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 fd
= open_dev(ent
->d_name
);
2404 if (!imsm_is_nvme_supported(fd
, 0)) {
2410 device_path
= diskfd_to_devpath(fd
);
2416 if (path_attached_to_hba(device_path
, hba
->path
)) {
2417 fd2devname(fd
, buf
);
2418 if (hba
->type
== SYS_DEV_VMD
)
2419 printf(" NVMe under VMD : %s", buf
);
2420 else if (hba
->type
== SYS_DEV_NVME
)
2421 printf(" NVMe Device : %s", buf
);
2422 if (!imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2424 printf(" (%s)\n", buf
);
2437 static void print_found_intel_controllers(struct sys_dev
*elem
)
2439 for (; elem
; elem
= elem
->next
) {
2440 pr_err("found Intel(R) ");
2441 if (elem
->type
== SYS_DEV_SATA
)
2442 fprintf(stderr
, "SATA ");
2443 else if (elem
->type
== SYS_DEV_SAS
)
2444 fprintf(stderr
, "SAS ");
2445 else if (elem
->type
== SYS_DEV_NVME
)
2446 fprintf(stderr
, "NVMe ");
2448 if (elem
->type
== SYS_DEV_VMD
)
2449 fprintf(stderr
, "VMD domain");
2451 fprintf(stderr
, "RAID controller");
2454 fprintf(stderr
, " at %s", elem
->pci_id
);
2455 fprintf(stderr
, ".\n");
2460 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2467 if ((dir
= opendir(hba_path
)) == NULL
)
2470 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2473 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2474 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2476 if (*port_count
== 0)
2478 else if (host
< host_base
)
2481 if (host
+ 1 > *port_count
+ host_base
)
2482 *port_count
= host
+ 1 - host_base
;
2488 static void print_imsm_capability(const struct imsm_orom
*orom
)
2490 printf(" Platform : Intel(R) ");
2491 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2492 printf("Matrix Storage Manager\n");
2493 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2494 printf("Virtual RAID on CPU\n");
2496 printf("Rapid Storage Technology%s\n",
2497 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2498 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2499 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2500 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2501 printf(" RAID Levels :%s%s%s%s%s\n",
2502 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2503 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2504 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2505 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2506 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2507 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2508 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2509 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2510 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2511 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2512 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2513 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2514 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2515 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2516 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2517 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2518 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2519 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2520 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2521 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2522 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2523 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2524 printf(" 2TB volumes :%s supported\n",
2525 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2526 printf(" 2TB disks :%s supported\n",
2527 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2528 printf(" Max Disks : %d\n", orom
->tds
);
2529 printf(" Max Volumes : %d per array, %d per %s\n",
2530 orom
->vpa
, orom
->vphba
,
2531 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2535 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2537 printf("MD_FIRMWARE_TYPE=imsm\n");
2538 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2539 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2540 orom
->hotfix_ver
, orom
->build
);
2541 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2542 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2543 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2544 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2545 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2546 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2547 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2548 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2549 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2550 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2551 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2552 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2553 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2554 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2555 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2556 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2557 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2558 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2559 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2560 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2561 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2562 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2563 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2564 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2565 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2566 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2567 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2568 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2571 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2573 /* There are two components to imsm platform support, the ahci SATA
2574 * controller and the option-rom. To find the SATA controller we
2575 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2576 * controller with the Intel vendor id is present. This approach
2577 * allows mdadm to leverage the kernel's ahci detection logic, with the
2578 * caveat that if ahci.ko is not loaded mdadm will not be able to
2579 * detect platform raid capabilities. The option-rom resides in a
2580 * platform "Adapter ROM". We scan for its signature to retrieve the
2581 * platform capabilities. If raid support is disabled in the BIOS the
2582 * option-rom capability structure will not be available.
2584 struct sys_dev
*list
, *hba
;
2589 if (enumerate_only
) {
2590 if (check_env("IMSM_NO_PLATFORM"))
2592 list
= find_intel_devices();
2595 for (hba
= list
; hba
; hba
= hba
->next
) {
2596 if (find_imsm_capability(hba
)) {
2606 list
= find_intel_devices();
2609 pr_err("no active Intel(R) RAID controller found.\n");
2611 } else if (verbose
> 0)
2612 print_found_intel_controllers(list
);
2614 for (hba
= list
; hba
; hba
= hba
->next
) {
2615 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2617 if (!find_imsm_capability(hba
)) {
2619 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2620 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2621 get_sys_dev_type(hba
->type
));
2627 if (controller_path
&& result
== 1) {
2628 pr_err("no active Intel(R) RAID controller found under %s\n",
2633 const struct orom_entry
*entry
;
2635 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2636 if (entry
->type
== SYS_DEV_VMD
) {
2637 print_imsm_capability(&entry
->orom
);
2638 printf(" 3rd party NVMe :%s supported\n",
2639 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2640 for (hba
= list
; hba
; hba
= hba
->next
) {
2641 if (hba
->type
== SYS_DEV_VMD
) {
2643 printf(" I/O Controller : %s (%s)\n",
2644 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2645 if (print_nvme_info(hba
)) {
2647 pr_err("failed to get devices attached to VMD domain.\n");
2656 print_imsm_capability(&entry
->orom
);
2657 if (entry
->type
== SYS_DEV_NVME
) {
2658 for (hba
= list
; hba
; hba
= hba
->next
) {
2659 if (hba
->type
== SYS_DEV_NVME
)
2660 print_nvme_info(hba
);
2666 struct devid_list
*devid
;
2667 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2668 hba
= device_by_id(devid
->devid
);
2672 printf(" I/O Controller : %s (%s)\n",
2673 hba
->path
, get_sys_dev_type(hba
->type
));
2674 if (hba
->type
== SYS_DEV_SATA
) {
2675 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2676 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2678 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2689 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2691 struct sys_dev
*list
, *hba
;
2694 list
= find_intel_devices();
2697 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2702 for (hba
= list
; hba
; hba
= hba
->next
) {
2703 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2705 if (!find_imsm_capability(hba
) && verbose
> 0) {
2707 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2708 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2714 const struct orom_entry
*entry
;
2716 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2717 if (entry
->type
== SYS_DEV_VMD
) {
2718 for (hba
= list
; hba
; hba
= hba
->next
)
2719 print_imsm_capability_export(&entry
->orom
);
2722 print_imsm_capability_export(&entry
->orom
);
2728 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2730 /* the imsm metadata format does not specify any host
2731 * identification information. We return -1 since we can never
2732 * confirm nor deny whether a given array is "meant" for this
2733 * host. We rely on compare_super and the 'family_num' fields to
2734 * exclude member disks that do not belong, and we rely on
2735 * mdadm.conf to specify the arrays that should be assembled.
2736 * Auto-assembly may still pick up "foreign" arrays.
2742 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2744 /* The uuid returned here is used for:
2745 * uuid to put into bitmap file (Create, Grow)
2746 * uuid for backup header when saving critical section (Grow)
2747 * comparing uuids when re-adding a device into an array
2748 * In these cases the uuid required is that of the data-array,
2749 * not the device-set.
2750 * uuid to recognise same set when adding a missing device back
2751 * to an array. This is a uuid for the device-set.
2753 * For each of these we can make do with a truncated
2754 * or hashed uuid rather than the original, as long as
2756 * In each case the uuid required is that of the data-array,
2757 * not the device-set.
2759 /* imsm does not track uuid's so we synthesis one using sha1 on
2760 * - The signature (Which is constant for all imsm array, but no matter)
2761 * - the orig_family_num of the container
2762 * - the index number of the volume
2763 * - the 'serial' number of the volume.
2764 * Hopefully these are all constant.
2766 struct intel_super
*super
= st
->sb
;
2769 struct sha1_ctx ctx
;
2770 struct imsm_dev
*dev
= NULL
;
2773 /* some mdadm versions failed to set ->orig_family_num, in which
2774 * case fall back to ->family_num. orig_family_num will be
2775 * fixed up with the first metadata update.
2777 family_num
= super
->anchor
->orig_family_num
;
2778 if (family_num
== 0)
2779 family_num
= super
->anchor
->family_num
;
2780 sha1_init_ctx(&ctx
);
2781 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2782 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2783 if (super
->current_vol
>= 0)
2784 dev
= get_imsm_dev(super
, super
->current_vol
);
2786 __u32 vol
= super
->current_vol
;
2787 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2788 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2790 sha1_finish_ctx(&ctx
, buf
);
2791 memcpy(uuid
, buf
, 4*4);
2796 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2798 __u8
*v
= get_imsm_version(mpb
);
2799 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2800 char major
[] = { 0, 0, 0 };
2801 char minor
[] = { 0 ,0, 0 };
2802 char patch
[] = { 0, 0, 0 };
2803 char *ver_parse
[] = { major
, minor
, patch
};
2807 while (*v
!= '\0' && v
< end
) {
2808 if (*v
!= '.' && j
< 2)
2809 ver_parse
[i
][j
++] = *v
;
2817 *m
= strtol(minor
, NULL
, 0);
2818 *p
= strtol(patch
, NULL
, 0);
2822 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2824 /* migr_strip_size when repairing or initializing parity */
2825 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2826 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2828 switch (get_imsm_raid_level(map
)) {
2833 return 128*1024 >> 9;
2837 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2839 /* migr_strip_size when rebuilding a degraded disk, no idea why
2840 * this is different than migr_strip_size_resync(), but it's good
2843 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2844 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2846 switch (get_imsm_raid_level(map
)) {
2849 if (map
->num_members
% map
->num_domains
== 0)
2850 return 128*1024 >> 9;
2854 return max((__u32
) 64*1024 >> 9, chunk
);
2856 return 128*1024 >> 9;
2860 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2862 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2863 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2864 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2865 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2867 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2870 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2872 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2873 int level
= get_imsm_raid_level(lo
);
2875 if (level
== 1 || level
== 10) {
2876 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2878 return hi
->num_domains
;
2880 return num_stripes_per_unit_resync(dev
);
2883 static __u8
imsm_num_data_members(struct imsm_map
*map
)
2885 /* named 'imsm_' because raid0, raid1 and raid10
2886 * counter-intuitively have the same number of data disks
2888 switch (get_imsm_raid_level(map
)) {
2890 return map
->num_members
;
2894 return map
->num_members
/2;
2896 return map
->num_members
- 1;
2898 dprintf("unsupported raid level\n");
2903 static unsigned long long calc_component_size(struct imsm_map
*map
,
2904 struct imsm_dev
*dev
)
2906 unsigned long long component_size
;
2907 unsigned long long dev_size
= imsm_dev_size(dev
);
2908 long long calc_dev_size
= 0;
2909 unsigned int member_disks
= imsm_num_data_members(map
);
2911 if (member_disks
== 0)
2914 component_size
= per_dev_array_size(map
);
2915 calc_dev_size
= component_size
* member_disks
;
2917 /* Component size is rounded to 1MB so difference between size from
2918 * metadata and size calculated from num_data_stripes equals up to
2919 * 2048 blocks per each device. If the difference is higher it means
2920 * that array size was expanded and num_data_stripes was not updated.
2922 if (llabs(calc_dev_size
- (long long)dev_size
) >
2923 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
2924 component_size
= dev_size
/ member_disks
;
2925 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
2926 component_size
/ map
->blocks_per_strip
,
2927 num_data_stripes(map
));
2930 return component_size
;
2933 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2935 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2936 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2938 switch(get_imsm_raid_level(map
)) {
2941 return chunk
* map
->num_domains
;
2943 return chunk
* map
->num_members
;
2949 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2951 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2952 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2953 __u32 strip
= block
/ chunk
;
2955 switch (get_imsm_raid_level(map
)) {
2958 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2959 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2961 return vol_stripe
* chunk
+ block
% chunk
;
2963 __u32 stripe
= strip
/ (map
->num_members
- 1);
2965 return stripe
* chunk
+ block
% chunk
;
2972 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2973 struct imsm_dev
*dev
)
2975 /* calculate the conversion factor between per member 'blocks'
2976 * (md/{resync,rebuild}_start) and imsm migration units, return
2977 * 0 for the 'not migrating' and 'unsupported migration' cases
2979 if (!dev
->vol
.migr_state
)
2982 switch (migr_type(dev
)) {
2983 case MIGR_GEN_MIGR
: {
2984 struct migr_record
*migr_rec
= super
->migr_rec
;
2985 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2990 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2991 __u32 stripes_per_unit
;
2992 __u32 blocks_per_unit
;
3001 /* yes, this is really the translation of migr_units to
3002 * per-member blocks in the 'resync' case
3004 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
3005 migr_chunk
= migr_strip_blocks_resync(dev
);
3006 disks
= imsm_num_data_members(map
);
3007 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
3008 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
3009 segment
= blocks_per_unit
/ stripe
;
3010 block_rel
= blocks_per_unit
- segment
* stripe
;
3011 parity_depth
= parity_segment_depth(dev
);
3012 block_map
= map_migr_block(dev
, block_rel
);
3013 return block_map
+ parity_depth
* segment
;
3015 case MIGR_REBUILD
: {
3016 __u32 stripes_per_unit
;
3019 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
3020 migr_chunk
= migr_strip_blocks_rebuild(dev
);
3021 return migr_chunk
* stripes_per_unit
;
3023 case MIGR_STATE_CHANGE
:
3029 static int imsm_level_to_layout(int level
)
3037 return ALGORITHM_LEFT_ASYMMETRIC
;
3044 /*******************************************************************************
3045 * Function: read_imsm_migr_rec
3046 * Description: Function reads imsm migration record from last sector of disk
3048 * fd : disk descriptor
3049 * super : metadata info
3053 ******************************************************************************/
3054 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
3057 unsigned int sector_size
= super
->sector_size
;
3058 unsigned long long dsize
;
3060 get_dev_size(fd
, NULL
, &dsize
);
3061 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
3063 pr_err("Cannot seek to anchor block: %s\n",
3067 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
3068 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3069 MIGR_REC_BUF_SECTORS
*sector_size
) {
3070 pr_err("Cannot read migr record block: %s\n",
3075 if (sector_size
== 4096)
3076 convert_from_4k_imsm_migr_rec(super
);
3082 static struct imsm_dev
*imsm_get_device_during_migration(
3083 struct intel_super
*super
)
3086 struct intel_dev
*dv
;
3088 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
3089 if (is_gen_migration(dv
->dev
))
3095 /*******************************************************************************
3096 * Function: load_imsm_migr_rec
3097 * Description: Function reads imsm migration record (it is stored at the last
3100 * super : imsm internal array info
3104 * -2 : no migration in progress
3105 ******************************************************************************/
3106 static int load_imsm_migr_rec(struct intel_super
*super
)
3112 struct imsm_dev
*dev
;
3113 struct imsm_map
*map
;
3117 /* find map under migration */
3118 dev
= imsm_get_device_during_migration(super
);
3119 /* nothing to load,no migration in progress?
3124 map
= get_imsm_map(dev
, MAP_0
);
3128 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3129 /* skip spare and failed disks
3133 /* read only from one of the first two slots
3135 slot
= get_imsm_disk_slot(map
, dl
->index
);
3136 if (slot
> 1 || slot
< 0)
3140 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3141 fd
= dev_open(nm
, O_RDONLY
);
3154 retval
= read_imsm_migr_rec(fd
, super
);
3161 /*******************************************************************************
3162 * function: imsm_create_metadata_checkpoint_update
3163 * Description: It creates update for checkpoint change.
3165 * super : imsm internal array info
3166 * u : pointer to prepared update
3169 * If length is equal to 0, input pointer u contains no update
3170 ******************************************************************************/
3171 static int imsm_create_metadata_checkpoint_update(
3172 struct intel_super
*super
,
3173 struct imsm_update_general_migration_checkpoint
**u
)
3176 int update_memory_size
= 0;
3178 dprintf("(enter)\n");
3184 /* size of all update data without anchor */
3185 update_memory_size
=
3186 sizeof(struct imsm_update_general_migration_checkpoint
);
3188 *u
= xcalloc(1, update_memory_size
);
3190 dprintf("error: cannot get memory\n");
3193 (*u
)->type
= update_general_migration_checkpoint
;
3194 (*u
)->curr_migr_unit
= current_migr_unit(super
->migr_rec
);
3195 dprintf("prepared for %llu\n", (*u
)->curr_migr_unit
);
3197 return update_memory_size
;
3200 static void imsm_update_metadata_locally(struct supertype
*st
,
3201 void *buf
, int len
);
3203 /*******************************************************************************
3204 * Function: write_imsm_migr_rec
3205 * Description: Function writes imsm migration record
3206 * (at the last sector of disk)
3208 * super : imsm internal array info
3212 ******************************************************************************/
3213 static int write_imsm_migr_rec(struct supertype
*st
)
3215 struct intel_super
*super
= st
->sb
;
3216 unsigned int sector_size
= super
->sector_size
;
3217 unsigned long long dsize
;
3221 struct imsm_update_general_migration_checkpoint
*u
;
3222 struct imsm_dev
*dev
;
3223 struct imsm_map
*map
;
3225 /* find map under migration */
3226 dev
= imsm_get_device_during_migration(super
);
3227 /* if no migration, write buffer anyway to clear migr_record
3228 * on disk based on first available device
3231 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3232 super
->current_vol
);
3234 map
= get_imsm_map(dev
, MAP_0
);
3236 if (sector_size
== 4096)
3237 convert_to_4k_imsm_migr_rec(super
);
3238 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3241 /* skip failed and spare devices */
3244 /* write to 2 first slots only */
3246 slot
= get_imsm_disk_slot(map
, sd
->index
);
3247 if (map
== NULL
|| slot
> 1 || slot
< 0)
3250 get_dev_size(sd
->fd
, NULL
, &dsize
);
3251 if (lseek64(sd
->fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*
3254 pr_err("Cannot seek to anchor block: %s\n",
3258 if ((unsigned int)write(sd
->fd
, super
->migr_rec_buf
,
3259 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3260 MIGR_REC_BUF_SECTORS
*sector_size
) {
3261 pr_err("Cannot write migr record block: %s\n",
3266 if (sector_size
== 4096)
3267 convert_from_4k_imsm_migr_rec(super
);
3268 /* update checkpoint information in metadata */
3269 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3271 dprintf("imsm: Cannot prepare update\n");
3274 /* update metadata locally */
3275 imsm_update_metadata_locally(st
, u
, len
);
3276 /* and possibly remotely */
3277 if (st
->update_tail
) {
3278 append_metadata_update(st
, u
, len
);
3279 /* during reshape we do all work inside metadata handler
3280 * manage_reshape(), so metadata update has to be triggered
3283 flush_metadata_updates(st
);
3284 st
->update_tail
= &st
->updates
;
3293 /* spare/missing disks activations are not allowe when
3294 * array/container performs reshape operation, because
3295 * all arrays in container works on the same disks set
3297 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3300 struct intel_dev
*i_dev
;
3301 struct imsm_dev
*dev
;
3303 /* check whole container
3305 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3307 if (is_gen_migration(dev
)) {
3308 /* No repair during any migration in container
3316 static unsigned long long imsm_component_size_alignment_check(int level
,
3318 unsigned int sector_size
,
3319 unsigned long long component_size
)
3321 unsigned int component_size_alignment
;
3323 /* check component size alignment
3325 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3327 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3328 level
, chunk_size
, component_size
,
3329 component_size_alignment
);
3331 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3332 dprintf("imsm: reported component size aligned from %llu ",
3334 component_size
-= component_size_alignment
;
3335 dprintf_cont("to %llu (%i).\n",
3336 component_size
, component_size_alignment
);
3339 return component_size
;
3342 /*******************************************************************************
3343 * Function: get_bitmap_header_sector
3344 * Description: Returns the sector where the bitmap header is placed.
3346 * st : supertype information
3347 * dev_idx : index of the device with bitmap
3350 * The sector where the bitmap header is placed
3351 ******************************************************************************/
3352 static unsigned long long get_bitmap_header_sector(struct intel_super
*super
,
3355 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3356 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3358 if (!super
->sector_size
) {
3359 dprintf("sector size is not set\n");
3363 return pba_of_lba0(map
) + calc_component_size(map
, dev
) +
3364 (IMSM_BITMAP_HEADER_OFFSET
/ super
->sector_size
);
3367 /*******************************************************************************
3368 * Function: get_bitmap_sector
3369 * Description: Returns the sector where the bitmap is placed.
3371 * st : supertype information
3372 * dev_idx : index of the device with bitmap
3375 * The sector where the bitmap is placed
3376 ******************************************************************************/
3377 static unsigned long long get_bitmap_sector(struct intel_super
*super
,
3380 if (!super
->sector_size
) {
3381 dprintf("sector size is not set\n");
3385 return get_bitmap_header_sector(super
, dev_idx
) +
3386 (IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
);
3389 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3391 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3392 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3394 return pba_of_lba0(map
) +
3395 (num_data_stripes(map
) * map
->blocks_per_strip
);
3398 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3400 struct intel_super
*super
= st
->sb
;
3401 struct migr_record
*migr_rec
= super
->migr_rec
;
3402 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3403 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3404 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3405 struct imsm_map
*map_to_analyse
= map
;
3407 int map_disks
= info
->array
.raid_disks
;
3409 memset(info
, 0, sizeof(*info
));
3411 map_to_analyse
= prev_map
;
3413 dl
= super
->current_disk
;
3415 info
->container_member
= super
->current_vol
;
3416 info
->array
.raid_disks
= map
->num_members
;
3417 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3418 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3419 info
->array
.md_minor
= -1;
3420 info
->array
.ctime
= 0;
3421 info
->array
.utime
= 0;
3422 info
->array
.chunk_size
=
3423 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3424 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3425 info
->custom_array_size
= imsm_dev_size(dev
);
3426 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3428 if (is_gen_migration(dev
)) {
3429 info
->reshape_active
= 1;
3430 info
->new_level
= get_imsm_raid_level(map
);
3431 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3432 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3433 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3434 if (info
->delta_disks
) {
3435 /* this needs to be applied to every array
3438 info
->reshape_active
= CONTAINER_RESHAPE
;
3440 /* We shape information that we give to md might have to be
3441 * modify to cope with md's requirement for reshaping arrays.
3442 * For example, when reshaping a RAID0, md requires it to be
3443 * presented as a degraded RAID4.
3444 * Also if a RAID0 is migrating to a RAID5 we need to specify
3445 * the array as already being RAID5, but the 'before' layout
3446 * is a RAID4-like layout.
