super : metadata info
3049 ******************************************************************************/
3050 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
3053 unsigned int sector_size
= super
->sector_size
;
3054 unsigned long long dsize
;
3056 get_dev_size(fd
, NULL
, &dsize
);
3057 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
3059 pr_err("Cannot seek to anchor block: %s\n",
3063 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
3064 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3065 MIGR_REC_BUF_SECTORS
*sector_size
) {
3066 pr_err("Cannot read migr record block: %s\n",
3071 if (sector_size
== 4096)
3072 convert_from_4k_imsm_migr_rec(super
);
3078 static struct imsm_dev
*imsm_get_device_during_migration(
3079 struct intel_super
*super
)
3082 struct intel_dev
*dv
;
3084 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
3085 if (is_gen_migration(dv
->dev
))
3091 /*******************************************************************************
3092 * Function: load_imsm_migr_rec
3093 * Description: Function reads imsm migration record (it is stored at the last
3096 * super : imsm internal array info
3100 * -2 : no migration in progress
3101 ******************************************************************************/
3102 static int load_imsm_migr_rec(struct intel_super
*super
)
3108 struct imsm_dev
*dev
;
3109 struct imsm_map
*map
;
3113 /* find map under migration */
3114 dev
= imsm_get_device_during_migration(super
);
3115 /* nothing to load,no migration in progress?
3120 map
= get_imsm_map(dev
, MAP_0
);
3124 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3125 /* skip spare and failed disks
3129 /* read only from one of the first two slots
3131 slot
= get_imsm_disk_slot(map
, dl
->index
);
3132 if (slot
> 1 || slot
< 0)
3136 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3137 fd
= dev_open(nm
, O_RDONLY
);
3150 retval
= read_imsm_migr_rec(fd
, super
);
3157 /*******************************************************************************
3158 * function: imsm_create_metadata_checkpoint_update
3159 * Description: It creates update for checkpoint change.
3161 * super : imsm internal array info
3162 * u : pointer to prepared update
3165 * If length is equal to 0, input pointer u contains no update
3166 ******************************************************************************/
3167 static int imsm_create_metadata_checkpoint_update(
3168 struct intel_super
*super
,
3169 struct imsm_update_general_migration_checkpoint
**u
)
3172 int update_memory_size
= 0;
3174 dprintf("(enter)\n");
3180 /* size of all update data without anchor */
3181 update_memory_size
=
3182 sizeof(struct imsm_update_general_migration_checkpoint
);
3184 *u
= xcalloc(1, update_memory_size
);
3186 dprintf("error: cannot get memory\n");
3189 (*u
)->type
= update_general_migration_checkpoint
;
3190 (*u
)->curr_migr_unit
= current_migr_unit(super
->migr_rec
);
3191 dprintf("prepared for %llu\n", (*u
)->curr_migr_unit
);
3193 return update_memory_size
;
3196 static void imsm_update_metadata_locally(struct supertype
*st
,
3197 void *buf
, int len
);
3199 /*******************************************************************************
3200 * Function: write_imsm_migr_rec
3201 * Description: Function writes imsm migration record
3202 * (at the last sector of disk)
3204 * super : imsm internal array info
3208 ******************************************************************************/
3209 static int write_imsm_migr_rec(struct supertype
*st
)
3211 struct intel_super
*super
= st
->sb
;
3212 unsigned int sector_size
= super
->sector_size
;
3213 unsigned long long dsize
;
3217 struct imsm_update_general_migration_checkpoint
*u
;
3218 struct imsm_dev
*dev
;
3219 struct imsm_map
*map
;
3221 /* find map under migration */
3222 dev
= imsm_get_device_during_migration(super
);
3223 /* if no migration, write buffer anyway to clear migr_record
3224 * on disk based on first available device
3227 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3228 super
->current_vol
);
3230 map
= get_imsm_map(dev
, MAP_0
);
3232 if (sector_size
== 4096)
3233 convert_to_4k_imsm_migr_rec(super
);
3234 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3237 /* skip failed and spare devices */
3240 /* write to 2 first slots only */
3242 slot
= get_imsm_disk_slot(map
, sd
->index
);
3243 if (map
== NULL
|| slot
> 1 || slot
< 0)
3246 get_dev_size(sd
->fd
, NULL
, &dsize
);
3247 if (lseek64(sd
->fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*
3250 pr_err("Cannot seek to anchor block: %s\n",
3254 if ((unsigned int)write(sd
->fd
, super
->migr_rec_buf
,
3255 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3256 MIGR_REC_BUF_SECTORS
*sector_size
) {
3257 pr_err("Cannot write migr record block: %s\n",
3262 if (sector_size
== 4096)
3263 convert_from_4k_imsm_migr_rec(super
);
3264 /* update checkpoint information in metadata */
3265 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3267 dprintf("imsm: Cannot prepare update\n");
3270 /* update metadata locally */
3271 imsm_update_metadata_locally(st
, u
, len
);
3272 /* and possibly remotely */
3273 if (st
->update_tail
) {
3274 append_metadata_update(st
, u
, len
);
3275 /* during reshape we do all work inside metadata handler
3276 * manage_reshape(), so metadata update has to be triggered
3279 flush_metadata_updates(st
);
3280 st
->update_tail
= &st
->updates
;
3289 /* spare/missing disks activations are not allowe when
3290 * array/container performs reshape operation, because
3291 * all arrays in container works on the same disks set
3293 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3296 struct intel_dev
*i_dev
;
3297 struct imsm_dev
*dev
;
3299 /* check whole container
3301 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3303 if (is_gen_migration(dev
)) {
3304 /* No repair during any migration in container
3312 static unsigned long long imsm_component_size_alignment_check(int level
,
3314 unsigned int sector_size
,
3315 unsigned long long component_size
)
3317 unsigned int component_size_alignment
;
3319 /* check component size alignment
3321 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3323 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3324 level
, chunk_size
, component_size
,
3325 component_size_alignment
);
3327 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3328 dprintf("imsm: reported component size aligned from %llu ",
3330 component_size
-= component_size_alignment
;
3331 dprintf_cont("to %llu (%i).\n",
3332 component_size
, component_size_alignment
);
3335 return component_size
;
3338 /*******************************************************************************
3339 * Function: get_bitmap_header_sector
3340 * Description: Returns the sector where the bitmap header is placed.
3342 * st : supertype information
3343 * dev_idx : index of the device with bitmap
3346 * The sector where the bitmap header is placed
3347 ******************************************************************************/
3348 static unsigned long long get_bitmap_header_sector(struct intel_super
*super
,
3351 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3352 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3354 if (!super
->sector_size
) {
3355 dprintf("sector size is not set\n");
3359 return pba_of_lba0(map
) + calc_component_size(map
, dev
) +
3360 (IMSM_BITMAP_HEADER_OFFSET
/ super
->sector_size
);
3363 /*******************************************************************************
3364 * Function: get_bitmap_sector
3365 * Description: Returns the sector where the bitmap is placed.
3367 * st : supertype information
3368 * dev_idx : index of the device with bitmap
3371 * The sector where the bitmap is placed
3372 ******************************************************************************/
3373 static unsigned long long get_bitmap_sector(struct intel_super
*super
,
3376 if (!super
->sector_size
) {
3377 dprintf("sector size is not set\n");
3381 return get_bitmap_header_sector(super
, dev_idx
) +
3382 (IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
);
3385 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3387 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3388 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3390 return pba_of_lba0(map
) +
3391 (num_data_stripes(map
) * map
->blocks_per_strip
);
3394 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3396 struct intel_super
*super
= st
->sb
;
3397 struct migr_record
*migr_rec
= super
->migr_rec
;
3398 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3399 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3400 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3401 struct imsm_map
*map_to_analyse
= map
;
3403 int map_disks
= info
->array
.raid_disks
;
3405 memset(info
, 0, sizeof(*info
));
3407 map_to_analyse
= prev_map
;
3409 dl
= super
->current_disk
;
3411 info
->container_member
= super
->current_vol
;
3412 info
->array
.raid_disks
= map
->num_members
;
3413 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3414 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3415 info
->array
.md_minor
= -1;
3416 info
->array
.ctime
= 0;
3417 info
->array
.utime
= 0;
3418 info
->array
.chunk_size
=
3419 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3420 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3421 info
->custom_array_size
= imsm_dev_size(dev
);
3422 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3424 if (is_gen_migration(dev
)) {
3425 info
->reshape_active
= 1;
3426 info
->new_level
= get_imsm_raid_level(map
);
3427 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3428 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3429 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3430 if (info
->delta_disks
) {
3431 /* this needs to be applied to every array
3434 info
->reshape_active
= CONTAINER_RESHAPE
;
3436 /* We shape information that we give to md might have to be
3437 * modify to cope with md's requirement for reshaping arrays.
3438 * For example, when reshaping a RAID0, md requires it to be
3439 * presented as a degraded RAID4.
3440 * Also if a RAID0 is migrating to a RAID5 we need to specify
3441 * the array as already being RAID5, but the 'before' layout
3442 * is a RAID4-like layout.
3444 switch (info
->array
.level
) {
3446 switch(info
->new_level
) {
3448 /* conversion is happening as RAID4 */
3449 info
->array
.level
= 4;
3450 info
->array
.raid_disks
+= 1;
3453 /* conversion is happening as RAID5 */
3454 info
->array
.level
= 5;
3455 info
->array
.layout
= ALGORITHM_PARITY_N
;
3456 info
->delta_disks
-= 1;
3459 /* FIXME error message */
3460 info
->array
.level
= UnSet
;
3466 info
->new_level
= UnSet
;
3467 info
->new_layout
= UnSet
;
3468 info
->new_chunk
= info
->array
.chunk_size
;
3469 info
->delta_disks
= 0;
3473 info
->disk
.major
= dl
->major
;
3474 info
->disk
.minor
= dl
->minor
;
3475 info
->disk
.number
= dl
->index
;
3476 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3480 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3481 info
->component_size
= calc_component_size(map
, dev
);
3482 info
->component_size
= imsm_component_size_alignment_check(
3484 info
->array
.chunk_size
,
3486 info
->component_size
);
3487 info
->bb
.supported
= 1;
3489 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3490 info
->recovery_start
= MaxSector
;
3492 if (info
->array
.level
== 5 &&
3493 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3494 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3495 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3496 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3497 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3498 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3500 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3502 } else if (info
->array
.level
<= 0) {
3503 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3505 if (dev
->rwh_policy
== RWH_BITMAP
) {
3506 info
->bitmap_offset
= get_bitmap_sector(super
, super
->current_vol
);
3507 info
->consistency_policy
= CONSISTENCY_POLICY_BITMAP
;
3509 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3513 info
->reshape_progress
= 0;
3514 info
->resync_start
= MaxSector
;
3515 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3516 !(info
->array
.state
& 1)) &&
3517 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3518 info
->resync_start
= 0;
3520 if (dev
->vol
.migr_state
) {
3521 switch (migr_type(dev
)) {
3524 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3526 __u64 units
= vol_curr_migr_unit(dev
);
3528 info
->resync_start
= blocks_per_unit
* units
;
3531 case MIGR_GEN_MIGR
: {
3532 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3534 __u64 units
= current_migr_unit(migr_rec
);
3537 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3539 (get_num_migr_units(migr_rec
)-1)) &&
3540 (super
->migr_rec
->rec_status
==
3541 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3544 info
->reshape_progress
= blocks_per_unit
* units
;
3546 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3547 (unsigned long long)units
,
3548 (unsigned long long)blocks_per_unit
,
3549 info
->reshape_progress
);
3551 used_disks
= imsm_num_data_members(prev_map
);
3552 if (used_disks
> 0) {
3553 info
->custom_array_size
= per_dev_array_size(map
) *
3558 /* we could emulate the checkpointing of
3559 * 'sync_action=check' migrations, but for now
3560 * we just immediately complete them
3563 /* this is handled by container_content_imsm() */
3564 case MIGR_STATE_CHANGE
:
3565 /* FIXME handle other migrations */
3567 /* we are not dirty, so... */
3568 info
->resync_start
= MaxSector
;
3572 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3573 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3575 info
->array
.major_version
= -1;
3576 info
->array
.minor_version
= -2;
3577 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3578 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3579 uuid_from_super_imsm(st
, info
->uuid
);
3583 for (i
=0; i
<map_disks
; i
++) {
3585 if (i
< info
->array
.raid_disks
) {
3586 struct imsm_disk
*dsk
;
3587 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3588 dsk
= get_imsm_disk(super
, j
);
3589 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3596 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3597 int failed
, int look_in_map
);
3599 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3602 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3604 if (is_gen_migration(dev
)) {
3607 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3609 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3610 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3611 if (map2
->map_state
!= map_state
) {
3612 map2
->map_state
= map_state
;
3613 super
->updates_pending
++;
3618 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3622 for (d
= super
->missing
; d
; d
= d
->next
)
3623 if (d
->index
== index
)
3628 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3630 struct intel_super
*super
= st
->sb
;
3631 struct imsm_disk
*disk
;
3632 int map_disks
= info
->array
.raid_disks
;
3633 int max_enough
= -1;
3635 struct imsm_super
*mpb
;
3637 if (super
->current_vol
>= 0) {
3638 getinfo_super_imsm_volume(st
, info
, map
);
3641 memset(info
, 0, sizeof(*info
));
3643 /* Set raid_disks to zero so that Assemble will always pull in valid
3646 info
->array
.raid_disks
= 0;
3647 info
->array
.level
= LEVEL_CONTAINER
;
3648 info
->array
.layout
= 0;
3649 info
->array
.md_minor
= -1;
3650 info
->array
.ctime
= 0; /* N/A for imsm */
3651 info
->array
.utime
= 0;
3652 info
->array
.chunk_size
= 0;
3654 info
->disk
.major
= 0;
3655 info
->disk
.minor
= 0;
3656 info
->disk
.raid_disk
= -1;
3657 info
->reshape_active
= 0;
3658 info
->array
.major_version
= -1;
3659 info
->array
.minor_version
= -2;
3660 strcpy(info
->text_version
, "imsm");
3661 info
->safe_mode_delay
= 0;
3662 info
->disk
.number
= -1;
3663 info
->disk
.state
= 0;
3665 info
->recovery_start
= MaxSector
;
3666 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3667 info
->bb
.supported
= 1;
3669 /* do we have the all the insync disks that we expect? */
3670 mpb
= super
->anchor
;
3671 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3673 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3674 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3675 int failed
, enough
, j
, missing
= 0;
3676 struct imsm_map
*map
;
3679 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3680 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3681 map
= get_imsm_map(dev
, MAP_0
);
3683 /* any newly missing disks?
3684 * (catches single-degraded vs double-degraded)
3686 for (j
= 0; j
< map
->num_members
; j
++) {
3687 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3688 __u32 idx
= ord_to_idx(ord
);
3690 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3691 info
->disk
.raid_disk
= j
;
3693 if (!(ord
& IMSM_ORD_REBUILD
) &&
3694 get_imsm_missing(super
, idx
)) {
3700 if (state
== IMSM_T_STATE_FAILED
)
3702 else if (state
== IMSM_T_STATE_DEGRADED
&&
3703 (state
!= map
->map_state
|| missing
))
3705 else /* we're normal, or already degraded */
3707 if (is_gen_migration(dev
) && missing
) {
3708 /* during general migration we need all disks
3709 * that process is running on.
3710 * No new missing disk is allowed.
3714 /* no more checks necessary
3718 /* in the missing/failed disk case check to see
3719 * if at least one array is runnable
3721 max_enough
= max(max_enough
, enough
);
3723 dprintf("enough: %d\n", max_enough
);
3724 info
->container_enough
= max_enough
;
3727 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3729 disk
= &super
->disks
->disk
;
3730 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3731 info
->component_size
= reserved
;
3732 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3733 /* we don't change info->disk.raid_disk here because
3734 * this state will be finalized in mdmon after we have
3735 * found the 'most fresh' version of the metadata
3737 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3738 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3739 0 : (1 << MD_DISK_SYNC
);
3742 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3743 * ->compare_super may have updated the 'num_raid_devs' field for spares
3745 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3746 uuid_from_super_imsm(st
, info
->uuid
);
3748 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3750 /* I don't know how to compute 'map' on imsm, so use safe default */
3753 for (i
= 0; i
< map_disks
; i
++)
3759 /* allocates memory and fills disk in mdinfo structure
3760 * for each disk in array */
3761 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3763 struct mdinfo
*mddev
;
3764 struct intel_super
*super
= st
->sb
;
3765 struct imsm_disk
*disk
;
3768 if (!super
|| !super
->disks
)
3771 mddev
= xcalloc(1, sizeof(*mddev
));
3775 tmp
= xcalloc(1, sizeof(*tmp
));
3777 tmp
->next
= mddev
->devs
;
3779 tmp
->disk
.number
= count
++;
3780 tmp
->disk
.major
= dl
->major
;
3781 tmp
->disk
.minor
= dl
->minor
;
3782 tmp
->disk
.state
= is_configured(disk
) ?
3783 (1 << MD_DISK_ACTIVE
) : 0;
3784 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3785 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3786 tmp
->disk
.raid_disk
= -1;
3792 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3793 char *update
, char *devname
, int verbose
,
3794 int uuid_set
, char *homehost
)
3796 /* For 'assemble' and 'force' we need to return non-zero if any
3797 * change was made. For others, the return value is ignored.
3798 * Update options are:
3799 * force-one : This device looks a bit old but needs to be included,
3800 * update age info appropriately.
3801 * assemble: clear any 'faulty' flag to allow this device to
3803 * force-array: Array is degraded but being forced, mark it clean
3804 * if that will be needed to assemble it.
3806 * newdev: not used ????
3807 * grow: Array has gained a new device - this is currently for
3809 * resync: mark as dirty so a resync will happen.
3810 * name: update the name - preserving the homehost
3811 * uuid: Change the uuid of the array to match watch is given
3813 * Following are not relevant for this imsm:
3814 * sparc2.2 : update from old dodgey metadata
3815 * super-minor: change the preferred_minor number
3816 * summaries: update redundant counters.
3817 * homehost: update the recorded homehost
3818 * _reshape_progress: record new reshape_progress position.
3821 struct intel_super
*super
= st
->sb
;
3822 struct imsm_super
*mpb
;
3824 /* we can only update container info */
3825 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3828 mpb
= super
->anchor
;
3830 if (strcmp(update
, "uuid") == 0) {
3831 /* We take this to mean that the family_num should be updated.
3832 * However that is much smaller than the uuid so we cannot really
3833 * allow an explicit uuid to be given. And it is hard to reliably
3835 * So if !uuid_set we know the current uuid is random and just used
3836 * the first 'int' and copy it to the other 3 positions.
3837 * Otherwise we require the 4 'int's to be the same as would be the
3838 * case if we are using a random uuid. So an explicit uuid will be
3839 * accepted as long as all for ints are the same... which shouldn't hurt
3842 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3845 if (info
->uuid
[0] != info
->uuid
[1] ||
3846 info
->uuid
[1] != info
->uuid
[2] ||
3847 info
->uuid
[2] != info
->uuid
[3])
3853 mpb
->orig_family_num
= info
->uuid
[0];
3854 } else if (strcmp(update
, "assemble") == 0)
3859 /* successful update? recompute checksum */
3861 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3866 static size_t disks_to_mpb_size(int disks
)
3870 size
= sizeof(struct imsm_super
);
3871 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3872 size
+= 2 * sizeof(struct imsm_dev
);
3873 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3874 size
+= (4 - 2) * sizeof(struct imsm_map
);
3875 /* 4 possible disk_ord_tbl's */
3876 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3877 /* maximum bbm log */
3878 size
+= sizeof(struct bbm_log
);
3883 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3884 unsigned long long data_offset
)
3886 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3889 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3892 static void free_devlist(struct intel_super
*super
)
3894 struct intel_dev
*dv
;
3896 while (super
->devlist
) {
3897 dv
= super
->devlist
->next
;
3898 free(super
->devlist
->dev
);
3899 free(super
->devlist
);
3900 super
->devlist
= dv
;
3904 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3906 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3909 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
,
3914 * 0 same, or first was empty, and second was copied
3915 * 1 second had wrong number
3917 * 3 wrong other info
3919 struct intel_super
*first
= st
->sb
;
3920 struct intel_super
*sec
= tst
->sb
;
3927 /* in platform dependent environment test if the disks
3928 * use the same Intel hba
3929 * If not on Intel hba at all, allow anything.
3931 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3932 if (first
->hba
->type
!= sec
->hba
->type
) {
3934 pr_err("HBAs of devices do not match %s != %s\n",
3935 get_sys_dev_type(first
->hba
->type
),
3936 get_sys_dev_type(sec
->hba
->type
));
3940 if (first
->orom
!= sec
->orom
) {
3942 pr_err("HBAs of devices do not match %s != %s\n",
3943 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3949 /* if an anchor does not have num_raid_devs set then it is a free
3952 if (first
->anchor
->num_raid_devs
> 0 &&
3953 sec
->anchor
->num_raid_devs
> 0) {
3954 /* Determine if these disks might ever have been
3955 * related. Further disambiguation can only take place
3956 * in load_super_imsm_all
3958 __u32 first_family
= first
->anchor
->orig_family_num
;
3959 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3961 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3962 MAX_SIGNATURE_LENGTH
) != 0)
3965 if (first_family
== 0)
3966 first_family
= first
->anchor
->family_num
;
3967 if (sec_family
== 0)
3968 sec_family
= sec
->anchor
->family_num
;
3970 if (first_family
!= sec_family
)
3975 /* if 'first' is a spare promote it to a populated mpb with sec's
3978 if (first
->anchor
->num_raid_devs
== 0 &&
3979 sec
->anchor
->num_raid_devs
> 0) {
3981 struct intel_dev
*dv
;
3982 struct imsm_dev
*dev
;
3984 /* we need to copy raid device info from sec if an allocation
3985 * fails here we don't associate the spare
3987 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3988 dv
= xmalloc(sizeof(*dv
));
3989 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3992 dv
->next
= first
->devlist
;
3993 first
->devlist
= dv
;
3995 if (i
< sec
->anchor
->num_raid_devs
) {
3996 /* allocation failure */
3997 free_devlist(first
);
3998 pr_err("imsm: failed to associate spare\n");
4001 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
4002 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
4003 first
->anchor
->family_num
= sec
->anchor
->family_num
;
4004 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
4005 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
4006 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
4012 static void fd2devname(int fd
, char *name
)
4016 char dname
[PATH_MAX
];
4021 if (fstat(fd
, &st
) != 0)
4023 sprintf(path
, "/sys/dev/block/%d:%d",
4024 major(st
.st_rdev
), minor(st
.st_rdev
));
4026 rv
= readlink(path
, dname
, sizeof(dname
)-1);
4031 nm
= strrchr(dname
, '/');
4034 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
4038 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
4041 char *name
= fd2kname(fd
);
4046 if (strncmp(name
, "nvme", 4) != 0)
4049 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
4051 return load_sys(path
, buf
, buf_len
);
4054 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
4056 static int imsm_read_serial(int fd
, char *devname
,
4057 __u8
*serial
, size_t serial_buf_len
)
4066 memset(buf
, 0, sizeof(buf
));
4068 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
4071 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
4073 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
4074 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
4075 fd2devname(fd
, (char *) serial
);
4081 pr_err("Failed to retrieve serial for %s\n",
4086 /* trim all whitespace and non-printable characters and convert
4089 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4092 /* ':' is reserved for use in placeholder serial
4093 * numbers for missing disks
4104 if (len
> serial_buf_len
) {
4105 /* truncate leading characters */
4106 dest
+= len
- serial_buf_len
;
4107 len
= serial_buf_len
;
4110 memset(serial
, 0, serial_buf_len
);
4111 memcpy(serial
, dest
, len
);
4116 static int serialcmp(__u8
*s1
, __u8
*s2
)
4118 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4121 static void serialcpy(__u8
*dest
, __u8
*src
)
4123 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4126 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4130 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4131 if (serialcmp(dl
->serial
, serial
) == 0)
4137 static struct imsm_disk
*
4138 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4142 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4143 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4145 if (serialcmp(disk
->serial
, serial
) == 0) {
4156 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4158 struct imsm_disk
*disk
;
4163 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4165 rv
= imsm_read_serial(fd
, devname
, serial
, MAX_RAID_SERIAL_LEN
);
4170 dl
= xcalloc(1, sizeof(*dl
));
4173 dl
->major
= major(stb
.st_rdev
);
4174 dl
->minor
= minor(stb
.st_rdev
);
4175 dl
->next
= super
->disks
;
4176 dl
->fd
= keep_fd
? fd
: -1;
4177 assert(super
->disks
== NULL
);
4179 serialcpy(dl
->serial
, serial
);
4182 fd2devname(fd
, name
);
4184 dl
->devname
= xstrdup(devname
);
4186 dl
->devname
= xstrdup(name
);
4188 /* look up this disk's index in the current anchor */
4189 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4192 /* only set index on disks that are a member of a
4193 * populated contianer, i.e. one with raid_devs
4195 if (is_failed(&dl
->disk
))
4197 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4204 /* When migrating map0 contains the 'destination' state while map1
4205 * contains the current state. When not migrating map0 contains the
4206 * current state. This routine assumes that map[0].map_state is set to
4207 * the current array state before being called.
4209 * Migration is indicated by one of the following states
4210 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4211 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4212 * map1state=unitialized)
4213 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4215 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4216 * map1state=degraded)
4217 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4220 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4221 __u8 to_state
, int migr_type
)
4223 struct imsm_map
*dest
;
4224 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4226 dev
->vol
.migr_state
= 1;
4227 set_migr_type(dev
, migr_type
);
4228 set_vol_curr_migr_unit(dev
, 0);
4229 dest
= get_imsm_map(dev
, MAP_1
);
4231 /* duplicate and then set the target end state in map[0] */
4232 memcpy(dest
, src
, sizeof_imsm_map(src
));
4233 if (migr_type
== MIGR_GEN_MIGR
) {
4237 for (i
= 0; i
< src
->num_members
; i
++) {
4238 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4239 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4243 if (migr_type
== MIGR_GEN_MIGR
)
4244 /* Clear migration record */
4245 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4247 src
->map_state
= to_state
;
4250 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4253 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4254 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4258 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4259 * completed in the last migration.