3448 switch (info
->array
.level
) {
3450 switch(info
->new_level
) {
3452 /* conversion is happening as RAID4 */
3453 info
->array
.level
= 4;
3454 info
->array
.raid_disks
+= 1;
3457 /* conversion is happening as RAID5 */
3458 info
->array
.level
= 5;
3459 info
->array
.layout
= ALGORITHM_PARITY_N
;
3460 info
->delta_disks
-= 1;
3463 /* FIXME error message */
3464 info
->array
.level
= UnSet
;
3470 info
->new_level
= UnSet
;
3471 info
->new_layout
= UnSet
;
3472 info
->new_chunk
= info
->array
.chunk_size
;
3473 info
->delta_disks
= 0;
3477 info
->disk
.major
= dl
->major
;
3478 info
->disk
.minor
= dl
->minor
;
3479 info
->disk
.number
= dl
->index
;
3480 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3484 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3485 info
->component_size
= calc_component_size(map
, dev
);
3486 info
->component_size
= imsm_component_size_alignment_check(
3488 info
->array
.chunk_size
,
3490 info
->component_size
);
3491 info
->bb
.supported
= 1;
3493 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3494 info
->recovery_start
= MaxSector
;
3496 if (info
->array
.level
== 5 &&
3497 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3498 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3499 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3500 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3501 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3502 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3504 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3506 } else if (info
->array
.level
<= 0) {
3507 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3509 if (dev
->rwh_policy
== RWH_BITMAP
) {
3510 info
->bitmap_offset
= get_bitmap_sector(super
, super
->current_vol
);
3511 info
->consistency_policy
= CONSISTENCY_POLICY_BITMAP
;
3513 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3517 info
->reshape_progress
= 0;
3518 info
->resync_start
= MaxSector
;
3519 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3520 !(info
->array
.state
& 1)) &&
3521 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3522 info
->resync_start
= 0;
3524 if (dev
->vol
.migr_state
) {
3525 switch (migr_type(dev
)) {
3528 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3530 __u64 units
= vol_curr_migr_unit(dev
);
3532 info
->resync_start
= blocks_per_unit
* units
;
3535 case MIGR_GEN_MIGR
: {
3536 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3538 __u64 units
= current_migr_unit(migr_rec
);
3541 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3543 (get_num_migr_units(migr_rec
)-1)) &&
3544 (super
->migr_rec
->rec_status
==
3545 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3548 info
->reshape_progress
= blocks_per_unit
* units
;
3550 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3551 (unsigned long long)units
,
3552 (unsigned long long)blocks_per_unit
,
3553 info
->reshape_progress
);
3555 used_disks
= imsm_num_data_members(prev_map
);
3556 if (used_disks
> 0) {
3557 info
->custom_array_size
= per_dev_array_size(map
) *
3562 /* we could emulate the checkpointing of
3563 * 'sync_action=check' migrations, but for now
3564 * we just immediately complete them
3567 /* this is handled by container_content_imsm() */
3568 case MIGR_STATE_CHANGE
:
3569 /* FIXME handle other migrations */
3571 /* we are not dirty, so... */
3572 info
->resync_start
= MaxSector
;
3576 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3577 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3579 info
->array
.major_version
= -1;
3580 info
->array
.minor_version
= -2;
3581 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3582 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3583 uuid_from_super_imsm(st
, info
->uuid
);
3587 for (i
=0; i
<map_disks
; i
++) {
3589 if (i
< info
->array
.raid_disks
) {
3590 struct imsm_disk
*dsk
;
3591 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3592 dsk
= get_imsm_disk(super
, j
);
3593 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3600 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3601 int failed
, int look_in_map
);
3603 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3606 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3608 if (is_gen_migration(dev
)) {
3611 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3613 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3614 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3615 if (map2
->map_state
!= map_state
) {
3616 map2
->map_state
= map_state
;
3617 super
->updates_pending
++;
3622 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3626 for (d
= super
->missing
; d
; d
= d
->next
)
3627 if (d
->index
== index
)
3632 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3634 struct intel_super
*super
= st
->sb
;
3635 struct imsm_disk
*disk
;
3636 int map_disks
= info
->array
.raid_disks
;
3637 int max_enough
= -1;
3639 struct imsm_super
*mpb
;
3641 if (super
->current_vol
>= 0) {
3642 getinfo_super_imsm_volume(st
, info
, map
);
3645 memset(info
, 0, sizeof(*info
));
3647 /* Set raid_disks to zero so that Assemble will always pull in valid
3650 info
->array
.raid_disks
= 0;
3651 info
->array
.level
= LEVEL_CONTAINER
;
3652 info
->array
.layout
= 0;
3653 info
->array
.md_minor
= -1;
3654 info
->array
.ctime
= 0; /* N/A for imsm */
3655 info
->array
.utime
= 0;
3656 info
->array
.chunk_size
= 0;
3658 info
->disk
.major
= 0;
3659 info
->disk
.minor
= 0;
3660 info
->disk
.raid_disk
= -1;
3661 info
->reshape_active
= 0;
3662 info
->array
.major_version
= -1;
3663 info
->array
.minor_version
= -2;
3664 strcpy(info
->text_version
, "imsm");
3665 info
->safe_mode_delay
= 0;
3666 info
->disk
.number
= -1;
3667 info
->disk
.state
= 0;
3669 info
->recovery_start
= MaxSector
;
3670 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3671 info
->bb
.supported
= 1;
3673 /* do we have the all the insync disks that we expect? */
3674 mpb
= super
->anchor
;
3675 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3677 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3678 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3679 int failed
, enough
, j
, missing
= 0;
3680 struct imsm_map
*map
;
3683 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3684 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3685 map
= get_imsm_map(dev
, MAP_0
);
3687 /* any newly missing disks?
3688 * (catches single-degraded vs double-degraded)
3690 for (j
= 0; j
< map
->num_members
; j
++) {
3691 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3692 __u32 idx
= ord_to_idx(ord
);
3694 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3695 info
->disk
.raid_disk
= j
;
3697 if (!(ord
& IMSM_ORD_REBUILD
) &&
3698 get_imsm_missing(super
, idx
)) {
3704 if (state
== IMSM_T_STATE_FAILED
)
3706 else if (state
== IMSM_T_STATE_DEGRADED
&&
3707 (state
!= map
->map_state
|| missing
))
3709 else /* we're normal, or already degraded */
3711 if (is_gen_migration(dev
) && missing
) {
3712 /* during general migration we need all disks
3713 * that process is running on.
3714 * No new missing disk is allowed.
3718 /* no more checks necessary
3722 /* in the missing/failed disk case check to see
3723 * if at least one array is runnable
3725 max_enough
= max(max_enough
, enough
);
3727 dprintf("enough: %d\n", max_enough
);
3728 info
->container_enough
= max_enough
;
3731 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3733 disk
= &super
->disks
->disk
;
3734 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3735 info
->component_size
= reserved
;
3736 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3737 /* we don't change info->disk.raid_disk here because
3738 * this state will be finalized in mdmon after we have
3739 * found the 'most fresh' version of the metadata
3741 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3742 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3743 0 : (1 << MD_DISK_SYNC
);
3746 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3747 * ->compare_super may have updated the 'num_raid_devs' field for spares
3749 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3750 uuid_from_super_imsm(st
, info
->uuid
);
3752 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3754 /* I don't know how to compute 'map' on imsm, so use safe default */
3757 for (i
= 0; i
< map_disks
; i
++)
3763 /* allocates memory and fills disk in mdinfo structure
3764 * for each disk in array */
3765 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3767 struct mdinfo
*mddev
;
3768 struct intel_super
*super
= st
->sb
;
3769 struct imsm_disk
*disk
;
3772 if (!super
|| !super
->disks
)
3775 mddev
= xcalloc(1, sizeof(*mddev
));
3779 tmp
= xcalloc(1, sizeof(*tmp
));
3781 tmp
->next
= mddev
->devs
;
3783 tmp
->disk
.number
= count
++;
3784 tmp
->disk
.major
= dl
->major
;
3785 tmp
->disk
.minor
= dl
->minor
;
3786 tmp
->disk
.state
= is_configured(disk
) ?
3787 (1 << MD_DISK_ACTIVE
) : 0;
3788 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3789 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3790 tmp
->disk
.raid_disk
= -1;
3796 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3797 char *update
, char *devname
, int verbose
,
3798 int uuid_set
, char *homehost
)
3800 /* For 'assemble' and 'force' we need to return non-zero if any
3801 * change was made. For others, the return value is ignored.
3802 * Update options are:
3803 * force-one : This device looks a bit old but needs to be included,
3804 * update age info appropriately.
3805 * assemble: clear any 'faulty' flag to allow this device to
3807 * force-array: Array is degraded but being forced, mark it clean
3808 * if that will be needed to assemble it.
3810 * newdev: not used ????
3811 * grow: Array has gained a new device - this is currently for
3813 * resync: mark as dirty so a resync will happen.
3814 * name: update the name - preserving the homehost
3815 * uuid: Change the uuid of the array to match watch is given
3817 * Following are not relevant for this imsm:
3818 * sparc2.2 : update from old dodgey metadata
3819 * super-minor: change the preferred_minor number
3820 * summaries: update redundant counters.
3821 * homehost: update the recorded homehost
3822 * _reshape_progress: record new reshape_progress position.
3825 struct intel_super
*super
= st
->sb
;
3826 struct imsm_super
*mpb
;
3828 /* we can only update container info */
3829 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3832 mpb
= super
->anchor
;
3834 if (strcmp(update
, "uuid") == 0) {
3835 /* We take this to mean that the family_num should be updated.
3836 * However that is much smaller than the uuid so we cannot really
3837 * allow an explicit uuid to be given. And it is hard to reliably
3839 * So if !uuid_set we know the current uuid is random and just used
3840 * the first 'int' and copy it to the other 3 positions.
3841 * Otherwise we require the 4 'int's to be the same as would be the
3842 * case if we are using a random uuid. So an explicit uuid will be
3843 * accepted as long as all for ints are the same... which shouldn't hurt
3846 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3849 if (info
->uuid
[0] != info
->uuid
[1] ||
3850 info
->uuid
[1] != info
->uuid
[2] ||
3851 info
->uuid
[2] != info
->uuid
[3])
3857 mpb
->orig_family_num
= info
->uuid
[0];
3858 } else if (strcmp(update
, "assemble") == 0)
3863 /* successful update? recompute checksum */
3865 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3870 static size_t disks_to_mpb_size(int disks
)
3874 size
= sizeof(struct imsm_super
);
3875 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3876 size
+= 2 * sizeof(struct imsm_dev
);
3877 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3878 size
+= (4 - 2) * sizeof(struct imsm_map
);
3879 /* 4 possible disk_ord_tbl's */
3880 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3881 /* maximum bbm log */
3882 size
+= sizeof(struct bbm_log
);
3887 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3888 unsigned long long data_offset
)
3890 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3893 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3896 static void free_devlist(struct intel_super
*super
)
3898 struct intel_dev
*dv
;
3900 while (super
->devlist
) {
3901 dv
= super
->devlist
->next
;
3902 free(super
->devlist
->dev
);
3903 free(super
->devlist
);
3904 super
->devlist
= dv
;
3908 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3910 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3913 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
,
3918 * 0 same, or first was empty, and second was copied
3919 * 1 second had wrong number
3921 * 3 wrong other info
3923 struct intel_super
*first
= st
->sb
;
3924 struct intel_super
*sec
= tst
->sb
;
3931 /* in platform dependent environment test if the disks
3932 * use the same Intel hba
3933 * If not on Intel hba at all, allow anything.
3935 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3936 if (first
->hba
->type
!= sec
->hba
->type
) {
3938 pr_err("HBAs of devices do not match %s != %s\n",
3939 get_sys_dev_type(first
->hba
->type
),
3940 get_sys_dev_type(sec
->hba
->type
));
3944 if (first
->orom
!= sec
->orom
) {
3946 pr_err("HBAs of devices do not match %s != %s\n",
3947 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3953 /* if an anchor does not have num_raid_devs set then it is a free
3956 if (first
->anchor
->num_raid_devs
> 0 &&
3957 sec
->anchor
->num_raid_devs
> 0) {
3958 /* Determine if these disks might ever have been
3959 * related. Further disambiguation can only take place
3960 * in load_super_imsm_all
3962 __u32 first_family
= first
->anchor
->orig_family_num
;
3963 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3965 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3966 MAX_SIGNATURE_LENGTH
) != 0)
3969 if (first_family
== 0)
3970 first_family
= first
->anchor
->family_num
;
3971 if (sec_family
== 0)
3972 sec_family
= sec
->anchor
->family_num
;
3974 if (first_family
!= sec_family
)
3979 /* if 'first' is a spare promote it to a populated mpb with sec's
3982 if (first
->anchor
->num_raid_devs
== 0 &&
3983 sec
->anchor
->num_raid_devs
> 0) {
3985 struct intel_dev
*dv
;
3986 struct imsm_dev
*dev
;
3988 /* we need to copy raid device info from sec if an allocation
3989 * fails here we don't associate the spare
3991 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3992 dv
= xmalloc(sizeof(*dv
));
3993 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3996 dv
->next
= first
->devlist
;
3997 first
->devlist
= dv
;
3999 if (i
< sec
->anchor
->num_raid_devs
) {
4000 /* allocation failure */
4001 free_devlist(first
);
4002 pr_err("imsm: failed to associate spare\n");
4005 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
4006 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
4007 first
->anchor
->family_num
= sec
->anchor
->family_num
;
4008 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
4009 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
4010 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
4016 static void fd2devname(int fd
, char *name
)
4020 char dname
[PATH_MAX
];
4025 if (fstat(fd
, &st
) != 0)
4027 sprintf(path
, "/sys/dev/block/%d:%d",
4028 major(st
.st_rdev
), minor(st
.st_rdev
));
4030 rv
= readlink(path
, dname
, sizeof(dname
)-1);
4035 nm
= strrchr(dname
, '/');
4038 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
4042 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
4045 char *name
= fd2kname(fd
);
4050 if (strncmp(name
, "nvme", 4) != 0)
4053 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
4055 return load_sys(path
, buf
, buf_len
);
4058 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
4060 static int imsm_read_serial(int fd
, char *devname
,
4061 __u8
*serial
, size_t serial_buf_len
)
4070 memset(buf
, 0, sizeof(buf
));
4072 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
4075 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
4077 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
4078 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
4079 fd2devname(fd
, (char *) serial
);
4085 pr_err("Failed to retrieve serial for %s\n",
4090 /* trim all whitespace and non-printable characters and convert
4093 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4096 /* ':' is reserved for use in placeholder serial
4097 * numbers for missing disks
4108 if (len
> serial_buf_len
) {
4109 /* truncate leading characters */
4110 dest
+= len
- serial_buf_len
;
4111 len
= serial_buf_len
;
4114 memset(serial
, 0, serial_buf_len
);
4115 memcpy(serial
, dest
, len
);
4120 static int serialcmp(__u8
*s1
, __u8
*s2
)
4122 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4125 static void serialcpy(__u8
*dest
, __u8
*src
)
4127 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4130 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4134 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4135 if (serialcmp(dl
->serial
, serial
) == 0)
4141 static struct imsm_disk
*
4142 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4146 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4147 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4149 if (serialcmp(disk
->serial
, serial
) == 0) {
4160 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4162 struct imsm_disk
*disk
;
4167 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4169 rv
= imsm_read_serial(fd
, devname
, serial
, MAX_RAID_SERIAL_LEN
);
4174 dl
= xcalloc(1, sizeof(*dl
));
4177 dl
->major
= major(stb
.st_rdev
);
4178 dl
->minor
= minor(stb
.st_rdev
);
4179 dl
->next
= super
->disks
;
4180 dl
->fd
= keep_fd
? fd
: -1;
4181 assert(super
->disks
== NULL
);
4183 serialcpy(dl
->serial
, serial
);
4186 fd2devname(fd
, name
);
4188 dl
->devname
= xstrdup(devname
);
4190 dl
->devname
= xstrdup(name
);
4192 /* look up this disk's index in the current anchor */
4193 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4196 /* only set index on disks that are a member of a
4197 * populated contianer, i.e. one with raid_devs
4199 if (is_failed(&dl
->disk
))
4201 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4208 /* When migrating map0 contains the 'destination' state while map1
4209 * contains the current state. When not migrating map0 contains the
4210 * current state. This routine assumes that map[0].map_state is set to
4211 * the current array state before being called.
4213 * Migration is indicated by one of the following states
4214 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4215 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4216 * map1state=unitialized)
4217 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4219 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4220 * map1state=degraded)
4221 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4224 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4225 __u8 to_state
, int migr_type
)
4227 struct imsm_map
*dest
;
4228 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4230 dev
->vol
.migr_state
= 1;
4231 set_migr_type(dev
, migr_type
);
4232 set_vol_curr_migr_unit(dev
, 0);
4233 dest
= get_imsm_map(dev
, MAP_1
);
4235 /* duplicate and then set the target end state in map[0] */
4236 memcpy(dest
, src
, sizeof_imsm_map(src
));
4237 if (migr_type
== MIGR_GEN_MIGR
) {
4241 for (i
= 0; i
< src
->num_members
; i
++) {
4242 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4243 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4247 if (migr_type
== MIGR_GEN_MIGR
)
4248 /* Clear migration record */
4249 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4251 src
->map_state
= to_state
;
4254 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4257 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4258 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4262 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4263 * completed in the last migration.
4265 * FIXME add support for raid-level-migration
4267 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4268 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4269 /* when final map state is other than expected
4270 * merge maps (not for migration)
4274 for (i
= 0; i
< prev
->num_members
; i
++)
4275 for (j
= 0; j
< map
->num_members
; j
++)
4276 /* during online capacity expansion
4277 * disks position can be changed
4278 * if takeover is used
4280 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4281 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4282 map
->disk_ord_tbl
[j
] |=
4283 prev
->disk_ord_tbl
[i
];
4286 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4287 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4290 dev
->vol
.migr_state
= 0;
4291 set_migr_type(dev
, 0);
4292 set_vol_curr_migr_unit(dev
, 0);
4293 map
->map_state
= map_state
;
4296 static int parse_raid_devices(struct intel_super
*super
)
4299 struct imsm_dev
*dev_new
;
4300 size_t len
, len_migr
;
4302 size_t space_needed
= 0;
4303 struct imsm_super
*mpb
= super
->anchor
;
4305 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4306 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4307 struct intel_dev
*dv
;
4309 len
= sizeof_imsm_dev(dev_iter
, 0);
4310 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4312 space_needed
+= len_migr
- len
;
4314 dv
= xmalloc(sizeof(*dv
));
4315 if (max_len
< len_migr
)
4317 if (max_len
> len_migr
)
4318 space_needed
+= max_len
- len_migr
;
4319 dev_new
= xmalloc(max_len
);
4320 imsm_copy_dev(dev_new
, dev_iter
);
4323 dv
->next
= super
->devlist
;
4324 super
->devlist
= dv
;
4327 /* ensure that super->buf is large enough when all raid devices
4330 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4333 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4334 super
->sector_size
);
4335 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4338 memcpy(buf
, super
->buf
, super
->len
);
4339 memset(buf
+ super
->len
, 0, len
- super
->len
);
4345 super
->extra_space
+= space_needed
;
4350 /*******************************************************************************
4351 * Function: check_mpb_migr_compatibility
4352 * Description: Function checks for unsupported migration features:
4353 * - migration optimization area (pba_of_lba0)
4354 * - descending reshape (ascending_migr)
4356 * super : imsm metadata information
4358 * 0 : migration is compatible
4359 * -1 : migration is not compatible
4360 ******************************************************************************/
4361 int check_mpb_migr_compatibility(struct intel_super
*super
)
4363 struct imsm_map
*map0
, *map1
;
4364 struct migr_record
*migr_rec
= super
->migr_rec
;
4367 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4368 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4371 dev_iter
->vol
.migr_state
== 1 &&
4372 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4373 /* This device is migrating */
4374 map0
= get_imsm_map(dev_iter
, MAP_0
);
4375 map1
= get_imsm_map(dev_iter
, MAP_1
);
4376 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4377 /* migration optimization area was used */
4379 if (migr_rec
->ascending_migr
== 0 &&
4380 migr_rec
->dest_depth_per_unit
> 0)
4381 /* descending reshape not supported yet */
4388 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4390 /* load_imsm_mpb - read matrix metadata
4391 * allocates super->mpb to be freed by free_imsm
4393 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4395 unsigned long long dsize
;
4396 unsigned long long sectors
;
4397 unsigned int sector_size
= super
->sector_size
;
4399 struct imsm_super
*anchor
;
4402 get_dev_size(fd
, NULL
, &dsize
);
4403 if (dsize
< 2*sector_size
) {
4405 pr_err("%s: device to small for imsm\n",
4410 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4412 pr_err("Cannot seek to anchor block on %s: %s\n",
4413 devname
, strerror(errno
));
4417 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4419 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4422 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4424 pr_err("Cannot read anchor block on %s: %s\n",
4425 devname
, strerror(errno
));
4430 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4432 pr_err("no IMSM anchor on %s\n", devname
);
4437 __free_imsm(super
, 0);
4438 /* reload capability and hba */
4440 /* capability and hba must be updated with new super allocation */
4441 find_intel_hba_capability(fd
, super
, devname
);
4442 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4443 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4445 pr_err("unable to allocate %zu byte mpb buffer\n",
4450 memcpy(super
->buf
, anchor
, sector_size
);
4452 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4455 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4456 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4457 pr_err("could not allocate migr_rec buffer\n");
4461 super
->clean_migration_record_by_mdmon
= 0;
4464 check_sum
= __gen_imsm_checksum(super
->anchor
);
4465 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4467 pr_err("IMSM checksum %x != %x on %s\n",
4469 __le32_to_cpu(super
->anchor
->check_sum
),
4477 /* read the extended mpb */
4478 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4480 pr_err("Cannot seek to extended mpb on %s: %s\n",
4481 devname
, strerror(errno
));
4485 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4486 super
->len
- sector_size
) != super
->len
- sector_size
) {
4488 pr_err("Cannot read extended mpb on %s: %s\n",
4489 devname
, strerror(errno
));
4493 check_sum
= __gen_imsm_checksum(super
->anchor
);
4494 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4496 pr_err("IMSM checksum %x != %x on %s\n",
4497 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4505 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4507 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4508 static void clear_hi(struct intel_super
*super
)
4510 struct imsm_super
*mpb
= super
->anchor
;
4512 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4514 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4515 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4516 disk
->total_blocks_hi
= 0;
4518 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4519 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4522 for (n
= 0; n
< 2; ++n
) {
4523 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4526 map
->pba_of_lba0_hi
= 0;
4527 map
->blocks_per_member_hi
= 0;
4528 map
->num_data_stripes_hi
= 0;
4534 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4538 err
= load_imsm_mpb(fd
, super
, devname
);
4541 if (super
->sector_size
== 4096)
4542 convert_from_4k(super
);
4543 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4546 err
= parse_raid_devices(super
);
4549 err
= load_bbm_log(super
);
4554 static void __free_imsm_disk(struct dl
*d
)
4566 static void free_imsm_disks(struct intel_super
*super
)
4570 while (super
->disks
) {
4572 super
->disks
= d
->next
;
4573 __free_imsm_disk(d
);
4575 while (super
->disk_mgmt_list
) {
4576 d
= super
->disk_mgmt_list
;
4577 super
->disk_mgmt_list
= d
->next
;
4578 __free_imsm_disk(d
);
4580 while (super
->missing
) {
4582 super
->missing
= d
->next
;
4583 __free_imsm_disk(d
);
4588 /* free all the pieces hanging off of a super pointer */
4589 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4591 struct intel_hba
*elem
, *next
;
4597 /* unlink capability description */
4599 if (super
->migr_rec_buf
) {
4600 free(super
->migr_rec_buf
);
4601 super
->migr_rec_buf
= NULL
;
4604 free_imsm_disks(super
);
4605 free_devlist(super
);
4609 free((void *)elem
->path
);
4615 free(super
->bbm_log
);
4619 static void free_imsm(struct intel_super
*super
)
4621 __free_imsm(super
, 1);
4622 free(super
->bb
.entries
);
4626 static void free_super_imsm(struct supertype
*st
)
4628 struct intel_super
*super
= st
->sb
;
4637 static struct intel_super
*alloc_super(void)
4639 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4641 super
->current_vol
= -1;
4642 super
->create_offset
= ~((unsigned long long) 0);
4644 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4645 sizeof(struct md_bb_entry
));
4646 if (!super
->bb
.entries
) {
4655 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4657 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4659 struct sys_dev
*hba_name
;
4662 if (fd
>= 0 && test_partition(fd
)) {
4663 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4667 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4672 hba_name
= find_disk_attached_hba(fd
, NULL
);
4675 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4679 rv
= attach_hba_to_super(super
, hba_name
);
4682 struct intel_hba
*hba
= super
->hba
;
4684 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4685 " but the container is assigned to Intel(R) %s %s (",
4687 get_sys_dev_type(hba_name
->type
),
4688 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4689 hba_name
->pci_id
? : "Err!",
4690 get_sys_dev_type(super
->hba
->type
),
4691 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4694 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4696 fprintf(stderr
, ", ");
4699 fprintf(stderr
, ").\n"
4700 " Mixing devices attached to different controllers is not allowed.\n");
4704 super
->orom
= find_imsm_capability(hba_name
);
4711 /* find_missing - helper routine for load_super_imsm_all that identifies
4712 * disks that have disappeared from the system. This routine relies on
4713 * the mpb being uptodate, which it is at load time.