4261 * FIXME add support for raid-level-migration
4263 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4264 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4265 /* when final map state is other than expected
4266 * merge maps (not for migration)
4270 for (i
= 0; i
< prev
->num_members
; i
++)
4271 for (j
= 0; j
< map
->num_members
; j
++)
4272 /* during online capacity expansion
4273 * disks position can be changed
4274 * if takeover is used
4276 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4277 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4278 map
->disk_ord_tbl
[j
] |=
4279 prev
->disk_ord_tbl
[i
];
4282 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4283 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4286 dev
->vol
.migr_state
= 0;
4287 set_migr_type(dev
, 0);
4288 set_vol_curr_migr_unit(dev
, 0);
4289 map
->map_state
= map_state
;
4292 static int parse_raid_devices(struct intel_super
*super
)
4295 struct imsm_dev
*dev_new
;
4296 size_t len
, len_migr
;
4298 size_t space_needed
= 0;
4299 struct imsm_super
*mpb
= super
->anchor
;
4301 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4302 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4303 struct intel_dev
*dv
;
4305 len
= sizeof_imsm_dev(dev_iter
, 0);
4306 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4308 space_needed
+= len_migr
- len
;
4310 dv
= xmalloc(sizeof(*dv
));
4311 if (max_len
< len_migr
)
4313 if (max_len
> len_migr
)
4314 space_needed
+= max_len
- len_migr
;
4315 dev_new
= xmalloc(max_len
);
4316 imsm_copy_dev(dev_new
, dev_iter
);
4319 dv
->next
= super
->devlist
;
4320 super
->devlist
= dv
;
4323 /* ensure that super->buf is large enough when all raid devices
4326 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4329 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4330 super
->sector_size
);
4331 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4334 memcpy(buf
, super
->buf
, super
->len
);
4335 memset(buf
+ super
->len
, 0, len
- super
->len
);
4341 super
->extra_space
+= space_needed
;
4346 /*******************************************************************************
4347 * Function: check_mpb_migr_compatibility
4348 * Description: Function checks for unsupported migration features:
4349 * - migration optimization area (pba_of_lba0)
4350 * - descending reshape (ascending_migr)
4352 * super : imsm metadata information
4354 * 0 : migration is compatible
4355 * -1 : migration is not compatible
4356 ******************************************************************************/
4357 int check_mpb_migr_compatibility(struct intel_super
*super
)
4359 struct imsm_map
*map0
, *map1
;
4360 struct migr_record
*migr_rec
= super
->migr_rec
;
4363 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4364 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4367 dev_iter
->vol
.migr_state
== 1 &&
4368 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4369 /* This device is migrating */
4370 map0
= get_imsm_map(dev_iter
, MAP_0
);
4371 map1
= get_imsm_map(dev_iter
, MAP_1
);
4372 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4373 /* migration optimization area was used */
4375 if (migr_rec
->ascending_migr
== 0 &&
4376 migr_rec
->dest_depth_per_unit
> 0)
4377 /* descending reshape not supported yet */
4384 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4386 /* load_imsm_mpb - read matrix metadata
4387 * allocates super->mpb to be freed by free_imsm
4389 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4391 unsigned long long dsize
;
4392 unsigned long long sectors
;
4393 unsigned int sector_size
= super
->sector_size
;
4395 struct imsm_super
*anchor
;
4398 get_dev_size(fd
, NULL
, &dsize
);
4399 if (dsize
< 2*sector_size
) {
4401 pr_err("%s: device to small for imsm\n",
4406 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4408 pr_err("Cannot seek to anchor block on %s: %s\n",
4409 devname
, strerror(errno
));
4413 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4415 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4418 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4420 pr_err("Cannot read anchor block on %s: %s\n",
4421 devname
, strerror(errno
));
4426 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4428 pr_err("no IMSM anchor on %s\n", devname
);
4433 __free_imsm(super
, 0);
4434 /* reload capability and hba */
4436 /* capability and hba must be updated with new super allocation */
4437 find_intel_hba_capability(fd
, super
, devname
);
4438 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4439 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4441 pr_err("unable to allocate %zu byte mpb buffer\n",
4446 memcpy(super
->buf
, anchor
, sector_size
);
4448 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4451 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4452 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4453 pr_err("could not allocate migr_rec buffer\n");
4457 super
->clean_migration_record_by_mdmon
= 0;
4460 check_sum
= __gen_imsm_checksum(super
->anchor
);
4461 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4463 pr_err("IMSM checksum %x != %x on %s\n",
4465 __le32_to_cpu(super
->anchor
->check_sum
),
4473 /* read the extended mpb */
4474 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4476 pr_err("Cannot seek to extended mpb on %s: %s\n",
4477 devname
, strerror(errno
));
4481 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4482 super
->len
- sector_size
) != super
->len
- sector_size
) {
4484 pr_err("Cannot read extended mpb on %s: %s\n",
4485 devname
, strerror(errno
));
4489 check_sum
= __gen_imsm_checksum(super
->anchor
);
4490 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4492 pr_err("IMSM checksum %x != %x on %s\n",
4493 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4501 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4503 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4504 static void clear_hi(struct intel_super
*super
)
4506 struct imsm_super
*mpb
= super
->anchor
;
4508 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4510 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4511 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4512 disk
->total_blocks_hi
= 0;
4514 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4515 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4518 for (n
= 0; n
< 2; ++n
) {
4519 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4522 map
->pba_of_lba0_hi
= 0;
4523 map
->blocks_per_member_hi
= 0;
4524 map
->num_data_stripes_hi
= 0;
4530 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4534 err
= load_imsm_mpb(fd
, super
, devname
);
4537 if (super
->sector_size
== 4096)
4538 convert_from_4k(super
);
4539 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4542 err
= parse_raid_devices(super
);
4545 err
= load_bbm_log(super
);
4550 static void __free_imsm_disk(struct dl
*d
)
4562 static void free_imsm_disks(struct intel_super
*super
)
4566 while (super
->disks
) {
4568 super
->disks
= d
->next
;
4569 __free_imsm_disk(d
);
4571 while (super
->disk_mgmt_list
) {
4572 d
= super
->disk_mgmt_list
;
4573 super
->disk_mgmt_list
= d
->next
;
4574 __free_imsm_disk(d
);
4576 while (super
->missing
) {
4578 super
->missing
= d
->next
;
4579 __free_imsm_disk(d
);
4584 /* free all the pieces hanging off of a super pointer */
4585 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4587 struct intel_hba
*elem
, *next
;
4593 /* unlink capability description */
4595 if (super
->migr_rec_buf
) {
4596 free(super
->migr_rec_buf
);
4597 super
->migr_rec_buf
= NULL
;
4600 free_imsm_disks(super
);
4601 free_devlist(super
);
4605 free((void *)elem
->path
);
4611 free(super
->bbm_log
);
4615 static void free_imsm(struct intel_super
*super
)
4617 __free_imsm(super
, 1);
4618 free(super
->bb
.entries
);
4622 static void free_super_imsm(struct supertype
*st
)
4624 struct intel_super
*super
= st
->sb
;
4633 static struct intel_super
*alloc_super(void)
4635 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4637 super
->current_vol
= -1;
4638 super
->create_offset
= ~((unsigned long long) 0);
4640 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4641 sizeof(struct md_bb_entry
));
4642 if (!super
->bb
.entries
) {
4651 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4653 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4655 struct sys_dev
*hba_name
;
4658 if (fd
>= 0 && test_partition(fd
)) {
4659 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4663 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4668 hba_name
= find_disk_attached_hba(fd
, NULL
);
4671 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4675 rv
= attach_hba_to_super(super
, hba_name
);
4678 struct intel_hba
*hba
= super
->hba
;
4680 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4681 " but the container is assigned to Intel(R) %s %s (",
4683 get_sys_dev_type(hba_name
->type
),
4684 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4685 hba_name
->pci_id
? : "Err!",
4686 get_sys_dev_type(super
->hba
->type
),
4687 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4690 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4692 fprintf(stderr
, ", ");
4695 fprintf(stderr
, ").\n"
4696 " Mixing devices attached to different controllers is not allowed.\n");
4700 super
->orom
= find_imsm_capability(hba_name
);
4707 /* find_missing - helper routine for load_super_imsm_all that identifies
4708 * disks that have disappeared from the system. This routine relies on
4709 * the mpb being uptodate, which it is at load time.
4711 static int find_missing(struct intel_super
*super
)
4714 struct imsm_super
*mpb
= super
->anchor
;
4716 struct imsm_disk
*disk
;
4718 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4719 disk
= __get_imsm_disk(mpb
, i
);
4720 dl
= serial_to_dl(disk
->serial
, super
);
4724 dl
= xmalloc(sizeof(*dl
));
4728 dl
->devname
= xstrdup("missing");
4730 serialcpy(dl
->serial
, disk
->serial
);
4733 dl
->next
= super
->missing
;
4734 super
->missing
= dl
;
4740 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4742 struct intel_disk
*idisk
= disk_list
;
4745 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4747 idisk
= idisk
->next
;
4753 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4754 struct intel_super
*super
,
4755 struct intel_disk
**disk_list
)
4757 struct imsm_disk
*d
= &super
->disks
->disk
;
4758 struct imsm_super
*mpb
= super
->anchor
;
4761 for (i
= 0; i
< tbl_size
; i
++) {
4762 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4763 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4765 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4766 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4767 dprintf("mpb from %d:%d matches %d:%d\n",
4768 super
->disks
->major
,
4769 super
->disks
->minor
,
4770 table
[i
]->disks
->major
,
4771 table
[i
]->disks
->minor
);
4775 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4776 is_configured(d
) == is_configured(tbl_d
)) &&
4777 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4778 /* current version of the mpb is a
4779 * better candidate than the one in
4780 * super_table, but copy over "cross
4781 * generational" status
4783 struct intel_disk
*idisk
;
4785 dprintf("mpb from %d:%d replaces %d:%d\n",
4786 super
->disks
->major
,
4787 super
->disks
->minor
,
4788 table
[i
]->disks
->major
,
4789 table
[i
]->disks
->minor
);
4791 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4792 if (idisk
&& is_failed(&idisk
->disk
))
4793 tbl_d
->status
|= FAILED_DISK
;
4796 struct intel_disk
*idisk
;
4797 struct imsm_disk
*disk
;
4799 /* tbl_mpb is more up to date, but copy
4800 * over cross generational status before
4803 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4804 if (disk
&& is_failed(disk
))
4805 d
->status
|= FAILED_DISK
;
4807 idisk
= disk_list_get(d
->serial
, *disk_list
);
4810 if (disk
&& is_configured(disk
))
4811 idisk
->disk
.status
|= CONFIGURED_DISK
;
4814 dprintf("mpb from %d:%d prefer %d:%d\n",
4815 super
->disks
->major
,
4816 super
->disks
->minor
,
4817 table
[i
]->disks
->major
,
4818 table
[i
]->disks
->minor
);
4826 table
[tbl_size
++] = super
;
4830 /* update/extend the merged list of imsm_disk records */
4831 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4832 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4833 struct intel_disk
*idisk
;
4835 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4837 idisk
->disk
.status
|= disk
->status
;
4838 if (is_configured(&idisk
->disk
) ||
4839 is_failed(&idisk
->disk
))
4840 idisk
->disk
.status
&= ~(SPARE_DISK
);
4842 idisk
= xcalloc(1, sizeof(*idisk
));
4843 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4844 idisk
->disk
= *disk
;
4845 idisk
->next
= *disk_list
;
4849 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4856 static struct intel_super
*
4857 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4860 struct imsm_super
*mpb
= super
->anchor
;
4864 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4865 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4866 struct intel_disk
*idisk
;
4868 idisk
= disk_list_get(disk
->serial
, disk_list
);
4870 if (idisk
->owner
== owner
||
4871 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4874 dprintf("'%.16s' owner %d != %d\n",
4875 disk
->serial
, idisk
->owner
,
4878 dprintf("unknown disk %x [%d]: %.16s\n",
4879 __le32_to_cpu(mpb
->family_num
), i
,
4885 if (ok_count
== mpb
->num_disks
)
4890 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4892 struct intel_super
*s
;
4894 for (s
= super_list
; s
; s
= s
->next
) {
4895 if (family_num
!= s
->anchor
->family_num
)
4897 pr_err("Conflict, offlining family %#x on '%s'\n",
4898 __le32_to_cpu(family_num
), s
->disks
->devname
);
4902 static struct intel_super
*
4903 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4905 struct intel_super
*super_table
[len
];
4906 struct intel_disk
*disk_list
= NULL
;
4907 struct intel_super
*champion
, *spare
;
4908 struct intel_super
*s
, **del
;
4913 memset(super_table
, 0, sizeof(super_table
));
4914 for (s
= *super_list
; s
; s
= s
->next
)
4915 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4917 for (i
= 0; i
< tbl_size
; i
++) {
4918 struct imsm_disk
*d
;
4919 struct intel_disk
*idisk
;
4920 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4923 d
= &s
->disks
->disk
;
4925 /* 'd' must appear in merged disk list for its
4926 * configuration to be valid
4928 idisk
= disk_list_get(d
->serial
, disk_list
);
4929 if (idisk
&& idisk
->owner
== i
)
4930 s
= validate_members(s
, disk_list
, i
);
4935 dprintf("marking family: %#x from %d:%d offline\n",
4937 super_table
[i
]->disks
->major
,
4938 super_table
[i
]->disks
->minor
);
4942 /* This is where the mdadm implementation differs from the Windows
4943 * driver which has no strict concept of a container. We can only
4944 * assemble one family from a container, so when returning a prodigal
4945 * array member to this system the code will not be able to disambiguate
4946 * the container contents that should be assembled ("foreign" versus
4947 * "local"). It requires user intervention to set the orig_family_num
4948 * to a new value to establish a new container. The Windows driver in
4949 * this situation fixes up the volume name in place and manages the
4950 * foreign array as an independent entity.
4955 for (i
= 0; i
< tbl_size
; i
++) {
4956 struct intel_super
*tbl_ent
= super_table
[i
];
4962 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4967 if (s
&& !is_spare
) {
4968 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4970 } else if (!s
&& !is_spare
)
4983 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4984 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4986 /* collect all dl's onto 'champion', and update them to
4987 * champion's version of the status
4989 for (s
= *super_list
; s
; s
= s
->next
) {
4990 struct imsm_super
*mpb
= champion
->anchor
;
4991 struct dl
*dl
= s
->disks
;
4996 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4998 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4999 struct imsm_disk
*disk
;
5001 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
5004 /* only set index on disks that are a member of
5005 * a populated contianer, i.e. one with
5008 if (is_failed(&dl
->disk
))
5010 else if (is_spare(&dl
->disk
))
5016 if (i
>= mpb
->num_disks
) {
5017 struct intel_disk
*idisk
;
5019 idisk
= disk_list_get(dl
->serial
, disk_list
);
5020 if (idisk
&& is_spare(&idisk
->disk
) &&
5021 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
5029 dl
->next
= champion
->disks
;
5030 champion
->disks
= dl
;
5034 /* delete 'champion' from super_list */
5035 for (del
= super_list
; *del
; ) {
5036 if (*del
== champion
) {
5037 *del
= (*del
)->next
;
5040 del
= &(*del
)->next
;
5042 champion
->next
= NULL
;
5046 struct intel_disk
*idisk
= disk_list
;
5048 disk_list
= disk_list
->next
;
5056 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
5057 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5058 int major
, int minor
, int keep_fd
);
5060 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5061 int *max
, int keep_fd
);
5063 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
5064 char *devname
, struct md_list
*devlist
,
5067 struct intel_super
*super_list
= NULL
;
5068 struct intel_super
*super
= NULL
;
5073 /* 'fd' is an opened container */
5074 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
5076 /* get super block from devlist devices */
5077 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
5080 /* all mpbs enter, maybe one leaves */
5081 super
= imsm_thunderdome(&super_list
, i
);
5087 if (find_missing(super
) != 0) {
5093 /* load migration record */
5094 err
= load_imsm_migr_rec(super
);
5096 /* migration is in progress,
5097 * but migr_rec cannot be loaded,
5103 /* Check migration compatibility */
5104 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5105 pr_err("Unsupported migration detected");
5107 fprintf(stderr
, " on %s\n", devname
);
5109 fprintf(stderr
, " (IMSM).\n");
5118 while (super_list
) {
5119 struct intel_super
*s
= super_list
;
5121 super_list
= super_list
->next
;
5130 strcpy(st
->container_devnm
, fd2devnm(fd
));
5132 st
->container_devnm
[0] = 0;
5133 if (err
== 0 && st
->ss
== NULL
) {
5134 st
->ss
= &super_imsm
;
5135 st
->minor_version
= 0;
5136 st
->max_devs
= IMSM_MAX_DEVICES
;
5142 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5143 int *max
, int keep_fd
)
5145 struct md_list
*tmpdev
;
5149 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5150 if (tmpdev
->used
!= 1)
5152 if (tmpdev
->container
== 1) {
5154 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5156 pr_err("cannot open device %s: %s\n",
5157 tmpdev
->devname
, strerror(errno
));
5161 err
= get_sra_super_block(fd
, super_list
,
5162 tmpdev
->devname
, &lmax
,
5171 int major
= major(tmpdev
->st_rdev
);
5172 int minor
= minor(tmpdev
->st_rdev
);
5173 err
= get_super_block(super_list
,
5190 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5191 int major
, int minor
, int keep_fd
)
5193 struct intel_super
*s
;
5205 sprintf(nm
, "%d:%d", major
, minor
);
5206 dfd
= dev_open(nm
, O_RDWR
);
5212 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5213 find_intel_hba_capability(dfd
, s
, devname
);
5214 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5216 /* retry the load if we might have raced against mdmon */
5217 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5218 for (retry
= 0; retry
< 3; retry
++) {
5220 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5226 s
->next
= *super_list
;
5234 if (dfd
>= 0 && !keep_fd
)
5241 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5248 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5252 if (sra
->array
.major_version
!= -1 ||
5253 sra
->array
.minor_version
!= -2 ||
5254 strcmp(sra
->text_version
, "imsm") != 0) {
5259 devnm
= fd2devnm(fd
);
5260 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5261 if (get_super_block(super_list
, devnm
, devname
,
5262 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5273 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5275 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5278 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5280 struct intel_super
*super
;
5284 if (test_partition(fd
))
5285 /* IMSM not allowed on partitions */
5288 free_super_imsm(st
);
5290 super
= alloc_super();
5291 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5294 /* Load hba and capabilities if they exist.
5295 * But do not preclude loading metadata in case capabilities or hba are
5296 * non-compliant and ignore_hw_compat is set.
5298 rv
= find_intel_hba_capability(fd
, super
, devname
);
5299 /* no orom/efi or non-intel hba of the disk */
5300 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5302 pr_err("No OROM/EFI properties for %s\n", devname
);
5306 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5308 /* retry the load if we might have raced against mdmon */
5310 struct mdstat_ent
*mdstat
= NULL
;
5311 char *name
= fd2kname(fd
);
5314 mdstat
= mdstat_by_component(name
);
5316 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5317 for (retry
= 0; retry
< 3; retry
++) {
5319 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5325 free_mdstat(mdstat
);
5330 pr_err("Failed to load all information sections on %s\n", devname
);
5336 if (st
->ss
== NULL
) {
5337 st
->ss
= &super_imsm
;
5338 st
->minor_version
= 0;
5339 st
->max_devs
= IMSM_MAX_DEVICES
;
5342 /* load migration record */
5343 if (load_imsm_migr_rec(super
) == 0) {
5344 /* Check for unsupported migration features */
5345 if (check_mpb_migr_compatibility(super
) != 0) {
5346 pr_err("Unsupported migration detected");
5348 fprintf(stderr
, " on %s\n", devname
);
5350 fprintf(stderr
, " (IMSM).\n");
5358 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5360 if (info
->level
== 1)
5362 return info
->chunk_size
>> 9;
5365 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5366 unsigned long long size
)
5368 if (info
->level
== 1)
5371 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5374 static void imsm_update_version_info(struct intel_super
*super
)
5376 /* update the version and attributes */
5377 struct imsm_super
*mpb
= super
->anchor
;
5379 struct imsm_dev
*dev
;
5380 struct imsm_map
*map
;
5383 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5384 dev
= get_imsm_dev(super
, i
);
5385 map
= get_imsm_map(dev
, MAP_0
);
5386 if (__le32_to_cpu(dev
->size_high
) > 0)
5387 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5389 /* FIXME detect when an array spans a port multiplier */
5391 mpb
->attributes
|= MPB_ATTRIB_PM
;
5394 if (mpb
->num_raid_devs
> 1 ||
5395 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5396 version
= MPB_VERSION_ATTRIBS
;
5397 switch (get_imsm_raid_level(map
)) {
5398 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5399 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5400 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5401 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5404 if (map
->num_members
>= 5)
5405 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5406 else if (dev
->status
== DEV_CLONE_N_GO
)
5407 version
= MPB_VERSION_CNG
;
5408 else if (get_imsm_raid_level(map
) == 5)
5409 version
= MPB_VERSION_RAID5
;
5410 else if (map
->num_members
>= 3)
5411 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5412 else if (get_imsm_raid_level(map
) == 1)
5413 version
= MPB_VERSION_RAID1
;
5415 version
= MPB_VERSION_RAID0
;
5417 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5421 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5423 struct imsm_super
*mpb
= super
->anchor
;
5424 char *reason
= NULL
;
5426 size_t len
= strlen(name
);
5430 while (isspace(start
[len
- 1]))
5432 while (*start
&& isspace(*start
))
5434 memmove(name
, start
, len
+ 1);
5437 if (len
> MAX_RAID_SERIAL_LEN
)
5438 reason
= "must be 16 characters or less";
5440 reason
= "must be a non-empty string";
5442 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5443 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5445 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5446 reason
= "already exists";
5451 if (reason
&& !quiet
)
5452 pr_err("imsm volume name %s\n", reason
);
5457 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5458 struct shape
*s
, char *name
,
5459 char *homehost
, int *uuid
,
5460 long long data_offset
)
5462 /* We are creating a volume inside a pre-existing container.
5463 * so st->sb is already set.
5465 struct intel_super
*super
= st
->sb
;
5466 unsigned int sector_size
= super
->sector_size
;
5467 struct imsm_super
*mpb
= super
->anchor
;
5468 struct intel_dev
*dv
;
5469 struct imsm_dev
*dev
;
5470 struct imsm_vol
*vol
;
5471 struct imsm_map
*map
;
5472 int idx
= mpb
->num_raid_devs
;
5475 unsigned long long array_blocks
;
5476 size_t size_old
, size_new
;
5477 unsigned long long num_data_stripes
;
5478 unsigned int data_disks
;
5479 unsigned long long size_per_member
;
5481 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5482 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5486 /* ensure the mpb is large enough for the new data */
5487 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5488 size_new
= disks_to_mpb_size(info
->nr_disks
);
5489 if (size_new
> size_old
) {
5491 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5493 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5494 pr_err("could not allocate new mpb\n");
5497 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5498 MIGR_REC_BUF_SECTORS
*
5499 MAX_SECTOR_SIZE
) != 0) {
5500 pr_err("could not allocate migr_rec buffer\n");
5506 memcpy(mpb_new
, mpb
, size_old
);
5509 super
->anchor
= mpb_new
;
5510 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5511 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5512 super
->len
= size_round
;
5514 super
->current_vol
= idx
;
5516 /* handle 'failed_disks' by either:
5517 * a) create dummy disk entries in the table if this the first
5518 * volume in the array. We add them here as this is the only
5519 * opportunity to add them. add_to_super_imsm_volume()
5520 * handles the non-failed disks and continues incrementing
5522 * b) validate that 'failed_disks' matches the current number
5523 * of missing disks if the container is populated
5525 if (super
->current_vol
== 0) {
5527 for (i
= 0; i
< info
->failed_disks
; i
++) {
5528 struct imsm_disk
*disk
;
5531 disk
= __get_imsm_disk(mpb
, i
);
5532 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5533 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5534 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5535 "missing:%d", (__u8
)i
);
5537 find_missing(super
);
5542 for (d
= super
->missing
; d
; d
= d
->next
)
5544 if (info
->failed_disks
> missing
) {
5545 pr_err("unable to add 'missing' disk to container\n");
5550 if (!check_name(super
, name
, 0))
5552 dv
= xmalloc(sizeof(*dv
));
5553 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5555 * Explicitly allow truncating to not confuse gcc's
5556 * -Werror=stringop-truncation
5558 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5559 memcpy(dev
->volume
, name
, namelen
);
5560 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5561 info
->layout
, info
->chunk_size
,
5562 s
->size
* BLOCKS_PER_KB
);
5563 data_disks
= get_data_disks(info
->level
, info
->layout
,
5565 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5566 size_per_member
= array_blocks
/ data_disks
;
5568 set_imsm_dev_size(dev
, array_blocks
);
5569 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5571 vol
->migr_state
= 0;
5572 set_migr_type(dev
, MIGR_INIT
);
5573 vol
->dirty
= !info
->state
;
5574 set_vol_curr_migr_unit(dev
, 0);
5575 map
= get_imsm_map(dev
, MAP_0
);
5576 set_pba_of_lba0(map
, super
->create_offset
);
5577 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5578 map
->failed_disk_num
= ~0;
5579 if (info
->level
> 0)
5580 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5581 : IMSM_T_STATE_UNINITIALIZED
);
5583 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5584 IMSM_T_STATE_NORMAL
;
5587 if (info
->level
== 1 && info
->raid_disks
> 2) {
5590 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5594 map
->raid_level
= info
->level
;
5595 if (info
->level
== 10) {
5596 map
->raid_level
= 1;
5597 map
->num_domains
= info
->raid_disks
/ 2;
5598 } else if (info
->level
== 1)
5599 map
->num_domains
= info
->raid_disks
;
5601 map
->num_domains
= 1;
5603 /* info->size is only int so use the 'size' parameter instead */
5604 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5605 num_data_stripes
/= map
->num_domains
;
5606 set_num_data_stripes(map
, num_data_stripes
);
5608 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5609 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5613 map
->num_members
= info
->raid_disks
;
5614 for (i
= 0; i
< map
->num_members
; i
++) {
5615 /* initialized in add_to_super */
5616 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5618 mpb
->num_raid_devs
++;
5619 mpb
->num_raid_devs_created
++;
5620 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5622 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5623 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5624 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5625 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5629 pr_err("imsm does not support consistency policy %s\n",
5630 map_num(consistency_policies
, s
->consistency_policy
));
5635 dv
->index
= super
->current_vol
;
5636 dv
->next
= super
->devlist
;
5637 super
->devlist
= dv
;
5639 imsm_update_version_info(super
);
5644 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5645 struct shape
*s
, char *name
,
5646 char *homehost
, int *uuid
,
5647 unsigned long long data_offset
)
5649 /* This is primarily called by Create when creating a new array.
5650 * We will then get add_to_super called for each component, and then
5651 * write_init_super called to write it out to each device.
5652 * For IMSM, Create can create on fresh devices or on a pre-existing
5654 * To create on a pre-existing array a different method will be called.
5655 * This one is just for fresh drives.
5657 struct intel_super
*super
;
5658 struct imsm_super
*mpb
;
5662 if (data_offset
!= INVALID_SECTORS
) {
5663 pr_err("data-offset not supported by imsm\n");
5668 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5672 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5674 mpb_size
= MAX_SECTOR_SIZE
;
5676 super
= alloc_super();
5678 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5683 pr_err("could not allocate superblock\n");
5686 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5687 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5688 pr_err("could not allocate migr_rec buffer\n");
5693 memset(super
->buf
, 0, mpb_size
);
5695 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5699 /* zeroing superblock */
5703 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5705 version
= (char *) mpb
->sig
;
5706 strcpy(version
, MPB_SIGNATURE
);
5707 version
+= strlen(MPB_SIGNATURE
);
5708 strcpy(version
, MPB_VERSION_RAID0
);
5713 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5715 unsigned int member_sector_size
;
5718 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5722 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5724 if (member_sector_size
!= super
->sector_size
)
5729 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5730 int fd
, char *devname
)
5732 struct intel_super
*super
= st
->sb
;
5733 struct imsm_super
*mpb
= super
->anchor
;
5734 struct imsm_disk
*_disk
;
5735 struct imsm_dev
*dev
;
5736 struct imsm_map
*map
;
5740 dev
= get_imsm_dev(super
, super
->current_vol
);
5741 map
= get_imsm_map(dev
, MAP_0
);
5743 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5744 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5750 /* we're doing autolayout so grab the pre-marked (in
5751 * validate_geometry) raid_disk
5753 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5754 if (dl
->raiddisk
== dk
->raid_disk
)
5757 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5758 if (dl
->major
== dk
->major
&&
5759 dl
->minor
== dk
->minor
)
5764 pr_err("%s is not a member of the same container\n", devname
);
5768 if (mpb
->num_disks
== 0)
5769 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5770 &super
->sector_size
))
5773 if (!drive_validate_sector_size(super
, dl
)) {
5774 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5778 /* add a pristine spare to the metadata */
5779 if (dl
->index
< 0) {
5780 dl
->index
= super
->anchor
->num_disks
;
5781 super
->anchor
->num_disks
++;
5783 /* Check the device has not already been added */
5784 slot
= get_imsm_disk_slot(map
, dl
->index
);
5786 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5787 pr_err("%s has been included in this array twice\n",
5791 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5792 dl
->disk
.status
= CONFIGURED_DISK
;
5794 /* update size of 'missing' disks to be at least as large as the
5795 * largest acitve member (we only have dummy missing disks when
5796 * creating the first volume)
5798 if (super
->current_vol
== 0) {
5799 for (df
= super
->missing
; df
; df
= df
->next
) {
5800 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5801 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5802 _disk
= __get_imsm_disk(mpb
, df
->index
);
5807 /* refresh unset/failed slots to point to valid 'missing' entries */
5808 for (df
= super
->missing
; df
; df
= df
->next
)
5809 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5810 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5812 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5814 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5815 if (is_gen_migration(dev
)) {
5816 struct imsm_map
*map2
= get_imsm_map(dev
,
5818 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5819 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5820 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5823 if ((unsigned)df
->index
==
5825 set_imsm_ord_tbl_ent(map2
,
5831 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5835 /* if we are creating the first raid device update the family number */
5836 if (super
->current_vol
== 0) {
5838 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5840 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5841 if (!_dev
|| !_disk
) {
5842 pr_err("BUG mpb setup error\n");
5848 sum
+= __gen_imsm_checksum(mpb
);
5849 mpb
->family_num
= __cpu_to_le32(sum
);
5850 mpb
->orig_family_num
= mpb
->family_num
;
5851 mpb
->creation_time
= __cpu_to_le64((__u64
)time(NULL
));
5853 super
->current_disk
= dl
;
5858 * Function marks disk as spare and restores disk serial
5859 * in case it was previously marked as failed by takeover operation
5861 * -1 : critical error
5862 * 0 : disk is marked as spare but serial is not set
5865 int mark_spare(struct dl
*disk
)
5867 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5874 if (!imsm_read_serial(disk
->fd
, NULL
, serial
, MAX_RAID_SERIAL_LEN
)) {
5875 /* Restore disk serial number, because takeover marks disk
5876 * as failed and adds to serial ':0' before it becomes
5879 serialcpy(disk
->serial
, serial
);
5880 serialcpy(disk
->disk
.serial
, serial
);
5883 disk
->disk
.status
= SPARE_DISK
;
5890 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
);
5892 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5893 int fd
, char *devname
,
5894 unsigned long long data_offset
)
5896 struct intel_super
*super
= st
->sb
;
5898 unsigned long long size
;
5899 unsigned int member_sector_size
;
5904 /* If we are on an RAID enabled platform check that the disk is
5905 * attached to the raid controller.
5906 * We do not need to test disks attachment for container based additions,
5907 * they shall be already tested when container was created/assembled.
5909 rv
= find_intel_hba_capability(fd
, super
, devname
);
5910 /* no orom/efi or non-intel hba of the disk */
5912 dprintf("capability: %p fd: %d ret: %d\n",
5913 super
->orom
, fd
, rv
);
5917 if (super
->current_vol
>= 0)
5918 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5921 dd
= xcalloc(sizeof(*dd
), 1);
5922 dd
->major
= major(stb
.st_rdev
);
5923 dd
->minor
= minor(stb
.st_rdev
);
5924 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5927 dd
->action
= DISK_ADD
;
5928 rv
= imsm_read_serial(fd
, devname
, dd
->serial
, MAX_RAID_SERIAL_LEN
);
5930 pr_err("failed to retrieve scsi serial, aborting\n");
5936 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5937 (super
->hba
->type
== SYS_DEV_VMD
))) {
5939 char *devpath
= diskfd_to_devpath(fd
);
5940 char controller_path
[PATH_MAX
];
5943 pr_err("failed to get devpath, aborting\n");
5950 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5953 if (!imsm_is_nvme_supported(dd
->fd
, 1)) {
5960 if (devpath_to_vendor(controller_path
) == 0x8086) {
5962 * If Intel's NVMe drive has serial ended with
5963 * "-A","-B","-1" or "-2" it means that this is "x8"
5964 * device (double drive on single PCIe card).
5965 * User should be warned about potential data loss.
5967 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5968 /* Skip empty character at the end */
5969 if (dd
->serial
[i
] == 0)
5972 if (((dd
->serial
[i
] == 'A') ||
5973 (dd
->serial
[i
] == 'B') ||
5974 (dd
->serial
[i
] == '1') ||
5975 (dd
->serial
[i
] == '2')) &&
5976 (dd
->serial
[i
-1] == '-'))
5977 pr_err("\tThe action you are about to take may put your data at risk.\n"
5978 "\tPlease note that x8 devices may consist of two separate x4 devices "
5979 "located on a single PCIe port.\n"
5980 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5983 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5984 !imsm_orom_has_tpv_support(super
->orom
)) {
5985 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5986 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
5993 get_dev_size(fd
, NULL
, &size
);
5994 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5996 if (super
->sector_size
== 0) {
5997 /* this a first device, so sector_size is not set yet */
5998 super
->sector_size
= member_sector_size
;
6001 /* clear migr_rec when adding disk to container */
6002 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6003 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
6005 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
6006 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
6007 MIGR_REC_BUF_SECTORS
*member_sector_size
)
6008 perror("Write migr_rec failed");
6012 serialcpy(dd
->disk
.serial
, dd
->serial
);
6013 set_total_blocks(&dd
->disk
, size
);
6014 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
6015 struct imsm_super
*mpb
= super
->anchor
;
6016 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6019 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
6020 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
6022 dd
->disk
.scsi_id
= __cpu_to_le32(0);
6024 if (st
->update_tail
) {
6025 dd
->next
= super
->disk_mgmt_list
;
6026 super
->disk_mgmt_list
= dd
;
6028 /* this is called outside of mdmon
6029 * write initial spare metadata
6030 * mdmon will overwrite it.
6032 dd
->next
= super
->disks
;
6034 write_super_imsm_spare(super
, dd
);
6040 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
6042 struct intel_super
*super
= st
->sb
;
6045 /* remove from super works only in mdmon - for communication
6046 * manager - monitor. Check if communication memory buffer
6049 if (!st
->update_tail
) {
6050 pr_err("shall be used in mdmon context only\n");
6053 dd
= xcalloc(1, sizeof(*dd
));
6054 dd
->major
= dk
->major
;
6055 dd
->minor
= dk
->minor
;
6058 dd
->action
= DISK_REMOVE
;
6060 dd
->next
= super
->disk_mgmt_list
;
6061 super
->disk_mgmt_list
= dd
;
6066 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
6069 char buf
[MAX_SECTOR_SIZE
];
6070 struct imsm_super anchor
;
6071 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
6074 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
)
6076 struct imsm_super
*mpb
= super
->anchor
;
6077 struct imsm_super
*spare
= &spare_record
.anchor
;
6083 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
6084 spare
->generation_num
= __cpu_to_le32(1UL);
6085 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
6086 spare
->num_disks
= 1;
6087 spare
->num_raid_devs
= 0;
6088 spare
->cache_size
= mpb
->cache_size
;
6089 spare
->pwr_cycle_count
= __cpu_to_le32(1);
6091 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
6092 MPB_SIGNATURE MPB_VERSION_RAID0
);
6094 spare
->disk
[0] = d
->disk
;
6095 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
6096 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6098 if (super
->sector_size
== 4096)
6099 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6101 sum
= __gen_imsm_checksum(spare
);
6102 spare
->family_num
= __cpu_to_le32(sum
);
6103 spare
->orig_family_num
= 0;
6104 sum
= __gen_imsm_checksum(spare
);
6105 spare
->check_sum
= __cpu_to_le32(sum
);
6107 if (store_imsm_mpb(d
->fd
, spare
)) {
6108 pr_err("failed for device %d:%d %s\n",
6109 d
->major
, d
->minor
, strerror(errno
));
6115 /* spare records have their own family number and do not have any defined raid
6118 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
6122 for (d
= super
->disks
; d
; d
= d
->next
) {
6126 if (write_super_imsm_spare(super
, d
))
6138 static int write_super_imsm(struct supertype
*st
, int doclose
)
6140 struct intel_super
*super
= st
->sb
;
6141 unsigned int sector_size
= super
->sector_size
;
6142 struct imsm_super
*mpb
= super
->anchor
;
6148 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6150 int clear_migration_record
= 1;
6153 /* 'generation' is incremented everytime the metadata is written */
6154 generation
= __le32_to_cpu(mpb
->generation_num
);
6156 mpb
->generation_num
= __cpu_to_le32(generation
);
6158 /* fix up cases where previous mdadm releases failed to set
6161 if (mpb
->orig_family_num
== 0)
6162 mpb
->orig_family_num
= mpb
->family_num
;
6164 for (d
= super
->disks
; d
; d
= d
->next
) {
6168 mpb
->disk
[d
->index
] = d
->disk
;
6172 for (d
= super
->missing
; d
; d
= d
->next
) {
6173 mpb
->disk
[d
->index
] = d
->disk
;
6176 mpb
->num_disks
= num_disks
;
6177 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6179 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6180 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6181 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6183 imsm_copy_dev(dev
, dev2
);
6184 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6186 if (is_gen_migration(dev2
))
6187 clear_migration_record
= 0;
6190 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6193 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6194 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6196 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6198 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6199 mpb_size
+= bbm_log_size
;
6200 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6203 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6206 /* recalculate checksum */
6207 sum
= __gen_imsm_checksum(mpb
);
6208 mpb
->check_sum
= __cpu_to_le32(sum
);
6210 if (super
->clean_migration_record_by_mdmon
) {
6211 clear_migration_record
= 1;
6212 super
->clean_migration_record_by_mdmon
= 0;
6214 if (clear_migration_record
)
6215 memset(super
->migr_rec_buf
, 0,
6216 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6218 if (sector_size
== 4096)
6219 convert_to_4k(super
);
6221 /* write the mpb for disks that compose raid devices */
6222 for (d
= super
->disks
; d
; d
= d
->next
) {
6223 if (d
->index
< 0 || is_failed(&d
->disk
))
6226 if (clear_migration_record
) {
6227 unsigned long long dsize
;
6229 get_dev_size(d
->fd
, NULL
, &dsize
);
6230 if (lseek64(d
->fd
, dsize
- sector_size
,
6232 if ((unsigned int)write(d
->fd
,
6233 super
->migr_rec_buf
,
6234 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6235 MIGR_REC_BUF_SECTORS
*sector_size
)
6236 perror("Write migr_rec failed");
6240 if (store_imsm_mpb(d
->fd
, mpb
))
6242 "failed for device %d:%d (fd: %d)%s\n",
6244 d
->fd
, strerror(errno
));
6253 return write_super_imsm_spares(super
, doclose
);
6258 static int create_array(struct supertype
*st
, int dev_idx
)
6261 struct imsm_update_create_array
*u
;
6262 struct intel_super
*super
= st
->sb
;
6263 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6264 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6265 struct disk_info
*inf
;
6266 struct imsm_disk
*disk
;
6269 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6270 sizeof(*inf
) * map
->num_members
;
6272 u
->type
= update_create_array
;
6273 u
->dev_idx
= dev_idx
;
6274 imsm_copy_dev(&u
->dev
, dev
);
6275 inf
= get_disk_info(u
);
6276 for (i
= 0; i
< map
->num_members
; i
++) {
6277 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6279 disk
= get_imsm_disk(super
, idx
);
6281 disk
= get_imsm_missing(super
, idx
);
6282 serialcpy(inf
[i
].serial
, disk
->serial
);
6284 append_metadata_update(st
, u
, len
);
6289 static int mgmt_disk(struct supertype
*st
)
6291 struct intel_super
*super
= st
->sb
;
6293 struct imsm_update_add_remove_disk
*u
;
6295 if (!super
->disk_mgmt_list
)
6300 u
->type
= update_add_remove_disk
;
6301 append_metadata_update(st
, u
, len
);
6306 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6308 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6310 struct ppl_header
*ppl_hdr
= buf
;
6313 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6315 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6317 perror("Failed to seek to PPL header location");
6321 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6323 perror("Write PPL header failed");
6332 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6334 struct intel_super
*super
= st
->sb
;
6336 struct ppl_header
*ppl_hdr
;
6339 /* first clear entire ppl space */
6340 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6344 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6346 pr_err("Failed to allocate PPL header buffer\n");
6350 memset(buf
, 0, PPL_HEADER_SIZE
);
6352 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6353 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6355 if (info
->mismatch_cnt
) {
6357 * We are overwriting an invalid ppl. Make one entry with wrong
6358 * checksum to prevent the kernel from skipping resync.
6360 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6361 ppl_hdr
->entries
[0].checksum
= ~0;
6364 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6370 static int is_rebuilding(struct imsm_dev
*dev
);
6372 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6373 struct mdinfo
*disk
)
6375 struct intel_super
*super
= st
->sb
;
6377 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6379 struct ppl_header
*ppl_hdr
= NULL
;
6381 struct imsm_dev
*dev
;
6384 unsigned long long ppl_offset
= 0;
6385 unsigned long long prev_gen_num
= 0;
6387 if (disk
->disk
.raid_disk
< 0)
6390 dev
= get_imsm_dev(super
, info
->container_member
);
6391 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6392 d
= get_imsm_dl_disk(super
, idx
);
6394 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6397 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6398 pr_err("Failed to allocate PPL header buffer\n");
6404 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6407 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6409 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6411 perror("Failed to seek to PPL header location");
6416 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6417 perror("Read PPL header failed");
6424 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6425 ppl_hdr
->checksum
= 0;
6427 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6428 dprintf("Wrong PPL header checksum on %s\n",
6433 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6434 /* previous was newest, it was already checked */
6438 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6439 super
->anchor
->orig_family_num
)) {
6440 dprintf("Wrong PPL header signature on %s\n",
6447 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6449 ppl_offset
+= PPL_HEADER_SIZE
;
6450 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6452 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6455 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6467 * Update metadata to use mutliple PPLs area (1MB).
6468 * This is done once for all RAID members
6470 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6471 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6473 struct mdinfo
*member_dev
;
6475 sprintf(subarray
, "%d", info
->container_member
);
6477 if (mdmon_running(st
->container_devnm
))
6478 st
->update_tail
= &st
->updates
;
6480 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6481 pr_err("Failed to update subarray %s\n",
6484 if (st
->update_tail
)
6485 flush_metadata_updates(st
);
6487 st
->ss
->sync_metadata(st
);
6488 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6489 for (member_dev
= info
->devs
; member_dev
;
6490 member_dev
= member_dev
->next
)
6491 member_dev
->ppl_size
=
6492 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6497 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6499 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6500 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6501 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6502 (is_rebuilding(dev
) &&
6503 vol_curr_migr_unit(dev
) == 0 &&
6504 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6505 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6507 info
->mismatch_cnt
++;
6508 } else if (ret
== 0 &&
6509 ppl_hdr
->entries_count
== 0 &&
6510 is_rebuilding(dev
) &&
6511 info
->resync_start
== 0) {
6513 * The header has no entries - add a single empty entry and
6514 * rewrite the header to prevent the kernel from going into
6515 * resync after an interrupted rebuild.
6517 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6518 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6526 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6528 struct intel_super
*super
= st
->sb
;
6532 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6533 info
->array
.level
!= 5)
6536 for (d
= super
->disks
; d
; d
= d
->next
) {
6537 if (d
->index
< 0 || is_failed(&d
->disk
))
6540 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6548 /*******************************************************************************
6549 * Function: write_init_bitmap_imsm_vol
6550 * Description: Write a bitmap header and prepares the area for the bitmap.
6552 * st : supertype information
6553 * vol_idx : the volume index to use
6558 ******************************************************************************/
6559 static int write_init_bitmap_imsm_vol(struct supertype
*st
, int vol_idx
)
6561 struct intel_super
*super
= st
->sb
;
6562 int prev_current_vol
= super
->current_vol
;
6566 super
->current_vol
= vol_idx
;
6567 for (d
= super
->disks
; d
; d
= d
->next
) {
6568 if (d
->index
< 0 || is_failed(&d
->disk
))
6570 ret
= st
->ss
->write_bitmap(st
, d
->fd
, NoUpdate
);
6574 super
->current_vol
= prev_current_vol
;
6578 /*******************************************************************************
6579 * Function: write_init_bitmap_imsm_all
6580 * Description: Write a bitmap header and prepares the area for the bitmap.
6581 * Operation is executed for volumes with CONSISTENCY_POLICY_BITMAP.
6583 * st : supertype information
6584 * info : info about the volume where the bitmap should be written
6585 * vol_idx : the volume index to use
6590 ******************************************************************************/
6591 static int write_init_bitmap_imsm_all(struct supertype
*st
, struct mdinfo
*info
,
6596 if (info
&& (info
->consistency_policy
== CONSISTENCY_POLICY_BITMAP
))
6597 ret
= write_init_bitmap_imsm_vol(st
, vol_idx
);
6602 static int write_init_super_imsm(struct supertype
*st
)
6604 struct intel_super
*super
= st
->sb
;
6605 int current_vol
= super
->current_vol
;
6609 getinfo_super_imsm(st
, &info
, NULL
);
6611 /* we are done with current_vol reset it to point st at the container */
6612 super
->current_vol
= -1;
6614 if (st
->update_tail
) {
6615 /* queue the recently created array / added disk
6616 * as a metadata update */
6618 /* determine if we are creating a volume or adding a disk */
6619 if (current_vol
< 0) {
6620 /* in the mgmt (add/remove) disk case we are running
6621 * in mdmon context, so don't close fd's
6625 /* adding the second volume to the array */
6626 rv
= write_init_ppl_imsm_all(st
, &info
);
6628 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6630 rv
= create_array(st
, current_vol
);
6634 for (d
= super
->disks
; d
; d
= d
->next
)
6635 Kill(d
->devname
, NULL
, 0, -1, 1);
6636 if (current_vol
>= 0) {
6637 rv
= write_init_ppl_imsm_all(st
, &info
);
6639 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6643 rv
= write_super_imsm(st
, 1);
6649 static int store_super_imsm(struct supertype
*st
, int fd
)
6651 struct intel_super
*super
= st
->sb
;
6652 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6657 if (super
->sector_size
== 4096)
6658 convert_to_4k(super
);
6659 return store_imsm_mpb(fd
, mpb
);
6662 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6663 int layout
, int raiddisks
, int chunk
,
6664 unsigned long long size
,
6665 unsigned long long data_offset
,
6667 unsigned long long *freesize
,
6671 unsigned long long ldsize
;
6672 struct intel_super
*super
;
6675 if (level
!= LEVEL_CONTAINER
)
6680 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6683 pr_err("imsm: Cannot open %s: %s\n",
6684 dev
, strerror(errno
));
6687 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6692 /* capabilities retrieve could be possible
6693 * note that there is no fd for the disks in array.
6695 super
= alloc_super();
6700 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6706 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6710 fd2devname(fd
, str
);
6711 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6712 fd
, str
, super
->orom
, rv
, raiddisks
);
6714 /* no orom/efi or non-intel hba of the disk */
6721 if (raiddisks
> super
->orom
->tds
) {
6723 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6724 raiddisks
, super
->orom
->tds
);
6728 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6729 (ldsize
>> 9) >> 32 > 0) {
6731 pr_err("%s exceeds maximum platform supported size\n", dev
);
6737 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6743 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6745 const unsigned long long base_start
= e
[*idx
].start
;
6746 unsigned long long end
= base_start
+ e
[*idx
].size
;
6749 if (base_start
== end
)
6753 for (i
= *idx
; i
< num_extents
; i
++) {
6754 /* extend overlapping extents */
6755 if (e
[i
].start
>= base_start
&&
6756 e
[i
].start
<= end
) {
6759 if (e
[i
].start
+ e
[i
].size
> end
)
6760 end
= e
[i
].start
+ e
[i
].size
;
6761 } else if (e
[i
].start
> end
) {
6767 return end
- base_start
;
6770 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6772 /* build a composite disk with all known extents and generate a new
6773 * 'maxsize' given the "all disks in an array must share a common start
6774 * offset" constraint
6776 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6780 unsigned long long pos
;
6781 unsigned long long start
= 0;
6782 unsigned long long maxsize
;
6783 unsigned long reserve
;
6785 /* coalesce and sort all extents. also, check to see if we need to
6786 * reserve space between member arrays
6789 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6792 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6795 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6800 while (i
< sum_extents
) {
6801 e
[j
].start
= e
[i
].start
;
6802 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6804 if (e
[j
-1].size
== 0)
6813 unsigned long long esize
;
6815 esize
= e
[i
].start
- pos
;
6816 if (esize
>= maxsize
) {
6821 pos
= e
[i
].start
+ e
[i
].size
;
6823 } while (e
[i
-1].size
);
6829 /* FIXME assumes volume at offset 0 is the first volume in a
6832 if (start_extent
> 0)
6833 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6837 if (maxsize
< reserve
)
6840 super
->create_offset
= ~((unsigned long long) 0);
6841 if (start
+ reserve
> super
->create_offset
)
6842 return 0; /* start overflows create_offset */
6843 super
->create_offset
= start
+ reserve
;
6845 return maxsize
- reserve
;
6848 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6850 if (level
< 0 || level
== 6 || level
== 4)
6853 /* if we have an orom prevent invalid raid levels */
6856 case 0: return imsm_orom_has_raid0(orom
);
6859 return imsm_orom_has_raid1e(orom
);
6860 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6861 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6862 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6865 return 1; /* not on an Intel RAID platform so anything goes */
6871 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6872 int dpa
, int verbose
)
6874 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6875 struct mdstat_ent
*memb
;
6881 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6882 if (memb
->metadata_version
&&
6883 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6884 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6885 !is_subarray(memb
->metadata_version
+9) &&
6887 struct dev_member
*dev
= memb
->members
;
6889 while(dev
&& (fd
< 0)) {
6890 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6891 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6893 fd
= open(path
, O_RDONLY
, 0);
6894 if (num
<= 0 || fd
< 0) {
6895 pr_vrb("Cannot open %s: %s\n",
6896 dev
->name
, strerror(errno
));
6902 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6903 struct mdstat_ent
*vol
;
6904 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6905 if (vol
->active
> 0 &&
6906 vol
->metadata_version
&&
6907 is_container_member(vol
, memb
->devnm
)) {
6912 if (*devlist
&& (found
< dpa
)) {
6913 dv
= xcalloc(1, sizeof(*dv
));
6914 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6915 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6918 dv
->next
= *devlist
;
6926 free_mdstat(mdstat
);
6931 static struct md_list
*
6932 get_loop_devices(void)
6935 struct md_list
*devlist
= NULL
;
6938 for(i
= 0; i
< 12; i
++) {
6939 dv
= xcalloc(1, sizeof(*dv
));
6940 dv
->devname
= xmalloc(40);
6941 sprintf(dv
->devname
, "/dev/loop%d", i
);
6949 static struct md_list
*
6950 get_devices(const char *hba_path
)
6952 struct md_list
*devlist
= NULL
;
6959 devlist
= get_loop_devices();
6962 /* scroll through /sys/dev/block looking for devices attached to
6965 dir
= opendir("/sys/dev/block");
6966 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6971 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6973 path
= devt_to_devpath(makedev(major
, minor
));
6976 if (!path_attached_to_hba(path
, hba_path
)) {
6983 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6985 fd2devname(fd
, buf
);
6988 pr_err("cannot open device: %s\n",
6993 dv
= xcalloc(1, sizeof(*dv
));
6994 dv
->devname
= xstrdup(buf
);
7001 devlist
= devlist
->next
;
7011 count_volumes_list(struct md_list
*devlist
, char *homehost
,
7012 int verbose
, int *found
)
7014 struct md_list
*tmpdev
;
7016 struct supertype
*st
;
7018 /* first walk the list of devices to find a consistent set
7019 * that match the criterea, if that is possible.
7020 * We flag the ones we like with 'used'.
7023 st
= match_metadata_desc_imsm("imsm");
7025 pr_vrb("cannot allocate memory for imsm supertype\n");
7029 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7030 char *devname
= tmpdev
->devname
;
7032 struct supertype
*tst
;
7034 if (tmpdev
->used
> 1)
7036 tst
= dup_super(st
);
7038 pr_vrb("cannot allocate memory for imsm supertype\n");
7041 tmpdev
->container
= 0;
7042 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
7044 dprintf("cannot open device %s: %s\n",
7045 devname
, strerror(errno
));
7047 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
7049 } else if (must_be_container(dfd
)) {
7050 struct supertype
*cst
;
7051 cst
= super_by_fd(dfd
, NULL
);
7053 dprintf("cannot recognize container type %s\n",
7056 } else if (tst
->ss
!= st
->ss
) {
7057 dprintf("non-imsm container - ignore it: %s\n",
7060 } else if (!tst
->ss
->load_container
||
7061 tst
->ss
->load_container(tst
, dfd
, NULL
))
7064 tmpdev
->container
= 1;
7067 cst
->ss
->free_super(cst
);
7069 tmpdev
->st_rdev
= rdev
;
7070 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
7071 dprintf("no RAID superblock on %s\n",
7074 } else if (tst
->ss
->compare_super
== NULL
) {
7075 dprintf("Cannot assemble %s metadata on %s\n",
7076 tst
->ss
->name
, devname
);
7082 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
7083 /* Ignore unrecognised devices during auto-assembly */
7088 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
7090 if (st
->minor_version
== -1)
7091 st
->minor_version
= tst
->minor_version
;
7093 if (memcmp(info
.uuid
, uuid_zero
,
7094 sizeof(int[4])) == 0) {
7095 /* this is a floating spare. It cannot define
7096 * an array unless there are no more arrays of
7097 * this type to be found. It can be included
7098 * in an array of this type though.