4715 static int find_missing(struct intel_super
*super
)
4718 struct imsm_super
*mpb
= super
->anchor
;
4720 struct imsm_disk
*disk
;
4722 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4723 disk
= __get_imsm_disk(mpb
, i
);
4724 dl
= serial_to_dl(disk
->serial
, super
);
4728 dl
= xmalloc(sizeof(*dl
));
4732 dl
->devname
= xstrdup("missing");
4734 serialcpy(dl
->serial
, disk
->serial
);
4737 dl
->next
= super
->missing
;
4738 super
->missing
= dl
;
4744 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4746 struct intel_disk
*idisk
= disk_list
;
4749 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4751 idisk
= idisk
->next
;
4757 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4758 struct intel_super
*super
,
4759 struct intel_disk
**disk_list
)
4761 struct imsm_disk
*d
= &super
->disks
->disk
;
4762 struct imsm_super
*mpb
= super
->anchor
;
4765 for (i
= 0; i
< tbl_size
; i
++) {
4766 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4767 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4769 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4770 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4771 dprintf("mpb from %d:%d matches %d:%d\n",
4772 super
->disks
->major
,
4773 super
->disks
->minor
,
4774 table
[i
]->disks
->major
,
4775 table
[i
]->disks
->minor
);
4779 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4780 is_configured(d
) == is_configured(tbl_d
)) &&
4781 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4782 /* current version of the mpb is a
4783 * better candidate than the one in
4784 * super_table, but copy over "cross
4785 * generational" status
4787 struct intel_disk
*idisk
;
4789 dprintf("mpb from %d:%d replaces %d:%d\n",
4790 super
->disks
->major
,
4791 super
->disks
->minor
,
4792 table
[i
]->disks
->major
,
4793 table
[i
]->disks
->minor
);
4795 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4796 if (idisk
&& is_failed(&idisk
->disk
))
4797 tbl_d
->status
|= FAILED_DISK
;
4800 struct intel_disk
*idisk
;
4801 struct imsm_disk
*disk
;
4803 /* tbl_mpb is more up to date, but copy
4804 * over cross generational status before
4807 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4808 if (disk
&& is_failed(disk
))
4809 d
->status
|= FAILED_DISK
;
4811 idisk
= disk_list_get(d
->serial
, *disk_list
);
4814 if (disk
&& is_configured(disk
))
4815 idisk
->disk
.status
|= CONFIGURED_DISK
;
4818 dprintf("mpb from %d:%d prefer %d:%d\n",
4819 super
->disks
->major
,
4820 super
->disks
->minor
,
4821 table
[i
]->disks
->major
,
4822 table
[i
]->disks
->minor
);
4830 table
[tbl_size
++] = super
;
4834 /* update/extend the merged list of imsm_disk records */
4835 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4836 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4837 struct intel_disk
*idisk
;
4839 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4841 idisk
->disk
.status
|= disk
->status
;
4842 if (is_configured(&idisk
->disk
) ||
4843 is_failed(&idisk
->disk
))
4844 idisk
->disk
.status
&= ~(SPARE_DISK
);
4846 idisk
= xcalloc(1, sizeof(*idisk
));
4847 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4848 idisk
->disk
= *disk
;
4849 idisk
->next
= *disk_list
;
4853 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4860 static struct intel_super
*
4861 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4864 struct imsm_super
*mpb
= super
->anchor
;
4868 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4869 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4870 struct intel_disk
*idisk
;
4872 idisk
= disk_list_get(disk
->serial
, disk_list
);
4874 if (idisk
->owner
== owner
||
4875 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4878 dprintf("'%.16s' owner %d != %d\n",
4879 disk
->serial
, idisk
->owner
,
4882 dprintf("unknown disk %x [%d]: %.16s\n",
4883 __le32_to_cpu(mpb
->family_num
), i
,
4889 if (ok_count
== mpb
->num_disks
)
4894 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4896 struct intel_super
*s
;
4898 for (s
= super_list
; s
; s
= s
->next
) {
4899 if (family_num
!= s
->anchor
->family_num
)
4901 pr_err("Conflict, offlining family %#x on '%s'\n",
4902 __le32_to_cpu(family_num
), s
->disks
->devname
);
4906 static struct intel_super
*
4907 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4909 struct intel_super
*super_table
[len
];
4910 struct intel_disk
*disk_list
= NULL
;
4911 struct intel_super
*champion
, *spare
;
4912 struct intel_super
*s
, **del
;
4917 memset(super_table
, 0, sizeof(super_table
));
4918 for (s
= *super_list
; s
; s
= s
->next
)
4919 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4921 for (i
= 0; i
< tbl_size
; i
++) {
4922 struct imsm_disk
*d
;
4923 struct intel_disk
*idisk
;
4924 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4927 d
= &s
->disks
->disk
;
4929 /* 'd' must appear in merged disk list for its
4930 * configuration to be valid
4932 idisk
= disk_list_get(d
->serial
, disk_list
);
4933 if (idisk
&& idisk
->owner
== i
)
4934 s
= validate_members(s
, disk_list
, i
);
4939 dprintf("marking family: %#x from %d:%d offline\n",
4941 super_table
[i
]->disks
->major
,
4942 super_table
[i
]->disks
->minor
);
4946 /* This is where the mdadm implementation differs from the Windows
4947 * driver which has no strict concept of a container. We can only
4948 * assemble one family from a container, so when returning a prodigal
4949 * array member to this system the code will not be able to disambiguate
4950 * the container contents that should be assembled ("foreign" versus
4951 * "local"). It requires user intervention to set the orig_family_num
4952 * to a new value to establish a new container. The Windows driver in
4953 * this situation fixes up the volume name in place and manages the
4954 * foreign array as an independent entity.
4959 for (i
= 0; i
< tbl_size
; i
++) {
4960 struct intel_super
*tbl_ent
= super_table
[i
];
4966 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4971 if (s
&& !is_spare
) {
4972 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4974 } else if (!s
&& !is_spare
)
4987 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4988 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4990 /* collect all dl's onto 'champion', and update them to
4991 * champion's version of the status
4993 for (s
= *super_list
; s
; s
= s
->next
) {
4994 struct imsm_super
*mpb
= champion
->anchor
;
4995 struct dl
*dl
= s
->disks
;
5000 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
5002 for (i
= 0; i
< mpb
->num_disks
; i
++) {
5003 struct imsm_disk
*disk
;
5005 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
5008 /* only set index on disks that are a member of
5009 * a populated contianer, i.e. one with
5012 if (is_failed(&dl
->disk
))
5014 else if (is_spare(&dl
->disk
))
5020 if (i
>= mpb
->num_disks
) {
5021 struct intel_disk
*idisk
;
5023 idisk
= disk_list_get(dl
->serial
, disk_list
);
5024 if (idisk
&& is_spare(&idisk
->disk
) &&
5025 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
5033 dl
->next
= champion
->disks
;
5034 champion
->disks
= dl
;
5038 /* delete 'champion' from super_list */
5039 for (del
= super_list
; *del
; ) {
5040 if (*del
== champion
) {
5041 *del
= (*del
)->next
;
5044 del
= &(*del
)->next
;
5046 champion
->next
= NULL
;
5050 struct intel_disk
*idisk
= disk_list
;
5052 disk_list
= disk_list
->next
;
5060 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
5061 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5062 int major
, int minor
, int keep_fd
);
5064 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5065 int *max
, int keep_fd
);
5067 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
5068 char *devname
, struct md_list
*devlist
,
5071 struct intel_super
*super_list
= NULL
;
5072 struct intel_super
*super
= NULL
;
5077 /* 'fd' is an opened container */
5078 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
5080 /* get super block from devlist devices */
5081 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
5084 /* all mpbs enter, maybe one leaves */
5085 super
= imsm_thunderdome(&super_list
, i
);
5091 if (find_missing(super
) != 0) {
5097 /* load migration record */
5098 err
= load_imsm_migr_rec(super
);
5100 /* migration is in progress,
5101 * but migr_rec cannot be loaded,
5107 /* Check migration compatibility */
5108 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5109 pr_err("Unsupported migration detected");
5111 fprintf(stderr
, " on %s\n", devname
);
5113 fprintf(stderr
, " (IMSM).\n");
5122 while (super_list
) {
5123 struct intel_super
*s
= super_list
;
5125 super_list
= super_list
->next
;
5134 strcpy(st
->container_devnm
, fd2devnm(fd
));
5136 st
->container_devnm
[0] = 0;
5137 if (err
== 0 && st
->ss
== NULL
) {
5138 st
->ss
= &super_imsm
;
5139 st
->minor_version
= 0;
5140 st
->max_devs
= IMSM_MAX_DEVICES
;
5146 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5147 int *max
, int keep_fd
)
5149 struct md_list
*tmpdev
;
5153 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5154 if (tmpdev
->used
!= 1)
5156 if (tmpdev
->container
== 1) {
5158 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5160 pr_err("cannot open device %s: %s\n",
5161 tmpdev
->devname
, strerror(errno
));
5165 err
= get_sra_super_block(fd
, super_list
,
5166 tmpdev
->devname
, &lmax
,
5175 int major
= major(tmpdev
->st_rdev
);
5176 int minor
= minor(tmpdev
->st_rdev
);
5177 err
= get_super_block(super_list
,
5194 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5195 int major
, int minor
, int keep_fd
)
5197 struct intel_super
*s
;
5209 sprintf(nm
, "%d:%d", major
, minor
);
5210 dfd
= dev_open(nm
, O_RDWR
);
5216 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5217 find_intel_hba_capability(dfd
, s
, devname
);
5218 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5220 /* retry the load if we might have raced against mdmon */
5221 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5222 for (retry
= 0; retry
< 3; retry
++) {
5224 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5230 s
->next
= *super_list
;
5238 if (dfd
>= 0 && !keep_fd
)
5245 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5252 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5256 if (sra
->array
.major_version
!= -1 ||
5257 sra
->array
.minor_version
!= -2 ||
5258 strcmp(sra
->text_version
, "imsm") != 0) {
5263 devnm
= fd2devnm(fd
);
5264 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5265 if (get_super_block(super_list
, devnm
, devname
,
5266 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5277 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5279 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5282 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5284 struct intel_super
*super
;
5288 if (test_partition(fd
))
5289 /* IMSM not allowed on partitions */
5292 free_super_imsm(st
);
5294 super
= alloc_super();
5295 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5298 /* Load hba and capabilities if they exist.
5299 * But do not preclude loading metadata in case capabilities or hba are
5300 * non-compliant and ignore_hw_compat is set.
5302 rv
= find_intel_hba_capability(fd
, super
, devname
);
5303 /* no orom/efi or non-intel hba of the disk */
5304 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5306 pr_err("No OROM/EFI properties for %s\n", devname
);
5310 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5312 /* retry the load if we might have raced against mdmon */
5314 struct mdstat_ent
*mdstat
= NULL
;
5315 char *name
= fd2kname(fd
);
5318 mdstat
= mdstat_by_component(name
);
5320 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5321 for (retry
= 0; retry
< 3; retry
++) {
5323 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5329 free_mdstat(mdstat
);
5334 pr_err("Failed to load all information sections on %s\n", devname
);
5340 if (st
->ss
== NULL
) {
5341 st
->ss
= &super_imsm
;
5342 st
->minor_version
= 0;
5343 st
->max_devs
= IMSM_MAX_DEVICES
;
5346 /* load migration record */
5347 if (load_imsm_migr_rec(super
) == 0) {
5348 /* Check for unsupported migration features */
5349 if (check_mpb_migr_compatibility(super
) != 0) {
5350 pr_err("Unsupported migration detected");
5352 fprintf(stderr
, " on %s\n", devname
);
5354 fprintf(stderr
, " (IMSM).\n");
5362 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5364 if (info
->level
== 1)
5366 return info
->chunk_size
>> 9;
5369 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5370 unsigned long long size
)
5372 if (info
->level
== 1)
5375 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5378 static void imsm_update_version_info(struct intel_super
*super
)
5380 /* update the version and attributes */
5381 struct imsm_super
*mpb
= super
->anchor
;
5383 struct imsm_dev
*dev
;
5384 struct imsm_map
*map
;
5387 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5388 dev
= get_imsm_dev(super
, i
);
5389 map
= get_imsm_map(dev
, MAP_0
);
5390 if (__le32_to_cpu(dev
->size_high
) > 0)
5391 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5393 /* FIXME detect when an array spans a port multiplier */
5395 mpb
->attributes
|= MPB_ATTRIB_PM
;
5398 if (mpb
->num_raid_devs
> 1 ||
5399 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5400 version
= MPB_VERSION_ATTRIBS
;
5401 switch (get_imsm_raid_level(map
)) {
5402 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5403 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5404 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5405 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5408 if (map
->num_members
>= 5)
5409 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5410 else if (dev
->status
== DEV_CLONE_N_GO
)
5411 version
= MPB_VERSION_CNG
;
5412 else if (get_imsm_raid_level(map
) == 5)
5413 version
= MPB_VERSION_RAID5
;
5414 else if (map
->num_members
>= 3)
5415 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5416 else if (get_imsm_raid_level(map
) == 1)
5417 version
= MPB_VERSION_RAID1
;
5419 version
= MPB_VERSION_RAID0
;
5421 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5425 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5427 struct imsm_super
*mpb
= super
->anchor
;
5428 char *reason
= NULL
;
5430 size_t len
= strlen(name
);
5434 while (isspace(start
[len
- 1]))
5436 while (*start
&& isspace(*start
))
5438 memmove(name
, start
, len
+ 1);
5441 if (len
> MAX_RAID_SERIAL_LEN
)
5442 reason
= "must be 16 characters or less";
5444 reason
= "must be a non-empty string";
5446 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5447 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5449 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5450 reason
= "already exists";
5455 if (reason
&& !quiet
)
5456 pr_err("imsm volume name %s\n", reason
);
5461 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5462 struct shape
*s
, char *name
,
5463 char *homehost
, int *uuid
,
5464 long long data_offset
)
5466 /* We are creating a volume inside a pre-existing container.
5467 * so st->sb is already set.
5469 struct intel_super
*super
= st
->sb
;
5470 unsigned int sector_size
= super
->sector_size
;
5471 struct imsm_super
*mpb
= super
->anchor
;
5472 struct intel_dev
*dv
;
5473 struct imsm_dev
*dev
;
5474 struct imsm_vol
*vol
;
5475 struct imsm_map
*map
;
5476 int idx
= mpb
->num_raid_devs
;
5479 unsigned long long array_blocks
;
5480 size_t size_old
, size_new
;
5481 unsigned long long num_data_stripes
;
5482 unsigned int data_disks
;
5483 unsigned long long size_per_member
;
5485 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5486 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5490 /* ensure the mpb is large enough for the new data */
5491 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5492 size_new
= disks_to_mpb_size(info
->nr_disks
);
5493 if (size_new
> size_old
) {
5495 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5497 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5498 pr_err("could not allocate new mpb\n");
5501 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5502 MIGR_REC_BUF_SECTORS
*
5503 MAX_SECTOR_SIZE
) != 0) {
5504 pr_err("could not allocate migr_rec buffer\n");
5510 memcpy(mpb_new
, mpb
, size_old
);
5513 super
->anchor
= mpb_new
;
5514 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5515 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5516 super
->len
= size_round
;
5518 super
->current_vol
= idx
;
5520 /* handle 'failed_disks' by either:
5521 * a) create dummy disk entries in the table if this the first
5522 * volume in the array. We add them here as this is the only
5523 * opportunity to add them. add_to_super_imsm_volume()
5524 * handles the non-failed disks and continues incrementing
5526 * b) validate that 'failed_disks' matches the current number
5527 * of missing disks if the container is populated
5529 if (super
->current_vol
== 0) {
5531 for (i
= 0; i
< info
->failed_disks
; i
++) {
5532 struct imsm_disk
*disk
;
5535 disk
= __get_imsm_disk(mpb
, i
);
5536 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5537 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5538 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5539 "missing:%d", (__u8
)i
);
5541 find_missing(super
);
5546 for (d
= super
->missing
; d
; d
= d
->next
)
5548 if (info
->failed_disks
> missing
) {
5549 pr_err("unable to add 'missing' disk to container\n");
5554 if (!check_name(super
, name
, 0))
5556 dv
= xmalloc(sizeof(*dv
));
5557 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5559 * Explicitly allow truncating to not confuse gcc's
5560 * -Werror=stringop-truncation
5562 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5563 memcpy(dev
->volume
, name
, namelen
);
5564 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5565 info
->layout
, info
->chunk_size
,
5566 s
->size
* BLOCKS_PER_KB
);
5567 data_disks
= get_data_disks(info
->level
, info
->layout
,
5569 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5570 size_per_member
= array_blocks
/ data_disks
;
5572 set_imsm_dev_size(dev
, array_blocks
);
5573 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5575 vol
->migr_state
= 0;
5576 set_migr_type(dev
, MIGR_INIT
);
5577 vol
->dirty
= !info
->state
;
5578 set_vol_curr_migr_unit(dev
, 0);
5579 map
= get_imsm_map(dev
, MAP_0
);
5580 set_pba_of_lba0(map
, super
->create_offset
);
5581 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5582 map
->failed_disk_num
= ~0;
5583 if (info
->level
> 0)
5584 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5585 : IMSM_T_STATE_UNINITIALIZED
);
5587 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5588 IMSM_T_STATE_NORMAL
;
5591 if (info
->level
== 1 && info
->raid_disks
> 2) {
5594 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5598 map
->raid_level
= info
->level
;
5599 if (info
->level
== 10) {
5600 map
->raid_level
= 1;
5601 map
->num_domains
= info
->raid_disks
/ 2;
5602 } else if (info
->level
== 1)
5603 map
->num_domains
= info
->raid_disks
;
5605 map
->num_domains
= 1;
5607 /* info->size is only int so use the 'size' parameter instead */
5608 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5609 num_data_stripes
/= map
->num_domains
;
5610 set_num_data_stripes(map
, num_data_stripes
);
5612 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5613 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5617 map
->num_members
= info
->raid_disks
;
5618 for (i
= 0; i
< map
->num_members
; i
++) {
5619 /* initialized in add_to_super */
5620 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5622 mpb
->num_raid_devs
++;
5623 mpb
->num_raid_devs_created
++;
5624 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5626 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5627 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5628 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5629 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5633 pr_err("imsm does not support consistency policy %s\n",
5634 map_num(consistency_policies
, s
->consistency_policy
));
5639 dv
->index
= super
->current_vol
;
5640 dv
->next
= super
->devlist
;
5641 super
->devlist
= dv
;
5643 imsm_update_version_info(super
);
5648 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5649 struct shape
*s
, char *name
,
5650 char *homehost
, int *uuid
,
5651 unsigned long long data_offset
)
5653 /* This is primarily called by Create when creating a new array.
5654 * We will then get add_to_super called for each component, and then
5655 * write_init_super called to write it out to each device.
5656 * For IMSM, Create can create on fresh devices or on a pre-existing
5658 * To create on a pre-existing array a different method will be called.
5659 * This one is just for fresh drives.