7104 if (st
->ss
!= tst
->ss
||
7105 st
->minor_version
!= tst
->minor_version
||
7106 st
->ss
->compare_super(st
, tst
, 1) != 0) {
7107 /* Some mismatch. If exactly one array matches this host,
7108 * we can resolve on that one.
7109 * Or, if we are auto assembling, we just ignore the second
7112 dprintf("superblock on %s doesn't match others - assembly aborted\n",
7118 dprintf("found: devname: %s\n", devname
);
7122 tst
->ss
->free_super(tst
);
7126 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
7127 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
7128 for (iter
= head
; iter
; iter
= iter
->next
) {
7129 dprintf("content->text_version: %s vol\n",
7130 iter
->text_version
);
7131 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
7132 /* do not assemble arrays with unsupported
7134 dprintf("Cannot activate member %s.\n",
7135 iter
->text_version
);
7142 dprintf("No valid super block on device list: err: %d %p\n",
7146 dprintf("no more devices to examine\n");
7149 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7150 if (tmpdev
->used
== 1 && tmpdev
->found
) {
7152 if (count
< tmpdev
->found
)
7155 count
-= tmpdev
->found
;
7158 if (tmpdev
->used
== 1)
7163 st
->ss
->free_super(st
);
7167 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
7170 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
7172 const struct orom_entry
*entry
;
7173 struct devid_list
*dv
, *devid_list
;
7178 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7179 if (strstr(idev
->path
, hba_path
))
7183 if (!idev
|| !idev
->dev_id
)
7186 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7188 if (!entry
|| !entry
->devid_list
)
7191 devid_list
= entry
->devid_list
;
7192 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7193 struct md_list
*devlist
;
7194 struct sys_dev
*device
= NULL
;
7199 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7201 device
= device_by_id(dv
->devid
);
7204 hpath
= device
->path
;
7208 devlist
= get_devices(hpath
);
7209 /* if no intel devices return zero volumes */
7210 if (devlist
== NULL
)
7213 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7215 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7216 if (devlist
== NULL
)
7220 count
+= count_volumes_list(devlist
,
7224 dprintf("found %d count: %d\n", found
, count
);
7227 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7230 struct md_list
*dv
= devlist
;
7231 devlist
= devlist
->next
;
7239 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7243 if (hba
->type
== SYS_DEV_VMD
) {
7244 struct sys_dev
*dev
;
7247 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7248 if (dev
->type
== SYS_DEV_VMD
)
7249 count
+= __count_volumes(dev
->path
, dpa
,
7254 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7257 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7259 /* up to 512 if the plaform supports it, otherwise the platform max.
7260 * 128 if no platform detected
7262 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7264 return min(512, (1 << fs
));
7268 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7269 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7271 /* check/set platform and metadata limits/defaults */
7272 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7273 pr_vrb("platform supports a maximum of %d disks per array\n",
7278 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7279 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7280 pr_vrb("platform does not support raid%d with %d disk%s\n",
7281 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7285 if (*chunk
== 0 || *chunk
== UnSet
)
7286 *chunk
= imsm_default_chunk(super
->orom
);
7288 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7289 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7293 if (layout
!= imsm_level_to_layout(level
)) {
7295 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7296 else if (level
== 10)
7297 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7299 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7304 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7305 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7306 pr_vrb("platform does not support a volume size over 2TB\n");
7313 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7314 * FIX ME add ahci details
7316 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7317 int layout
, int raiddisks
, int *chunk
,
7318 unsigned long long size
,
7319 unsigned long long data_offset
,
7321 unsigned long long *freesize
,
7325 struct intel_super
*super
= st
->sb
;
7326 struct imsm_super
*mpb
;
7328 unsigned long long pos
= 0;
7329 unsigned long long maxsize
;
7333 /* We must have the container info already read in. */
7337 mpb
= super
->anchor
;
7339 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7340 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7344 /* General test: make sure there is space for
7345 * 'raiddisks' device extents of size 'size' at a given
7348 unsigned long long minsize
= size
;
7349 unsigned long long start_offset
= MaxSector
;
7352 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7353 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7358 e
= get_extents(super
, dl
, 0);
7361 unsigned long long esize
;
7362 esize
= e
[i
].start
- pos
;
7363 if (esize
>= minsize
)
7365 if (found
&& start_offset
== MaxSector
) {
7368 } else if (found
&& pos
!= start_offset
) {
7372 pos
= e
[i
].start
+ e
[i
].size
;
7374 } while (e
[i
-1].size
);
7379 if (dcnt
< raiddisks
) {
7381 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7388 /* This device must be a member of the set */
7389 if (!stat_is_blkdev(dev
, &rdev
))
7391 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7392 if (dl
->major
== (int)major(rdev
) &&
7393 dl
->minor
== (int)minor(rdev
))
7398 pr_err("%s is not in the same imsm set\n", dev
);
7400 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7401 /* If a volume is present then the current creation attempt
7402 * cannot incorporate new spares because the orom may not
7403 * understand this configuration (all member disks must be
7404 * members of each array in the container).
7406 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7407 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7409 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7410 mpb
->num_disks
!= raiddisks
) {
7411 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7415 /* retrieve the largest free space block */
7416 e
= get_extents(super
, dl
, 0);
7421 unsigned long long esize
;
7423 esize
= e
[i
].start
- pos
;
7424 if (esize
>= maxsize
)
7426 pos
= e
[i
].start
+ e
[i
].size
;
7428 } while (e
[i
-1].size
);
7433 pr_err("unable to determine free space for: %s\n",
7437 if (maxsize
< size
) {
7439 pr_err("%s not enough space (%llu < %llu)\n",
7440 dev
, maxsize
, size
);
7444 /* count total number of extents for merge */
7446 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7448 i
+= dl
->extent_cnt
;
7450 maxsize
= merge_extents(super
, i
);
7452 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7453 pr_err("attempting to create a second volume with size less then remaining space.\n");
7455 if (maxsize
< size
|| maxsize
== 0) {
7458 pr_err("no free space left on device. Aborting...\n");
7460 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7466 *freesize
= maxsize
;
7469 int count
= count_volumes(super
->hba
,
7470 super
->orom
->dpa
, verbose
);
7471 if (super
->orom
->vphba
<= count
) {
7472 pr_vrb("platform does not support more than %d raid volumes.\n",
7473 super
->orom
->vphba
);
7480 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7481 unsigned long long size
, int chunk
,
7482 unsigned long long *freesize
)
7484 struct intel_super
*super
= st
->sb
;
7485 struct imsm_super
*mpb
= super
->anchor
;
7490 unsigned long long maxsize
;
7491 unsigned long long minsize
;
7495 /* find the largest common start free region of the possible disks */
7499 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7505 /* don't activate new spares if we are orom constrained
7506 * and there is already a volume active in the container
7508 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7511 e
= get_extents(super
, dl
, 0);
7514 for (i
= 1; e
[i
-1].size
; i
++)
7522 maxsize
= merge_extents(super
, extent_cnt
);
7526 minsize
= chunk
* 2;
7528 if (cnt
< raiddisks
||
7529 (super
->orom
&& used
&& used
!= raiddisks
) ||
7530 maxsize
< minsize
||
7532 pr_err("not enough devices with space to create array.\n");
7533 return 0; /* No enough free spaces large enough */
7544 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7545 pr_err("attempting to create a second volume with size less then remaining space.\n");
7547 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7549 dl
->raiddisk
= cnt
++;
7553 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7558 static int reserve_space(struct supertype
*st
, int raiddisks
,
7559 unsigned long long size
, int chunk
,
7560 unsigned long long *freesize
)
7562 struct intel_super
*super
= st
->sb
;
7567 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7570 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7572 dl
->raiddisk
= cnt
++;
7579 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7580 int raiddisks
, int *chunk
, unsigned long long size
,
7581 unsigned long long data_offset
,
7582 char *dev
, unsigned long long *freesize
,
7583 int consistency_policy
, int verbose
)
7590 * if given unused devices create a container
7591 * if given given devices in a container create a member volume
7593 if (level
== LEVEL_CONTAINER
) {
7594 /* Must be a fresh device to add to a container */
7595 return validate_geometry_imsm_container(st
, level
, layout
,
7604 * Size is given in sectors.
7606 if (size
&& (size
< 2048)) {
7607 pr_err("Given size must be greater than 1M.\n");
7608 /* Depends on algorithm in Create.c :
7609 * if container was given (dev == NULL) return -1,
7610 * if block device was given ( dev != NULL) return 0.
7612 return dev
? -1 : 0;
7617 struct intel_super
*super
= st
->sb
;
7618 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7619 raiddisks
, chunk
, size
,
7622 /* we are being asked to automatically layout a
7623 * new volume based on the current contents of
7624 * the container. If the the parameters can be
7625 * satisfied reserve_space will record the disks,
7626 * start offset, and size of the volume to be
7627 * created. add_to_super and getinfo_super
7628 * detect when autolayout is in progress.
7630 /* assuming that freesize is always given when array is
7632 if (super
->orom
&& freesize
) {
7634 count
= count_volumes(super
->hba
,
7635 super
->orom
->dpa
, verbose
);
7636 if (super
->orom
->vphba
<= count
) {
7637 pr_vrb("platform does not support more than %d raid volumes.\n",
7638 super
->orom
->vphba
);
7643 return reserve_space(st
, raiddisks
, size
,
7649 /* creating in a given container */
7650 return validate_geometry_imsm_volume(st
, level
, layout
,
7651 raiddisks
, chunk
, size
,
7653 dev
, freesize
, verbose
);
7656 /* This device needs to be a device in an 'imsm' container */
7657 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7660 pr_err("Cannot create this array on device %s\n",
7665 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7667 pr_err("Cannot open %s: %s\n",
7668 dev
, strerror(errno
));
7671 /* Well, it is in use by someone, maybe an 'imsm' container. */
7672 cfd
= open_container(fd
);
7676 pr_err("Cannot use %s: It is busy\n",
7680 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7681 if (sra
&& sra
->array
.major_version
== -1 &&
7682 strcmp(sra
->text_version
, "imsm") == 0)
7686 /* This is a member of a imsm container. Load the container
7687 * and try to create a volume
7689 struct intel_super
*super
;
7691 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7693 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7695 return validate_geometry_imsm_volume(st
, level
, layout
,
7697 size
, data_offset
, dev
,
7704 pr_err("failed container membership check\n");
7710 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7712 struct intel_super
*super
= st
->sb
;
7714 if (level
&& *level
== UnSet
)
7715 *level
= LEVEL_CONTAINER
;
7717 if (level
&& layout
&& *layout
== UnSet
)
7718 *layout
= imsm_level_to_layout(*level
);
7720 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7721 *chunk
= imsm_default_chunk(super
->orom
);
7724 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7726 static int kill_subarray_imsm(struct supertype
*st
, char *subarray_id
)
7728 /* remove the subarray currently referenced by subarray_id */
7730 struct intel_dev
**dp
;
7731 struct intel_super
*super
= st
->sb
;
7732 __u8 current_vol
= strtoul(subarray_id
, NULL
, 10);
7733 struct imsm_super
*mpb
= super
->anchor
;
7735 if (mpb
->num_raid_devs
== 0)
7738 /* block deletions that would change the uuid of active subarrays
7740 * FIXME when immutable ids are available, but note that we'll
7741 * also need to fixup the invalidated/active subarray indexes in
7744 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7747 if (i
< current_vol
)
7749 sprintf(subarray
, "%u", i
);
7750 if (is_subarray_active(subarray
, st
->devnm
)) {
7751 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7758 if (st
->update_tail
) {
7759 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7761 u
->type
= update_kill_array
;
7762 u
->dev_idx
= current_vol
;
7763 append_metadata_update(st
, u
, sizeof(*u
));
7768 for (dp
= &super
->devlist
; *dp
;)
7769 if ((*dp
)->index
== current_vol
) {
7772 handle_missing(super
, (*dp
)->dev
);
7773 if ((*dp
)->index
> current_vol
)
7778 /* no more raid devices, all active components are now spares,
7779 * but of course failed are still failed
7781 if (--mpb
->num_raid_devs
== 0) {
7784 for (d
= super
->disks
; d
; d
= d
->next
)
7789 super
->updates_pending
++;
7794 static int get_rwh_policy_from_update(char *update
)
7796 if (strcmp(update
, "ppl") == 0)
7797 return RWH_MULTIPLE_DISTRIBUTED
;
7798 else if (strcmp(update
, "no-ppl") == 0)
7799 return RWH_MULTIPLE_OFF
;
7800 else if (strcmp(update
, "bitmap") == 0)
7802 else if (strcmp(update
, "no-bitmap") == 0)
7807 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7808 char *update
, struct mddev_ident
*ident
)
7810 /* update the subarray currently referenced by ->current_vol */
7811 struct intel_super
*super
= st
->sb
;
7812 struct imsm_super
*mpb
= super
->anchor
;
7814 if (strcmp(update
, "name") == 0) {
7815 char *name
= ident
->name
;
7819 if (is_subarray_active(subarray
, st
->devnm
)) {
7820 pr_err("Unable to update name of active subarray\n");
7824 if (!check_name(super
, name
, 0))
7827 vol
= strtoul(subarray
, &ep
, 10);
7828 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7831 if (st
->update_tail
) {
7832 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7834 u
->type
= update_rename_array
;
7836 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7837 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7838 append_metadata_update(st
, u
, sizeof(*u
));
7840 struct imsm_dev
*dev
;
7843 dev
= get_imsm_dev(super
, vol
);
7844 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7845 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7846 memcpy(dev
->volume
, name
, namelen
);
7847 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7848 dev
= get_imsm_dev(super
, i
);
7849 handle_missing(super
, dev
);
7851 super
->updates_pending
++;
7853 } else if (get_rwh_policy_from_update(update
) != -1) {
7856 int vol
= strtoul(subarray
, &ep
, 10);
7858 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7861 new_policy
= get_rwh_policy_from_update(update
);
7863 if (st
->update_tail
) {
7864 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7866 u
->type
= update_rwh_policy
;
7868 u
->new_policy
= new_policy
;
7869 append_metadata_update(st
, u
, sizeof(*u
));
7871 struct imsm_dev
*dev
;
7873 dev
= get_imsm_dev(super
, vol
);
7874 dev
->rwh_policy
= new_policy
;
7875 super
->updates_pending
++;
7877 if (new_policy
== RWH_BITMAP
)
7878 return write_init_bitmap_imsm_vol(st
, vol
);
7885 static int is_gen_migration(struct imsm_dev
*dev
)
7890 if (!dev
->vol
.migr_state
)
7893 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7899 static int is_rebuilding(struct imsm_dev
*dev
)
7901 struct imsm_map
*migr_map
;
7903 if (!dev
->vol
.migr_state
)
7906 if (migr_type(dev
) != MIGR_REBUILD
)
7909 migr_map
= get_imsm_map(dev
, MAP_1
);
7911 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7917 static int is_initializing(struct imsm_dev
*dev
)
7919 struct imsm_map
*migr_map
;
7921 if (!dev
->vol
.migr_state
)
7924 if (migr_type(dev
) != MIGR_INIT
)
7927 migr_map
= get_imsm_map(dev
, MAP_1
);
7929 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7935 static void update_recovery_start(struct intel_super
*super
,
7936 struct imsm_dev
*dev
,
7937 struct mdinfo
*array
)
7939 struct mdinfo
*rebuild
= NULL
;
7943 if (!is_rebuilding(dev
))
7946 /* Find the rebuild target, but punt on the dual rebuild case */
7947 for (d
= array
->devs
; d
; d
= d
->next
)
7948 if (d
->recovery_start
== 0) {
7955 /* (?) none of the disks are marked with
7956 * IMSM_ORD_REBUILD, so assume they are missing and the
7957 * disk_ord_tbl was not correctly updated
7959 dprintf("failed to locate out-of-sync disk\n");
7963 units
= vol_curr_migr_unit(dev
);
7964 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7967 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7969 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7971 /* Given a container loaded by load_super_imsm_all,
7972 * extract information about all the arrays into
7974 * If 'subarray' is given, just extract info about that array.
7976 * For each imsm_dev create an mdinfo, fill it in,
7977 * then look for matching devices in super->disks
7978 * and create appropriate device mdinfo.
7980 struct intel_super
*super
= st
->sb
;
7981 struct imsm_super
*mpb
= super
->anchor
;
7982 struct mdinfo
*rest
= NULL
;
7986 int spare_disks
= 0;
7987 int current_vol
= super
->current_vol
;
7989 /* do not assemble arrays when not all attributes are supported */
7990 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7992 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7995 /* count spare devices, not used in maps
7997 for (d
= super
->disks
; d
; d
= d
->next
)
8001 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8002 struct imsm_dev
*dev
;
8003 struct imsm_map
*map
;
8004 struct imsm_map
*map2
;
8005 struct mdinfo
*this;
8012 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
8015 dev
= get_imsm_dev(super
, i
);
8016 map
= get_imsm_map(dev
, MAP_0
);
8017 map2
= get_imsm_map(dev
, MAP_1
);
8018 level
= get_imsm_raid_level(map
);
8020 /* do not publish arrays that are in the middle of an
8021 * unsupported migration
8023 if (dev
->vol
.migr_state
&&
8024 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
8025 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
8029 /* do not publish arrays that are not support by controller's
8033 this = xmalloc(sizeof(*this));
8035 super
->current_vol
= i
;
8036 getinfo_super_imsm_volume(st
, this, NULL
);
8038 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
8039 /* mdadm does not support all metadata features- set the bit in all arrays state */
8040 if (!validate_geometry_imsm_orom(super
,
8041 level
, /* RAID level */
8042 imsm_level_to_layout(level
),
8043 map
->num_members
, /* raid disks */
8044 &chunk
, imsm_dev_size(dev
),
8046 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
8048 this->array
.state
|=
8049 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8050 (1<<MD_SB_BLOCK_VOLUME
);
8053 /* if array has bad blocks, set suitable bit in all arrays state */
8055 this->array
.state
|=
8056 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8057 (1<<MD_SB_BLOCK_VOLUME
);
8059 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
8060 unsigned long long recovery_start
;
8061 struct mdinfo
*info_d
;
8069 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
8070 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
8071 for (d
= super
->disks
; d
; d
= d
->next
)
8072 if (d
->index
== idx
)
8075 recovery_start
= MaxSector
;
8078 if (d
&& is_failed(&d
->disk
))
8080 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
8082 if (!(ord
& IMSM_ORD_REBUILD
))
8083 this->array
.working_disks
++;
8085 * if we skip some disks the array will be assmebled degraded;
8086 * reset resync start to avoid a dirty-degraded
8087 * situation when performing the intial sync
8092 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
8093 if ((!able_to_resync(level
, missing
) ||
8094 recovery_start
== 0))
8095 this->resync_start
= MaxSector
;
8098 * FIXME handle dirty degraded
8105 info_d
= xcalloc(1, sizeof(*info_d
));
8106 info_d
->next
= this->devs
;
8107 this->devs
= info_d
;
8109 info_d
->disk
.number
= d
->index
;
8110 info_d
->disk
.major
= d
->major
;
8111 info_d
->disk
.minor
= d
->minor
;
8112 info_d
->disk
.raid_disk
= slot
;
8113 info_d
->recovery_start
= recovery_start
;
8115 if (slot
< map2
->num_members
)
8116 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8118 this->array
.spare_disks
++;
8120 if (slot
< map
->num_members
)
8121 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8123 this->array
.spare_disks
++;
8126 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
8127 info_d
->data_offset
= pba_of_lba0(map
);
8128 info_d
->component_size
= calc_component_size(map
, dev
);
8130 if (map
->raid_level
== 5) {
8131 info_d
->ppl_sector
= this->ppl_sector
;
8132 info_d
->ppl_size
= this->ppl_size
;
8133 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
8134 recovery_start
== 0)
8135 this->resync_start
= 0;
8138 info_d
->bb
.supported
= 1;
8139 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
8140 info_d
->data_offset
,
8141 info_d
->component_size
,
8144 /* now that the disk list is up-to-date fixup recovery_start */
8145 update_recovery_start(super
, dev
, this);
8146 this->array
.spare_disks
+= spare_disks
;
8148 /* check for reshape */
8149 if (this->reshape_active
== 1)
8150 recover_backup_imsm(st
, this);
8154 super
->current_vol
= current_vol
;
8158 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
8159 int failed
, int look_in_map
)
8161 struct imsm_map
*map
;
8163 map
= get_imsm_map(dev
, look_in_map
);
8166 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
8167 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
8169 switch (get_imsm_raid_level(map
)) {
8171 return IMSM_T_STATE_FAILED
;
8174 if (failed
< map
->num_members
)
8175 return IMSM_T_STATE_DEGRADED
;
8177 return IMSM_T_STATE_FAILED
;
8182 * check to see if any mirrors have failed, otherwise we
8183 * are degraded. Even numbered slots are mirrored on
8187 /* gcc -Os complains that this is unused */
8188 int insync
= insync
;
8190 for (i
= 0; i
< map
->num_members
; i
++) {
8191 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8192 int idx
= ord_to_idx(ord
);
8193 struct imsm_disk
*disk
;
8195 /* reset the potential in-sync count on even-numbered
8196 * slots. num_copies is always 2 for imsm raid10
8201 disk
= get_imsm_disk(super
, idx
);
8202 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8205 /* no in-sync disks left in this mirror the
8209 return IMSM_T_STATE_FAILED
;
8212 return IMSM_T_STATE_DEGRADED
;
8216 return IMSM_T_STATE_DEGRADED
;
8218 return IMSM_T_STATE_FAILED
;
8224 return map
->map_state
;
8227 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8232 struct imsm_disk
*disk
;
8233 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8234 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8235 struct imsm_map
*map_for_loop
;
8240 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8241 * disks that are being rebuilt. New failures are recorded to
8242 * map[0]. So we look through all the disks we started with and
8243 * see if any failures are still present, or if any new ones
8247 if (prev
&& (map
->num_members
< prev
->num_members
))
8248 map_for_loop
= prev
;
8250 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8252 /* when MAP_X is passed both maps failures are counted
8255 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8256 i
< prev
->num_members
) {
8257 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8258 idx_1
= ord_to_idx(ord
);
8260 disk
= get_imsm_disk(super
, idx_1
);
8261 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8264 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8265 i
< map
->num_members
) {
8266 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8267 idx
= ord_to_idx(ord
);
8270 disk
= get_imsm_disk(super
, idx
);
8271 if (!disk
|| is_failed(disk
) ||
8272 ord
& IMSM_ORD_REBUILD
)
8281 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8284 struct intel_super
*super
= c
->sb
;
8285 struct imsm_super
*mpb
= super
->anchor
;
8286 struct imsm_update_prealloc_bb_mem u
;
8288 if (atoi(inst
) >= mpb
->num_raid_devs
) {
8289 pr_err("subarry index %d, out of range\n", atoi(inst
));
8293 dprintf("imsm: open_new %s\n", inst
);
8294 a
->info
.container_member
= atoi(inst
);
8296 u
.type
= update_prealloc_badblocks_mem
;
8297 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8302 static int is_resyncing(struct imsm_dev
*dev
)
8304 struct imsm_map
*migr_map
;
8306 if (!dev
->vol
.migr_state
)
8309 if (migr_type(dev
) == MIGR_INIT
||
8310 migr_type(dev
) == MIGR_REPAIR
)
8313 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8316 migr_map
= get_imsm_map(dev
, MAP_1
);
8318 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8319 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8325 /* return true if we recorded new information */
8326 static int mark_failure(struct intel_super
*super
,
8327 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8331 struct imsm_map
*map
;
8332 char buf
[MAX_RAID_SERIAL_LEN
+3];
8333 unsigned int len
, shift
= 0;
8335 /* new failures are always set in map[0] */
8336 map
= get_imsm_map(dev
, MAP_0
);
8338 slot
= get_imsm_disk_slot(map
, idx
);
8342 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8343 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8346 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8347 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8349 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8350 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8351 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8353 disk
->status
|= FAILED_DISK
;
8354 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8355 /* mark failures in second map if second map exists and this disk
8357 * This is valid for migration, initialization and rebuild
8359 if (dev
->vol
.migr_state
) {
8360 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8361 int slot2
= get_imsm_disk_slot(map2
, idx
);
8363 if (slot2
< map2
->num_members
&& slot2
>= 0)
8364 set_imsm_ord_tbl_ent(map2
, slot2
,
8365 idx
| IMSM_ORD_REBUILD
);
8367 if (map
->failed_disk_num
== 0xff ||
8368 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8369 map
->failed_disk_num
= slot
;
8371 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8376 static void mark_missing(struct intel_super
*super
,
8377 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8379 mark_failure(super
, dev
, disk
, idx
);
8381 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8384 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8385 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8388 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8392 if (!super
->missing
)
8395 /* When orom adds replacement for missing disk it does
8396 * not remove entry of missing disk, but just updates map with
8397 * new added disk. So it is not enough just to test if there is
8398 * any missing disk, we have to look if there are any failed disks
8399 * in map to stop migration */
8401 dprintf("imsm: mark missing\n");
8402 /* end process for initialization and rebuild only
8404 if (is_gen_migration(dev
) == 0) {
8405 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8409 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8410 struct imsm_map
*map1
;
8411 int i
, ord
, ord_map1
;
8414 for (i
= 0; i
< map
->num_members
; i
++) {
8415 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8416 if (!(ord
& IMSM_ORD_REBUILD
))
8419 map1
= get_imsm_map(dev
, MAP_1
);
8423 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8424 if (ord_map1
& IMSM_ORD_REBUILD
)
8429 map_state
= imsm_check_degraded(super
, dev
,
8431 end_migration(dev
, super
, map_state
);
8435 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8436 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8437 super
->updates_pending
++;
8440 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8443 unsigned long long array_blocks
;
8444 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8445 int used_disks
= imsm_num_data_members(map
);
8447 if (used_disks
== 0) {
8448 /* when problems occures
8449 * return current array_blocks value
8451 array_blocks
= imsm_dev_size(dev
);
8453 return array_blocks
;
8456 /* set array size in metadata
8459 /* OLCE size change is caused by added disks
8461 array_blocks
= per_dev_array_size(map
) * used_disks
;
8463 /* Online Volume Size Change
8464 * Using available free space
8466 array_blocks
= new_size
;
8468 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8469 set_imsm_dev_size(dev
, array_blocks
);
8471 return array_blocks
;
8474 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8476 static void imsm_progress_container_reshape(struct intel_super
*super
)
8478 /* if no device has a migr_state, but some device has a
8479 * different number of members than the previous device, start
8480 * changing the number of devices in this device to match
8483 struct imsm_super
*mpb
= super
->anchor
;
8484 int prev_disks
= -1;
8488 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8489 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8490 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8491 struct imsm_map
*map2
;
8492 int prev_num_members
;
8494 if (dev
->vol
.migr_state
)
8497 if (prev_disks
== -1)
8498 prev_disks
= map
->num_members
;
8499 if (prev_disks
== map
->num_members
)
8502 /* OK, this array needs to enter reshape mode.