5661 struct intel_super
*super
;
5662 struct imsm_super
*mpb
;
5666 if (data_offset
!= INVALID_SECTORS
) {
5667 pr_err("data-offset not supported by imsm\n");
5672 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5676 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5678 mpb_size
= MAX_SECTOR_SIZE
;
5680 super
= alloc_super();
5682 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5687 pr_err("could not allocate superblock\n");
5690 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5691 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5692 pr_err("could not allocate migr_rec buffer\n");
5697 memset(super
->buf
, 0, mpb_size
);
5699 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5703 /* zeroing superblock */
5707 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5709 version
= (char *) mpb
->sig
;
5710 strcpy(version
, MPB_SIGNATURE
);
5711 version
+= strlen(MPB_SIGNATURE
);
5712 strcpy(version
, MPB_VERSION_RAID0
);
5717 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5719 unsigned int member_sector_size
;
5722 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5726 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5728 if (member_sector_size
!= super
->sector_size
)
5733 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5734 int fd
, char *devname
)
5736 struct intel_super
*super
= st
->sb
;
5737 struct imsm_super
*mpb
= super
->anchor
;
5738 struct imsm_disk
*_disk
;
5739 struct imsm_dev
*dev
;
5740 struct imsm_map
*map
;
5744 dev
= get_imsm_dev(super
, super
->current_vol
);
5745 map
= get_imsm_map(dev
, MAP_0
);
5747 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5748 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5754 /* we're doing autolayout so grab the pre-marked (in
5755 * validate_geometry) raid_disk
5757 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5758 if (dl
->raiddisk
== dk
->raid_disk
)
5761 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5762 if (dl
->major
== dk
->major
&&
5763 dl
->minor
== dk
->minor
)
5768 pr_err("%s is not a member of the same container\n", devname
);
5772 if (mpb
->num_disks
== 0)
5773 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5774 &super
->sector_size
))
5777 if (!drive_validate_sector_size(super
, dl
)) {
5778 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5782 /* add a pristine spare to the metadata */
5783 if (dl
->index
< 0) {
5784 dl
->index
= super
->anchor
->num_disks
;
5785 super
->anchor
->num_disks
++;
5787 /* Check the device has not already been added */
5788 slot
= get_imsm_disk_slot(map
, dl
->index
);
5790 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5791 pr_err("%s has been included in this array twice\n",
5795 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5796 dl
->disk
.status
= CONFIGURED_DISK
;
5798 /* update size of 'missing' disks to be at least as large as the
5799 * largest acitve member (we only have dummy missing disks when
5800 * creating the first volume)
5802 if (super
->current_vol
== 0) {
5803 for (df
= super
->missing
; df
; df
= df
->next
) {
5804 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5805 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5806 _disk
= __get_imsm_disk(mpb
, df
->index
);
5811 /* refresh unset/failed slots to point to valid 'missing' entries */
5812 for (df
= super
->missing
; df
; df
= df
->next
)
5813 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5814 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5816 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5818 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5819 if (is_gen_migration(dev
)) {
5820 struct imsm_map
*map2
= get_imsm_map(dev
,
5822 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5823 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5824 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5827 if ((unsigned)df
->index
==
5829 set_imsm_ord_tbl_ent(map2
,
5835 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5839 /* if we are creating the first raid device update the family number */
5840 if (super
->current_vol
== 0) {
5842 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5844 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5845 if (!_dev
|| !_disk
) {
5846 pr_err("BUG mpb setup error\n");
5852 sum
+= __gen_imsm_checksum(mpb
);
5853 mpb
->family_num
= __cpu_to_le32(sum
);
5854 mpb
->orig_family_num
= mpb
->family_num
;
5855 mpb
->creation_time
= __cpu_to_le64((__u64
)time(NULL
));
5857 super
->current_disk
= dl
;
5862 * Function marks disk as spare and restores disk serial
5863 * in case it was previously marked as failed by takeover operation
5865 * -1 : critical error
5866 * 0 : disk is marked as spare but serial is not set
5869 int mark_spare(struct dl
*disk
)
5871 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5878 if (!imsm_read_serial(disk
->fd
, NULL
, serial
, MAX_RAID_SERIAL_LEN
)) {
5879 /* Restore disk serial number, because takeover marks disk
5880 * as failed and adds to serial ':0' before it becomes
5883 serialcpy(disk
->serial
, serial
);
5884 serialcpy(disk
->disk
.serial
, serial
);
5887 disk
->disk
.status
= SPARE_DISK
;
5894 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
);
5896 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5897 int fd
, char *devname
,
5898 unsigned long long data_offset
)
5900 struct intel_super
*super
= st
->sb
;
5902 unsigned long long size
;
5903 unsigned int member_sector_size
;
5908 /* If we are on an RAID enabled platform check that the disk is
5909 * attached to the raid controller.
5910 * We do not need to test disks attachment for container based additions,
5911 * they shall be already tested when container was created/assembled.
5913 rv
= find_intel_hba_capability(fd
, super
, devname
);
5914 /* no orom/efi or non-intel hba of the disk */
5916 dprintf("capability: %p fd: %d ret: %d\n",
5917 super
->orom
, fd
, rv
);
5921 if (super
->current_vol
>= 0)
5922 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5925 dd
= xcalloc(sizeof(*dd
), 1);
5926 dd
->major
= major(stb
.st_rdev
);
5927 dd
->minor
= minor(stb
.st_rdev
);
5928 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5931 dd
->action
= DISK_ADD
;
5932 rv
= imsm_read_serial(fd
, devname
, dd
->serial
, MAX_RAID_SERIAL_LEN
);
5934 pr_err("failed to retrieve scsi serial, aborting\n");
5940 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5941 (super
->hba
->type
== SYS_DEV_VMD
))) {
5943 char *devpath
= diskfd_to_devpath(fd
);
5944 char controller_path
[PATH_MAX
];
5945 char *controller_name
;
5948 pr_err("failed to get devpath, aborting\n");
5955 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5957 controller_name
= basename(devpath
);
5958 if (is_multipath_nvme(fd
))
5959 pr_err("%s controller supports Multi-Path I/O, Intel (R) VROC does not support multipathing\n", controller_name
);
5963 if (!imsm_is_nvme_supported(dd
->fd
, 1)) {
5970 if (devpath_to_vendor(controller_path
) == 0x8086) {
5972 * If Intel's NVMe drive has serial ended with
5973 * "-A","-B","-1" or "-2" it means that this is "x8"
5974 * device (double drive on single PCIe card).
5975 * User should be warned about potential data loss.
5977 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5978 /* Skip empty character at the end */
5979 if (dd
->serial
[i
] == 0)
5982 if (((dd
->serial
[i
] == 'A') ||
5983 (dd
->serial
[i
] == 'B') ||
5984 (dd
->serial
[i
] == '1') ||
5985 (dd
->serial
[i
] == '2')) &&
5986 (dd
->serial
[i
-1] == '-'))
5987 pr_err("\tThe action you are about to take may put your data at risk.\n"
5988 "\tPlease note that x8 devices may consist of two separate x4 devices "
5989 "located on a single PCIe port.\n"
5990 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5993 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5994 !imsm_orom_has_tpv_support(super
->orom
)) {
5995 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5996 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
6003 get_dev_size(fd
, NULL
, &size
);
6004 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
6006 if (super
->sector_size
== 0) {
6007 /* this a first device, so sector_size is not set yet */
6008 super
->sector_size
= member_sector_size
;
6011 /* clear migr_rec when adding disk to container */
6012 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6013 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
6015 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
6016 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
6017 MIGR_REC_BUF_SECTORS
*member_sector_size
)
6018 perror("Write migr_rec failed");
6022 serialcpy(dd
->disk
.serial
, dd
->serial
);
6023 set_total_blocks(&dd
->disk
, size
);
6024 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
6025 struct imsm_super
*mpb
= super
->anchor
;
6026 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6029 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
6030 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
6032 dd
->disk
.scsi_id
= __cpu_to_le32(0);
6034 if (st
->update_tail
) {
6035 dd
->next
= super
->disk_mgmt_list
;
6036 super
->disk_mgmt_list
= dd
;
6038 /* this is called outside of mdmon
6039 * write initial spare metadata
6040 * mdmon will overwrite it.
6042 dd
->next
= super
->disks
;
6044 write_super_imsm_spare(super
, dd
);
6050 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
6052 struct intel_super
*super
= st
->sb
;
6055 /* remove from super works only in mdmon - for communication
6056 * manager - monitor. Check if communication memory buffer
6059 if (!st
->update_tail
) {
6060 pr_err("shall be used in mdmon context only\n");
6063 dd
= xcalloc(1, sizeof(*dd
));
6064 dd
->major
= dk
->major
;
6065 dd
->minor
= dk
->minor
;
6068 dd
->action
= DISK_REMOVE
;
6070 dd
->next
= super
->disk_mgmt_list
;
6071 super
->disk_mgmt_list
= dd
;
6076 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
6079 char buf
[MAX_SECTOR_SIZE
];
6080 struct imsm_super anchor
;
6081 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
6084 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
)
6086 struct imsm_super
*mpb
= super
->anchor
;
6087 struct imsm_super
*spare
= &spare_record
.anchor
;
6093 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
6094 spare
->generation_num
= __cpu_to_le32(1UL);
6095 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
6096 spare
->num_disks
= 1;
6097 spare
->num_raid_devs
= 0;
6098 spare
->cache_size
= mpb
->cache_size
;
6099 spare
->pwr_cycle_count
= __cpu_to_le32(1);
6101 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
6102 MPB_SIGNATURE MPB_VERSION_RAID0
);
6104 spare
->disk
[0] = d
->disk
;
6105 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
6106 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6108 if (super
->sector_size
== 4096)
6109 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6111 sum
= __gen_imsm_checksum(spare
);
6112 spare
->family_num
= __cpu_to_le32(sum
);
6113 spare
->orig_family_num
= 0;
6114 sum
= __gen_imsm_checksum(spare
);
6115 spare
->check_sum
= __cpu_to_le32(sum
);
6117 if (store_imsm_mpb(d
->fd
, spare
)) {
6118 pr_err("failed for device %d:%d %s\n",
6119 d
->major
, d
->minor
, strerror(errno
));
6125 /* spare records have their own family number and do not have any defined raid
6128 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
6132 for (d
= super
->disks
; d
; d
= d
->next
) {
6136 if (write_super_imsm_spare(super
, d
))
6148 static int write_super_imsm(struct supertype
*st
, int doclose
)
6150 struct intel_super
*super
= st
->sb
;
6151 unsigned int sector_size
= super
->sector_size
;
6152 struct imsm_super
*mpb
= super
->anchor
;
6158 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6160 int clear_migration_record
= 1;
6163 /* 'generation' is incremented everytime the metadata is written */
6164 generation
= __le32_to_cpu(mpb
->generation_num
);
6166 mpb
->generation_num
= __cpu_to_le32(generation
);
6168 /* fix up cases where previous mdadm releases failed to set
6171 if (mpb
->orig_family_num
== 0)
6172 mpb
->orig_family_num
= mpb
->family_num
;
6174 for (d
= super
->disks
; d
; d
= d
->next
) {
6178 mpb
->disk
[d
->index
] = d
->disk
;
6182 for (d
= super
->missing
; d
; d
= d
->next
) {
6183 mpb
->disk
[d
->index
] = d
->disk
;
6186 mpb
->num_disks
= num_disks
;
6187 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6189 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6190 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6191 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6193 imsm_copy_dev(dev
, dev2
);
6194 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6196 if (is_gen_migration(dev2
))
6197 clear_migration_record
= 0;
6200 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6203 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6204 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6206 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6208 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6209 mpb_size
+= bbm_log_size
;
6210 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6213 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6216 /* recalculate checksum */
6217 sum
= __gen_imsm_checksum(mpb
);
6218 mpb
->check_sum
= __cpu_to_le32(sum
);
6220 if (super
->clean_migration_record_by_mdmon
) {
6221 clear_migration_record
= 1;
6222 super
->clean_migration_record_by_mdmon
= 0;
6224 if (clear_migration_record
)
6225 memset(super
->migr_rec_buf
, 0,
6226 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6228 if (sector_size
== 4096)
6229 convert_to_4k(super
);
6231 /* write the mpb for disks that compose raid devices */
6232 for (d
= super
->disks
; d
; d
= d
->next
) {
6233 if (d
->index
< 0 || is_failed(&d
->disk
))
6236 if (clear_migration_record
) {
6237 unsigned long long dsize
;
6239 get_dev_size(d
->fd
, NULL
, &dsize
);
6240 if (lseek64(d
->fd
, dsize
- sector_size
,
6242 if ((unsigned int)write(d
->fd
,
6243 super
->migr_rec_buf
,
6244 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6245 MIGR_REC_BUF_SECTORS
*sector_size
)
6246 perror("Write migr_rec failed");
6250 if (store_imsm_mpb(d
->fd
, mpb
))
6252 "failed for device %d:%d (fd: %d)%s\n",
6254 d
->fd
, strerror(errno
));
6263 return write_super_imsm_spares(super
, doclose
);
6268 static int create_array(struct supertype
*st
, int dev_idx
)
6271 struct imsm_update_create_array
*u
;
6272 struct intel_super
*super
= st
->sb
;
6273 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6274 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6275 struct disk_info
*inf
;
6276 struct imsm_disk
*disk
;
6279 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6280 sizeof(*inf
) * map
->num_members
;
6282 u
->type
= update_create_array
;
6283 u
->dev_idx
= dev_idx
;
6284 imsm_copy_dev(&u
->dev
, dev
);
6285 inf
= get_disk_info(u
);
6286 for (i
= 0; i
< map
->num_members
; i
++) {
6287 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6289 disk
= get_imsm_disk(super
, idx
);
6291 disk
= get_imsm_missing(super
, idx
);
6292 serialcpy(inf
[i
].serial
, disk
->serial
);
6294 append_metadata_update(st
, u
, len
);
6299 static int mgmt_disk(struct supertype
*st
)
6301 struct intel_super
*super
= st
->sb
;
6303 struct imsm_update_add_remove_disk
*u
;
6305 if (!super
->disk_mgmt_list
)
6310 u
->type
= update_add_remove_disk
;
6311 append_metadata_update(st
, u
, len
);
6316 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6318 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6320 struct ppl_header
*ppl_hdr
= buf
;
6323 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6325 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6327 perror("Failed to seek to PPL header location");
6331 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6333 perror("Write PPL header failed");
6342 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6344 struct intel_super
*super
= st
->sb
;
6346 struct ppl_header
*ppl_hdr
;
6349 /* first clear entire ppl space */
6350 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6354 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6356 pr_err("Failed to allocate PPL header buffer\n");
6360 memset(buf
, 0, PPL_HEADER_SIZE
);
6362 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6363 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6365 if (info
->mismatch_cnt
) {
6367 * We are overwriting an invalid ppl. Make one entry with wrong
6368 * checksum to prevent the kernel from skipping resync.
6370 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6371 ppl_hdr
->entries
[0].checksum
= ~0;
6374 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6380 static int is_rebuilding(struct imsm_dev
*dev
);
6382 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6383 struct mdinfo
*disk
)
6385 struct intel_super
*super
= st
->sb
;
6387 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6389 struct ppl_header
*ppl_hdr
= NULL
;
6391 struct imsm_dev
*dev
;
6394 unsigned long long ppl_offset
= 0;
6395 unsigned long long prev_gen_num
= 0;
6397 if (disk
->disk
.raid_disk
< 0)
6400 dev
= get_imsm_dev(super
, info
->container_member
);
6401 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6402 d
= get_imsm_dl_disk(super
, idx
);
6404 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6407 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6408 pr_err("Failed to allocate PPL header buffer\n");
6414 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6417 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6419 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6421 perror("Failed to seek to PPL header location");
6426 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6427 perror("Read PPL header failed");
6434 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6435 ppl_hdr
->checksum
= 0;
6437 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6438 dprintf("Wrong PPL header checksum on %s\n",
6443 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6444 /* previous was newest, it was already checked */
6448 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6449 super
->anchor
->orig_family_num
)) {
6450 dprintf("Wrong PPL header signature on %s\n",
6457 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6459 ppl_offset
+= PPL_HEADER_SIZE
;
6460 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6462 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6465 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6477 * Update metadata to use mutliple PPLs area (1MB).
6478 * This is done once for all RAID members
6480 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6481 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6483 struct mdinfo
*member_dev
;
6485 sprintf(subarray
, "%d", info
->container_member
);
6487 if (mdmon_running(st
->container_devnm
))
6488 st
->update_tail
= &st
->updates
;
6490 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6491 pr_err("Failed to update subarray %s\n",
6494 if (st
->update_tail
)
6495 flush_metadata_updates(st
);
6497 st
->ss
->sync_metadata(st
);
6498 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6499 for (member_dev
= info
->devs
; member_dev
;
6500 member_dev
= member_dev
->next
)
6501 member_dev
->ppl_size
=
6502 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6507 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6509 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6510 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6511 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6512 (is_rebuilding(dev
) &&
6513 vol_curr_migr_unit(dev
) == 0 &&
6514 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6515 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6517 info
->mismatch_cnt
++;
6518 } else if (ret
== 0 &&
6519 ppl_hdr
->entries_count
== 0 &&
6520 is_rebuilding(dev
) &&
6521 info
->resync_start
== 0) {
6523 * The header has no entries - add a single empty entry and
6524 * rewrite the header to prevent the kernel from going into
6525 * resync after an interrupted rebuild.
6527 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6528 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6536 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6538 struct intel_super
*super
= st
->sb
;
6542 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6543 info
->array
.level
!= 5)
6546 for (d
= super
->disks
; d
; d
= d
->next
) {
6547 if (d
->index
< 0 || is_failed(&d
->disk
))
6550 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6558 /*******************************************************************************
6559 * Function: write_init_bitmap_imsm_vol
6560 * Description: Write a bitmap header and prepares the area for the bitmap.
6562 * st : supertype information
6563 * vol_idx : the volume index to use
6568 ******************************************************************************/
6569 static int write_init_bitmap_imsm_vol(struct supertype
*st
, int vol_idx
)
6571 struct intel_super
*super
= st
->sb
;
6572 int prev_current_vol
= super
->current_vol
;
6576 super
->current_vol
= vol_idx
;
6577 for (d
= super
->disks
; d
; d
= d
->next
) {
6578 if (d
->index
< 0 || is_failed(&d
->disk
))
6580 ret
= st
->ss
->write_bitmap(st
, d
->fd
, NoUpdate
);
6584 super
->current_vol
= prev_current_vol
;
6588 /*******************************************************************************
6589 * Function: write_init_bitmap_imsm_all
6590 * Description: Write a bitmap header and prepares the area for the bitmap.
6591 * Operation is executed for volumes with CONSISTENCY_POLICY_BITMAP.
6593 * st : supertype information
6594 * info : info about the volume where the bitmap should be written
6595 * vol_idx : the volume index to use
6600 ******************************************************************************/
6601 static int write_init_bitmap_imsm_all(struct supertype
*st
, struct mdinfo
*info
,
6606 if (info
&& (info
->consistency_policy
== CONSISTENCY_POLICY_BITMAP
))
6607 ret
= write_init_bitmap_imsm_vol(st
, vol_idx
);
6612 static int write_init_super_imsm(struct supertype
*st
)
6614 struct intel_super
*super
= st
->sb
;
6615 int current_vol
= super
->current_vol
;
6619 getinfo_super_imsm(st
, &info
, NULL
);
6621 /* we are done with current_vol reset it to point st at the container */
6622 super
->current_vol
= -1;
6624 if (st
->update_tail
) {
6625 /* queue the recently created array / added disk
6626 * as a metadata update */
6628 /* determine if we are creating a volume or adding a disk */
6629 if (current_vol
< 0) {
6630 /* in the mgmt (add/remove) disk case we are running
6631 * in mdmon context, so don't close fd's
6635 /* adding the second volume to the array */
6636 rv
= write_init_ppl_imsm_all(st
, &info
);
6638 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6640 rv
= create_array(st
, current_vol
);
6644 for (d
= super
->disks
; d
; d
= d
->next
)
6645 Kill(d
->devname
, NULL
, 0, -1, 1);
6646 if (current_vol
>= 0) {
6647 rv
= write_init_ppl_imsm_all(st
, &info
);
6649 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6653 rv
= write_super_imsm(st
, 1);
6659 static int store_super_imsm(struct supertype
*st
, int fd
)
6661 struct intel_super
*super
= st
->sb
;
6662 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6667 if (super
->sector_size
== 4096)
6668 convert_to_4k(super
);
6669 return store_imsm_mpb(fd
, mpb
);
6672 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6673 int layout
, int raiddisks
, int chunk
,
6674 unsigned long long size
,
6675 unsigned long long data_offset
,
6677 unsigned long long *freesize
,
6681 unsigned long long ldsize
;
6682 struct intel_super
*super
;
6685 if (level
!= LEVEL_CONTAINER
)
6690 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6693 pr_err("imsm: Cannot open %s: %s\n",
6694 dev
, strerror(errno
));
6697 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6702 /* capabilities retrieve could be possible
6703 * note that there is no fd for the disks in array.
6705 super
= alloc_super();
6710 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6716 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6720 fd2devname(fd
, str
);
6721 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6722 fd
, str
, super
->orom
, rv
, raiddisks
);
6724 /* no orom/efi or non-intel hba of the disk */
6731 if (raiddisks
> super
->orom
->tds
) {
6733 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6734 raiddisks
, super
->orom
->tds
);
6738 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6739 (ldsize
>> 9) >> 32 > 0) {
6741 pr_err("%s exceeds maximum platform supported size\n", dev
);
6747 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6753 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6755 const unsigned long long base_start
= e
[*idx
].start
;
6756 unsigned long long end
= base_start
+ e
[*idx
].size
;
6759 if (base_start
== end
)
6763 for (i
= *idx
; i
< num_extents
; i
++) {
6764 /* extend overlapping extents */
6765 if (e
[i
].start
>= base_start
&&
6766 e
[i
].start
<= end
) {
6769 if (e
[i
].start
+ e
[i
].size
> end
)
6770 end
= e
[i
].start
+ e
[i
].size
;
6771 } else if (e
[i
].start
> end
) {
6777 return end
- base_start
;
6780 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6782 /* build a composite disk with all known extents and generate a new
6783 * 'maxsize' given the "all disks in an array must share a common start
6784 * offset" constraint
6786 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6790 unsigned long long pos
;
6791 unsigned long long start
= 0;
6792 unsigned long long maxsize
;
6793 unsigned long reserve
;
6795 /* coalesce and sort all extents. also, check to see if we need to
6796 * reserve space between member arrays
6799 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6802 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6805 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6810 while (i
< sum_extents
) {
6811 e
[j
].start
= e
[i
].start
;
6812 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6814 if (e
[j
-1].size
== 0)
6823 unsigned long long esize
;
6825 esize
= e
[i
].start
- pos
;
6826 if (esize
>= maxsize
) {
6831 pos
= e
[i
].start
+ e
[i
].size
;
6833 } while (e
[i
-1].size
);
6839 /* FIXME assumes volume at offset 0 is the first volume in a
6842 if (start_extent
> 0)
6843 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6847 if (maxsize
< reserve
)
6850 super
->create_offset
= ~((unsigned long long) 0);
6851 if (start
+ reserve
> super
->create_offset
)
6852 return 0; /* start overflows create_offset */
6853 super
->create_offset
= start
+ reserve
;
6855 return maxsize
- reserve
;
6858 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6860 if (level
< 0 || level
== 6 || level
== 4)
6863 /* if we have an orom prevent invalid raid levels */
6866 case 0: return imsm_orom_has_raid0(orom
);
6869 return imsm_orom_has_raid1e(orom
);
6870 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6871 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6872 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6875 return 1; /* not on an Intel RAID platform so anything goes */
6881 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6882 int dpa
, int verbose
)
6884 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6885 struct mdstat_ent
*memb
;
6891 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6892 if (memb
->metadata_version
&&
6893 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6894 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6895 !is_subarray(memb
->metadata_version
+9) &&
6897 struct dev_member
*dev
= memb
->members
;
6899 while(dev
&& (fd
< 0)) {
6900 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6901 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6903 fd
= open(path
, O_RDONLY
, 0);
6904 if (num
<= 0 || fd
< 0) {
6905 pr_vrb("Cannot open %s: %s\n",
6906 dev
->name
, strerror(errno
));
6912 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6913 struct mdstat_ent
*vol
;
6914 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6915 if (vol
->active
> 0 &&
6916 vol
->metadata_version
&&
6917 is_container_member(vol
, memb
->devnm
)) {
6922 if (*devlist
&& (found
< dpa
)) {
6923 dv
= xcalloc(1, sizeof(*dv
));
6924 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6925 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6928 dv
->next
= *devlist
;
6936 free_mdstat(mdstat
);
6941 static struct md_list
*
6942 get_loop_devices(void)
6945 struct md_list
*devlist
= NULL
;
6948 for(i
= 0; i
< 12; i
++) {
6949 dv
= xcalloc(1, sizeof(*dv
));
6950 dv
->devname
= xmalloc(40);
6951 sprintf(dv
->devname
, "/dev/loop%d", i
);
6959 static struct md_list
*
6960 get_devices(const char *hba_path
)
6962 struct md_list
*devlist
= NULL
;
6969 devlist
= get_loop_devices();
6972 /* scroll through /sys/dev/block looking for devices attached to
6975 dir
= opendir("/sys/dev/block");
6976 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6981 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6983 path
= devt_to_devpath(makedev(major
, minor
));
6986 if (!path_attached_to_hba(path
, hba_path
)) {
6993 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6995 fd2devname(fd
, buf
);
6998 pr_err("cannot open device: %s\n",
7003 dv
= xcalloc(1, sizeof(*dv
));
7004 dv
->devname
= xstrdup(buf
);
7011 devlist
= devlist
->next
;
7021 count_volumes_list(struct md_list
*devlist
, char *homehost
,
7022 int verbose
, int *found
)
7024 struct md_list
*tmpdev
;
7026 struct supertype
*st
;
7028 /* first walk the list of devices to find a consistent set
7029 * that match the criterea, if that is possible.
7030 * We flag the ones we like with 'used'.
7033 st
= match_metadata_desc_imsm("imsm");
7035 pr_vrb("cannot allocate memory for imsm supertype\n");
7039 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7040 char *devname
= tmpdev
->devname
;
7042 struct supertype
*tst
;
7044 if (tmpdev
->used
> 1)
7046 tst
= dup_super(st
);
7048 pr_vrb("cannot allocate memory for imsm supertype\n");
7051 tmpdev
->container
= 0;
7052 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
7054 dprintf("cannot open device %s: %s\n",
7055 devname
, strerror(errno
));
7057 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
7059 } else if (must_be_container(dfd
)) {
7060 struct supertype
*cst
;
7061 cst
= super_by_fd(dfd
, NULL
);
7063 dprintf("cannot recognize container type %s\n",
7066 } else if (tst
->ss
!= st
->ss
) {
7067 dprintf("non-imsm container - ignore it: %s\n",
7070 } else if (!tst
->ss
->load_container
||
7071 tst
->ss
->load_container(tst
, dfd
, NULL
))
7074 tmpdev
->container
= 1;
7077 cst
->ss
->free_super(cst
);
7079 tmpdev
->st_rdev
= rdev
;
7080 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
7081 dprintf("no RAID superblock on %s\n",
7084 } else if (tst
->ss
->compare_super
== NULL
) {
7085 dprintf("Cannot assemble %s metadata on %s\n",
7086 tst
->ss
->name
, devname
);
7092 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
7093 /* Ignore unrecognised devices during auto-assembly */
7098 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
7100 if (st
->minor_version
== -1)
7101 st
->minor_version
= tst
->minor_version
;
7103 if (memcmp(info
.uuid
, uuid_zero
,
7104 sizeof(int[4])) == 0) {
7105 /* this is a floating spare. It cannot define
7106 * an array unless there are no more arrays of
7107 * this type to be found. It can be included
7108 * in an array of this type though.
7114 if (st
->ss
!= tst
->ss
||
7115 st
->minor_version
!= tst
->minor_version
||
7116 st
->ss
->compare_super(st
, tst
, 1) != 0) {
7117 /* Some mismatch. If exactly one array matches this host,
7118 * we can resolve on that one.
7119 * Or, if we are auto assembling, we just ignore the second
7122 dprintf("superblock on %s doesn't match others - assembly aborted\n",
7128 dprintf("found: devname: %s\n", devname
);
7132 tst
->ss
->free_super(tst
);
7136 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
7137 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
7138 for (iter
= head
; iter
; iter
= iter
->next
) {
7139 dprintf("content->text_version: %s vol\n",
7140 iter
->text_version
);
7141 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
7142 /* do not assemble arrays with unsupported
7144 dprintf("Cannot activate member %s.\n",
7145 iter
->text_version
);
7152 dprintf("No valid super block on device list: err: %d %p\n",
7156 dprintf("no more devices to examine\n");
7159 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7160 if (tmpdev
->used
== 1 && tmpdev
->found
) {
7162 if (count
< tmpdev
->found
)
7165 count
-= tmpdev
->found
;
7168 if (tmpdev
->used
== 1)
7173 st
->ss
->free_super(st
);
7177 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
7180 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
7182 const struct orom_entry
*entry
;
7183 struct devid_list
*dv
, *devid_list
;
7188 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7189 if (strstr(idev
->path
, hba_path
))
7193 if (!idev
|| !idev
->dev_id
)
7196 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7198 if (!entry
|| !entry
->devid_list
)
7201 devid_list
= entry
->devid_list
;
7202 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7203 struct md_list
*devlist
;
7204 struct sys_dev
*device
= NULL
;
7209 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7211 device
= device_by_id(dv
->devid
);
7214 hpath
= device
->path
;
7218 devlist
= get_devices(hpath
);
7219 /* if no intel devices return zero volumes */
7220 if (devlist
== NULL
)
7223 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7225 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7226 if (devlist
== NULL
)
7230 count
+= count_volumes_list(devlist
,
7234 dprintf("found %d count: %d\n", found
, count
);
7237 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7240 struct md_list
*dv
= devlist
;
7241 devlist
= devlist
->next
;
7249 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7253 if (hba
->type
== SYS_DEV_VMD
) {
7254 struct sys_dev
*dev
;
7257 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7258 if (dev
->type
== SYS_DEV_VMD
)
7259 count
+= __count_volumes(dev
->path
, dpa
,
7264 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7267 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7269 /* up to 512 if the plaform supports it, otherwise the platform max.
7270 * 128 if no platform detected
7272 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7274 return min(512, (1 << fs
));
7278 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7279 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7281 /* check/set platform and metadata limits/defaults */
7282 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7283 pr_vrb("platform supports a maximum of %d disks per array\n",
7288 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7289 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7290 pr_vrb("platform does not support raid%d with %d disk%s\n",
7291 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7295 if (*chunk
== 0 || *chunk
== UnSet
)
7296 *chunk
= imsm_default_chunk(super
->orom
);
7298 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7299 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7303 if (layout
!= imsm_level_to_layout(level
)) {
7305 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7306 else if (level
== 10)
7307 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7309 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7314 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7315 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7316 pr_vrb("platform does not support a volume size over 2TB\n");
7323 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7324 * FIX ME add ahci details
7326 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7327 int layout
, int raiddisks
, int *chunk
,
7328 unsigned long long size
,
7329 unsigned long long data_offset
,
7331 unsigned long long *freesize
,
7335 struct intel_super
*super
= st
->sb
;
7336 struct imsm_super
*mpb
;
7338 unsigned long long pos
= 0;
7339 unsigned long long maxsize
;
7343 /* We must have the container info already read in. */
7347 mpb
= super
->anchor
;
7349 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7350 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7354 /* General test: make sure there is space for
7355 * 'raiddisks' device extents of size 'size' at a given
7358 unsigned long long minsize
= size
;
7359 unsigned long long start_offset
= MaxSector
;
7362 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7363 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7368 e
= get_extents(super
, dl
, 0);
7371 unsigned long long esize
;
7372 esize
= e
[i
].start
- pos
;
7373 if (esize
>= minsize
)
7375 if (found
&& start_offset
== MaxSector
) {
7378 } else if (found
&& pos
!= start_offset
) {
7382 pos
= e
[i
].start
+ e
[i
].size
;
7384 } while (e
[i
-1].size
);
7389 if (dcnt
< raiddisks
) {
7391 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7398 /* This device must be a member of the set */
7399 if (!stat_is_blkdev(dev
, &rdev
))
7401 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7402 if (dl
->major
== (int)major(rdev
) &&
7403 dl
->minor
== (int)minor(rdev
))
7408 pr_err("%s is not in the same imsm set\n", dev
);
7410 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7411 /* If a volume is present then the current creation attempt
7412 * cannot incorporate new spares because the orom may not
7413 * understand this configuration (all member disks must be
7414 * members of each array in the container).
7416 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7417 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7419 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7420 mpb
->num_disks
!= raiddisks
) {
7421 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7425 /* retrieve the largest free space block */
7426 e
= get_extents(super
, dl
, 0);
7431 unsigned long long esize
;
7433 esize
= e
[i
].start
- pos
;
7434 if (esize
>= maxsize
)
7436 pos
= e
[i
].start
+ e
[i
].size
;
7438 } while (e
[i
-1].size
);
7443 pr_err("unable to determine free space for: %s\n",
7447 if (maxsize
< size
) {
7449 pr_err("%s not enough space (%llu < %llu)\n",
7450 dev
, maxsize
, size
);
7454 /* count total number of extents for merge */
7456 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7458 i
+= dl
->extent_cnt
;
7460 maxsize
= merge_extents(super
, i
);
7462 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7463 pr_err("attempting to create a second volume with size less then remaining space.\n");
7465 if (maxsize
< size
|| maxsize
== 0) {
7468 pr_err("no free space left on device. Aborting...\n");
7470 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7476 *freesize
= maxsize
;
7479 int count
= count_volumes(super
->hba
,
7480 super
->orom
->dpa
, verbose
);
7481 if (super
->orom
->vphba
<= count
) {
7482 pr_vrb("platform does not support more than %d raid volumes.\n",
7483 super
->orom
->vphba
);
7490 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7491 unsigned long long size
, int chunk
,
7492 unsigned long long *freesize
)
7494 struct intel_super
*super
= st
->sb
;
7495 struct imsm_super
*mpb
= super
->anchor
;
7500 unsigned long long maxsize
;
7501 unsigned long long minsize
;
7505 /* find the largest common start free region of the possible disks */
7509 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7515 /* don't activate new spares if we are orom constrained
7516 * and there is already a volume active in the container
7518 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7521 e
= get_extents(super
, dl
, 0);
7524 for (i
= 1; e
[i
-1].size
; i
++)
7532 maxsize
= merge_extents(super
, extent_cnt
);
7536 minsize
= chunk
* 2;
7538 if (cnt
< raiddisks
||
7539 (super
->orom
&& used
&& used
!= raiddisks
) ||
7540 maxsize
< minsize
||
7542 pr_err("not enough devices with space to create array.\n");
7543 return 0; /* No enough free spaces large enough */
7554 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7555 pr_err("attempting to create a second volume with size less then remaining space.\n");
7557 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7559 dl
->raiddisk
= cnt
++;
7563 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7568 static int reserve_space(struct supertype
*st
, int raiddisks
,
7569 unsigned long long size
, int chunk
,
7570 unsigned long long *freesize
)
7572 struct intel_super
*super
= st
->sb
;
7577 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7580 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7582 dl
->raiddisk
= cnt
++;
7589 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7590 int raiddisks
, int *chunk
, unsigned long long size
,
7591 unsigned long long data_offset
,
7592 char *dev
, unsigned long long *freesize
,
7593 int consistency_policy
, int verbose
)
7600 * if given unused devices create a container
7601 * if given given devices in a container create a member volume
7603 if (level
== LEVEL_CONTAINER
) {
7604 /* Must be a fresh device to add to a container */
7605 return validate_geometry_imsm_container(st
, level
, layout
,
7614 * Size is given in sectors.
7616 if (size
&& (size
< 2048)) {
7617 pr_err("Given size must be greater than 1M.\n");
7618 /* Depends on algorithm in Create.c :
7619 * if container was given (dev == NULL) return -1,
7620 * if block device was given ( dev != NULL) return 0.
7622 return dev
? -1 : 0;
7627 struct intel_super
*super
= st
->sb
;
7628 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7629 raiddisks
, chunk
, size
,
7632 /* we are being asked to automatically layout a
7633 * new volume based on the current contents of
7634 * the container. If the the parameters can be
7635 * satisfied reserve_space will record the disks,
7636 * start offset, and size of the volume to be
7637 * created. add_to_super and getinfo_super
7638 * detect when autolayout is in progress.
7640 /* assuming that freesize is always given when array is
7642 if (super
->orom
&& freesize
) {
7644 count
= count_volumes(super
->hba
,
7645 super
->orom
->dpa
, verbose
);
7646 if (super
->orom
->vphba
<= count
) {
7647 pr_vrb("platform does not support more than %d raid volumes.\n",
7648 super
->orom
->vphba
);
7653 return reserve_space(st
, raiddisks
, size
,
7659 /* creating in a given container */
7660 return validate_geometry_imsm_volume(st
, level
, layout
,
7661 raiddisks
, chunk
, size
,
7663 dev
, freesize
, verbose
);
7666 /* This device needs to be a device in an 'imsm' container */
7667 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7670 pr_err("Cannot create this array on device %s\n",
7675 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7677 pr_err("Cannot open %s: %s\n",
7678 dev
, strerror(errno
));
7681 /* Well, it is in use by someone, maybe an 'imsm' container. */
7682 cfd
= open_container(fd
);
7686 pr_err("Cannot use %s: It is busy\n",
7690 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7691 if (sra
&& sra
->array
.major_version
== -1 &&
7692 strcmp(sra
->text_version
, "imsm") == 0)
7696 /* This is a member of a imsm container. Load the container
7697 * and try to create a volume
7699 struct intel_super
*super
;
7701 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7703 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7705 return validate_geometry_imsm_volume(st
, level
, layout
,
7707 size
, data_offset
, dev
,
7714 pr_err("failed container membership check\n");
7720 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7722 struct intel_super
*super
= st
->sb
;
7724 if (level
&& *level
== UnSet
)
7725 *level
= LEVEL_CONTAINER
;
7727 if (level
&& layout
&& *layout
== UnSet
)
7728 *layout
= imsm_level_to_layout(*level
);
7730 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7731 *chunk
= imsm_default_chunk(super
->orom
);
7734 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7736 static int kill_subarray_imsm(struct supertype
*st
, char *subarray_id
)
7738 /* remove the subarray currently referenced by subarray_id */
7740 struct intel_dev
**dp
;
7741 struct intel_super
*super
= st
->sb
;
7742 __u8 current_vol
= strtoul(subarray_id
, NULL
, 10);
7743 struct imsm_super
*mpb
= super
->anchor
;
7745 if (mpb
->num_raid_devs
== 0)
7748 /* block deletions that would change the uuid of active subarrays
7750 * FIXME when immutable ids are available, but note that we'll
7751 * also need to fixup the invalidated/active subarray indexes in
7754 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7757 if (i
< current_vol
)
7759 sprintf(subarray
, "%u", i
);
7760 if (is_subarray_active(subarray
, st
->devnm
)) {
7761 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7768 if (st
->update_tail
) {
7769 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7771 u
->type
= update_kill_array
;
7772 u
->dev_idx
= current_vol
;
7773 append_metadata_update(st
, u
, sizeof(*u
));
7778 for (dp
= &super
->devlist
; *dp
;)
7779 if ((*dp
)->index
== current_vol
) {
7782 handle_missing(super
, (*dp
)->dev
);
7783 if ((*dp
)->index
> current_vol
)
7788 /* no more raid devices, all active components are now spares,
7789 * but of course failed are still failed
7791 if (--mpb
->num_raid_devs
== 0) {
7794 for (d
= super
->disks
; d
; d
= d
->next
)
7799 super
->updates_pending
++;
7804 static int get_rwh_policy_from_update(char *update
)
7806 if (strcmp(update
, "ppl") == 0)
7807 return RWH_MULTIPLE_DISTRIBUTED
;
7808 else if (strcmp(update
, "no-ppl") == 0)
7809 return RWH_MULTIPLE_OFF
;
7810 else if (strcmp(update
, "bitmap") == 0)
7812 else if (strcmp(update
, "no-bitmap") == 0)
7817 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7818 char *update
, struct mddev_ident
*ident
)
7820 /* update the subarray currently referenced by ->current_vol */
7821 struct intel_super
*super
= st
->sb
;
7822 struct imsm_super
*mpb
= super
->anchor
;
7824 if (strcmp(update
, "name") == 0) {
7825 char *name
= ident
->name
;
7829 if (is_subarray_active(subarray
, st
->devnm
)) {
7830 pr_err("Unable to update name of active subarray\n");
7834 if (!check_name(super
, name
, 0))
7837 vol
= strtoul(subarray
, &ep
, 10);
7838 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7841 if (st
->update_tail
) {
7842 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7844 u
->type
= update_rename_array
;
7846 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7847 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7848 append_metadata_update(st
, u
, sizeof(*u
));
7850 struct imsm_dev
*dev
;
7853 dev
= get_imsm_dev(super
, vol
);
7854 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7855 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7856 memcpy(dev
->volume
, name
, namelen
);
7857 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7858 dev
= get_imsm_dev(super
, i
);
7859 handle_missing(super
, dev
);
7861 super
->updates_pending
++;
7863 } else if (get_rwh_policy_from_update(update
) != -1) {
7866 int vol
= strtoul(subarray
, &ep
, 10);
7868 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7871 new_policy
= get_rwh_policy_from_update(update
);
7873 if (st
->update_tail
) {
7874 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7876 u
->type
= update_rwh_policy
;
7878 u
->new_policy
= new_policy
;
7879 append_metadata_update(st
, u
, sizeof(*u
));
7881 struct imsm_dev
*dev
;
7883 dev
= get_imsm_dev(super
, vol
);
7884 dev
->rwh_policy
= new_policy
;
7885 super
->updates_pending
++;
7887 if (new_policy
== RWH_BITMAP
)
7888 return write_init_bitmap_imsm_vol(st
, vol
);
7895 static int is_gen_migration(struct imsm_dev
*dev
)
7900 if (!dev
->vol
.migr_state
)
7903 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7909 static int is_rebuilding(struct imsm_dev
*dev
)
7911 struct imsm_map
*migr_map
;
7913 if (!dev
->vol
.migr_state
)
7916 if (migr_type(dev
) != MIGR_REBUILD
)
7919 migr_map
= get_imsm_map(dev
, MAP_1
);
7921 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7927 static int is_initializing(struct imsm_dev
*dev
)
7929 struct imsm_map
*migr_map
;
7931 if (!dev
->vol
.migr_state
)
7934 if (migr_type(dev
) != MIGR_INIT
)
7937 migr_map
= get_imsm_map(dev
, MAP_1
);
7939 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7945 static void update_recovery_start(struct intel_super
*super
,
7946 struct imsm_dev
*dev
,
7947 struct mdinfo
*array
)
7949 struct mdinfo
*rebuild
= NULL
;
7953 if (!is_rebuilding(dev
))
7956 /* Find the rebuild target, but punt on the dual rebuild case */
7957 for (d
= array
->devs
; d
; d
= d
->next
)
7958 if (d
->recovery_start
== 0) {
7965 /* (?) none of the disks are marked with
7966 * IMSM_ORD_REBUILD, so assume they are missing and the
7967 * disk_ord_tbl was not correctly updated
7969 dprintf("failed to locate out-of-sync disk\n");
7973 units
= vol_curr_migr_unit(dev
);
7974 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7977 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7979 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7981 /* Given a container loaded by load_super_imsm_all,
7982 * extract information about all the arrays into
7984 * If 'subarray' is given, just extract info about that array.
7986 * For each imsm_dev create an mdinfo, fill it in,
7987 * then look for matching devices in super->disks
7988 * and create appropriate device mdinfo.