8503 * i.e it needs a migr_state
8506 copy_map_size
= sizeof_imsm_map(map
);
8507 prev_num_members
= map
->num_members
;
8508 map
->num_members
= prev_disks
;
8509 dev
->vol
.migr_state
= 1;
8510 set_vol_curr_migr_unit(dev
, 0);
8511 set_migr_type(dev
, MIGR_GEN_MIGR
);
8512 for (i
= prev_num_members
;
8513 i
< map
->num_members
; i
++)
8514 set_imsm_ord_tbl_ent(map
, i
, i
);
8515 map2
= get_imsm_map(dev
, MAP_1
);
8516 /* Copy the current map */
8517 memcpy(map2
, map
, copy_map_size
);
8518 map2
->num_members
= prev_num_members
;
8520 imsm_set_array_size(dev
, -1);
8521 super
->clean_migration_record_by_mdmon
= 1;
8522 super
->updates_pending
++;
8526 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8527 * states are handled in imsm_set_disk() with one exception, when a
8528 * resync is stopped due to a new failure this routine will set the
8529 * 'degraded' state for the array.
8531 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8533 int inst
= a
->info
.container_member
;
8534 struct intel_super
*super
= a
->container
->sb
;
8535 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8536 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8537 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8538 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8539 __u32 blocks_per_unit
;
8541 if (dev
->vol
.migr_state
&&
8542 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8543 /* array state change is blocked due to reshape action
8545 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8546 * - finish the reshape (if last_checkpoint is big and action != reshape)
8547 * - update vol_curr_migr_unit
8549 if (a
->curr_action
== reshape
) {
8550 /* still reshaping, maybe update vol_curr_migr_unit */
8551 goto mark_checkpoint
;
8553 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8554 /* for some reason we aborted the reshape.
8556 * disable automatic metadata rollback
8557 * user action is required to recover process
8560 struct imsm_map
*map2
=
8561 get_imsm_map(dev
, MAP_1
);
8562 dev
->vol
.migr_state
= 0;
8563 set_migr_type(dev
, 0);
8564 set_vol_curr_migr_unit(dev
, 0);
8566 sizeof_imsm_map(map2
));
8567 super
->updates_pending
++;
8570 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8571 unsigned long long array_blocks
;
8575 used_disks
= imsm_num_data_members(map
);
8576 if (used_disks
> 0) {
8578 per_dev_array_size(map
) *
8581 round_size_to_mb(array_blocks
,
8583 a
->info
.custom_array_size
= array_blocks
;
8584 /* encourage manager to update array
8588 a
->check_reshape
= 1;
8590 /* finalize online capacity expansion/reshape */
8591 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8593 mdi
->disk
.raid_disk
,
8596 imsm_progress_container_reshape(super
);
8601 /* before we activate this array handle any missing disks */
8602 if (consistent
== 2)
8603 handle_missing(super
, dev
);
8605 if (consistent
== 2 &&
8606 (!is_resync_complete(&a
->info
) ||
8607 map_state
!= IMSM_T_STATE_NORMAL
||
8608 dev
->vol
.migr_state
))
8611 if (is_resync_complete(&a
->info
)) {
8612 /* complete intialization / resync,
8613 * recovery and interrupted recovery is completed in
8616 if (is_resyncing(dev
)) {
8617 dprintf("imsm: mark resync done\n");
8618 end_migration(dev
, super
, map_state
);
8619 super
->updates_pending
++;
8620 a
->last_checkpoint
= 0;
8622 } else if ((!is_resyncing(dev
) && !failed
) &&
8623 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8624 /* mark the start of the init process if nothing is failed */
8625 dprintf("imsm: mark resync start\n");
8626 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8627 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8629 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8630 super
->updates_pending
++;
8634 /* skip checkpointing for general migration,
8635 * it is controlled in mdadm
8637 if (is_gen_migration(dev
))
8638 goto skip_mark_checkpoint
;
8640 /* check if we can update vol_curr_migr_unit from resync_start,
8643 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8644 if (blocks_per_unit
) {
8645 set_vol_curr_migr_unit(dev
,
8646 a
->last_checkpoint
/ blocks_per_unit
);
8647 dprintf("imsm: mark checkpoint (%llu)\n",
8648 vol_curr_migr_unit(dev
));
8649 super
->updates_pending
++;
8652 skip_mark_checkpoint
:
8653 /* mark dirty / clean */
8654 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8655 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8656 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8658 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8660 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8661 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8662 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8663 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8665 super
->updates_pending
++;
8671 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8673 int inst
= a
->info
.container_member
;
8674 struct intel_super
*super
= a
->container
->sb
;
8675 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8676 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8678 if (slot
> map
->num_members
) {
8679 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8680 slot
, map
->num_members
- 1);
8687 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8690 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8692 int inst
= a
->info
.container_member
;
8693 struct intel_super
*super
= a
->container
->sb
;
8694 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8695 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8696 struct imsm_disk
*disk
;
8698 int recovery_not_finished
= 0;
8702 int rebuild_done
= 0;
8705 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8709 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8710 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8712 /* check for new failures */
8713 if (disk
&& (state
& DS_FAULTY
)) {
8714 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8715 super
->updates_pending
++;
8718 /* check if in_sync */
8719 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8720 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8722 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8724 super
->updates_pending
++;
8727 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8728 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8730 /* check if recovery complete, newly degraded, or failed */
8731 dprintf("imsm: Detected transition to state ");
8732 switch (map_state
) {
8733 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8734 dprintf("normal: ");
8735 if (is_rebuilding(dev
)) {
8736 dprintf_cont("while rebuilding");
8737 /* check if recovery is really finished */
8738 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8739 if (mdi
->recovery_start
!= MaxSector
) {
8740 recovery_not_finished
= 1;
8743 if (recovery_not_finished
) {
8745 dprintf("Rebuild has not finished yet, state not changed");
8746 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8747 a
->last_checkpoint
= mdi
->recovery_start
;
8748 super
->updates_pending
++;
8752 end_migration(dev
, super
, map_state
);
8753 map
->failed_disk_num
= ~0;
8754 super
->updates_pending
++;
8755 a
->last_checkpoint
= 0;
8758 if (is_gen_migration(dev
)) {
8759 dprintf_cont("while general migration");
8760 if (a
->last_checkpoint
>= a
->info
.component_size
)
8761 end_migration(dev
, super
, map_state
);
8763 map
->map_state
= map_state
;
8764 map
->failed_disk_num
= ~0;
8765 super
->updates_pending
++;
8769 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8770 dprintf_cont("degraded: ");
8771 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8772 dprintf_cont("mark degraded");
8773 map
->map_state
= map_state
;
8774 super
->updates_pending
++;
8775 a
->last_checkpoint
= 0;
8778 if (is_rebuilding(dev
)) {
8779 dprintf_cont("while rebuilding ");
8780 if (state
& DS_FAULTY
) {
8781 dprintf_cont("removing failed drive ");
8782 if (n
== map
->failed_disk_num
) {
8783 dprintf_cont("end migration");
8784 end_migration(dev
, super
, map_state
);
8785 a
->last_checkpoint
= 0;
8787 dprintf_cont("fail detected during rebuild, changing map state");
8788 map
->map_state
= map_state
;
8790 super
->updates_pending
++;
8796 /* check if recovery is really finished */
8797 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8798 if (mdi
->recovery_start
!= MaxSector
) {
8799 recovery_not_finished
= 1;
8802 if (recovery_not_finished
) {
8804 dprintf_cont("Rebuild has not finished yet");
8805 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8806 a
->last_checkpoint
=
8807 mdi
->recovery_start
;
8808 super
->updates_pending
++;
8813 dprintf_cont(" Rebuild done, still degraded");
8814 end_migration(dev
, super
, map_state
);
8815 a
->last_checkpoint
= 0;
8816 super
->updates_pending
++;
8818 for (i
= 0; i
< map
->num_members
; i
++) {
8819 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8821 if (idx
& IMSM_ORD_REBUILD
)
8822 map
->failed_disk_num
= i
;
8824 super
->updates_pending
++;
8827 if (is_gen_migration(dev
)) {
8828 dprintf_cont("while general migration");
8829 if (a
->last_checkpoint
>= a
->info
.component_size
)
8830 end_migration(dev
, super
, map_state
);
8832 map
->map_state
= map_state
;
8833 manage_second_map(super
, dev
);
8835 super
->updates_pending
++;
8838 if (is_initializing(dev
)) {
8839 dprintf_cont("while initialization.");
8840 map
->map_state
= map_state
;
8841 super
->updates_pending
++;
8845 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8846 dprintf_cont("failed: ");
8847 if (is_gen_migration(dev
)) {
8848 dprintf_cont("while general migration");
8849 map
->map_state
= map_state
;
8850 super
->updates_pending
++;
8853 if (map
->map_state
!= map_state
) {
8854 dprintf_cont("mark failed");
8855 end_migration(dev
, super
, map_state
);
8856 super
->updates_pending
++;
8857 a
->last_checkpoint
= 0;
8862 dprintf_cont("state %i\n", map_state
);
8867 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8870 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8871 unsigned long long dsize
;
8872 unsigned long long sectors
;
8873 unsigned int sector_size
;
8875 get_dev_sector_size(fd
, NULL
, §or_size
);
8876 get_dev_size(fd
, NULL
, &dsize
);
8878 if (mpb_size
> sector_size
) {
8879 /* -1 to account for anchor */
8880 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8882 /* write the extended mpb to the sectors preceeding the anchor */
8883 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8887 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8888 sector_size
* sectors
) != sector_size
* sectors
)
8892 /* first block is stored on second to last sector of the disk */
8893 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8896 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8902 static void imsm_sync_metadata(struct supertype
*container
)
8904 struct intel_super
*super
= container
->sb
;
8906 dprintf("sync metadata: %d\n", super
->updates_pending
);
8907 if (!super
->updates_pending
)
8910 write_super_imsm(container
, 0);
8912 super
->updates_pending
= 0;
8915 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8917 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8918 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8921 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8925 if (dl
&& is_failed(&dl
->disk
))
8929 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8934 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8935 struct active_array
*a
, int activate_new
,
8936 struct mdinfo
*additional_test_list
)
8938 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8939 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8940 struct imsm_super
*mpb
= super
->anchor
;
8941 struct imsm_map
*map
;
8942 unsigned long long pos
;
8947 __u32 array_start
= 0;
8948 __u32 array_end
= 0;
8950 struct mdinfo
*test_list
;
8952 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8953 /* If in this array, skip */
8954 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8955 if (d
->state_fd
>= 0 &&
8956 d
->disk
.major
== dl
->major
&&
8957 d
->disk
.minor
== dl
->minor
) {
8958 dprintf("%x:%x already in array\n",
8959 dl
->major
, dl
->minor
);
8964 test_list
= additional_test_list
;
8966 if (test_list
->disk
.major
== dl
->major
&&
8967 test_list
->disk
.minor
== dl
->minor
) {
8968 dprintf("%x:%x already in additional test list\n",
8969 dl
->major
, dl
->minor
);
8972 test_list
= test_list
->next
;
8977 /* skip in use or failed drives */
8978 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8980 dprintf("%x:%x status (failed: %d index: %d)\n",
8981 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8985 /* skip pure spares when we are looking for partially
8986 * assimilated drives
8988 if (dl
->index
== -1 && !activate_new
)
8991 if (!drive_validate_sector_size(super
, dl
))
8994 /* Does this unused device have the requisite free space?
8995 * It needs to be able to cover all member volumes
8997 ex
= get_extents(super
, dl
, 1);
8999 dprintf("cannot get extents\n");
9002 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9003 dev
= get_imsm_dev(super
, i
);
9004 map
= get_imsm_map(dev
, MAP_0
);
9006 /* check if this disk is already a member of
9009 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
9015 array_start
= pba_of_lba0(map
);
9016 array_end
= array_start
+
9017 per_dev_array_size(map
) - 1;
9020 /* check that we can start at pba_of_lba0 with
9021 * num_data_stripes*blocks_per_stripe of space
9023 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
9027 pos
= ex
[j
].start
+ ex
[j
].size
;
9029 } while (ex
[j
-1].size
);
9036 if (i
< mpb
->num_raid_devs
) {
9037 dprintf("%x:%x does not have %u to %u available\n",
9038 dl
->major
, dl
->minor
, array_start
, array_end
);
9048 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
9050 struct imsm_dev
*dev2
;
9051 struct imsm_map
*map
;
9057 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
9059 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
9060 if (state
== IMSM_T_STATE_FAILED
) {
9061 map
= get_imsm_map(dev2
, MAP_0
);
9064 for (slot
= 0; slot
< map
->num_members
; slot
++) {
9066 * Check if failed disks are deleted from intel
9067 * disk list or are marked to be deleted
9069 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
9070 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
9072 * Do not rebuild the array if failed disks
9073 * from failed sub-array are not removed from
9077 is_failed(&idisk
->disk
) &&
9078 (idisk
->action
!= DISK_REMOVE
))
9086 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
9087 struct metadata_update
**updates
)
9090 * Find a device with unused free space and use it to replace a
9091 * failed/vacant region in an array. We replace failed regions one a
9092 * array at a time. The result is that a new spare disk will be added
9093 * to the first failed array and after the monitor has finished
9094 * propagating failures the remainder will be consumed.
9096 * FIXME add a capability for mdmon to request spares from another
9100 struct intel_super
*super
= a
->container
->sb
;
9101 int inst
= a
->info
.container_member
;
9102 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
9103 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9104 int failed
= a
->info
.array
.raid_disks
;
9105 struct mdinfo
*rv
= NULL
;
9108 struct metadata_update
*mu
;
9110 struct imsm_update_activate_spare
*u
;
9115 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
9116 if ((d
->curr_state
& DS_FAULTY
) &&
9118 /* wait for Removal to happen */
9120 if (d
->state_fd
>= 0)
9124 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
9125 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
9127 if (imsm_reshape_blocks_arrays_changes(super
))
9130 /* Cannot activate another spare if rebuild is in progress already
9132 if (is_rebuilding(dev
)) {
9133 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
9137 if (a
->info
.array
.level
== 4)
9138 /* No repair for takeovered array
9139 * imsm doesn't support raid4
9143 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
9144 IMSM_T_STATE_DEGRADED
)
9147 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
9148 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
9153 * If there are any failed disks check state of the other volume.
9154 * Block rebuild if the another one is failed until failed disks
9155 * are removed from container.
9158 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
9159 MAX_RAID_SERIAL_LEN
, dev
->volume
);
9160 /* check if states of the other volumes allow for rebuild */
9161 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
9163 allowed
= imsm_rebuild_allowed(a
->container
,
9171 /* For each slot, if it is not working, find a spare */
9172 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
9173 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9174 if (d
->disk
.raid_disk
== i
)
9176 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
9177 if (d
&& (d
->state_fd
>= 0))
9181 * OK, this device needs recovery. Try to re-add the
9182 * previous occupant of this slot, if this fails see if
9183 * we can continue the assimilation of a spare that was
9184 * partially assimilated, finally try to activate a new
9187 dl
= imsm_readd(super
, i
, a
);
9189 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9191 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9195 /* found a usable disk with enough space */
9196 di
= xcalloc(1, sizeof(*di
));
9198 /* dl->index will be -1 in the case we are activating a
9199 * pristine spare. imsm_process_update() will create a
9200 * new index in this case. Once a disk is found to be
9201 * failed in all member arrays it is kicked from the
9204 di
->disk
.number
= dl
->index
;
9206 /* (ab)use di->devs to store a pointer to the device
9209 di
->devs
= (struct mdinfo
*) dl
;
9211 di
->disk
.raid_disk
= i
;
9212 di
->disk
.major
= dl
->major
;
9213 di
->disk
.minor
= dl
->minor
;
9215 di
->recovery_start
= 0;
9216 di
->data_offset
= pba_of_lba0(map
);
9217 di
->component_size
= a
->info
.component_size
;
9218 di
->container_member
= inst
;
9219 di
->bb
.supported
= 1;
9220 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9221 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9222 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9224 super
->random
= random32();
9228 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9229 i
, di
->data_offset
);
9233 /* No spares found */
9235 /* Now 'rv' has a list of devices to return.
9236 * Create a metadata_update record to update the
9237 * disk_ord_tbl for the array
9239 mu
= xmalloc(sizeof(*mu
));
9240 mu
->buf
= xcalloc(num_spares
,
9241 sizeof(struct imsm_update_activate_spare
));
9243 mu
->space_list
= NULL
;
9244 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9245 mu
->next
= *updates
;
9246 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9248 for (di
= rv
; di
; di
= di
->next
) {
9249 u
->type
= update_activate_spare
;
9250 u
->dl
= (struct dl
*) di
->devs
;
9252 u
->slot
= di
->disk
.raid_disk
;
9263 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9265 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9266 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9267 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9268 struct disk_info
*inf
= get_disk_info(u
);
9269 struct imsm_disk
*disk
;
9273 for (i
= 0; i
< map
->num_members
; i
++) {
9274 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9275 for (j
= 0; j
< new_map
->num_members
; j
++)
9276 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9283 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9287 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9288 if (dl
->major
== major
&& dl
->minor
== minor
)
9293 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9299 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9300 if (dl
->major
== major
&& dl
->minor
== minor
) {
9303 prev
->next
= dl
->next
;
9305 super
->disks
= dl
->next
;
9307 __free_imsm_disk(dl
);
9308 dprintf("removed %x:%x\n", major
, minor
);
9316 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9318 static int add_remove_disk_update(struct intel_super
*super
)
9320 int check_degraded
= 0;
9323 /* add/remove some spares to/from the metadata/contrainer */
9324 while (super
->disk_mgmt_list
) {
9325 struct dl
*disk_cfg
;
9327 disk_cfg
= super
->disk_mgmt_list
;
9328 super
->disk_mgmt_list
= disk_cfg
->next
;
9329 disk_cfg
->next
= NULL
;
9331 if (disk_cfg
->action
== DISK_ADD
) {
9332 disk_cfg
->next
= super
->disks
;
9333 super
->disks
= disk_cfg
;
9335 dprintf("added %x:%x\n",
9336 disk_cfg
->major
, disk_cfg
->minor
);
9337 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9338 dprintf("Disk remove action processed: %x.%x\n",
9339 disk_cfg
->major
, disk_cfg
->minor
);
9340 disk
= get_disk_super(super
,
9344 /* store action status */
9345 disk
->action
= DISK_REMOVE
;
9346 /* remove spare disks only */
9347 if (disk
->index
== -1) {
9348 remove_disk_super(super
,
9352 disk_cfg
->fd
= disk
->fd
;
9356 /* release allocate disk structure */
9357 __free_imsm_disk(disk_cfg
);
9360 return check_degraded
;
9363 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9364 struct intel_super
*super
,
9367 struct intel_dev
*id
;
9368 void **tofree
= NULL
;
9371 dprintf("(enter)\n");
9372 if (u
->subdev
< 0 || u
->subdev
> 1) {
9373 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9376 if (space_list
== NULL
|| *space_list
== NULL
) {
9377 dprintf("imsm: Error: Memory is not allocated\n");
9381 for (id
= super
->devlist
; id
; id
= id
->next
) {
9382 if (id
->index
== (unsigned)u
->subdev
) {
9383 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9384 struct imsm_map
*map
;
9385 struct imsm_dev
*new_dev
=
9386 (struct imsm_dev
*)*space_list
;
9387 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9389 struct dl
*new_disk
;
9391 if (new_dev
== NULL
)
9393 *space_list
= **space_list
;
9394 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9395 map
= get_imsm_map(new_dev
, MAP_0
);
9397 dprintf("imsm: Error: migration in progress");
9401 to_state
= map
->map_state
;
9402 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9404 /* this should not happen */
9405 if (u
->new_disks
[0] < 0) {
9406 map
->failed_disk_num
=
9407 map
->num_members
- 1;
9408 to_state
= IMSM_T_STATE_DEGRADED
;
9410 to_state
= IMSM_T_STATE_NORMAL
;
9412 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9413 if (u
->new_level
> -1)
9414 map
->raid_level
= u
->new_level
;
9415 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9416 if ((u
->new_level
== 5) &&
9417 (migr_map
->raid_level
== 0)) {
9418 int ord
= map
->num_members
- 1;
9419 migr_map
->num_members
--;
9420 if (u
->new_disks
[0] < 0)
9421 ord
|= IMSM_ORD_REBUILD
;
9422 set_imsm_ord_tbl_ent(map
,
9423 map
->num_members
- 1,
9427 tofree
= (void **)dev
;
9429 /* update chunk size
9431 if (u
->new_chunksize
> 0) {
9432 unsigned long long num_data_stripes
;
9433 struct imsm_map
*dest_map
=
9434 get_imsm_map(dev
, MAP_0
);
9436 imsm_num_data_members(dest_map
);
9438 if (used_disks
== 0)
9441 map
->blocks_per_strip
=
9442 __cpu_to_le16(u
->new_chunksize
* 2);
9444 imsm_dev_size(dev
) / used_disks
;
9445 num_data_stripes
/= map
->blocks_per_strip
;
9446 num_data_stripes
/= map
->num_domains
;
9447 set_num_data_stripes(map
, num_data_stripes
);
9450 /* ensure blocks_per_member has valid value
9452 set_blocks_per_member(map
,
9453 per_dev_array_size(map
) +
9454 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9458 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9459 migr_map
->raid_level
== map
->raid_level
)
9462 if (u
->new_disks
[0] >= 0) {
9465 new_disk
= get_disk_super(super
,
9466 major(u
->new_disks
[0]),
9467 minor(u
->new_disks
[0]));
9468 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9469 major(u
->new_disks
[0]),
9470 minor(u
->new_disks
[0]),
9471 new_disk
, new_disk
->index
);
9472 if (new_disk
== NULL
)
9473 goto error_disk_add
;
9475 new_disk
->index
= map
->num_members
- 1;
9476 /* slot to fill in autolayout
9478 new_disk
->raiddisk
= new_disk
->index
;
9479 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9480 new_disk
->disk
.status
&= ~SPARE_DISK
;
9482 goto error_disk_add
;
9485 *tofree
= *space_list
;
9486 /* calculate new size
9488 imsm_set_array_size(new_dev
, -1);
9495 *space_list
= tofree
;
9499 dprintf("Error: imsm: Cannot find disk.\n");
9503 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9504 struct intel_super
*super
)
9506 struct intel_dev
*id
;
9509 dprintf("(enter)\n");
9510 if (u
->subdev
< 0 || u
->subdev
> 1) {
9511 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9515 for (id
= super
->devlist
; id
; id
= id
->next
) {
9516 if (id
->index
== (unsigned)u
->subdev
) {
9517 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9518 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9519 int used_disks
= imsm_num_data_members(map
);
9520 unsigned long long blocks_per_member
;
9521 unsigned long long num_data_stripes
;
9522 unsigned long long new_size_per_disk
;
9524 if (used_disks
== 0)
9527 /* calculate new size
9529 new_size_per_disk
= u
->new_size
/ used_disks
;
9530 blocks_per_member
= new_size_per_disk
+
9531 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9532 num_data_stripes
= new_size_per_disk
/
9533 map
->blocks_per_strip
;
9534 num_data_stripes
/= map
->num_domains
;
9535 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9536 u
->new_size
, new_size_per_disk
,
9538 set_blocks_per_member(map
, blocks_per_member
);
9539 set_num_data_stripes(map
, num_data_stripes
);
9540 imsm_set_array_size(dev
, u
->new_size
);
9550 static int prepare_spare_to_activate(struct supertype
*st
,
9551 struct imsm_update_activate_spare
*u
)
9553 struct intel_super
*super
= st
->sb
;
9554 int prev_current_vol
= super
->current_vol
;
9555 struct active_array
*a
;
9558 for (a
= st
->arrays
; a
; a
= a
->next
)
9560 * Additional initialization (adding bitmap header, filling
9561 * the bitmap area with '1's to force initial rebuild for a whole
9562 * data-area) is required when adding the spare to the volume
9563 * with write-intent bitmap.
9565 if (a
->info
.container_member
== u
->array
&&
9566 a
->info
.consistency_policy
== CONSISTENCY_POLICY_BITMAP
) {
9569 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9575 super
->current_vol
= u
->array
;
9576 if (st
->ss
->write_bitmap(st
, dl
->fd
, NoUpdate
))
9578 super
->current_vol
= prev_current_vol
;
9583 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9584 struct intel_super
*super
,
9585 struct active_array
*active_array
)
9587 struct imsm_super
*mpb
= super
->anchor
;
9588 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9589 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9590 struct imsm_map
*migr_map
;
9591 struct active_array
*a
;
9592 struct imsm_disk
*disk
;
9599 int second_map_created
= 0;
9601 for (; u
; u
= u
->next
) {
9602 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9607 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9612 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9617 /* count failures (excluding rebuilds and the victim)
9618 * to determine map[0] state
9621 for (i
= 0; i
< map
->num_members
; i
++) {
9624 disk
= get_imsm_disk(super
,
9625 get_imsm_disk_idx(dev
, i
, MAP_X
));
9626 if (!disk
|| is_failed(disk
))
9630 /* adding a pristine spare, assign a new index */
9631 if (dl
->index
< 0) {
9632 dl
->index
= super
->anchor
->num_disks
;
9633 super
->anchor
->num_disks
++;
9636 disk
->status
|= CONFIGURED_DISK
;
9637 disk
->status
&= ~SPARE_DISK
;
9640 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9641 if (!second_map_created
) {
9642 second_map_created
= 1;
9643 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9644 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9646 map
->map_state
= to_state
;
9647 migr_map
= get_imsm_map(dev
, MAP_1
);
9648 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9649 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9650 dl
->index
| IMSM_ORD_REBUILD
);
9652 /* update the family_num to mark a new container
9653 * generation, being careful to record the existing
9654 * family_num in orig_family_num to clean up after
9655 * earlier mdadm versions that neglected to set it.