7990 struct intel_super
*super
= st
->sb
;
7991 struct imsm_super
*mpb
= super
->anchor
;
7992 struct mdinfo
*rest
= NULL
;
7996 int spare_disks
= 0;
7997 int current_vol
= super
->current_vol
;
7999 /* do not assemble arrays when not all attributes are supported */
8000 if (imsm_check_attributes(mpb
->attributes
) == 0) {
8002 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
8005 /* count spare devices, not used in maps
8007 for (d
= super
->disks
; d
; d
= d
->next
)
8011 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8012 struct imsm_dev
*dev
;
8013 struct imsm_map
*map
;
8014 struct imsm_map
*map2
;
8015 struct mdinfo
*this;
8022 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
8025 dev
= get_imsm_dev(super
, i
);
8026 map
= get_imsm_map(dev
, MAP_0
);
8027 map2
= get_imsm_map(dev
, MAP_1
);
8028 level
= get_imsm_raid_level(map
);
8030 /* do not publish arrays that are in the middle of an
8031 * unsupported migration
8033 if (dev
->vol
.migr_state
&&
8034 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
8035 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
8039 /* do not publish arrays that are not support by controller's
8043 this = xmalloc(sizeof(*this));
8045 super
->current_vol
= i
;
8046 getinfo_super_imsm_volume(st
, this, NULL
);
8048 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
8049 /* mdadm does not support all metadata features- set the bit in all arrays state */
8050 if (!validate_geometry_imsm_orom(super
,
8051 level
, /* RAID level */
8052 imsm_level_to_layout(level
),
8053 map
->num_members
, /* raid disks */
8054 &chunk
, imsm_dev_size(dev
),
8056 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
8058 this->array
.state
|=
8059 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8060 (1<<MD_SB_BLOCK_VOLUME
);
8063 /* if array has bad blocks, set suitable bit in all arrays state */
8065 this->array
.state
|=
8066 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8067 (1<<MD_SB_BLOCK_VOLUME
);
8069 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
8070 unsigned long long recovery_start
;
8071 struct mdinfo
*info_d
;
8079 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
8080 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
8081 for (d
= super
->disks
; d
; d
= d
->next
)
8082 if (d
->index
== idx
)
8085 recovery_start
= MaxSector
;
8088 if (d
&& is_failed(&d
->disk
))
8090 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
8092 if (!(ord
& IMSM_ORD_REBUILD
))
8093 this->array
.working_disks
++;
8095 * if we skip some disks the array will be assmebled degraded;
8096 * reset resync start to avoid a dirty-degraded
8097 * situation when performing the intial sync
8102 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
8103 if ((!able_to_resync(level
, missing
) ||
8104 recovery_start
== 0))
8105 this->resync_start
= MaxSector
;
8108 * FIXME handle dirty degraded
8115 info_d
= xcalloc(1, sizeof(*info_d
));
8116 info_d
->next
= this->devs
;
8117 this->devs
= info_d
;
8119 info_d
->disk
.number
= d
->index
;
8120 info_d
->disk
.major
= d
->major
;
8121 info_d
->disk
.minor
= d
->minor
;
8122 info_d
->disk
.raid_disk
= slot
;
8123 info_d
->recovery_start
= recovery_start
;
8125 if (slot
< map2
->num_members
)
8126 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8128 this->array
.spare_disks
++;
8130 if (slot
< map
->num_members
)
8131 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8133 this->array
.spare_disks
++;
8136 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
8137 info_d
->data_offset
= pba_of_lba0(map
);
8138 info_d
->component_size
= calc_component_size(map
, dev
);
8140 if (map
->raid_level
== 5) {
8141 info_d
->ppl_sector
= this->ppl_sector
;
8142 info_d
->ppl_size
= this->ppl_size
;
8143 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
8144 recovery_start
== 0)
8145 this->resync_start
= 0;
8148 info_d
->bb
.supported
= 1;
8149 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
8150 info_d
->data_offset
,
8151 info_d
->component_size
,
8154 /* now that the disk list is up-to-date fixup recovery_start */
8155 update_recovery_start(super
, dev
, this);
8156 this->array
.spare_disks
+= spare_disks
;
8158 /* check for reshape */
8159 if (this->reshape_active
== 1)
8160 recover_backup_imsm(st
, this);
8164 super
->current_vol
= current_vol
;
8168 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
8169 int failed
, int look_in_map
)
8171 struct imsm_map
*map
;
8173 map
= get_imsm_map(dev
, look_in_map
);
8176 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
8177 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
8179 switch (get_imsm_raid_level(map
)) {
8181 return IMSM_T_STATE_FAILED
;
8184 if (failed
< map
->num_members
)
8185 return IMSM_T_STATE_DEGRADED
;
8187 return IMSM_T_STATE_FAILED
;
8192 * check to see if any mirrors have failed, otherwise we
8193 * are degraded. Even numbered slots are mirrored on
8197 /* gcc -Os complains that this is unused */
8198 int insync
= insync
;
8200 for (i
= 0; i
< map
->num_members
; i
++) {
8201 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8202 int idx
= ord_to_idx(ord
);
8203 struct imsm_disk
*disk
;
8205 /* reset the potential in-sync count on even-numbered
8206 * slots. num_copies is always 2 for imsm raid10
8211 disk
= get_imsm_disk(super
, idx
);
8212 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8215 /* no in-sync disks left in this mirror the
8219 return IMSM_T_STATE_FAILED
;
8222 return IMSM_T_STATE_DEGRADED
;
8226 return IMSM_T_STATE_DEGRADED
;
8228 return IMSM_T_STATE_FAILED
;
8234 return map
->map_state
;
8237 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8242 struct imsm_disk
*disk
;
8243 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8244 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8245 struct imsm_map
*map_for_loop
;
8250 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8251 * disks that are being rebuilt. New failures are recorded to
8252 * map[0]. So we look through all the disks we started with and
8253 * see if any failures are still present, or if any new ones
8257 if (prev
&& (map
->num_members
< prev
->num_members
))
8258 map_for_loop
= prev
;
8260 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8262 /* when MAP_X is passed both maps failures are counted
8265 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8266 i
< prev
->num_members
) {
8267 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8268 idx_1
= ord_to_idx(ord
);
8270 disk
= get_imsm_disk(super
, idx_1
);
8271 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8274 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8275 i
< map
->num_members
) {
8276 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8277 idx
= ord_to_idx(ord
);
8280 disk
= get_imsm_disk(super
, idx
);
8281 if (!disk
|| is_failed(disk
) ||
8282 ord
& IMSM_ORD_REBUILD
)
8291 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8294 struct intel_super
*super
= c
->sb
;
8295 struct imsm_super
*mpb
= super
->anchor
;
8296 struct imsm_update_prealloc_bb_mem u
;
8298 if (atoi(inst
) >= mpb
->num_raid_devs
) {
8299 pr_err("subarry index %d, out of range\n", atoi(inst
));
8303 dprintf("imsm: open_new %s\n", inst
);
8304 a
->info
.container_member
= atoi(inst
);
8306 u
.type
= update_prealloc_badblocks_mem
;
8307 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8312 static int is_resyncing(struct imsm_dev
*dev
)
8314 struct imsm_map
*migr_map
;
8316 if (!dev
->vol
.migr_state
)
8319 if (migr_type(dev
) == MIGR_INIT
||
8320 migr_type(dev
) == MIGR_REPAIR
)
8323 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8326 migr_map
= get_imsm_map(dev
, MAP_1
);
8328 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8329 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8335 /* return true if we recorded new information */
8336 static int mark_failure(struct intel_super
*super
,
8337 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8341 struct imsm_map
*map
;
8342 char buf
[MAX_RAID_SERIAL_LEN
+3];
8343 unsigned int len
, shift
= 0;
8345 /* new failures are always set in map[0] */
8346 map
= get_imsm_map(dev
, MAP_0
);
8348 slot
= get_imsm_disk_slot(map
, idx
);
8352 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8353 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8356 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8357 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8359 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8360 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8361 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8363 disk
->status
|= FAILED_DISK
;
8364 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8365 /* mark failures in second map if second map exists and this disk
8367 * This is valid for migration, initialization and rebuild
8369 if (dev
->vol
.migr_state
) {
8370 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8371 int slot2
= get_imsm_disk_slot(map2
, idx
);
8373 if (slot2
< map2
->num_members
&& slot2
>= 0)
8374 set_imsm_ord_tbl_ent(map2
, slot2
,
8375 idx
| IMSM_ORD_REBUILD
);
8377 if (map
->failed_disk_num
== 0xff ||
8378 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8379 map
->failed_disk_num
= slot
;
8381 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8386 static void mark_missing(struct intel_super
*super
,
8387 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8389 mark_failure(super
, dev
, disk
, idx
);
8391 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8394 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8395 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8398 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8402 if (!super
->missing
)
8405 /* When orom adds replacement for missing disk it does
8406 * not remove entry of missing disk, but just updates map with
8407 * new added disk. So it is not enough just to test if there is
8408 * any missing disk, we have to look if there are any failed disks
8409 * in map to stop migration */
8411 dprintf("imsm: mark missing\n");
8412 /* end process for initialization and rebuild only
8414 if (is_gen_migration(dev
) == 0) {
8415 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8419 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8420 struct imsm_map
*map1
;
8421 int i
, ord
, ord_map1
;
8424 for (i
= 0; i
< map
->num_members
; i
++) {
8425 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8426 if (!(ord
& IMSM_ORD_REBUILD
))
8429 map1
= get_imsm_map(dev
, MAP_1
);
8433 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8434 if (ord_map1
& IMSM_ORD_REBUILD
)
8439 map_state
= imsm_check_degraded(super
, dev
,
8441 end_migration(dev
, super
, map_state
);
8445 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8446 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8447 super
->updates_pending
++;
8450 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8453 unsigned long long array_blocks
;
8454 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8455 int used_disks
= imsm_num_data_members(map
);
8457 if (used_disks
== 0) {
8458 /* when problems occures
8459 * return current array_blocks value
8461 array_blocks
= imsm_dev_size(dev
);
8463 return array_blocks
;
8466 /* set array size in metadata
8469 /* OLCE size change is caused by added disks
8471 array_blocks
= per_dev_array_size(map
) * used_disks
;
8473 /* Online Volume Size Change
8474 * Using available free space
8476 array_blocks
= new_size
;
8478 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8479 set_imsm_dev_size(dev
, array_blocks
);
8481 return array_blocks
;
8484 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8486 static void imsm_progress_container_reshape(struct intel_super
*super
)
8488 /* if no device has a migr_state, but some device has a
8489 * different number of members than the previous device, start
8490 * changing the number of devices in this device to match
8493 struct imsm_super
*mpb
= super
->anchor
;
8494 int prev_disks
= -1;
8498 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8499 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8500 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8501 struct imsm_map
*map2
;
8502 int prev_num_members
;
8504 if (dev
->vol
.migr_state
)
8507 if (prev_disks
== -1)
8508 prev_disks
= map
->num_members
;
8509 if (prev_disks
== map
->num_members
)
8512 /* OK, this array needs to enter reshape mode.
8513 * i.e it needs a migr_state
8516 copy_map_size
= sizeof_imsm_map(map
);
8517 prev_num_members
= map
->num_members
;
8518 map
->num_members
= prev_disks
;
8519 dev
->vol
.migr_state
= 1;
8520 set_vol_curr_migr_unit(dev
, 0);
8521 set_migr_type(dev
, MIGR_GEN_MIGR
);
8522 for (i
= prev_num_members
;
8523 i
< map
->num_members
; i
++)
8524 set_imsm_ord_tbl_ent(map
, i
, i
);
8525 map2
= get_imsm_map(dev
, MAP_1
);
8526 /* Copy the current map */
8527 memcpy(map2
, map
, copy_map_size
);
8528 map2
->num_members
= prev_num_members
;
8530 imsm_set_array_size(dev
, -1);
8531 super
->clean_migration_record_by_mdmon
= 1;
8532 super
->updates_pending
++;
8536 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8537 * states are handled in imsm_set_disk() with one exception, when a
8538 * resync is stopped due to a new failure this routine will set the
8539 * 'degraded' state for the array.
8541 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8543 int inst
= a
->info
.container_member
;
8544 struct intel_super
*super
= a
->container
->sb
;
8545 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8546 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8547 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8548 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8549 __u32 blocks_per_unit
;
8551 if (dev
->vol
.migr_state
&&
8552 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8553 /* array state change is blocked due to reshape action
8555 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8556 * - finish the reshape (if last_checkpoint is big and action != reshape)
8557 * - update vol_curr_migr_unit
8559 if (a
->curr_action
== reshape
) {
8560 /* still reshaping, maybe update vol_curr_migr_unit */
8561 goto mark_checkpoint
;
8563 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8564 /* for some reason we aborted the reshape.
8566 * disable automatic metadata rollback
8567 * user action is required to recover process
8570 struct imsm_map
*map2
=
8571 get_imsm_map(dev
, MAP_1
);
8572 dev
->vol
.migr_state
= 0;
8573 set_migr_type(dev
, 0);
8574 set_vol_curr_migr_unit(dev
, 0);
8576 sizeof_imsm_map(map2
));
8577 super
->updates_pending
++;
8580 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8581 unsigned long long array_blocks
;
8585 used_disks
= imsm_num_data_members(map
);
8586 if (used_disks
> 0) {
8588 per_dev_array_size(map
) *
8591 round_size_to_mb(array_blocks
,
8593 a
->info
.custom_array_size
= array_blocks
;
8594 /* encourage manager to update array
8598 a
->check_reshape
= 1;
8600 /* finalize online capacity expansion/reshape */
8601 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8603 mdi
->disk
.raid_disk
,
8606 imsm_progress_container_reshape(super
);
8611 /* before we activate this array handle any missing disks */
8612 if (consistent
== 2)
8613 handle_missing(super
, dev
);
8615 if (consistent
== 2 &&
8616 (!is_resync_complete(&a
->info
) ||
8617 map_state
!= IMSM_T_STATE_NORMAL
||
8618 dev
->vol
.migr_state
))
8621 if (is_resync_complete(&a
->info
)) {
8622 /* complete intialization / resync,
8623 * recovery and interrupted recovery is completed in
8626 if (is_resyncing(dev
)) {
8627 dprintf("imsm: mark resync done\n");
8628 end_migration(dev
, super
, map_state
);
8629 super
->updates_pending
++;
8630 a
->last_checkpoint
= 0;
8632 } else if ((!is_resyncing(dev
) && !failed
) &&
8633 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8634 /* mark the start of the init process if nothing is failed */
8635 dprintf("imsm: mark resync start\n");
8636 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8637 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8639 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8640 super
->updates_pending
++;
8644 /* skip checkpointing for general migration,
8645 * it is controlled in mdadm
8647 if (is_gen_migration(dev
))
8648 goto skip_mark_checkpoint
;
8650 /* check if we can update vol_curr_migr_unit from resync_start,
8653 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8654 if (blocks_per_unit
) {
8655 set_vol_curr_migr_unit(dev
,
8656 a
->last_checkpoint
/ blocks_per_unit
);
8657 dprintf("imsm: mark checkpoint (%llu)\n",
8658 vol_curr_migr_unit(dev
));
8659 super
->updates_pending
++;
8662 skip_mark_checkpoint
:
8663 /* mark dirty / clean */
8664 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8665 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8666 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8668 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8670 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8671 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8672 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8673 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8675 super
->updates_pending
++;
8681 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8683 int inst
= a
->info
.container_member
;
8684 struct intel_super
*super
= a
->container
->sb
;
8685 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8686 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8688 if (slot
> map
->num_members
) {
8689 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8690 slot
, map
->num_members
- 1);
8697 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8700 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8702 int inst
= a
->info
.container_member
;
8703 struct intel_super
*super
= a
->container
->sb
;
8704 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8705 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8706 struct imsm_disk
*disk
;
8708 int recovery_not_finished
= 0;
8712 int rebuild_done
= 0;
8715 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8719 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8720 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8722 /* check for new failures */
8723 if (disk
&& (state
& DS_FAULTY
)) {
8724 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8725 super
->updates_pending
++;
8728 /* check if in_sync */
8729 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8730 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8732 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8734 super
->updates_pending
++;
8737 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8738 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8740 /* check if recovery complete, newly degraded, or failed */
8741 dprintf("imsm: Detected transition to state ");
8742 switch (map_state
) {
8743 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8744 dprintf("normal: ");
8745 if (is_rebuilding(dev
)) {
8746 dprintf_cont("while rebuilding");
8747 /* check if recovery is really finished */
8748 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8749 if (mdi
->recovery_start
!= MaxSector
) {
8750 recovery_not_finished
= 1;
8753 if (recovery_not_finished
) {
8755 dprintf("Rebuild has not finished yet, state not changed");
8756 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8757 a
->last_checkpoint
= mdi
->recovery_start
;
8758 super
->updates_pending
++;
8762 end_migration(dev
, super
, map_state
);
8763 map
->failed_disk_num
= ~0;
8764 super
->updates_pending
++;
8765 a
->last_checkpoint
= 0;
8768 if (is_gen_migration(dev
)) {
8769 dprintf_cont("while general migration");
8770 if (a
->last_checkpoint
>= a
->info
.component_size
)
8771 end_migration(dev
, super
, map_state
);
8773 map
->map_state
= map_state
;
8774 map
->failed_disk_num
= ~0;
8775 super
->updates_pending
++;
8779 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8780 dprintf_cont("degraded: ");
8781 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8782 dprintf_cont("mark degraded");
8783 map
->map_state
= map_state
;
8784 super
->updates_pending
++;
8785 a
->last_checkpoint
= 0;
8788 if (is_rebuilding(dev
)) {
8789 dprintf_cont("while rebuilding ");
8790 if (state
& DS_FAULTY
) {
8791 dprintf_cont("removing failed drive ");
8792 if (n
== map
->failed_disk_num
) {
8793 dprintf_cont("end migration");
8794 end_migration(dev
, super
, map_state
);
8795 a
->last_checkpoint
= 0;
8797 dprintf_cont("fail detected during rebuild, changing map state");
8798 map
->map_state
= map_state
;
8800 super
->updates_pending
++;
8806 /* check if recovery is really finished */
8807 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8808 if (mdi
->recovery_start
!= MaxSector
) {
8809 recovery_not_finished
= 1;
8812 if (recovery_not_finished
) {
8814 dprintf_cont("Rebuild has not finished yet");
8815 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8816 a
->last_checkpoint
=
8817 mdi
->recovery_start
;
8818 super
->updates_pending
++;
8823 dprintf_cont(" Rebuild done, still degraded");
8824 end_migration(dev
, super
, map_state
);
8825 a
->last_checkpoint
= 0;
8826 super
->updates_pending
++;
8828 for (i
= 0; i
< map
->num_members
; i
++) {
8829 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8831 if (idx
& IMSM_ORD_REBUILD
)
8832 map
->failed_disk_num
= i
;
8834 super
->updates_pending
++;
8837 if (is_gen_migration(dev
)) {
8838 dprintf_cont("while general migration");
8839 if (a
->last_checkpoint
>= a
->info
.component_size
)
8840 end_migration(dev
, super
, map_state
);
8842 map
->map_state
= map_state
;
8843 manage_second_map(super
, dev
);
8845 super
->updates_pending
++;
8848 if (is_initializing(dev
)) {
8849 dprintf_cont("while initialization.");
8850 map
->map_state
= map_state
;
8851 super
->updates_pending
++;
8855 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8856 dprintf_cont("failed: ");
8857 if (is_gen_migration(dev
)) {
8858 dprintf_cont("while general migration");
8859 map
->map_state
= map_state
;
8860 super
->updates_pending
++;
8863 if (map
->map_state
!= map_state
) {
8864 dprintf_cont("mark failed");
8865 end_migration(dev
, super
, map_state
);
8866 super
->updates_pending
++;
8867 a
->last_checkpoint
= 0;
8872 dprintf_cont("state %i\n", map_state
);
8877 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8880 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8881 unsigned long long dsize
;
8882 unsigned long long sectors
;
8883 unsigned int sector_size
;
8885 get_dev_sector_size(fd
, NULL
, §or_size
);
8886 get_dev_size(fd
, NULL
, &dsize
);
8888 if (mpb_size
> sector_size
) {
8889 /* -1 to account for anchor */
8890 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8892 /* write the extended mpb to the sectors preceeding the anchor */
8893 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8897 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8898 sector_size
* sectors
) != sector_size
* sectors
)
8902 /* first block is stored on second to last sector of the disk */
8903 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8906 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8912 static void imsm_sync_metadata(struct supertype
*container
)
8914 struct intel_super
*super
= container
->sb
;
8916 dprintf("sync metadata: %d\n", super
->updates_pending
);
8917 if (!super
->updates_pending
)
8920 write_super_imsm(container
, 0);
8922 super
->updates_pending
= 0;
8925 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8927 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8928 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8931 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8935 if (dl
&& is_failed(&dl
->disk
))
8939 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8944 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8945 struct active_array
*a
, int activate_new
,
8946 struct mdinfo
*additional_test_list
)
8948 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8949 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8950 struct imsm_super
*mpb
= super
->anchor
;
8951 struct imsm_map
*map
;
8952 unsigned long long pos
;
8957 __u32 array_start
= 0;
8958 __u32 array_end
= 0;
8960 struct mdinfo
*test_list
;
8962 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8963 /* If in this array, skip */
8964 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8965 if (d
->state_fd
>= 0 &&
8966 d
->disk
.major
== dl
->major
&&
8967 d
->disk
.minor
== dl
->minor
) {
8968 dprintf("%x:%x already in array\n",
8969 dl
->major
, dl
->minor
);
8974 test_list
= additional_test_list
;
8976 if (test_list
->disk
.major
== dl
->major
&&
8977 test_list
->disk
.minor
== dl
->minor
) {
8978 dprintf("%x:%x already in additional test list\n",
8979 dl
->major
, dl
->minor
);
8982 test_list
= test_list
->next
;
8987 /* skip in use or failed drives */
8988 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8990 dprintf("%x:%x status (failed: %d index: %d)\n",
8991 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8995 /* skip pure spares when we are looking for partially
8996 * assimilated drives
8998 if (dl
->index
== -1 && !activate_new
)
9001 if (!drive_validate_sector_size(super
, dl
))
9004 /* Does this unused device have the requisite free space?
9005 * It needs to be able to cover all member volumes
9007 ex
= get_extents(super
, dl
, 1);
9009 dprintf("cannot get extents\n");
9012 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9013 dev
= get_imsm_dev(super
, i
);
9014 map
= get_imsm_map(dev
, MAP_0
);
9016 /* check if this disk is already a member of
9019 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
9025 array_start
= pba_of_lba0(map
);
9026 array_end
= array_start
+
9027 per_dev_array_size(map
) - 1;
9030 /* check that we can start at pba_of_lba0 with
9031 * num_data_stripes*blocks_per_stripe of space
9033 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
9037 pos
= ex
[j
].start
+ ex
[j
].size
;
9039 } while (ex
[j
-1].size
);
9046 if (i
< mpb
->num_raid_devs
) {
9047 dprintf("%x:%x does not have %u to %u available\n",
9048 dl
->major
, dl
->minor
, array_start
, array_end
);
9058 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
9060 struct imsm_dev
*dev2
;
9061 struct imsm_map
*map
;
9067 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
9069 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
9070 if (state
== IMSM_T_STATE_FAILED
) {
9071 map
= get_imsm_map(dev2
, MAP_0
);
9074 for (slot
= 0; slot
< map
->num_members
; slot
++) {
9076 * Check if failed disks are deleted from intel
9077 * disk list or are marked to be deleted
9079 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
9080 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
9082 * Do not rebuild the array if failed disks
9083 * from failed sub-array are not removed from
9087 is_failed(&idisk
->disk
) &&
9088 (idisk
->action
!= DISK_REMOVE
))
9096 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
9097 struct metadata_update
**updates
)
9100 * Find a device with unused free space and use it to replace a
9101 * failed/vacant region in an array. We replace failed regions one a
9102 * array at a time. The result is that a new spare disk will be added
9103 * to the first failed array and after the monitor has finished
9104 * propagating failures the remainder will be consumed.
9106 * FIXME add a capability for mdmon to request spares from another
9110 struct intel_super
*super
= a
->container
->sb
;
9111 int inst
= a
->info
.container_member
;
9112 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
9113 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9114 int failed
= a
->info
.array
.raid_disks
;
9115 struct mdinfo
*rv
= NULL
;
9118 struct metadata_update
*mu
;
9120 struct imsm_update_activate_spare
*u
;
9125 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
9126 if ((d
->curr_state
& DS_FAULTY
) &&
9128 /* wait for Removal to happen */
9130 if (d
->state_fd
>= 0)
9134 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
9135 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
9137 if (imsm_reshape_blocks_arrays_changes(super
))
9140 /* Cannot activate another spare if rebuild is in progress already
9142 if (is_rebuilding(dev
)) {
9143 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
9147 if (a
->info
.array
.level
== 4)
9148 /* No repair for takeovered array
9149 * imsm doesn't support raid4
9153 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
9154 IMSM_T_STATE_DEGRADED
)
9157 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
9158 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
9163 * If there are any failed disks check state of the other volume.
9164 * Block rebuild if the another one is failed until failed disks
9165 * are removed from container.
9168 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
9169 MAX_RAID_SERIAL_LEN
, dev
->volume
);
9170 /* check if states of the other volumes allow for rebuild */
9171 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
9173 allowed
= imsm_rebuild_allowed(a
->container
,
9181 /* For each slot, if it is not working, find a spare */
9182 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
9183 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9184 if (d
->disk
.raid_disk
== i
)
9186 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
9187 if (d
&& (d
->state_fd
>= 0))
9191 * OK, this device needs recovery. Try to re-add the
9192 * previous occupant of this slot, if this fails see if
9193 * we can continue the assimilation of a spare that was
9194 * partially assimilated, finally try to activate a new
9197 dl
= imsm_readd(super
, i
, a
);
9199 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9201 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9205 /* found a usable disk with enough space */
9206 di
= xcalloc(1, sizeof(*di
));
9208 /* dl->index will be -1 in the case we are activating a
9209 * pristine spare. imsm_process_update() will create a
9210 * new index in this case. Once a disk is found to be
9211 * failed in all member arrays it is kicked from the
9214 di
->disk
.number
= dl
->index
;
9216 /* (ab)use di->devs to store a pointer to the device
9219 di
->devs
= (struct mdinfo
*) dl
;
9221 di
->disk
.raid_disk
= i
;
9222 di
->disk
.major
= dl
->major
;
9223 di
->disk
.minor
= dl
->minor
;
9225 di
->recovery_start
= 0;
9226 di
->data_offset
= pba_of_lba0(map
);
9227 di
->component_size
= a
->info
.component_size
;
9228 di
->container_member
= inst
;
9229 di
->bb
.supported
= 1;
9230 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9231 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9232 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9234 super
->random
= random32();
9238 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9239 i
, di
->data_offset
);
9243 /* No spares found */
9245 /* Now 'rv' has a list of devices to return.
9246 * Create a metadata_update record to update the
9247 * disk_ord_tbl for the array
9249 mu
= xmalloc(sizeof(*mu
));
9250 mu
->buf
= xcalloc(num_spares
,
9251 sizeof(struct imsm_update_activate_spare
));
9253 mu
->space_list
= NULL
;
9254 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9255 mu
->next
= *updates
;
9256 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9258 for (di
= rv
; di
; di
= di
->next
) {
9259 u
->type
= update_activate_spare
;
9260 u
->dl
= (struct dl
*) di
->devs
;
9262 u
->slot
= di
->disk
.raid_disk
;
9273 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9275 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9276 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9277 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9278 struct disk_info
*inf
= get_disk_info(u
);
9279 struct imsm_disk
*disk
;
9283 for (i
= 0; i
< map
->num_members
; i
++) {
9284 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9285 for (j
= 0; j
< new_map
->num_members
; j
++)
9286 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9293 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9297 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9298 if (dl
->major
== major
&& dl
->minor
== minor
)
9303 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9309 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9310 if (dl
->major
== major
&& dl
->minor
== minor
) {
9313 prev
->next
= dl
->next
;
9315 super
->disks
= dl
->next
;
9317 __free_imsm_disk(dl
);
9318 dprintf("removed %x:%x\n", major
, minor
);
9326 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9328 static int add_remove_disk_update(struct intel_super
*super
)
9330 int check_degraded
= 0;
9333 /* add/remove some spares to/from the metadata/contrainer */
9334 while (super
->disk_mgmt_list
) {
9335 struct dl
*disk_cfg
;
9337 disk_cfg
= super
->disk_mgmt_list
;
9338 super
->disk_mgmt_list
= disk_cfg
->next
;
9339 disk_cfg
->next
= NULL
;
9341 if (disk_cfg
->action
== DISK_ADD
) {
9342 disk_cfg
->next
= super
->disks
;
9343 super
->disks
= disk_cfg
;
9345 dprintf("added %x:%x\n",
9346 disk_cfg
->major
, disk_cfg
->minor
);
9347 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9348 dprintf("Disk remove action processed: %x.%x\n",
9349 disk_cfg
->major
, disk_cfg
->minor
);
9350 disk
= get_disk_super(super
,
9354 /* store action status */
9355 disk
->action
= DISK_REMOVE
;
9356 /* remove spare disks only */
9357 if (disk
->index
== -1) {
9358 remove_disk_super(super
,
9362 disk_cfg
->fd
= disk
->fd
;
9366 /* release allocate disk structure */
9367 __free_imsm_disk(disk_cfg
);
9370 return check_degraded
;
9373 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9374 struct intel_super
*super
,
9377 struct intel_dev
*id
;
9378 void **tofree
= NULL
;
9381 dprintf("(enter)\n");
9382 if (u
->subdev
< 0 || u
->subdev
> 1) {
9383 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9386 if (space_list
== NULL
|| *space_list
== NULL
) {
9387 dprintf("imsm: Error: Memory is not allocated\n");
9391 for (id
= super
->devlist
; id
; id
= id
->next
) {
9392 if (id
->index
== (unsigned)u
->subdev
) {
9393 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9394 struct imsm_map
*map
;
9395 struct imsm_dev
*new_dev
=
9396 (struct imsm_dev
*)*space_list
;
9397 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9399 struct dl
*new_disk
;
9401 if (new_dev
== NULL
)
9403 *space_list
= **space_list
;
9404 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9405 map
= get_imsm_map(new_dev
, MAP_0
);
9407 dprintf("imsm: Error: migration in progress");
9411 to_state
= map
->map_state
;
9412 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9414 /* this should not happen */
9415 if (u
->new_disks
[0] < 0) {
9416 map
->failed_disk_num
=
9417 map
->num_members
- 1;
9418 to_state
= IMSM_T_STATE_DEGRADED
;
9420 to_state
= IMSM_T_STATE_NORMAL
;
9422 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9423 if (u
->new_level
> -1)
9424 map
->raid_level
= u
->new_level
;
9425 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9426 if ((u
->new_level
== 5) &&
9427 (migr_map
->raid_level
== 0)) {
9428 int ord
= map
->num_members
- 1;
9429 migr_map
->num_members
--;
9430 if (u
->new_disks
[0] < 0)
9431 ord
|= IMSM_ORD_REBUILD
;
9432 set_imsm_ord_tbl_ent(map
,
9433 map
->num_members
- 1,
9437 tofree
= (void **)dev
;
9439 /* update chunk size
9441 if (u
->new_chunksize
> 0) {
9442 unsigned long long num_data_stripes
;
9443 struct imsm_map
*dest_map
=
9444 get_imsm_map(dev
, MAP_0
);
9446 imsm_num_data_members(dest_map
);
9448 if (used_disks
== 0)
9451 map
->blocks_per_strip
=
9452 __cpu_to_le16(u
->new_chunksize
* 2);
9454 imsm_dev_size(dev
) / used_disks
;
9455 num_data_stripes
/= map
->blocks_per_strip
;
9456 num_data_stripes
/= map
->num_domains
;
9457 set_num_data_stripes(map
, num_data_stripes
);
9460 /* ensure blocks_per_member has valid value
9462 set_blocks_per_member(map
,
9463 per_dev_array_size(map
) +
9464 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9468 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9469 migr_map
->raid_level
== map
->raid_level
)
9472 if (u
->new_disks
[0] >= 0) {
9475 new_disk
= get_disk_super(super
,
9476 major(u
->new_disks
[0]),
9477 minor(u
->new_disks
[0]));
9478 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9479 major(u
->new_disks
[0]),
9480 minor(u
->new_disks
[0]),
9481 new_disk
, new_disk
->index
);
9482 if (new_disk
== NULL
)
9483 goto error_disk_add
;
9485 new_disk
->index
= map
->num_members
- 1;
9486 /* slot to fill in autolayout
9488 new_disk
->raiddisk
= new_disk
->index
;
9489 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9490 new_disk
->disk
.status
&= ~SPARE_DISK
;
9492 goto error_disk_add
;
9495 *tofree
= *space_list
;
9496 /* calculate new size
9498 imsm_set_array_size(new_dev
, -1);
9505 *space_list
= tofree
;
9509 dprintf("Error: imsm: Cannot find disk.\n");
9513 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9514 struct intel_super
*super
)
9516 struct intel_dev
*id
;
9519 dprintf("(enter)\n");
9520 if (u
->subdev
< 0 || u
->subdev
> 1) {
9521 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9525 for (id
= super
->devlist
; id
; id
= id
->next
) {
9526 if (id
->index
== (unsigned)u
->subdev
) {
9527 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9528 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9529 int used_disks
= imsm_num_data_members(map
);
9530 unsigned long long blocks_per_member
;
9531 unsigned long long num_data_stripes
;
9532 unsigned long long new_size_per_disk
;
9534 if (used_disks
== 0)
9537 /* calculate new size
9539 new_size_per_disk
= u
->new_size
/ used_disks
;
9540 blocks_per_member
= new_size_per_disk
+
9541 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9542 num_data_stripes
= new_size_per_disk
/
9543 map
->blocks_per_strip
;
9544 num_data_stripes
/= map
->num_domains
;
9545 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9546 u
->new_size
, new_size_per_disk
,
9548 set_blocks_per_member(map
, blocks_per_member
);
9549 set_num_data_stripes(map
, num_data_stripes
);
9550 imsm_set_array_size(dev
, u
->new_size
);
9560 static int prepare_spare_to_activate(struct supertype
*st
,
9561 struct imsm_update_activate_spare
*u
)
9563 struct intel_super
*super
= st
->sb
;
9564 int prev_current_vol
= super
->current_vol
;
9565 struct active_array
*a
;
9568 for (a
= st
->arrays
; a
; a
= a
->next
)
9570 * Additional initialization (adding bitmap header, filling
9571 * the bitmap area with '1's to force initial rebuild for a whole
9572 * data-area) is required when adding the spare to the volume
9573 * with write-intent bitmap.
9575 if (a
->info
.container_member
== u
->array
&&
9576 a
->info
.consistency_policy
== CONSISTENCY_POLICY_BITMAP
) {
9579 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9585 super
->current_vol
= u
->array
;
9586 if (st
->ss
->write_bitmap(st
, dl
->fd
, NoUpdate
))
9588 super
->current_vol
= prev_current_vol
;
9593 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9594 struct intel_super
*super
,
9595 struct active_array
*active_array
)
9597 struct imsm_super
*mpb
= super
->anchor
;
9598 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9599 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9600 struct imsm_map
*migr_map
;
9601 struct active_array
*a
;
9602 struct imsm_disk
*disk
;
9609 int second_map_created
= 0;
9611 for (; u
; u
= u
->next
) {
9612 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9617 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9622 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9627 /* count failures (excluding rebuilds and the victim)
9628 * to determine map[0] state
9631 for (i
= 0; i
< map
->num_members
; i
++) {
9634 disk
= get_imsm_disk(super
,
9635 get_imsm_disk_idx(dev
, i
, MAP_X
));
9636 if (!disk
|| is_failed(disk
))
9640 /* adding a pristine spare, assign a new index */
9641 if (dl
->index
< 0) {
9642 dl
->index
= super
->anchor
->num_disks
;
9643 super
->anchor
->num_disks
++;
9646 disk
->status
|= CONFIGURED_DISK
;
9647 disk
->status
&= ~SPARE_DISK
;
9650 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9651 if (!second_map_created
) {
9652 second_map_created
= 1;
9653 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9654 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9656 map
->map_state
= to_state
;
9657 migr_map
= get_imsm_map(dev
, MAP_1
);
9658 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9659 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9660 dl
->index
| IMSM_ORD_REBUILD
);
9662 /* update the family_num to mark a new container
9663 * generation, being careful to record the existing
9664 * family_num in orig_family_num to clean up after
9665 * earlier mdadm versions that neglected to set it.
9667 if (mpb
->orig_family_num
== 0)
9668 mpb
->orig_family_num
= mpb
->family_num
;
9669 mpb
->family_num
+= super
->random
;
9671 /* count arrays using the victim in the metadata */
9673 for (a
= active_array
; a
; a
= a
->next
) {
9674 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9675 map
= get_imsm_map(dev
, MAP_0
);
9677 if (get_imsm_disk_slot(map
, victim
) >= 0)
9681 /* delete the victim if it is no longer being
9687 /* We know that 'manager' isn't touching anything,
9688 * so it is safe to delete
9690 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9691 if ((*dlp
)->index
== victim
)
9694 /* victim may be on the missing list */
9696 for (dlp
= &super
->missing
; *dlp
;
9697 dlp
= &(*dlp
)->next
)
9698 if ((*dlp
)->index
== victim
)
9700 imsm_delete(super
, dlp
, victim
);
9707 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9708 struct intel_super
*super
,
9711 struct dl
*new_disk
;
9712 struct intel_dev
*id
;
9714 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9715 int disk_count
= u
->old_raid_disks
;
9716 void **tofree
= NULL
;
9717 int devices_to_reshape
= 1;
9718 struct imsm_super
*mpb
= super
->anchor
;
9720 unsigned int dev_id
;
9722 dprintf("(enter)\n");
9724 /* enable spares to use in array */
9725 for (i
= 0; i
< delta_disks
; i
++) {
9726 new_disk
= get_disk_super(super
,
9727 major(u
->new_disks
[i
]),
9728 minor(u
->new_disks
[i
]));
9729 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9730 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9731 new_disk
, new_disk
->index
);
9732 if (new_disk
== NULL
||
9733 (new_disk
->index
>= 0 &&
9734 new_disk
->index
< u
->old_raid_disks
))
9735 goto update_reshape_exit
;
9736 new_disk
->index
= disk_count
++;
9737 /* slot to fill in autolayout
9739 new_disk
->raiddisk
= new_disk
->index
;
9740 new_disk
->disk
.status
|=
9742 new_disk
->disk
.status
&= ~SPARE_DISK
;
9745 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9746 mpb
->num_raid_devs
);
9747 /* manage changes in volume
9749 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9750 void **sp
= *space_list
;
9751 struct imsm_dev
*newdev
;
9752 struct imsm_map
*newmap
, *oldmap
;
9754 for (id
= super
->devlist
; id
; id
= id
->next
) {
9755 if (id
->index
== dev_id
)
9764 /* Copy the dev, but not (all of) the map */
9765 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9766 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9767 newmap
= get_imsm_map(newdev
, MAP_0
);
9768 /* Copy the current map */
9769 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9770 /* update one device only
9772 if (devices_to_reshape
) {
9773 dprintf("imsm: modifying subdev: %i\n",
9775 devices_to_reshape
--;
9776 newdev
->vol
.migr_state
= 1;
9777 set_vol_curr_migr_unit(newdev
, 0);
9778 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9779 newmap
->num_members
= u
->new_raid_disks
;
9780 for (i
= 0; i
< delta_disks
; i
++) {
9781 set_imsm_ord_tbl_ent(newmap
,
9782 u
->old_raid_disks
+ i
,
9783 u
->old_raid_disks
+ i
);
9785 /* New map is correct, now need to save old map
9787 newmap
= get_imsm_map(newdev
, MAP_1
);
9788 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9790 imsm_set_array_size(newdev
, -1);
9793 sp
= (void **)id
->dev
;
9798 /* Clear migration record */
9799 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9802 *space_list
= tofree
;
9805 update_reshape_exit
:
9810 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9811 struct intel_super
*super
,
9814 struct imsm_dev
*dev
= NULL
;
9815 struct intel_dev
*dv
;
9816 struct imsm_dev
*dev_new
;
9817 struct imsm_map
*map
;
9821 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9822 if (dv
->index
== (unsigned int)u
->subarray
) {
9830 map
= get_imsm_map(dev
, MAP_0
);
9832 if (u
->direction
== R10_TO_R0
) {
9833 unsigned long long num_data_stripes
;
9835 /* Number of failed disks must be half of initial disk number */
9836 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9837 (map
->num_members
/ 2))
9840 /* iterate through devices to mark removed disks as spare */
9841 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9842 if (dm
->disk
.status
& FAILED_DISK
) {
9843 int idx
= dm
->index
;
9844 /* update indexes on the disk list */
9845 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9846 the index values will end up being correct.... NB */
9847 for (du
= super
->disks
; du
; du
= du
->next
)
9848 if (du
->index
> idx
)
9850 /* mark as spare disk */
9855 map
->num_members
= map
->num_members
/ 2;
9856 map
->map_state
= IMSM_T_STATE_NORMAL
;
9857 map
->num_domains
= 1;
9858 map
->raid_level
= 0;
9859 map
->failed_disk_num
= -1;
9860 num_data_stripes
= imsm_dev_size(dev
) / 2;
9861 num_data_stripes
/= map
->blocks_per_strip
;
9862 set_num_data_stripes(map
, num_data_stripes
);
9865 if (u
->direction
== R0_TO_R10
) {
9867 unsigned long long num_data_stripes
;
9869 /* update slots in current disk list */
9870 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9874 /* create new *missing* disks */
9875 for (i
= 0; i
< map
->num_members
; i
++) {
9876 space
= *space_list
;
9879 *space_list
= *space
;
9881 memcpy(du
, super
->disks
, sizeof(*du
));
9885 du
->index
= (i
* 2) + 1;
9886 sprintf((char *)du
->disk
.serial
,
9887 " MISSING_%d", du
->index
);
9888 sprintf((char *)du
->serial
,
9889 "MISSING_%d", du
->index
);
9890 du
->next
= super
->missing
;
9891 super
->missing
= du
;
9893 /* create new dev and map */
9894 space
= *space_list
;
9897 *space_list
= *space
;
9898 dev_new
= (void *)space
;
9899 memcpy(dev_new
, dev
, sizeof(*dev
));
9900 /* update new map */
9901 map
= get_imsm_map(dev_new
, MAP_0
);
9902 map
->num_members
= map
->num_members
* 2;
9903 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9904 map
->num_domains
= 2;
9905 map
->raid_level
= 1;
9906 num_data_stripes
= imsm_dev_size(dev
) / 2;
9907 num_data_stripes
/= map
->blocks_per_strip
;
9908 num_data_stripes
/= map
->num_domains
;
9909 set_num_data_stripes(map
, num_data_stripes
);
9911 /* replace dev<->dev_new */
9914 /* update disk order table */
9915 for (du
= super
->disks
; du
; du
= du
->next
)
9917 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9918 for (du
= super
->missing
; du
; du
= du
->next
)
9919 if (du
->index
>= 0) {
9920 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9921 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9927 static void imsm_process_update(struct supertype
*st
,
9928 struct metadata_update
*update
)
9931 * crack open the metadata_update envelope to find the update record
9932 * update can be one of:
9933 * update_reshape_container_disks - all the arrays in the container
9934 * are being reshaped to have more devices. We need to mark
9935 * the arrays for general migration and convert selected spares
9936 * into active devices.
9937 * update_activate_spare - a spare device has replaced a failed
9938 * device in an array, update the disk_ord_tbl. If this disk is
9939 * present in all member arrays then also clear the SPARE_DISK
9941 * update_create_array
9943 * update_rename_array
9944 * update_add_remove_disk
9946 struct intel_super
*super
= st
->sb
;
9947 struct imsm_super
*mpb
;
9948 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9950 /* update requires a larger buf but the allocation failed */
9951 if (super
->next_len
&& !super
->next_buf
) {
9952 super
->next_len
= 0;
9956 if (super
->next_buf
) {
9957 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9959 super
->len
= super
->next_len
;
9960 super
->buf
= super
->next_buf
;
9962 super
->next_len
= 0;
9963 super
->next_buf
= NULL
;
9966 mpb
= super
->anchor
;
9969 case update_general_migration_checkpoint
: {
9970 struct intel_dev
*id
;
9971 struct imsm_update_general_migration_checkpoint
*u
=
9972 (void *)update
->buf
;
9974 dprintf("called for update_general_migration_checkpoint\n");
9976 /* find device under general migration */
9977 for (id
= super
->devlist
; id
; id
= id
->next
) {
9978 if (is_gen_migration(id
->dev
)) {
9979 set_vol_curr_migr_unit(id
->dev
,
9981 super
->updates_pending
++;
9986 case update_takeover
: {
9987 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9988 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9989 imsm_update_version_info(super
);
9990 super
->updates_pending
++;
9995 case update_reshape_container_disks
: {
9996 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9997 if (apply_reshape_container_disks_update(
9998 u
, super
, &update
->space_list
))
9999 super
->updates_pending
++;
10002 case update_reshape_migration
: {
10003 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10004 if (apply_reshape_migration_update(
10005 u
, super
, &update
->space_list
))
10006 super
->updates_pending
++;
10009 case update_size_change
: {
10010 struct imsm_update_size_change
*u
= (void *)update
->buf
;
10011 if (apply_size_change_update(u
, super
))
10012 super
->updates_pending
++;
10015 case update_activate_spare
: {
10016 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
10018 if (prepare_spare_to_activate(st
, u
) &&
10019 apply_update_activate_spare(u
, super
, st
->arrays
))
10020 super
->updates_pending
++;
10023 case update_create_array
: {
10024 /* someone wants to create a new array, we need to be aware of
10025 * a few races/collisions:
10026 * 1/ 'Create' called by two separate instances of mdadm
10027 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
10028 * devices that have since been assimilated via
10030 * In the event this update can not be carried out mdadm will
10031 * (FIX ME) notice that its update did not take hold.