9657 if (mpb
->orig_family_num
== 0)
9658 mpb
->orig_family_num
= mpb
->family_num
;
9659 mpb
->family_num
+= super
->random
;
9661 /* count arrays using the victim in the metadata */
9663 for (a
= active_array
; a
; a
= a
->next
) {
9664 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9665 map
= get_imsm_map(dev
, MAP_0
);
9667 if (get_imsm_disk_slot(map
, victim
) >= 0)
9671 /* delete the victim if it is no longer being
9677 /* We know that 'manager' isn't touching anything,
9678 * so it is safe to delete
9680 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9681 if ((*dlp
)->index
== victim
)
9684 /* victim may be on the missing list */
9686 for (dlp
= &super
->missing
; *dlp
;
9687 dlp
= &(*dlp
)->next
)
9688 if ((*dlp
)->index
== victim
)
9690 imsm_delete(super
, dlp
, victim
);
9697 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9698 struct intel_super
*super
,
9701 struct dl
*new_disk
;
9702 struct intel_dev
*id
;
9704 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9705 int disk_count
= u
->old_raid_disks
;
9706 void **tofree
= NULL
;
9707 int devices_to_reshape
= 1;
9708 struct imsm_super
*mpb
= super
->anchor
;
9710 unsigned int dev_id
;
9712 dprintf("(enter)\n");
9714 /* enable spares to use in array */
9715 for (i
= 0; i
< delta_disks
; i
++) {
9716 new_disk
= get_disk_super(super
,
9717 major(u
->new_disks
[i
]),
9718 minor(u
->new_disks
[i
]));
9719 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9720 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9721 new_disk
, new_disk
->index
);
9722 if (new_disk
== NULL
||
9723 (new_disk
->index
>= 0 &&
9724 new_disk
->index
< u
->old_raid_disks
))
9725 goto update_reshape_exit
;
9726 new_disk
->index
= disk_count
++;
9727 /* slot to fill in autolayout
9729 new_disk
->raiddisk
= new_disk
->index
;
9730 new_disk
->disk
.status
|=
9732 new_disk
->disk
.status
&= ~SPARE_DISK
;
9735 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9736 mpb
->num_raid_devs
);
9737 /* manage changes in volume
9739 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9740 void **sp
= *space_list
;
9741 struct imsm_dev
*newdev
;
9742 struct imsm_map
*newmap
, *oldmap
;
9744 for (id
= super
->devlist
; id
; id
= id
->next
) {
9745 if (id
->index
== dev_id
)
9754 /* Copy the dev, but not (all of) the map */
9755 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9756 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9757 newmap
= get_imsm_map(newdev
, MAP_0
);
9758 /* Copy the current map */
9759 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9760 /* update one device only
9762 if (devices_to_reshape
) {
9763 dprintf("imsm: modifying subdev: %i\n",
9765 devices_to_reshape
--;
9766 newdev
->vol
.migr_state
= 1;
9767 set_vol_curr_migr_unit(newdev
, 0);
9768 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9769 newmap
->num_members
= u
->new_raid_disks
;
9770 for (i
= 0; i
< delta_disks
; i
++) {
9771 set_imsm_ord_tbl_ent(newmap
,
9772 u
->old_raid_disks
+ i
,
9773 u
->old_raid_disks
+ i
);
9775 /* New map is correct, now need to save old map
9777 newmap
= get_imsm_map(newdev
, MAP_1
);
9778 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9780 imsm_set_array_size(newdev
, -1);
9783 sp
= (void **)id
->dev
;
9788 /* Clear migration record */
9789 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9792 *space_list
= tofree
;
9795 update_reshape_exit
:
9800 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9801 struct intel_super
*super
,
9804 struct imsm_dev
*dev
= NULL
;
9805 struct intel_dev
*dv
;
9806 struct imsm_dev
*dev_new
;
9807 struct imsm_map
*map
;
9811 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9812 if (dv
->index
== (unsigned int)u
->subarray
) {
9820 map
= get_imsm_map(dev
, MAP_0
);
9822 if (u
->direction
== R10_TO_R0
) {
9823 unsigned long long num_data_stripes
;
9825 /* Number of failed disks must be half of initial disk number */
9826 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9827 (map
->num_members
/ 2))
9830 /* iterate through devices to mark removed disks as spare */
9831 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9832 if (dm
->disk
.status
& FAILED_DISK
) {
9833 int idx
= dm
->index
;
9834 /* update indexes on the disk list */
9835 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9836 the index values will end up being correct.... NB */
9837 for (du
= super
->disks
; du
; du
= du
->next
)
9838 if (du
->index
> idx
)
9840 /* mark as spare disk */
9845 map
->num_members
= map
->num_members
/ 2;
9846 map
->map_state
= IMSM_T_STATE_NORMAL
;
9847 map
->num_domains
= 1;
9848 map
->raid_level
= 0;
9849 map
->failed_disk_num
= -1;
9850 num_data_stripes
= imsm_dev_size(dev
) / 2;
9851 num_data_stripes
/= map
->blocks_per_strip
;
9852 set_num_data_stripes(map
, num_data_stripes
);
9855 if (u
->direction
== R0_TO_R10
) {
9857 unsigned long long num_data_stripes
;
9859 /* update slots in current disk list */
9860 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9864 /* create new *missing* disks */
9865 for (i
= 0; i
< map
->num_members
; i
++) {
9866 space
= *space_list
;
9869 *space_list
= *space
;
9871 memcpy(du
, super
->disks
, sizeof(*du
));
9875 du
->index
= (i
* 2) + 1;
9876 sprintf((char *)du
->disk
.serial
,
9877 " MISSING_%d", du
->index
);
9878 sprintf((char *)du
->serial
,
9879 "MISSING_%d", du
->index
);
9880 du
->next
= super
->missing
;
9881 super
->missing
= du
;
9883 /* create new dev and map */
9884 space
= *space_list
;
9887 *space_list
= *space
;
9888 dev_new
= (void *)space
;
9889 memcpy(dev_new
, dev
, sizeof(*dev
));
9890 /* update new map */
9891 map
= get_imsm_map(dev_new
, MAP_0
);
9892 map
->num_members
= map
->num_members
* 2;
9893 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9894 map
->num_domains
= 2;
9895 map
->raid_level
= 1;
9896 num_data_stripes
= imsm_dev_size(dev
) / 2;
9897 num_data_stripes
/= map
->blocks_per_strip
;
9898 num_data_stripes
/= map
->num_domains
;
9899 set_num_data_stripes(map
, num_data_stripes
);
9901 /* replace dev<->dev_new */
9904 /* update disk order table */
9905 for (du
= super
->disks
; du
; du
= du
->next
)
9907 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9908 for (du
= super
->missing
; du
; du
= du
->next
)
9909 if (du
->index
>= 0) {
9910 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9911 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9917 static void imsm_process_update(struct supertype
*st
,
9918 struct metadata_update
*update
)
9921 * crack open the metadata_update envelope to find the update record
9922 * update can be one of:
9923 * update_reshape_container_disks - all the arrays in the container
9924 * are being reshaped to have more devices. We need to mark
9925 * the arrays for general migration and convert selected spares
9926 * into active devices.
9927 * update_activate_spare - a spare device has replaced a failed
9928 * device in an array, update the disk_ord_tbl. If this disk is
9929 * present in all member arrays then also clear the SPARE_DISK
9931 * update_create_array
9933 * update_rename_array
9934 * update_add_remove_disk
9936 struct intel_super
*super
= st
->sb
;
9937 struct imsm_super
*mpb
;
9938 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9940 /* update requires a larger buf but the allocation failed */
9941 if (super
->next_len
&& !super
->next_buf
) {
9942 super
->next_len
= 0;
9946 if (super
->next_buf
) {
9947 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9949 super
->len
= super
->next_len
;
9950 super
->buf
= super
->next_buf
;
9952 super
->next_len
= 0;
9953 super
->next_buf
= NULL
;
9956 mpb
= super
->anchor
;
9959 case update_general_migration_checkpoint
: {
9960 struct intel_dev
*id
;
9961 struct imsm_update_general_migration_checkpoint
*u
=
9962 (void *)update
->buf
;
9964 dprintf("called for update_general_migration_checkpoint\n");
9966 /* find device under general migration */
9967 for (id
= super
->devlist
; id
; id
= id
->next
) {
9968 if (is_gen_migration(id
->dev
)) {
9969 set_vol_curr_migr_unit(id
->dev
,
9971 super
->updates_pending
++;
9976 case update_takeover
: {
9977 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9978 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9979 imsm_update_version_info(super
);
9980 super
->updates_pending
++;
9985 case update_reshape_container_disks
: {
9986 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9987 if (apply_reshape_container_disks_update(
9988 u
, super
, &update
->space_list
))
9989 super
->updates_pending
++;
9992 case update_reshape_migration
: {
9993 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9994 if (apply_reshape_migration_update(
9995 u
, super
, &update
->space_list
))
9996 super
->updates_pending
++;
9999 case update_size_change
: {
10000 struct imsm_update_size_change
*u
= (void *)update
->buf
;
10001 if (apply_size_change_update(u
, super
))
10002 super
->updates_pending
++;
10005 case update_activate_spare
: {
10006 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
10008 if (prepare_spare_to_activate(st
, u
) &&
10009 apply_update_activate_spare(u
, super
, st
->arrays
))
10010 super
->updates_pending
++;
10013 case update_create_array
: {
10014 /* someone wants to create a new array, we need to be aware of
10015 * a few races/collisions:
10016 * 1/ 'Create' called by two separate instances of mdadm
10017 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
10018 * devices that have since been assimilated via
10020 * In the event this update can not be carried out mdadm will
10021 * (FIX ME) notice that its update did not take hold.
10023 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10024 struct intel_dev
*dv
;
10025 struct imsm_dev
*dev
;
10026 struct imsm_map
*map
, *new_map
;
10027 unsigned long long start
, end
;
10028 unsigned long long new_start
, new_end
;
10030 struct disk_info
*inf
;
10033 /* handle racing creates: first come first serve */
10034 if (u
->dev_idx
< mpb
->num_raid_devs
) {
10035 dprintf("subarray %d already defined\n", u
->dev_idx
);
10039 /* check update is next in sequence */
10040 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
10041 dprintf("can not create array %d expected index %d\n",
10042 u
->dev_idx
, mpb
->num_raid_devs
);
10046 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
10047 new_start
= pba_of_lba0(new_map
);
10048 new_end
= new_start
+ per_dev_array_size(new_map
);
10049 inf
= get_disk_info(u
);
10051 /* handle activate_spare versus create race:
10052 * check to make sure that overlapping arrays do not include
10053 * overalpping disks
10055 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10056 dev
= get_imsm_dev(super
, i
);
10057 map
= get_imsm_map(dev
, MAP_0
);
10058 start
= pba_of_lba0(map
);
10059 end
= start
+ per_dev_array_size(map
);
10060 if ((new_start
>= start
&& new_start
<= end
) ||
10061 (start
>= new_start
&& start
<= new_end
))
10066 if (disks_overlap(super
, i
, u
)) {
10067 dprintf("arrays overlap\n");
10072 /* check that prepare update was successful */
10073 if (!update
->space
) {
10074 dprintf("prepare update failed\n");
10078 /* check that all disks are still active before committing
10079 * changes. FIXME: could we instead handle this by creating a
10080 * degraded array? That's probably not what the user expects,
10081 * so better to drop this update on the floor.
10083 for (i
= 0; i
< new_map
->num_members
; i
++) {
10084 dl
= serial_to_dl(inf
[i
].serial
, super
);
10086 dprintf("disk disappeared\n");
10091 super
->updates_pending
++;
10093 /* convert spares to members and fixup ord_tbl */
10094 for (i
= 0; i
< new_map
->num_members
; i
++) {
10095 dl
= serial_to_dl(inf
[i
].serial
, super
);
10096 if (dl
->index
== -1) {
10097 dl
->index
= mpb
->num_disks
;
10099 dl
->disk
.status
|= CONFIGURED_DISK
;
10100 dl
->disk
.status
&= ~SPARE_DISK
;
10102 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
10105 dv
= update
->space
;
10107 update
->space
= NULL
;
10108 imsm_copy_dev(dev
, &u
->dev
);
10109 dv
->index
= u
->dev_idx
;
10110 dv
->next
= super
->devlist
;
10111 super
->devlist
= dv
;
10112 mpb
->num_raid_devs
++;
10114 imsm_update_version_info(super
);
10117 /* mdmon knows how to release update->space, but not
10118 * ((struct intel_dev *) update->space)->dev
10120 if (update
->space
) {
10121 dv
= update
->space
;
10126 case update_kill_array
: {
10127 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
10128 int victim
= u
->dev_idx
;
10129 struct active_array
*a
;
10130 struct intel_dev
**dp
;
10131 struct imsm_dev
*dev
;
10133 /* sanity check that we are not affecting the uuid of
10134 * active arrays, or deleting an active array
10136 * FIXME when immutable ids are available, but note that
10137 * we'll also need to fixup the invalidated/active
10138 * subarray indexes in mdstat
10140 for (a
= st
->arrays
; a
; a
= a
->next
)
10141 if (a
->info
.container_member
>= victim
)
10143 /* by definition if mdmon is running at least one array
10144 * is active in the container, so checking
10145 * mpb->num_raid_devs is just extra paranoia
10147 dev
= get_imsm_dev(super
, victim
);
10148 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
10149 dprintf("failed to delete subarray-%d\n", victim
);
10153 for (dp
= &super
->devlist
; *dp
;)
10154 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
10157 if ((*dp
)->index
> (unsigned)victim
)
10161 mpb
->num_raid_devs
--;
10162 super
->updates_pending
++;
10165 case update_rename_array
: {
10166 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
10167 char name
[MAX_RAID_SERIAL_LEN
+1];
10168 int target
= u
->dev_idx
;
10169 struct active_array
*a
;
10170 struct imsm_dev
*dev
;
10172 /* sanity check that we are not affecting the uuid of
10175 memset(name
, 0, sizeof(name
));
10176 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
10177 name
[MAX_RAID_SERIAL_LEN
] = '\0';
10178 for (a
= st
->arrays
; a
; a
= a
->next
)
10179 if (a
->info
.container_member
== target
)
10181 dev
= get_imsm_dev(super
, u
->dev_idx
);
10182 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
10183 dprintf("failed to rename subarray-%d\n", target
);
10187 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
10188 super
->updates_pending
++;
10191 case update_add_remove_disk
: {
10192 /* we may be able to repair some arrays if disks are
10193 * being added, check the status of add_remove_disk
10194 * if discs has been added.
10196 if (add_remove_disk_update(super
)) {
10197 struct active_array
*a
;
10199 super
->updates_pending
++;
10200 for (a
= st
->arrays
; a
; a
= a
->next
)
10201 a
->check_degraded
= 1;
10205 case update_prealloc_badblocks_mem
:
10207 case update_rwh_policy
: {
10208 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
10209 int target
= u
->dev_idx
;
10210 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
10212 dprintf("could not find subarray-%d\n", target
);
10216 if (dev
->rwh_policy
!= u
->new_policy
) {
10217 dev
->rwh_policy
= u
->new_policy
;
10218 super
->updates_pending
++;
10223 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10227 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10229 static int imsm_prepare_update(struct supertype
*st
,
10230 struct metadata_update
*update
)
10233 * Allocate space to hold new disk entries, raid-device entries or a new
10234 * mpb if necessary. The manager synchronously waits for updates to
10235 * complete in the monitor, so new mpb buffers allocated here can be
10236 * integrated by the monitor thread without worrying about live pointers
10237 * in the manager thread.
10239 enum imsm_update_type type
;
10240 struct intel_super
*super
= st
->sb
;
10241 unsigned int sector_size
= super
->sector_size
;
10242 struct imsm_super
*mpb
= super
->anchor
;
10246 if (update
->len
< (int)sizeof(type
))
10249 type
= *(enum imsm_update_type
*) update
->buf
;
10252 case update_general_migration_checkpoint
:
10253 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10255 dprintf("called for update_general_migration_checkpoint\n");
10257 case update_takeover
: {
10258 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10259 if (update
->len
< (int)sizeof(*u
))
10261 if (u
->direction
== R0_TO_R10
) {
10262 void **tail
= (void **)&update
->space_list
;
10263 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10264 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10265 int num_members
= map
->num_members
;
10268 /* allocate memory for added disks */
10269 for (i
= 0; i
< num_members
; i
++) {
10270 size
= sizeof(struct dl
);
10271 space
= xmalloc(size
);
10276 /* allocate memory for new device */
10277 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10278 (num_members
* sizeof(__u32
));
10279 space
= xmalloc(size
);
10283 len
= disks_to_mpb_size(num_members
* 2);
10288 case update_reshape_container_disks
: {
10289 /* Every raid device in the container is about to
10290 * gain some more devices, and we will enter a
10292 * So each 'imsm_map' will be bigger, and the imsm_vol
10293 * will now hold 2 of them.
10294 * Thus we need new 'struct imsm_dev' allocations sized
10295 * as sizeof_imsm_dev but with more devices in both maps.
10297 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10298 struct intel_dev
*dl
;
10299 void **space_tail
= (void**)&update
->space_list
;
10301 if (update
->len
< (int)sizeof(*u
))
10304 dprintf("for update_reshape\n");
10306 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10307 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10309 if (u
->new_raid_disks
> u
->old_raid_disks
)
10310 size
+= sizeof(__u32
)*2*
10311 (u
->new_raid_disks
- u
->old_raid_disks
);
10315 *space_tail
= NULL
;
10318 len
= disks_to_mpb_size(u
->new_raid_disks
);
10319 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10322 case update_reshape_migration
: {
10323 /* for migration level 0->5 we need to add disks
10324 * so the same as for container operation we will copy
10325 * device to the bigger location.
10326 * in memory prepared device and new disk area are prepared
10327 * for usage in process update
10329 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10330 struct intel_dev
*id
;
10331 void **space_tail
= (void **)&update
->space_list
;
10334 int current_level
= -1;
10336 if (update
->len
< (int)sizeof(*u
))
10339 dprintf("for update_reshape\n");
10341 /* add space for bigger array in update
10343 for (id
= super
->devlist
; id
; id
= id
->next
) {
10344 if (id
->index
== (unsigned)u
->subdev
) {
10345 size
= sizeof_imsm_dev(id
->dev
, 1);
10346 if (u
->new_raid_disks
> u
->old_raid_disks
)
10347 size
+= sizeof(__u32
)*2*
10348 (u
->new_raid_disks
- u
->old_raid_disks
);
10352 *space_tail
= NULL
;
10356 if (update
->space_list
== NULL
)
10359 /* add space for disk in update
10361 size
= sizeof(struct dl
);
10365 *space_tail
= NULL
;
10367 /* add spare device to update
10369 for (id
= super
->devlist
; id
; id
= id
->next
)
10370 if (id
->index
== (unsigned)u
->subdev
) {
10371 struct imsm_dev
*dev
;
10372 struct imsm_map
*map
;
10374 dev
= get_imsm_dev(super
, u
->subdev
);
10375 map
= get_imsm_map(dev
, MAP_0
);
10376 current_level
= map
->raid_level
;
10379 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10380 struct mdinfo
*spares
;
10382 spares
= get_spares_for_grow(st
);
10385 struct mdinfo
*dev
;
10387 dev
= spares
->devs
;
10390 makedev(dev
->disk
.major
,
10392 dl
= get_disk_super(super
,
10395 dl
->index
= u
->old_raid_disks
;
10398 sysfs_free(spares
);
10401 len
= disks_to_mpb_size(u
->new_raid_disks
);
10402 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10405 case update_size_change
: {
10406 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10410 case update_activate_spare
: {
10411 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10415 case update_create_array
: {
10416 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10417 struct intel_dev
*dv
;
10418 struct imsm_dev
*dev
= &u
->dev
;
10419 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10421 struct disk_info
*inf
;
10425 if (update
->len
< (int)sizeof(*u
))
10428 inf
= get_disk_info(u
);
10429 len
= sizeof_imsm_dev(dev
, 1);
10430 /* allocate a new super->devlist entry */
10431 dv
= xmalloc(sizeof(*dv
));
10432 dv
->dev
= xmalloc(len
);
10433 update
->space
= dv
;
10435 /* count how many spares will be converted to members */
10436 for (i
= 0; i
< map
->num_members
; i
++) {
10437 dl
= serial_to_dl(inf
[i
].serial
, super
);
10439 /* hmm maybe it failed?, nothing we can do about
10444 if (count_memberships(dl
, super
) == 0)
10447 len
+= activate
* sizeof(struct imsm_disk
);
10450 case update_kill_array
: {
10451 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10455 case update_rename_array
: {
10456 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10460 case update_add_remove_disk
:
10461 /* no update->len needed */
10463 case update_prealloc_badblocks_mem
:
10464 super
->extra_space
+= sizeof(struct bbm_log
) -
10465 get_imsm_bbm_log_size(super
->bbm_log
);
10467 case update_rwh_policy
: {
10468 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10476 /* check if we need a larger metadata buffer */
10477 if (super
->next_buf
)
10478 buf_len
= super
->next_len
;
10480 buf_len
= super
->len
;
10482 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10483 /* ok we need a larger buf than what is currently allocated
10484 * if this allocation fails process_update will notice that
10485 * ->next_len is set and ->next_buf is NULL
10487 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10488 super
->extra_space
+ len
, sector_size
);
10489 if (super
->next_buf
)
10490 free(super
->next_buf
);
10492 super
->next_len
= buf_len
;
10493 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10494 memset(super
->next_buf
, 0, buf_len
);
10496 super
->next_buf
= NULL
;
10501 /* must be called while manager is quiesced */
10502 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10504 struct imsm_super
*mpb
= super
->anchor
;
10506 struct imsm_dev
*dev
;
10507 struct imsm_map
*map
;
10508 unsigned int i
, j
, num_members
;
10509 __u32 ord
, ord_map0
;
10510 struct bbm_log
*log
= super
->bbm_log
;
10512 dprintf("deleting device[%d] from imsm_super\n", index
);
10514 /* shift all indexes down one */
10515 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10516 if (iter
->index
> (int)index
)
10518 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10519 if (iter
->index
> (int)index
)
10522 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10523 dev
= get_imsm_dev(super
, i
);
10524 map
= get_imsm_map(dev
, MAP_0
);
10525 num_members
= map
->num_members
;
10526 for (j
= 0; j
< num_members
; j
++) {
10527 /* update ord entries being careful not to propagate
10528 * ord-flags to the first map
10530 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10531 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10533 if (ord_to_idx(ord
) <= index
)
10536 map
= get_imsm_map(dev
, MAP_0
);
10537 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10538 map
= get_imsm_map(dev
, MAP_1
);
10540 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10544 for (i
= 0; i
< log
->entry_count
; i
++) {
10545 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10547 if (entry
->disk_ordinal
<= index
)
10549 entry
->disk_ordinal
--;
10553 super
->updates_pending
++;
10555 struct dl
*dl
= *dlp
;
10557 *dlp
= (*dlp
)->next
;
10558 __free_imsm_disk(dl
);
10562 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10563 struct intel_super
*super
,
10564 struct imsm_dev
*dev
)
10570 struct imsm_map
*map
;
10573 ret_val
= raid_disks
/2;
10574 /* check map if all disks pairs not failed
10577 map
= get_imsm_map(dev
, MAP_0
);
10578 for (i
= 0; i
< ret_val
; i
++) {
10579 int degradation
= 0;
10580 if (get_imsm_disk(super
, i
) == NULL
)
10582 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10584 if (degradation
== 2)
10587 map
= get_imsm_map(dev
, MAP_1
);
10588 /* if there is no second map
10589 * result can be returned
10593 /* check degradation in second map
10595 for (i
= 0; i
< ret_val
; i
++) {
10596 int degradation
= 0;
10597 if (get_imsm_disk(super
, i
) == NULL
)
10599 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10601 if (degradation
== 2)
10615 /*******************************************************************************
10616 * Function: validate_container_imsm
10617 * Description: This routine validates container after assemble,
10618 * eg. if devices in container are under the same controller.
10621 * info : linked list with info about devices used in array
10625 ******************************************************************************/
10626 int validate_container_imsm(struct mdinfo
*info
)
10628 if (check_env("IMSM_NO_PLATFORM"))
10631 struct sys_dev
*idev
;
10632 struct sys_dev
*hba
= NULL
;
10633 struct sys_dev
*intel_devices
= find_intel_devices();
10634 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10635 info
->disk
.minor
));
10637 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10638 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10647 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10648 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10652 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10653 struct mdinfo
*dev
;
10655 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10656 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10658 struct sys_dev
*hba2
= NULL
;
10659 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10660 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10668 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10669 get_orom_by_device_id(hba2
->dev_id
);
10671 if (hba2
&& hba
->type
!= hba2
->type
) {
10672 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10673 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10677 if (orom
!= orom2
) {
10678 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10679 " This operation is not supported and can lead to data loss.\n");
10684 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10685 " This operation is not supported and can lead to data loss.\n");
10693 /*******************************************************************************
10694 * Function: imsm_record_badblock
10695 * Description: This routine stores new bad block record in BBM log
10698 * a : array containing a bad block
10699 * slot : disk number containing a bad block
10700 * sector : bad block sector
10701 * length : bad block sectors range
10705 ******************************************************************************/
10706 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10707 unsigned long long sector
, int length
)
10709 struct intel_super
*super
= a
->container
->sb
;
10713 ord
= imsm_disk_slot_to_ord(a
, slot
);
10717 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10720 super
->updates_pending
++;
10724 /*******************************************************************************
10725 * Function: imsm_clear_badblock
10726 * Description: This routine clears bad block record from BBM log
10729 * a : array containing a bad block
10730 * slot : disk number containing a bad block
10731 * sector : bad block sector
10732 * length : bad block sectors range
10736 ******************************************************************************/
10737 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10738 unsigned long long sector
, int length
)
10740 struct intel_super
*super
= a
->container
->sb
;
10744 ord
= imsm_disk_slot_to_ord(a
, slot
);
10748 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10750 super
->updates_pending
++;
10754 /*******************************************************************************
10755 * Function: imsm_get_badblocks
10756 * Description: This routine get list of bad blocks for an array
10760 * slot : disk number
10762 * bb : structure containing bad blocks
10764 ******************************************************************************/
10765 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10767 int inst
= a
->info
.container_member
;
10768 struct intel_super
*super
= a
->container
->sb
;
10769 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10770 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10773 ord
= imsm_disk_slot_to_ord(a
, slot
);
10777 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10778 per_dev_array_size(map
), &super
->bb
);
10782 /*******************************************************************************
10783 * Function: examine_badblocks_imsm
10784 * Description: Prints list of bad blocks on a disk to the standard output
10787 * st : metadata handler
10788 * fd : open file descriptor for device
10789 * devname : device name
10793 ******************************************************************************/
10794 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10796 struct intel_super
*super
= st
->sb
;
10797 struct bbm_log
*log
= super
->bbm_log
;
10798 struct dl
*d
= NULL
;
10801 for (d
= super
->disks
; d
; d
= d
->next
) {
10802 if (strcmp(d
->devname
, devname
) == 0)
10806 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10807 pr_err("%s doesn't appear to be part of a raid array\n",
10814 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10816 for (i
= 0; i
< log
->entry_count
; i
++) {
10817 if (entry
[i
].disk_ordinal
== d
->index
) {
10818 unsigned long long sector
= __le48_to_cpu(
10819 &entry
[i
].defective_block_start
);
10820 int cnt
= entry
[i
].marked_count
+ 1;
10823 printf("Bad-blocks on %s:\n", devname
);
10827 printf("%20llu for %d sectors\n", sector
, cnt
);
10833 printf("No bad-blocks list configured on %s\n", devname
);
10837 /*******************************************************************************
10838 * Function: init_migr_record_imsm
10839 * Description: Function inits imsm migration record
10841 * super : imsm internal array info
10842 * dev : device under migration
10843 * info : general array info to find the smallest device
10846 ******************************************************************************/
10847 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10848 struct mdinfo
*info
)
10850 struct intel_super
*super
= st
->sb
;
10851 struct migr_record
*migr_rec
= super
->migr_rec
;
10852 int new_data_disks
;
10853 unsigned long long dsize
, dev_sectors
;
10854 long long unsigned min_dev_sectors
= -1LLU;
10855 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10856 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10857 unsigned long long num_migr_units
;
10858 unsigned long long array_blocks
;
10859 struct dl
*dl_disk
= NULL
;
10861 memset(migr_rec
, 0, sizeof(struct migr_record
));
10862 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10864 /* only ascending reshape supported now */
10865 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10867 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10868 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10869 migr_rec
->dest_depth_per_unit
*=
10870 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10871 new_data_disks
= imsm_num_data_members(map_dest
);
10872 migr_rec
->blocks_per_unit
=
10873 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10874 migr_rec
->dest_depth_per_unit
=
10875 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10876 array_blocks
= info
->component_size
* new_data_disks
;
10878 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10880 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10882 set_num_migr_units(migr_rec
, num_migr_units
);
10884 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10885 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10887 /* Find the smallest dev */
10888 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
10889 /* ignore spares in container */
10890 if (dl_disk
->index
< 0)
10892 get_dev_size(dl_disk
->fd
, NULL
, &dsize
);
10893 dev_sectors
= dsize
/ 512;
10894 if (dev_sectors
< min_dev_sectors
)
10895 min_dev_sectors
= dev_sectors
;
10897 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
10898 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10900 write_imsm_migr_rec(st
);
10905 /*******************************************************************************
10906 * Function: save_backup_imsm
10907 * Description: Function saves critical data stripes to Migration Copy Area
10908 * and updates the current migration unit status.
10909 * Use restore_stripes() to form a destination stripe,
10910 * and to write it to the Copy Area.
10912 * st : supertype information
10913 * dev : imsm device that backup is saved for
10914 * info : general array info
10915 * buf : input buffer
10916 * length : length of data to backup (blocks_per_unit)
10920 ******************************************************************************/
10921 int save_backup_imsm(struct supertype
*st
,
10922 struct imsm_dev
*dev
,
10923 struct mdinfo
*info
,
10928 struct intel_super
*super
= st
->sb
;
10929 unsigned long long *target_offsets
;
10932 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10933 int new_disks
= map_dest
->num_members
;
10934 int dest_layout
= 0;
10936 unsigned long long start
;
10937 int data_disks
= imsm_num_data_members(map_dest
);
10939 targets
= xmalloc(new_disks
* sizeof(int));
10941 for (i
= 0; i
< new_disks
; i
++) {
10942 struct dl
*dl_disk
= get_imsm_dl_disk(super
, i
);
10944 targets
[i
] = dl_disk
->fd
;
10947 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10949 start
= info
->reshape_progress
* 512;
10950 for (i
= 0; i
< new_disks
; i
++) {
10951 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
10952 /* move back copy area adderss, it will be moved forward
10953 * in restore_stripes() using start input variable
10955 target_offsets
[i
] -= start
/data_disks
;
10958 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10959 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10961 if (restore_stripes(targets
, /* list of dest devices */
10962 target_offsets
, /* migration record offsets */
10965 map_dest
->raid_level
,
10967 -1, /* source backup file descriptor */
10968 0, /* input buf offset
10969 * always 0 buf is already offseted */
10973 pr_err("Error restoring stripes\n");
10983 free(target_offsets
);
10988 /*******************************************************************************
10989 * Function: save_checkpoint_imsm
10990 * Description: Function called for current unit status update
10991 * in the migration record. It writes it to disk.
10993 * super : imsm internal array info
10994 * info : general array info
10998 * 2: failure, means no valid migration record
10999 * / no general migration in progress /
11000 ******************************************************************************/
11001 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
11003 struct intel_super
*super
= st
->sb
;
11004 unsigned long long blocks_per_unit
;
11005 unsigned long long curr_migr_unit
;
11007 if (load_imsm_migr_rec(super
) != 0) {
11008 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
11012 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
11013 if (blocks_per_unit
== 0) {
11014 dprintf("imsm: no migration in progress.\n");
11017 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
11018 /* check if array is alligned to copy area
11019 * if it is not alligned, add one to current migration unit value
11020 * this can happend on array reshape finish only
11022 if (info
->reshape_progress
% blocks_per_unit
)
11025 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
11026 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
11027 set_migr_dest_1st_member_lba(super
->migr_rec
,
11028 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
11030 if (write_imsm_migr_rec(st
) < 0) {
11031 dprintf("imsm: Cannot write migration record outside backup area\n");
11038 /*******************************************************************************
11039 * Function: recover_backup_imsm
11040 * Description: Function recovers critical data from the Migration Copy Area
11041 * while assembling an array.
11043 * super : imsm internal array info
11044 * info : general array info
11046 * 0 : success (or there is no data to recover)
11048 ******************************************************************************/
11049 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
11051 struct intel_super
*super
= st
->sb
;
11052 struct migr_record
*migr_rec
= super
->migr_rec
;
11053 struct imsm_map
*map_dest
;
11054 struct intel_dev
*id
= NULL
;
11055 unsigned long long read_offset
;
11056 unsigned long long write_offset
;
11058 int new_disks
, err
;
11061 unsigned int sector_size
= super
->sector_size
;
11062 unsigned long long curr_migr_unit
= current_migr_unit(migr_rec
);
11063 unsigned long long num_migr_units
= get_num_migr_units(migr_rec
);
11065 int skipped_disks
= 0;
11066 struct dl
*dl_disk
;
11068 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
11072 /* recover data only during assemblation */
11073 if (strncmp(buffer
, "inactive", 8) != 0)
11075 /* no data to recover */
11076 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
11078 if (curr_migr_unit
>= num_migr_units
)
11081 /* find device during reshape */
11082 for (id
= super
->devlist
; id
; id
= id
->next
)
11083 if (is_gen_migration(id
->dev
))
11088 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
11089 new_disks
= map_dest
->num_members
;
11091 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
11093 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
11094 pba_of_lba0(map_dest
)) * 512;
11096 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11097 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
11100 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
11101 if (dl_disk
->index
< 0)
11104 if (dl_disk
->fd
< 0) {
11108 if (lseek64(dl_disk
->fd
, read_offset
, SEEK_SET
) < 0) {
11109 pr_err("Cannot seek to block: %s\n",
11114 if (read(dl_disk
->fd
, buf
, unit_len
) != unit_len
) {
11115 pr_err("Cannot read copy area block: %s\n",
11120 if (lseek64(dl_disk
->fd
, write_offset
, SEEK_SET
) < 0) {
11121 pr_err("Cannot seek to block: %s\n",
11126 if (write(dl_disk
->fd
, buf
, unit_len
) != unit_len
) {
11127 pr_err("Cannot restore block: %s\n",
11134 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11138 pr_err("Cannot restore data from backup. Too many failed disks\n");
11142 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11143 /* ignore error == 2, this can mean end of reshape here
11145 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11154 static char disk_by_path
[] = "/dev/disk/by-path/";
11156 static const char *imsm_get_disk_controller_domain(const char *path
)
11158 char disk_path
[PATH_MAX
];
11162 strcpy(disk_path
, disk_by_path
);
11163 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
11164 if (stat(disk_path
, &st
) == 0) {
11165 struct sys_dev
* hba
;
11168 path
= devt_to_devpath(st
.st_rdev
);
11171 hba
= find_disk_attached_hba(-1, path
);
11172 if (hba
&& hba
->type
== SYS_DEV_SAS
)
11174 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
11176 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
11178 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
11182 dprintf("path: %s hba: %s attached: %s\n",
11183 path
, (hba
) ? hba
->path
: "NULL", drv
);
11189 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11191 static char devnm
[32];
11192 char subdev_name
[20];
11193 struct mdstat_ent
*mdstat
;
11195 sprintf(subdev_name
, "%d", subdev
);
11196 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11200 strcpy(devnm
, mdstat
->devnm
);
11201 free_mdstat(mdstat
);
11205 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11206 struct geo_params
*geo
,
11207 int *old_raid_disks
,
11210 /* currently we only support increasing the number of devices
11211 * for a container. This increases the number of device for each
11212 * member array. They must all be RAID0 or RAID5.
11215 struct mdinfo
*info
, *member
;
11216 int devices_that_can_grow
= 0;
11218 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11220 if (geo
->size
> 0 ||
11221 geo
->level
!= UnSet
||
11222 geo
->layout
!= UnSet
||
11223 geo
->chunksize
!= 0 ||
11224 geo
->raid_disks
== UnSet
) {
11225 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11229 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11230 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11234 info
= container_content_imsm(st
, NULL
);
11235 for (member
= info
; member
; member
= member
->next
) {
11238 dprintf("imsm: checking device_num: %i\n",
11239 member
->container_member
);
11241 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11242 /* we work on container for Online Capacity Expansion
11243 * only so raid_disks has to grow
11245 dprintf("imsm: for container operation raid disks increase is required\n");
11249 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11250 /* we cannot use this container with other raid level
11252 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11253 info
->array
.level
);
11256 /* check for platform support
11257 * for this raid level configuration
11259 struct intel_super
*super
= st
->sb
;
11260 if (!is_raid_level_supported(super
->orom
,
11261 member
->array
.level
,
11262 geo
->raid_disks
)) {
11263 dprintf("platform does not support raid%d with %d disk%s\n",
11266 geo
->raid_disks
> 1 ? "s" : "");
11269 /* check if component size is aligned to chunk size
11271 if (info
->component_size
%
11272 (info
->array
.chunk_size
/512)) {
11273 dprintf("Component size is not aligned to chunk size\n");
11278 if (*old_raid_disks
&&
11279 info
->array
.raid_disks
!= *old_raid_disks
)
11281 *old_raid_disks
= info
->array
.raid_disks
;
11283 /* All raid5 and raid0 volumes in container
11284 * have to be ready for Online Capacity Expansion
11285 * so they need to be assembled. We have already
11286 * checked that no recovery etc is happening.
11288 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11289 st
->container_devnm
);
11290 if (result
== NULL
) {
11291 dprintf("imsm: cannot find array\n");
11294 devices_that_can_grow
++;
11297 if (!member
&& devices_that_can_grow
)
11301 dprintf("Container operation allowed\n");
11303 dprintf("Error: %i\n", ret_val
);
11308 /* Function: get_spares_for_grow
11309 * Description: Allocates memory and creates list of spare devices
11310 * avaliable in container. Checks if spare drive size is acceptable.
11311 * Parameters: Pointer to the supertype structure
11312 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11315 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11317 struct spare_criteria sc
;
11319 get_spare_criteria_imsm(st
, &sc
);
11320 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11323 /******************************************************************************
11324 * function: imsm_create_metadata_update_for_reshape
11325 * Function creates update for whole IMSM container.
11327 ******************************************************************************/
11328 static int imsm_create_metadata_update_for_reshape(
11329 struct supertype
*st
,
11330 struct geo_params
*geo
,
11331 int old_raid_disks
,
11332 struct imsm_update_reshape
**updatep
)
11334 struct intel_super
*super
= st
->sb
;
11335 struct imsm_super
*mpb
= super
->anchor
;
11336 int update_memory_size
;
11337 struct imsm_update_reshape
*u
;
11338 struct mdinfo
*spares
;
11341 struct mdinfo
*dev
;
11343 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11345 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11347 /* size of all update data without anchor */
11348 update_memory_size
= sizeof(struct imsm_update_reshape
);
11350 /* now add space for spare disks that we need to add. */
11351 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11353 u
= xcalloc(1, update_memory_size
);
11354 u
->type
= update_reshape_container_disks
;
11355 u
->old_raid_disks
= old_raid_disks
;
11356 u
->new_raid_disks
= geo
->raid_disks
;
11358 /* now get spare disks list
11360 spares
= get_spares_for_grow(st
);
11362 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11363 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11368 /* we have got spares
11369 * update disk list in imsm_disk list table in anchor
11371 dprintf("imsm: %i spares are available.\n\n",
11372 spares
->array
.spare_disks
);
11374 dev
= spares
->devs
;
11375 for (i
= 0; i
< delta_disks
; i
++) {
11380 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11382 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11383 dl
->index
= mpb
->num_disks
;
11391 sysfs_free(spares
);
11393 dprintf("imsm: reshape update preparation :");
11394 if (i
== delta_disks
) {
11395 dprintf_cont(" OK\n");
11397 return update_memory_size
;
11400 dprintf_cont(" Error\n");
11405 /******************************************************************************
11406 * function: imsm_create_metadata_update_for_size_change()
11407 * Creates update for IMSM array for array size change.
11409 ******************************************************************************/
11410 static int imsm_create_metadata_update_for_size_change(
11411 struct supertype
*st
,
11412 struct geo_params
*geo
,
11413 struct imsm_update_size_change
**updatep
)
11415 struct intel_super
*super
= st
->sb
;
11416 int update_memory_size
;
11417 struct imsm_update_size_change
*u
;
11419 dprintf("(enter) New size = %llu\n", geo
->size
);
11421 /* size of all update data without anchor */
11422 update_memory_size
= sizeof(struct imsm_update_size_change
);
11424 u
= xcalloc(1, update_memory_size
);
11425 u
->type
= update_size_change
;
11426 u
->subdev
= super
->current_vol
;
11427 u
->new_size
= geo
->size
;
11429 dprintf("imsm: reshape update preparation : OK\n");
11432 return update_memory_size
;
11435 /******************************************************************************
11436 * function: imsm_create_metadata_update_for_migration()
11437 * Creates update for IMSM array.
11439 ******************************************************************************/
11440 static int imsm_create_metadata_update_for_migration(
11441 struct supertype
*st
,
11442 struct geo_params
*geo
,
11443 struct imsm_update_reshape_migration
**updatep
)
11445 struct intel_super
*super
= st
->sb
;
11446 int update_memory_size
;
11447 struct imsm_update_reshape_migration
*u
;
11448 struct imsm_dev
*dev
;
11449 int previous_level
= -1;
11451 dprintf("(enter) New Level = %i\n", geo
->level
);
11453 /* size of all update data without anchor */
11454 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11456 u
= xcalloc(1, update_memory_size
);
11457 u
->type
= update_reshape_migration
;
11458 u
->subdev
= super
->current_vol
;
11459 u
->new_level
= geo
->level
;
11460 u
->new_layout
= geo
->layout
;
11461 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11462 u
->new_disks
[0] = -1;
11463 u
->new_chunksize
= -1;
11465 dev
= get_imsm_dev(super
, u
->subdev
);
11467 struct imsm_map
*map
;
11469 map
= get_imsm_map(dev
, MAP_0
);
11471 int current_chunk_size
=
11472 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11474 if (geo
->chunksize
!= current_chunk_size
) {
11475 u
->new_chunksize
= geo
->chunksize
/ 1024;
11476 dprintf("imsm: chunk size change from %i to %i\n",
11477 current_chunk_size
, u
->new_chunksize
);
11479 previous_level
= map
->raid_level
;
11482 if (geo
->level
== 5 && previous_level
== 0) {
11483 struct mdinfo
*spares
= NULL
;
11485 u
->new_raid_disks
++;
11486 spares
= get_spares_for_grow(st
);
11487 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11489 sysfs_free(spares
);
11490 update_memory_size
= 0;
11491 pr_err("cannot get spare device for requested migration\n");
11494 sysfs_free(spares
);
11496 dprintf("imsm: reshape update preparation : OK\n");
11499 return update_memory_size
;
11502 static void imsm_update_metadata_locally(struct supertype
*st
,
11503 void *buf
, int len
)
11505 struct metadata_update mu
;
11510 mu
.space_list
= NULL
;
11512 if (imsm_prepare_update(st
, &mu
))
11513 imsm_process_update(st
, &mu
);
11515 while (mu
.space_list
) {
11516 void **space
= mu
.space_list
;
11517 mu
.space_list
= *space
;
11522 /***************************************************************************
11523 * Function: imsm_analyze_change
11524 * Description: Function analyze change for single volume
11525 * and validate if transition is supported
11526 * Parameters: Geometry parameters, supertype structure,
11527 * metadata change direction (apply/rollback)
11528 * Returns: Operation type code on success, -1 if fail
11529 ****************************************************************************/
11530 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11531 struct geo_params
*geo
,
11534 struct mdinfo info
;
11536 int check_devs
= 0;
11538 /* number of added/removed disks in operation result */
11539 int devNumChange
= 0;
11540 /* imsm compatible layout value for array geometry verification */
11541 int imsm_layout
= -1;
11543 struct imsm_dev
*dev
;
11544 struct imsm_map
*map
;
11545 struct intel_super
*super
;
11546 unsigned long long current_size
;
11547 unsigned long long free_size
;
11548 unsigned long long max_size
;
11551 getinfo_super_imsm_volume(st
, &info
, NULL
);
11552 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11553 geo
->level
!= UnSet
) {
11554 switch (info
.array
.level
) {
11556 if (geo
->level
== 5) {
11557 change
= CH_MIGRATION
;
11558 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11559 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11561 goto analyse_change_exit
;
11563 imsm_layout
= geo
->layout
;
11565 devNumChange
= 1; /* parity disk added */
11566 } else if (geo
->level
== 10) {
11567 change
= CH_TAKEOVER
;
11569 devNumChange
= 2; /* two mirrors added */
11570 imsm_layout
= 0x102; /* imsm supported layout */
11575 if (geo
->level
== 0) {
11576 change
= CH_TAKEOVER
;
11578 devNumChange
= -(geo
->raid_disks
/2);
11579 imsm_layout
= 0; /* imsm raid0 layout */
11583 if (change
== -1) {
11584 pr_err("Error. Level Migration from %d to %d not supported!\n",
11585 info
.array
.level
, geo
->level
);
11586 goto analyse_change_exit
;
11589 geo
->level
= info
.array
.level
;
11591 if (geo
->layout
!= info
.array
.layout
&&
11592 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11593 change
= CH_MIGRATION
;
11594 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11595 geo
->layout
== 5) {
11596 /* reshape 5 -> 4 */
11597 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11598 geo
->layout
== 0) {
11599 /* reshape 4 -> 5 */
11603 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11604 info
.array
.layout
, geo
->layout
);
11606 goto analyse_change_exit
;
11609 geo
->layout
= info
.array
.layout
;
11610 if (imsm_layout
== -1)
11611 imsm_layout
= info
.array
.layout
;
11614 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11615 geo
->chunksize
!= info
.array
.chunk_size
) {
11616 if (info
.array
.level
== 10) {
11617 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11619 goto analyse_change_exit
;
11620 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11621 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11622 geo
->chunksize
/1024, info
.component_size
/2);
11624 goto analyse_change_exit
;
11626 change
= CH_MIGRATION
;
11628 geo
->chunksize
= info
.array
.chunk_size
;
11631 chunk
= geo
->chunksize
/ 1024;
11634 dev
= get_imsm_dev(super
, super
->current_vol
);
11635 map
= get_imsm_map(dev
, MAP_0
);
11636 data_disks
= imsm_num_data_members(map
);
11637 /* compute current size per disk member
11639 current_size
= info
.custom_array_size
/ data_disks
;
11641 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11642 /* align component size
11644 geo
->size
= imsm_component_size_alignment_check(
11645 get_imsm_raid_level(dev
->vol
.map
),
11646 chunk
* 1024, super
->sector_size
,
11648 if (geo
->size
== 0) {
11649 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11651 goto analyse_change_exit
;
11655 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11656 if (change
!= -1) {
11657 pr_err("Error. Size change should be the only one at a time.\n");
11659 goto analyse_change_exit
;
11661 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11662 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11663 super
->current_vol
, st
->devnm
);
11664 goto analyse_change_exit
;
11666 /* check the maximum available size
11668 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11669 0, chunk
, &free_size
);
11671 /* Cannot find maximum available space
11675 max_size
= free_size
+ current_size
;
11676 /* align component size
11678 max_size
= imsm_component_size_alignment_check(
11679 get_imsm_raid_level(dev
->vol
.map
),
11680 chunk
* 1024, super
->sector_size
,
11683 if (geo
->size
== MAX_SIZE
) {
11684 /* requested size change to the maximum available size
11686 if (max_size
== 0) {
11687 pr_err("Error. Cannot find maximum available space.\n");
11689 goto analyse_change_exit
;
11691 geo
->size
= max_size
;
11694 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11695 /* accept size for rollback only
11698 /* round size due to metadata compatibility
11700 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11701 << SECT_PER_MB_SHIFT
;
11702 dprintf("Prepare update for size change to %llu\n",
11704 if (current_size
>= geo
->size
) {
11705 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11706 current_size
, geo
->size
);
11707 goto analyse_change_exit
;
11709 if (max_size
&& geo
->size
> max_size
) {
11710 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11711 max_size
, geo
->size
);
11712 goto analyse_change_exit
;
11715 geo
->size
*= data_disks
;
11716 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11717 change
= CH_ARRAY_SIZE
;
11719 if (!validate_geometry_imsm(st
,
11722 geo
->raid_disks
+ devNumChange
,
11724 geo
->size
, INVALID_SECTORS
,
11725 0, 0, info
.consistency_policy
, 1))
11729 struct intel_super
*super
= st
->sb
;
11730 struct imsm_super
*mpb
= super
->anchor
;
11732 if (mpb
->num_raid_devs
> 1) {
11733 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11739 analyse_change_exit
:
11740 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11741 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11742 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11748 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11750 struct intel_super
*super
= st
->sb
;
11751 struct imsm_update_takeover
*u
;
11753 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11755 u
->type
= update_takeover
;
11756 u
->subarray
= super
->current_vol
;
11758 /* 10->0 transition */
11759 if (geo
->level
== 0)
11760 u
->direction
= R10_TO_R0
;
11762 /* 0->10 transition */
11763 if (geo
->level
== 10)
11764 u
->direction
= R0_TO_R10
;
11766 /* update metadata locally */
11767 imsm_update_metadata_locally(st
, u
,
11768 sizeof(struct imsm_update_takeover
));
11769 /* and possibly remotely */
11770 if (st
->update_tail
)
11771 append_metadata_update(st
, u
,
11772 sizeof(struct imsm_update_takeover
));
11779 /* Flush size update if size calculated by num_data_stripes is higher than
11780 * imsm_dev_size to eliminate differences during reshape.
11781 * Mdmon will recalculate them correctly.
11782 * If subarray index is not set then check whole container.
11784 * 0 - no error occurred
11785 * 1 - error detected
11787 static int imsm_fix_size_mismatch(struct supertype
*st
, int subarray_index
)
11789 struct intel_super
*super
= st
->sb
;
11790 int tmp
= super
->current_vol
;
11794 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
11795 if (subarray_index
>= 0 && i
!= subarray_index
)
11797 super
->current_vol
= i
;
11798 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11799 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11800 unsigned int disc_count
= imsm_num_data_members(map
);
11801 struct geo_params geo
;
11802 struct imsm_update_size_change
*update
;
11803 unsigned long long calc_size
= per_dev_array_size(map
) * disc_count
;
11804 unsigned long long d_size
= imsm_dev_size(dev
);
11807 if (calc_size
== d_size
|| dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
11810 /* There is a difference, verify that imsm_dev_size is
11811 * rounded correctly and push update.
11813 if (d_size
!= round_size_to_mb(d_size
, disc_count
)) {
11814 dprintf("imsm: Size of volume %d is not rounded correctly\n",
11818 memset(&geo
, 0, sizeof(struct geo_params
));
11820 u_size
= imsm_create_metadata_update_for_size_change(st
, &geo
,
11823 dprintf("imsm: Cannot prepare size change update\n");
11826 imsm_update_metadata_locally(st
, update
, u_size
);
11827 if (st
->update_tail
) {
11828 append_metadata_update(st
, update
, u_size
);
11829 flush_metadata_updates(st
);
11830 st
->update_tail
= &st
->updates
;
11832 imsm_sync_metadata(st
);
11837 super
->current_vol
= tmp
;
11841 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11843 int layout
, int chunksize
, int raid_disks
,
11844 int delta_disks
, char *backup
, char *dev
,
11845 int direction
, int verbose
)
11848 struct geo_params geo
;
11850 dprintf("(enter)\n");
11852 memset(&geo
, 0, sizeof(struct geo_params
));
11854 geo
.dev_name
= dev
;
11855 strcpy(geo
.devnm
, st
->devnm
);
11858 geo
.layout
= layout
;
11859 geo
.chunksize
= chunksize
;
11860 geo
.raid_disks
= raid_disks
;
11861 if (delta_disks
!= UnSet
)
11862 geo
.raid_disks
+= delta_disks
;
11864 dprintf("for level : %i\n", geo
.level
);
11865 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11867 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11868 /* On container level we can only increase number of devices. */
11869 dprintf("imsm: info: Container operation\n");
11870 int old_raid_disks
= 0;
11872 if (imsm_reshape_is_allowed_on_container(
11873 st
, &geo
, &old_raid_disks
, direction
)) {
11874 struct imsm_update_reshape
*u
= NULL
;
11877 if (imsm_fix_size_mismatch(st
, -1)) {
11878 dprintf("imsm: Cannot fix size mismatch\n");
11879 goto exit_imsm_reshape_super
;
11882 len
= imsm_create_metadata_update_for_reshape(
11883 st
, &geo
, old_raid_disks
, &u
);
11886 dprintf("imsm: Cannot prepare update\n");
11887 goto exit_imsm_reshape_super
;
11891 /* update metadata locally */
11892 imsm_update_metadata_locally(st
, u
, len
);
11893 /* and possibly remotely */
11894 if (st
->update_tail
)
11895 append_metadata_update(st
, u
, len
);
11900 pr_err("(imsm) Operation is not allowed on this container\n");
11903 /* On volume level we support following operations
11904 * - takeover: raid10 -> raid0; raid0 -> raid10
11905 * - chunk size migration
11906 * - migration: raid5 -> raid0; raid0 -> raid5
11908 struct intel_super
*super
= st
->sb
;
11909 struct intel_dev
*dev
= super
->devlist
;
11911 dprintf("imsm: info: Volume operation\n");
11912 /* find requested device */
11915 imsm_find_array_devnm_by_subdev(
11916 dev
->index
, st
->container_devnm
);
11917 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11922 pr_err("Cannot find %s (%s) subarray\n",
11923 geo
.dev_name
, geo
.devnm
);
11924 goto exit_imsm_reshape_super
;
11926 super
->current_vol
= dev
->index
;
11927 change
= imsm_analyze_change(st
, &geo
, direction
);
11930 ret_val
= imsm_takeover(st
, &geo
);
11932 case CH_MIGRATION
: {
11933 struct imsm_update_reshape_migration
*u
= NULL
;
11935 imsm_create_metadata_update_for_migration(
11938 dprintf("imsm: Cannot prepare update\n");
11942 /* update metadata locally */
11943 imsm_update_metadata_locally(st
, u
, len
);
11944 /* and possibly remotely */
11945 if (st
->update_tail
)
11946 append_metadata_update(st
, u
, len
);
11951 case CH_ARRAY_SIZE
: {
11952 struct imsm_update_size_change
*u
= NULL
;
11954 imsm_create_metadata_update_for_size_change(
11957 dprintf("imsm: Cannot prepare update\n");
11961 /* update metadata locally */
11962 imsm_update_metadata_locally(st
, u
, len
);
11963 /* and possibly remotely */
11964 if (st
->update_tail
)
11965 append_metadata_update(st
, u
, len
);
11975 exit_imsm_reshape_super
:
11976 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11980 #define COMPLETED_OK 0
11981 #define COMPLETED_NONE 1
11982 #define COMPLETED_DELAYED 2
11984 static int read_completed(int fd
, unsigned long long *val
)
11989 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11993 ret
= COMPLETED_OK
;
11994 if (strncmp(buf
, "none", 4) == 0) {
11995 ret
= COMPLETED_NONE
;
11996 } else if (strncmp(buf
, "delayed", 7) == 0) {
11997 ret
= COMPLETED_DELAYED
;
12000 *val
= strtoull(buf
, &ep
, 0);
12001 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
12007 /*******************************************************************************
12008 * Function: wait_for_reshape_imsm
12009 * Description: Function writes new sync_max value and waits until
12010 * reshape process reach new position
12012 * sra : general array info
12013 * ndata : number of disks in new array's layout
12016 * 1 : there is no reshape in progress,
12018 ******************************************************************************/
12019 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
12021 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
12023 unsigned long long completed
;
12024 /* to_complete : new sync_max position */
12025 unsigned long long to_complete
= sra
->reshape_progress
;
12026 unsigned long long position_to_set
= to_complete
/ ndata
;
12029 dprintf("cannot open reshape_position\n");
12034 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
12036 dprintf("cannot read reshape_position (no reshape in progres)\n");
12045 if (completed
> position_to_set
) {
12046 dprintf("wrong next position to set %llu (%llu)\n",
12047 to_complete
, position_to_set
);
12051 dprintf("Position set: %llu\n", position_to_set
);
12052 if (sysfs_set_num(sra
, NULL
, "sync_max",
12053 position_to_set
) != 0) {
12054 dprintf("cannot set reshape position to %llu\n",
12063 int timeout
= 3000;
12065 sysfs_wait(fd
, &timeout
);
12066 if (sysfs_get_str(sra
, NULL
, "sync_action",
12068 strncmp(action
, "reshape", 7) != 0) {
12069 if (strncmp(action
, "idle", 4) == 0)
12075 rc
= read_completed(fd
, &completed
);
12077 dprintf("cannot read reshape_position (in loop)\n");
12080 } else if (rc
== COMPLETED_NONE
)
12082 } while (completed
< position_to_set
);
12088 /*******************************************************************************
12089 * Function: check_degradation_change
12090 * Description: Check that array hasn't become failed.
12092 * info : for sysfs access
12093 * sources : source disks descriptors
12094 * degraded: previous degradation level
12096 * degradation level
12097 ******************************************************************************/
12098 int check_degradation_change(struct mdinfo
*info
,
12102 unsigned long long new_degraded
;
12105 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
12106 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
12107 /* check each device to ensure it is still working */
12110 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
12111 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
12113 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
12116 if (sysfs_get_str(info
,
12117 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
12118 strstr(sbuf
, "faulty") ||
12119 strstr(sbuf
, "in_sync") == NULL
) {
12120 /* this device is dead */
12121 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
12122 if (sd
->disk
.raid_disk
>= 0 &&
12123 sources
[sd
->disk
.raid_disk
] >= 0) {
12125 sd
->disk
.raid_disk
]);
12126 sources
[sd
->disk
.raid_disk
] =
12135 return new_degraded
;
12138 /*******************************************************************************
12139 * Function: imsm_manage_reshape
12140 * Description: Function finds array under reshape and it manages reshape
12141 * process. It creates stripes backups (if required) and sets
12144 * afd : Backup handle (nattive) - not used
12145 * sra : general array info
12146 * reshape : reshape parameters - not used
12147 * st : supertype structure
12148 * blocks : size of critical section [blocks]
12149 * fds : table of source device descriptor
12150 * offsets : start of array (offest per devices)
12152 * destfd : table of destination device descriptor
12153 * destoffsets : table of destination offsets (per device)
12155 * 1 : success, reshape is done
12157 ******************************************************************************/
12158 static int imsm_manage_reshape(
12159 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
12160 struct supertype
*st
, unsigned long backup_blocks
,
12161 int *fds
, unsigned long long *offsets
,
12162 int dests
, int *destfd
, unsigned long long *destoffsets
)
12165 struct intel_super
*super
= st
->sb
;
12166 struct intel_dev
*dv
;
12167 unsigned int sector_size
= super
->sector_size
;
12168 struct imsm_dev
*dev
= NULL
;
12169 struct imsm_map
*map_src
, *map_dest
;
12170 int migr_vol_qan
= 0;
12171 int ndata
, odata
; /* [bytes] */
12172 int chunk
; /* [bytes] */
12173 struct migr_record
*migr_rec
;
12175 unsigned int buf_size
; /* [bytes] */
12176 unsigned long long max_position
; /* array size [bytes] */
12177 unsigned long long next_step
; /* [blocks]/[bytes] */
12178 unsigned long long old_data_stripe_length
;
12179 unsigned long long start_src
; /* [bytes] */
12180 unsigned long long start
; /* [bytes] */
12181 unsigned long long start_buf_shift
; /* [bytes] */
12183 int source_layout
= 0;
12184 int subarray_index
= -1;
12189 if (!fds
|| !offsets
)
12192 /* Find volume during the reshape */
12193 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
12194 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12195 dv
->dev
->vol
.migr_state
== 1) {
12198 subarray_index
= dv
->index
;
12201 /* Only one volume can migrate at the same time */
12202 if (migr_vol_qan
!= 1) {
12203 pr_err("%s", migr_vol_qan
?
12204 "Number of migrating volumes greater than 1\n" :
12205 "There is no volume during migrationg\n");
12209 map_dest
= get_imsm_map(dev
, MAP_0
);
12210 map_src
= get_imsm_map(dev
, MAP_1
);
12211 if (map_src
== NULL
)
12214 ndata
= imsm_num_data_members(map_dest
);
12215 odata
= imsm_num_data_members(map_src
);
12217 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12218 old_data_stripe_length
= odata
* chunk
;
12220 migr_rec
= super
->migr_rec
;
12222 /* initialize migration record for start condition */
12223 if (sra
->reshape_progress
== 0)
12224 init_migr_record_imsm(st
, dev
, sra
);
12226 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12227 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12230 /* Save checkpoint to update migration record for current
12231 * reshape position (in md). It can be farther than current
12232 * reshape position in metadata.
12234 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12235 /* ignore error == 2, this can mean end of reshape here
12237 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12242 /* size for data */
12243 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12244 /* extend buffer size for parity disk */
12245 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12246 /* add space for stripe alignment */
12247 buf_size
+= old_data_stripe_length
;
12248 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12249 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12253 max_position
= sra
->component_size
* ndata
;
12254 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12256 while (current_migr_unit(migr_rec
) <
12257 get_num_migr_units(migr_rec
)) {
12258 /* current reshape position [blocks] */
12259 unsigned long long current_position
=
12260 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12261 * current_migr_unit(migr_rec
);
12262 unsigned long long border
;
12264 /* Check that array hasn't become failed.
12266 degraded
= check_degradation_change(sra
, fds
, degraded
);
12267 if (degraded
> 1) {
12268 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12272 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12274 if ((current_position
+ next_step
) > max_position
)
12275 next_step
= max_position
- current_position
;
12277 start
= current_position
* 512;
12279 /* align reading start to old geometry */
12280 start_buf_shift
= start
% old_data_stripe_length
;
12281 start_src
= start
- start_buf_shift
;
12283 border
= (start_src
/ odata
) - (start
/ ndata
);
12285 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12286 /* save critical stripes to buf
12287 * start - start address of current unit
12288 * to backup [bytes]
12289 * start_src - start address of current unit
12290 * to backup alligned to source array
12293 unsigned long long next_step_filler
;
12294 unsigned long long copy_length
= next_step
* 512;
12296 /* allign copy area length to stripe in old geometry */
12297 next_step_filler
= ((copy_length
+ start_buf_shift
)
12298 % old_data_stripe_length
);
12299 if (next_step_filler
)
12300 next_step_filler
= (old_data_stripe_length
12301 - next_step_filler
);
12302 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12303 start
, start_src
, copy_length
,
12304 start_buf_shift
, next_step_filler
);
12306 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12307 chunk
, map_src
->raid_level
,
12308 source_layout
, 0, NULL
, start_src
,
12310 next_step_filler
+ start_buf_shift
,
12312 dprintf("imsm: Cannot save stripes to buffer\n");
12315 /* Convert data to destination format and store it
12316 * in backup general migration area
12318 if (save_backup_imsm(st
, dev
, sra
,
12319 buf
+ start_buf_shift
, copy_length
)) {
12320 dprintf("imsm: Cannot save stripes to target devices\n");
12323 if (save_checkpoint_imsm(st
, sra
,
12324 UNIT_SRC_IN_CP_AREA
)) {
12325 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12329 /* set next step to use whole border area */
12330 border
/= next_step
;
12332 next_step
*= border
;
12334 /* When data backed up, checkpoint stored,
12335 * kick the kernel to reshape unit of data
12337 next_step
= next_step
+ sra
->reshape_progress
;
12338 /* limit next step to array max position */
12339 if (next_step
> max_position
)
12340 next_step
= max_position
;
12341 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12342 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12343 sra
->reshape_progress
= next_step
;
12345 /* wait until reshape finish */
12346 if (wait_for_reshape_imsm(sra
, ndata
)) {
12347 dprintf("wait_for_reshape_imsm returned error!\n");
12353 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12354 /* ignore error == 2, this can mean end of reshape here
12356 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12362 /* clear migr_rec on disks after successful migration */
12365 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12366 for (d
= super
->disks
; d
; d
= d
->next
) {
12367 if (d
->index
< 0 || is_failed(&d
->disk
))
12369 unsigned long long dsize
;
12371 get_dev_size(d
->fd
, NULL
, &dsize
);
12372 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12374 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12375 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12376 MIGR_REC_BUF_SECTORS
*sector_size
)
12377 perror("Write migr_rec failed");
12381 /* return '1' if done */
12384 /* After the reshape eliminate size mismatch in metadata.
12385 * Don't update md/component_size here, volume hasn't
12386 * to take whole space. It is allowed by kernel.
12387 * md/component_size will be set propoperly after next assembly.
12389 imsm_fix_size_mismatch(st
, subarray_index
);
12393 /* See Grow.c: abort_reshape() for further explanation */
12394 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12395 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12396 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12401 /*******************************************************************************
12402 * Function: calculate_bitmap_min_chunksize
12403 * Description: Calculates the minimal valid bitmap chunk size
12405 * max_bits : indicate how many bits can be used for the bitmap
12406 * data_area_size : the size of the data area covered by the bitmap
12409 * The bitmap chunk size
12410 ******************************************************************************/
12411 static unsigned long long
12412 calculate_bitmap_min_chunksize(unsigned long long max_bits
,
12413 unsigned long long data_area_size
)
12415 unsigned long long min_chunk
=
12416 4096; /* sub-page chunks don't work yet.. */
12417 unsigned long long bits
= data_area_size
/ min_chunk
+ 1;
12419 while (bits
> max_bits
) {
12421 bits
= (bits
+ 1) / 2;
12426 /*******************************************************************************
12427 * Function: calculate_bitmap_chunksize
12428 * Description: Calculates the bitmap chunk size for the given device
12430 * st : supertype information
12431 * dev : device for the bitmap
12434 * The bitmap chunk size
12435 ******************************************************************************/
12436 static unsigned long long calculate_bitmap_chunksize(struct supertype
*st
,
12437 struct imsm_dev
*dev
)
12439 struct intel_super
*super
= st
->sb
;
12440 unsigned long long min_chunksize
;
12441 unsigned long long result
= IMSM_DEFAULT_BITMAP_CHUNKSIZE
;
12442 size_t dev_size
= imsm_dev_size(dev
);
12444 min_chunksize
= calculate_bitmap_min_chunksize(
12445 IMSM_BITMAP_AREA_SIZE
* super
->sector_size
, dev_size
);
12447 if (result
< min_chunksize
)
12448 result
= min_chunksize
;
12453 /*******************************************************************************
12454 * Function: init_bitmap_header
12455 * Description: Initialize the bitmap header structure
12457 * st : supertype information
12458 * bms : bitmap header struct to initialize
12459 * dev : device for the bitmap
12464 ******************************************************************************/
12465 static int init_bitmap_header(struct supertype
*st
, struct bitmap_super_s
*bms
,
12466 struct imsm_dev
*dev
)
12473 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
12474 bms
->version
= __cpu_to_le32(BITMAP_MAJOR_HI
);
12475 bms
->daemon_sleep
= __cpu_to_le32(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
);
12476 bms
->sync_size
= __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
);
12477 bms
->write_behind
= __cpu_to_le32(0);
12479 uuid_from_super_imsm(st
, vol_uuid
);
12480 memcpy(bms
->uuid
, vol_uuid
, 16);
12482 bms
->chunksize
= calculate_bitmap_chunksize(st
, dev
);
12487 /*******************************************************************************
12488 * Function: validate_internal_bitmap_for_drive
12489 * Description: Verify if the bitmap header for a given drive.
12491 * st : supertype information
12492 * offset : The offset from the beginning of the drive where to look for
12493 * the bitmap header.
12494 * d : the drive info
12499 ******************************************************************************/
12500 static int validate_internal_bitmap_for_drive(struct supertype
*st
,
12501 unsigned long long offset
,
12504 struct intel_super
*super
= st
->sb
;
12507 bitmap_super_t
*bms
;
12515 if (posix_memalign(&read_buf
, MAX_SECTOR_SIZE
, IMSM_BITMAP_HEADER_SIZE
))
12520 fd
= open(d
->devname
, O_RDONLY
, 0);
12522 dprintf("cannot open the device %s\n", d
->devname
);
12527 if (lseek64(fd
, offset
* super
->sector_size
, SEEK_SET
) < 0)
12529 if (read(fd
, read_buf
, IMSM_BITMAP_HEADER_SIZE
) !=
12530 IMSM_BITMAP_HEADER_SIZE
)
12533 uuid_from_super_imsm(st
, vol_uuid
);
12536 if ((bms
->magic
!= __cpu_to_le32(BITMAP_MAGIC
)) ||
12537 (bms
->version
!= __cpu_to_le32(BITMAP_MAJOR_HI
)) ||
12538 (!same_uuid((int *)bms
->uuid
, vol_uuid
, st
->ss
->swapuuid
))) {
12539 dprintf("wrong bitmap header detected\n");
12545 if ((d
->fd
< 0) && (fd
>= 0))
12553 /*******************************************************************************
12554 * Function: validate_internal_bitmap_imsm
12555 * Description: Verify if the bitmap header is in place and with proper data.
12557 * st : supertype information
12560 * 0 : success or device w/o RWH_BITMAP
12562 ******************************************************************************/
12563 static int validate_internal_bitmap_imsm(struct supertype
*st
)
12565 struct intel_super
*super
= st
->sb
;
12566 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
12567 unsigned long long offset
;
12573 if (dev
->rwh_policy
!= RWH_BITMAP
)
12576 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12577 for (d
= super
->disks
; d
; d
= d
->next
) {
12578 if (d
->index
< 0 || is_failed(&d
->disk
))
12581 if (validate_internal_bitmap_for_drive(st
, offset
, d
)) {
12582 pr_err("imsm: bitmap validation failed\n");
12589 /*******************************************************************************
12590 * Function: add_internal_bitmap_imsm
12591 * Description: Mark the volume to use the bitmap and updates the chunk size value.
12593 * st : supertype information
12594 * chunkp : bitmap chunk size
12595 * delay : not used for imsm
12596 * write_behind : not used for imsm
12597 * size : not used for imsm
12598 * may_change : not used for imsm
12599 * amajor : not used for imsm
12604 ******************************************************************************/
12605 static int add_internal_bitmap_imsm(struct supertype
*st
, int *chunkp
,
12606 int delay
, int write_behind
,
12607 unsigned long long size
, int may_change
,
12610 struct intel_super
*super
= st
->sb
;
12611 int vol_idx
= super
->current_vol
;
12612 struct imsm_dev
*dev
;
12614 if (!super
->devlist
|| vol_idx
== -1 || !chunkp
)
12617 dev
= get_imsm_dev(super
, vol_idx
);
12620 dprintf("cannot find the device for volume index %d\n",
12624 dev
->rwh_policy
= RWH_BITMAP
;
12626 *chunkp
= calculate_bitmap_chunksize(st
, dev
);
12631 /*******************************************************************************
12632 * Function: locate_bitmap_imsm
12633 * Description: Seek 'fd' to start of write-intent-bitmap.
12635 * st : supertype information
12636 * fd : file descriptor for the device
12637 * node_num : not used for imsm
12642 ******************************************************************************/
12643 static int locate_bitmap_imsm(struct supertype
*st
, int fd
, int node_num
)
12645 struct intel_super
*super
= st
->sb
;
12646 unsigned long long offset
;
12647 int vol_idx
= super
->current_vol
;
12649 if (!super
->devlist
|| vol_idx
== -1)
12652 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12653 dprintf("bitmap header offset is %llu\n", offset
);
12655 lseek64(fd
, offset
<< 9, 0);
12660 /*******************************************************************************
12661 * Function: write_init_bitmap_imsm
12662 * Description: Write a bitmap header and prepares the area for the bitmap.
12664 * st : supertype information
12665 * fd : file descriptor for the device
12666 * update : not used for imsm
12671 ******************************************************************************/
12672 static int write_init_bitmap_imsm(struct supertype
*st
, int fd
,
12673 enum bitmap_update update
)
12675 struct intel_super
*super
= st
->sb
;
12676 int vol_idx
= super
->current_vol
;
12678 unsigned long long offset
;
12679 bitmap_super_t bms
= { 0 };
12680 size_t written
= 0;
12685 if (!super
->devlist
|| !super
->sector_size
|| vol_idx
== -1)
12688 struct imsm_dev
*dev
= get_imsm_dev(super
, vol_idx
);
12690 /* first clear the space for bitmap header */
12691 unsigned long long bitmap_area_start
=
12692 get_bitmap_header_sector(super
, vol_idx
);
12694 dprintf("zeroing area start (%llu) and size (%u)\n", bitmap_area_start
,
12695 IMSM_BITMAP_AND_HEADER_SIZE
/ super
->sector_size
);
12696 if (zero_disk_range(fd
, bitmap_area_start
,
12697 IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
)) {
12698 pr_err("imsm: cannot zeroing the space for the bitmap\n");
12702 /* The bitmap area should be filled with "1"s to perform initial
12705 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
))
12707 memset(buf
, 0xFF, MAX_SECTOR_SIZE
);
12708 offset
= get_bitmap_sector(super
, vol_idx
);
12709 lseek64(fd
, offset
<< 9, 0);
12710 while (written
< IMSM_BITMAP_AREA_SIZE
) {
12711 to_write
= IMSM_BITMAP_AREA_SIZE
- written
;
12712 if (to_write
> MAX_SECTOR_SIZE
)
12713 to_write
= MAX_SECTOR_SIZE
;
12714 rv_num
= write(fd
, buf
, MAX_SECTOR_SIZE
);
12715 if (rv_num
!= MAX_SECTOR_SIZE
) {
12717 dprintf("cannot initialize bitmap area\n");
12723 /* write a bitmap header */
12724 init_bitmap_header(st
, &bms
, dev
);
12725 memset(buf
, 0, MAX_SECTOR_SIZE
);
12726 memcpy(buf
, &bms
, sizeof(bitmap_super_t
));
12727 if (locate_bitmap_imsm(st
, fd
, 0)) {
12729 dprintf("cannot locate the bitmap\n");
12732 if (write(fd
, buf
, MAX_SECTOR_SIZE
) != MAX_SECTOR_SIZE
) {
12734 dprintf("cannot write the bitmap header\n");
12745 /*******************************************************************************
12746 * Function: is_vol_to_setup_bitmap
12747 * Description: Checks if a bitmap should be activated on the dev.
12749 * info : info about the volume to setup the bitmap
12750 * dev : the device to check against bitmap creation
12753 * 0 : bitmap should be set up on the device
12755 ******************************************************************************/
12756 static int is_vol_to_setup_bitmap(struct mdinfo
*info
, struct imsm_dev
*dev
)
12761 if ((strcmp((char *)dev
->volume
, info
->name
) == 0) &&
12762 (dev
->rwh_policy
== RWH_BITMAP
))
12768 /*******************************************************************************
12769 * Function: set_bitmap_sysfs
12770 * Description: Set the sysfs atributes of a given volume to activate the bitmap.
12772 * info : info about the volume where the bitmap should be setup
12773 * chunksize : bitmap chunk size
12774 * location : location of the bitmap
12779 ******************************************************************************/
12780 static int set_bitmap_sysfs(struct mdinfo
*info
, unsigned long long chunksize
,
12783 /* The bitmap/metadata is set to external to allow changing of value for
12784 * bitmap/location. When external is used, the kernel will treat an offset
12785 * related to the device's first lba (in opposition to the "internal" case
12786 * when this value is related to the beginning of the superblock).
12788 if (sysfs_set_str(info
, NULL
, "bitmap/metadata", "external")) {
12789 dprintf("failed to set bitmap/metadata\n");
12793 /* It can only be changed when no bitmap is active.
12794 * Should be bigger than 512 and must be power of 2.
12795 * It is expecting the value in bytes.
12797 if (sysfs_set_num(info
, NULL
, "bitmap/chunksize",
12798 __cpu_to_le32(chunksize
))) {
12799 dprintf("failed to set bitmap/chunksize\n");
12803 /* It is expecting the value in sectors. */
12804 if (sysfs_set_num(info
, NULL
, "bitmap/space",
12805 __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
))) {
12806 dprintf("failed to set bitmap/space\n");
12810 /* Determines the delay between the bitmap updates.
12811 * It is expecting the value in seconds.
12813 if (sysfs_set_num(info
, NULL
, "bitmap/time_base",
12814 __cpu_to_le64(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
))) {
12815 dprintf("failed to set bitmap/time_base\n");
12819 /* It is expecting the value in sectors with a sign at the beginning. */
12820 if (sysfs_set_str(info
, NULL
, "bitmap/location", location
)) {
12821 dprintf("failed to set bitmap/location\n");
12828 /*******************************************************************************
12829 * Function: set_bitmap_imsm
12830 * Description: Setup the bitmap for the given volume
12832 * st : supertype information
12833 * info : info about the volume where the bitmap should be setup
12838 ******************************************************************************/
12839 static int set_bitmap_imsm(struct supertype
*st
, struct mdinfo
*info
)
12841 struct intel_super
*super
= st
->sb
;
12842 int prev_current_vol
= super
->current_vol
;
12843 struct imsm_dev
*dev
;
12845 char location
[16] = "";
12846 unsigned long long chunksize
;
12847 struct intel_dev
*dev_it
;
12849 for (dev_it
= super
->devlist
; dev_it
; dev_it
= dev_it
->next
) {
12850 super
->current_vol
= dev_it
->index
;
12851 dev
= get_imsm_dev(super
, super
->current_vol
);
12853 if (is_vol_to_setup_bitmap(info
, dev
)) {
12854 if (validate_internal_bitmap_imsm(st
)) {
12855 dprintf("bitmap header validation failed\n");
12859 chunksize
= calculate_bitmap_chunksize(st
, dev
);
12860 dprintf("chunk size is %llu\n", chunksize
);
12862 snprintf(location
, sizeof(location
), "+%llu",
12863 get_bitmap_sector(super
, super
->current_vol
));
12864 dprintf("bitmap offset is %s\n", location
);
12866 if (set_bitmap_sysfs(info
, chunksize
, location
)) {
12867 dprintf("cannot setup the bitmap\n");
12874 super
->current_vol
= prev_current_vol
;
12878 struct superswitch super_imsm
= {
12879 .examine_super
= examine_super_imsm
,
12880 .brief_examine_super
= brief_examine_super_imsm
,
12881 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12882 .export_examine_super
= export_examine_super_imsm
,
12883 .detail_super
= detail_super_imsm
,
12884 .brief_detail_super
= brief_detail_super_imsm
,
12885 .write_init_super
= write_init_super_imsm
,
12886 .validate_geometry
= validate_geometry_imsm
,
12887 .add_to_super
= add_to_super_imsm
,
12888 .remove_from_super
= remove_from_super_imsm
,
12889 .detail_platform
= detail_platform_imsm
,
12890 .export_detail_platform
= export_detail_platform_imsm
,
12891 .kill_subarray
= kill_subarray_imsm
,
12892 .update_subarray
= update_subarray_imsm
,
12893 .load_container
= load_container_imsm
,
12894 .default_geometry
= default_geometry_imsm
,
12895 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12896 .reshape_super
= imsm_reshape_super
,
12897 .manage_reshape
= imsm_manage_reshape
,
12898 .recover_backup
= recover_backup_imsm
,
12899 .examine_badblocks
= examine_badblocks_imsm
,
12900 .match_home
= match_home_imsm
,
12901 .uuid_from_super
= uuid_from_super_imsm
,
12902 .getinfo_super
= getinfo_super_imsm
,
12903 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12904 .update_super
= update_super_imsm
,
12906 .avail_size
= avail_size_imsm
,
12907 .get_spare_criteria
= get_spare_criteria_imsm
,
12909 .compare_super
= compare_super_imsm
,
12911 .load_super
= load_super_imsm
,
12912 .init_super
= init_super_imsm
,
12913 .store_super
= store_super_imsm
,
12914 .free_super
= free_super_imsm
,
12915 .match_metadata_desc
= match_metadata_desc_imsm
,
12916 .container_content
= container_content_imsm
,
12917 .validate_container
= validate_container_imsm
,
12919 .add_internal_bitmap
= add_internal_bitmap_imsm
,
12920 .locate_bitmap
= locate_bitmap_imsm
,
12921 .write_bitmap
= write_init_bitmap_imsm
,
12922 .set_bitmap
= set_bitmap_imsm
,
12924 .write_init_ppl
= write_init_ppl_imsm
,
12925 .validate_ppl
= validate_ppl_imsm
,
12931 .open_new
= imsm_open_new
,
12932 .set_array_state
= imsm_set_array_state
,
12933 .set_disk
= imsm_set_disk
,
12934 .sync_metadata
= imsm_sync_metadata
,
12935 .activate_spare
= imsm_activate_spare
,
12936 .process_update
= imsm_process_update
,
12937 .prepare_update
= imsm_prepare_update
,
12938 .record_bad_block
= imsm_record_badblock
,
12939 .clear_bad_block
= imsm_clear_badblock
,
12940 .get_bad_blocks
= imsm_get_badblocks
,