10033 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10034 struct intel_dev
*dv
;
10035 struct imsm_dev
*dev
;
10036 struct imsm_map
*map
, *new_map
;
10037 unsigned long long start
, end
;
10038 unsigned long long new_start
, new_end
;
10040 struct disk_info
*inf
;
10043 /* handle racing creates: first come first serve */
10044 if (u
->dev_idx
< mpb
->num_raid_devs
) {
10045 dprintf("subarray %d already defined\n", u
->dev_idx
);
10049 /* check update is next in sequence */
10050 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
10051 dprintf("can not create array %d expected index %d\n",
10052 u
->dev_idx
, mpb
->num_raid_devs
);
10056 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
10057 new_start
= pba_of_lba0(new_map
);
10058 new_end
= new_start
+ per_dev_array_size(new_map
);
10059 inf
= get_disk_info(u
);
10061 /* handle activate_spare versus create race:
10062 * check to make sure that overlapping arrays do not include
10063 * overalpping disks
10065 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10066 dev
= get_imsm_dev(super
, i
);
10067 map
= get_imsm_map(dev
, MAP_0
);
10068 start
= pba_of_lba0(map
);
10069 end
= start
+ per_dev_array_size(map
);
10070 if ((new_start
>= start
&& new_start
<= end
) ||
10071 (start
>= new_start
&& start
<= new_end
))
10076 if (disks_overlap(super
, i
, u
)) {
10077 dprintf("arrays overlap\n");
10082 /* check that prepare update was successful */
10083 if (!update
->space
) {
10084 dprintf("prepare update failed\n");
10088 /* check that all disks are still active before committing
10089 * changes. FIXME: could we instead handle this by creating a
10090 * degraded array? That's probably not what the user expects,
10091 * so better to drop this update on the floor.
10093 for (i
= 0; i
< new_map
->num_members
; i
++) {
10094 dl
= serial_to_dl(inf
[i
].serial
, super
);
10096 dprintf("disk disappeared\n");
10101 super
->updates_pending
++;
10103 /* convert spares to members and fixup ord_tbl */
10104 for (i
= 0; i
< new_map
->num_members
; i
++) {
10105 dl
= serial_to_dl(inf
[i
].serial
, super
);
10106 if (dl
->index
== -1) {
10107 dl
->index
= mpb
->num_disks
;
10109 dl
->disk
.status
|= CONFIGURED_DISK
;
10110 dl
->disk
.status
&= ~SPARE_DISK
;
10112 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
10115 dv
= update
->space
;
10117 update
->space
= NULL
;
10118 imsm_copy_dev(dev
, &u
->dev
);
10119 dv
->index
= u
->dev_idx
;
10120 dv
->next
= super
->devlist
;
10121 super
->devlist
= dv
;
10122 mpb
->num_raid_devs
++;
10124 imsm_update_version_info(super
);
10127 /* mdmon knows how to release update->space, but not
10128 * ((struct intel_dev *) update->space)->dev
10130 if (update
->space
) {
10131 dv
= update
->space
;
10136 case update_kill_array
: {
10137 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
10138 int victim
= u
->dev_idx
;
10139 struct active_array
*a
;
10140 struct intel_dev
**dp
;
10141 struct imsm_dev
*dev
;
10143 /* sanity check that we are not affecting the uuid of
10144 * active arrays, or deleting an active array
10146 * FIXME when immutable ids are available, but note that
10147 * we'll also need to fixup the invalidated/active
10148 * subarray indexes in mdstat
10150 for (a
= st
->arrays
; a
; a
= a
->next
)
10151 if (a
->info
.container_member
>= victim
)
10153 /* by definition if mdmon is running at least one array
10154 * is active in the container, so checking
10155 * mpb->num_raid_devs is just extra paranoia
10157 dev
= get_imsm_dev(super
, victim
);
10158 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
10159 dprintf("failed to delete subarray-%d\n", victim
);
10163 for (dp
= &super
->devlist
; *dp
;)
10164 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
10167 if ((*dp
)->index
> (unsigned)victim
)
10171 mpb
->num_raid_devs
--;
10172 super
->updates_pending
++;
10175 case update_rename_array
: {
10176 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
10177 char name
[MAX_RAID_SERIAL_LEN
+1];
10178 int target
= u
->dev_idx
;
10179 struct active_array
*a
;
10180 struct imsm_dev
*dev
;
10182 /* sanity check that we are not affecting the uuid of
10185 memset(name
, 0, sizeof(name
));
10186 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
10187 name
[MAX_RAID_SERIAL_LEN
] = '\0';
10188 for (a
= st
->arrays
; a
; a
= a
->next
)
10189 if (a
->info
.container_member
== target
)
10191 dev
= get_imsm_dev(super
, u
->dev_idx
);
10192 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
10193 dprintf("failed to rename subarray-%d\n", target
);
10197 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
10198 super
->updates_pending
++;
10201 case update_add_remove_disk
: {
10202 /* we may be able to repair some arrays if disks are
10203 * being added, check the status of add_remove_disk
10204 * if discs has been added.
10206 if (add_remove_disk_update(super
)) {
10207 struct active_array
*a
;
10209 super
->updates_pending
++;
10210 for (a
= st
->arrays
; a
; a
= a
->next
)
10211 a
->check_degraded
= 1;
10215 case update_prealloc_badblocks_mem
:
10217 case update_rwh_policy
: {
10218 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
10219 int target
= u
->dev_idx
;
10220 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
10222 dprintf("could not find subarray-%d\n", target
);
10226 if (dev
->rwh_policy
!= u
->new_policy
) {
10227 dev
->rwh_policy
= u
->new_policy
;
10228 super
->updates_pending
++;
10233 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10237 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10239 static int imsm_prepare_update(struct supertype
*st
,
10240 struct metadata_update
*update
)
10243 * Allocate space to hold new disk entries, raid-device entries or a new
10244 * mpb if necessary. The manager synchronously waits for updates to
10245 * complete in the monitor, so new mpb buffers allocated here can be
10246 * integrated by the monitor thread without worrying about live pointers
10247 * in the manager thread.
10249 enum imsm_update_type type
;
10250 struct intel_super
*super
= st
->sb
;
10251 unsigned int sector_size
= super
->sector_size
;
10252 struct imsm_super
*mpb
= super
->anchor
;
10256 if (update
->len
< (int)sizeof(type
))
10259 type
= *(enum imsm_update_type
*) update
->buf
;
10262 case update_general_migration_checkpoint
:
10263 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10265 dprintf("called for update_general_migration_checkpoint\n");
10267 case update_takeover
: {
10268 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10269 if (update
->len
< (int)sizeof(*u
))
10271 if (u
->direction
== R0_TO_R10
) {
10272 void **tail
= (void **)&update
->space_list
;
10273 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10274 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10275 int num_members
= map
->num_members
;
10278 /* allocate memory for added disks */
10279 for (i
= 0; i
< num_members
; i
++) {
10280 size
= sizeof(struct dl
);
10281 space
= xmalloc(size
);
10286 /* allocate memory for new device */
10287 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10288 (num_members
* sizeof(__u32
));
10289 space
= xmalloc(size
);
10293 len
= disks_to_mpb_size(num_members
* 2);
10298 case update_reshape_container_disks
: {
10299 /* Every raid device in the container is about to
10300 * gain some more devices, and we will enter a
10302 * So each 'imsm_map' will be bigger, and the imsm_vol
10303 * will now hold 2 of them.
10304 * Thus we need new 'struct imsm_dev' allocations sized
10305 * as sizeof_imsm_dev but with more devices in both maps.
10307 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10308 struct intel_dev
*dl
;
10309 void **space_tail
= (void**)&update
->space_list
;
10311 if (update
->len
< (int)sizeof(*u
))
10314 dprintf("for update_reshape\n");
10316 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10317 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10319 if (u
->new_raid_disks
> u
->old_raid_disks
)
10320 size
+= sizeof(__u32
)*2*
10321 (u
->new_raid_disks
- u
->old_raid_disks
);
10325 *space_tail
= NULL
;
10328 len
= disks_to_mpb_size(u
->new_raid_disks
);
10329 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10332 case update_reshape_migration
: {
10333 /* for migration level 0->5 we need to add disks
10334 * so the same as for container operation we will copy
10335 * device to the bigger location.
10336 * in memory prepared device and new disk area are prepared
10337 * for usage in process update
10339 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10340 struct intel_dev
*id
;
10341 void **space_tail
= (void **)&update
->space_list
;
10344 int current_level
= -1;
10346 if (update
->len
< (int)sizeof(*u
))
10349 dprintf("for update_reshape\n");
10351 /* add space for bigger array in update
10353 for (id
= super
->devlist
; id
; id
= id
->next
) {
10354 if (id
->index
== (unsigned)u
->subdev
) {
10355 size
= sizeof_imsm_dev(id
->dev
, 1);
10356 if (u
->new_raid_disks
> u
->old_raid_disks
)
10357 size
+= sizeof(__u32
)*2*
10358 (u
->new_raid_disks
- u
->old_raid_disks
);
10362 *space_tail
= NULL
;
10366 if (update
->space_list
== NULL
)
10369 /* add space for disk in update
10371 size
= sizeof(struct dl
);
10375 *space_tail
= NULL
;
10377 /* add spare device to update
10379 for (id
= super
->devlist
; id
; id
= id
->next
)
10380 if (id
->index
== (unsigned)u
->subdev
) {
10381 struct imsm_dev
*dev
;
10382 struct imsm_map
*map
;
10384 dev
= get_imsm_dev(super
, u
->subdev
);
10385 map
= get_imsm_map(dev
, MAP_0
);
10386 current_level
= map
->raid_level
;
10389 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10390 struct mdinfo
*spares
;
10392 spares
= get_spares_for_grow(st
);
10395 struct mdinfo
*dev
;
10397 dev
= spares
->devs
;
10400 makedev(dev
->disk
.major
,
10402 dl
= get_disk_super(super
,
10405 dl
->index
= u
->old_raid_disks
;
10408 sysfs_free(spares
);
10411 len
= disks_to_mpb_size(u
->new_raid_disks
);
10412 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10415 case update_size_change
: {
10416 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10420 case update_activate_spare
: {
10421 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10425 case update_create_array
: {
10426 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10427 struct intel_dev
*dv
;
10428 struct imsm_dev
*dev
= &u
->dev
;
10429 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10431 struct disk_info
*inf
;
10435 if (update
->len
< (int)sizeof(*u
))
10438 inf
= get_disk_info(u
);
10439 len
= sizeof_imsm_dev(dev
, 1);
10440 /* allocate a new super->devlist entry */
10441 dv
= xmalloc(sizeof(*dv
));
10442 dv
->dev
= xmalloc(len
);
10443 update
->space
= dv
;
10445 /* count how many spares will be converted to members */
10446 for (i
= 0; i
< map
->num_members
; i
++) {
10447 dl
= serial_to_dl(inf
[i
].serial
, super
);
10449 /* hmm maybe it failed?, nothing we can do about
10454 if (count_memberships(dl
, super
) == 0)
10457 len
+= activate
* sizeof(struct imsm_disk
);
10460 case update_kill_array
: {
10461 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10465 case update_rename_array
: {
10466 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10470 case update_add_remove_disk
:
10471 /* no update->len needed */
10473 case update_prealloc_badblocks_mem
:
10474 super
->extra_space
+= sizeof(struct bbm_log
) -
10475 get_imsm_bbm_log_size(super
->bbm_log
);
10477 case update_rwh_policy
: {
10478 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10486 /* check if we need a larger metadata buffer */
10487 if (super
->next_buf
)
10488 buf_len
= super
->next_len
;
10490 buf_len
= super
->len
;
10492 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10493 /* ok we need a larger buf than what is currently allocated
10494 * if this allocation fails process_update will notice that
10495 * ->next_len is set and ->next_buf is NULL
10497 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10498 super
->extra_space
+ len
, sector_size
);
10499 if (super
->next_buf
)
10500 free(super
->next_buf
);
10502 super
->next_len
= buf_len
;
10503 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10504 memset(super
->next_buf
, 0, buf_len
);
10506 super
->next_buf
= NULL
;
10511 /* must be called while manager is quiesced */
10512 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10514 struct imsm_super
*mpb
= super
->anchor
;
10516 struct imsm_dev
*dev
;
10517 struct imsm_map
*map
;
10518 unsigned int i
, j
, num_members
;
10519 __u32 ord
, ord_map0
;
10520 struct bbm_log
*log
= super
->bbm_log
;
10522 dprintf("deleting device[%d] from imsm_super\n", index
);
10524 /* shift all indexes down one */
10525 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10526 if (iter
->index
> (int)index
)
10528 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10529 if (iter
->index
> (int)index
)
10532 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10533 dev
= get_imsm_dev(super
, i
);
10534 map
= get_imsm_map(dev
, MAP_0
);
10535 num_members
= map
->num_members
;
10536 for (j
= 0; j
< num_members
; j
++) {
10537 /* update ord entries being careful not to propagate
10538 * ord-flags to the first map
10540 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10541 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10543 if (ord_to_idx(ord
) <= index
)
10546 map
= get_imsm_map(dev
, MAP_0
);
10547 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10548 map
= get_imsm_map(dev
, MAP_1
);
10550 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10554 for (i
= 0; i
< log
->entry_count
; i
++) {
10555 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10557 if (entry
->disk_ordinal
<= index
)
10559 entry
->disk_ordinal
--;
10563 super
->updates_pending
++;
10565 struct dl
*dl
= *dlp
;
10567 *dlp
= (*dlp
)->next
;
10568 __free_imsm_disk(dl
);
10572 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10573 struct intel_super
*super
,
10574 struct imsm_dev
*dev
)
10580 struct imsm_map
*map
;
10583 ret_val
= raid_disks
/2;
10584 /* check map if all disks pairs not failed
10587 map
= get_imsm_map(dev
, MAP_0
);
10588 for (i
= 0; i
< ret_val
; i
++) {
10589 int degradation
= 0;
10590 if (get_imsm_disk(super
, i
) == NULL
)
10592 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10594 if (degradation
== 2)
10597 map
= get_imsm_map(dev
, MAP_1
);
10598 /* if there is no second map
10599 * result can be returned
10603 /* check degradation in second map
10605 for (i
= 0; i
< ret_val
; i
++) {
10606 int degradation
= 0;
10607 if (get_imsm_disk(super
, i
) == NULL
)
10609 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10611 if (degradation
== 2)
10625 /*******************************************************************************
10626 * Function: validate_container_imsm
10627 * Description: This routine validates container after assemble,
10628 * eg. if devices in container are under the same controller.
10631 * info : linked list with info about devices used in array
10635 ******************************************************************************/
10636 int validate_container_imsm(struct mdinfo
*info
)
10638 if (check_env("IMSM_NO_PLATFORM"))
10641 struct sys_dev
*idev
;
10642 struct sys_dev
*hba
= NULL
;
10643 struct sys_dev
*intel_devices
= find_intel_devices();
10644 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10645 info
->disk
.minor
));
10647 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10648 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10657 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10658 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10662 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10663 struct mdinfo
*dev
;
10665 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10666 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10668 struct sys_dev
*hba2
= NULL
;
10669 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10670 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10678 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10679 get_orom_by_device_id(hba2
->dev_id
);
10681 if (hba2
&& hba
->type
!= hba2
->type
) {
10682 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10683 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10687 if (orom
!= orom2
) {
10688 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10689 " This operation is not supported and can lead to data loss.\n");
10694 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10695 " This operation is not supported and can lead to data loss.\n");
10703 /*******************************************************************************
10704 * Function: imsm_record_badblock
10705 * Description: This routine stores new bad block record in BBM log
10708 * a : array containing a bad block
10709 * slot : disk number containing a bad block
10710 * sector : bad block sector
10711 * length : bad block sectors range
10715 ******************************************************************************/
10716 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10717 unsigned long long sector
, int length
)
10719 struct intel_super
*super
= a
->container
->sb
;
10723 ord
= imsm_disk_slot_to_ord(a
, slot
);
10727 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10730 super
->updates_pending
++;
10734 /*******************************************************************************
10735 * Function: imsm_clear_badblock
10736 * Description: This routine clears bad block record from BBM log
10739 * a : array containing a bad block
10740 * slot : disk number containing a bad block
10741 * sector : bad block sector
10742 * length : bad block sectors range
10746 ******************************************************************************/
10747 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10748 unsigned long long sector
, int length
)
10750 struct intel_super
*super
= a
->container
->sb
;
10754 ord
= imsm_disk_slot_to_ord(a
, slot
);
10758 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10760 super
->updates_pending
++;
10764 /*******************************************************************************
10765 * Function: imsm_get_badblocks
10766 * Description: This routine get list of bad blocks for an array
10770 * slot : disk number
10772 * bb : structure containing bad blocks
10774 ******************************************************************************/
10775 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10777 int inst
= a
->info
.container_member
;
10778 struct intel_super
*super
= a
->container
->sb
;
10779 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10780 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10783 ord
= imsm_disk_slot_to_ord(a
, slot
);
10787 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10788 per_dev_array_size(map
), &super
->bb
);
10792 /*******************************************************************************
10793 * Function: examine_badblocks_imsm
10794 * Description: Prints list of bad blocks on a disk to the standard output
10797 * st : metadata handler
10798 * fd : open file descriptor for device
10799 * devname : device name
10803 ******************************************************************************/
10804 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10806 struct intel_super
*super
= st
->sb
;
10807 struct bbm_log
*log
= super
->bbm_log
;
10808 struct dl
*d
= NULL
;
10811 for (d
= super
->disks
; d
; d
= d
->next
) {
10812 if (strcmp(d
->devname
, devname
) == 0)
10816 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10817 pr_err("%s doesn't appear to be part of a raid array\n",
10824 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10826 for (i
= 0; i
< log
->entry_count
; i
++) {
10827 if (entry
[i
].disk_ordinal
== d
->index
) {
10828 unsigned long long sector
= __le48_to_cpu(
10829 &entry
[i
].defective_block_start
);
10830 int cnt
= entry
[i
].marked_count
+ 1;
10833 printf("Bad-blocks on %s:\n", devname
);
10837 printf("%20llu for %d sectors\n", sector
, cnt
);
10843 printf("No bad-blocks list configured on %s\n", devname
);
10847 /*******************************************************************************
10848 * Function: init_migr_record_imsm
10849 * Description: Function inits imsm migration record
10851 * super : imsm internal array info
10852 * dev : device under migration
10853 * info : general array info to find the smallest device
10856 ******************************************************************************/
10857 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10858 struct mdinfo
*info
)
10860 struct intel_super
*super
= st
->sb
;
10861 struct migr_record
*migr_rec
= super
->migr_rec
;
10862 int new_data_disks
;
10863 unsigned long long dsize
, dev_sectors
;
10864 long long unsigned min_dev_sectors
= -1LLU;
10865 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10866 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10867 unsigned long long num_migr_units
;
10868 unsigned long long array_blocks
;
10869 struct dl
*dl_disk
= NULL
;
10871 memset(migr_rec
, 0, sizeof(struct migr_record
));
10872 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10874 /* only ascending reshape supported now */
10875 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10877 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10878 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10879 migr_rec
->dest_depth_per_unit
*=
10880 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10881 new_data_disks
= imsm_num_data_members(map_dest
);
10882 migr_rec
->blocks_per_unit
=
10883 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10884 migr_rec
->dest_depth_per_unit
=
10885 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10886 array_blocks
= info
->component_size
* new_data_disks
;
10888 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10890 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10892 set_num_migr_units(migr_rec
, num_migr_units
);
10894 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10895 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10897 /* Find the smallest dev */
10898 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
10899 /* ignore spares in container */
10900 if (dl_disk
->index
< 0)
10902 get_dev_size(dl_disk
->fd
, NULL
, &dsize
);
10903 dev_sectors
= dsize
/ 512;
10904 if (dev_sectors
< min_dev_sectors
)
10905 min_dev_sectors
= dev_sectors
;
10907 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
10908 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10910 write_imsm_migr_rec(st
);
10915 /*******************************************************************************
10916 * Function: save_backup_imsm
10917 * Description: Function saves critical data stripes to Migration Copy Area
10918 * and updates the current migration unit status.
10919 * Use restore_stripes() to form a destination stripe,
10920 * and to write it to the Copy Area.
10922 * st : supertype information
10923 * dev : imsm device that backup is saved for
10924 * info : general array info
10925 * buf : input buffer
10926 * length : length of data to backup (blocks_per_unit)
10930 ******************************************************************************/
10931 int save_backup_imsm(struct supertype
*st
,
10932 struct imsm_dev
*dev
,
10933 struct mdinfo
*info
,
10938 struct intel_super
*super
= st
->sb
;
10939 unsigned long long *target_offsets
;
10942 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10943 int new_disks
= map_dest
->num_members
;
10944 int dest_layout
= 0;
10946 unsigned long long start
;
10947 int data_disks
= imsm_num_data_members(map_dest
);
10949 targets
= xmalloc(new_disks
* sizeof(int));
10951 for (i
= 0; i
< new_disks
; i
++) {
10952 struct dl
*dl_disk
= get_imsm_dl_disk(super
, i
);
10954 targets
[i
] = dl_disk
->fd
;
10957 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10959 start
= info
->reshape_progress
* 512;
10960 for (i
= 0; i
< new_disks
; i
++) {
10961 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
10962 /* move back copy area adderss, it will be moved forward
10963 * in restore_stripes() using start input variable
10965 target_offsets
[i
] -= start
/data_disks
;
10968 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10969 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10971 if (restore_stripes(targets
, /* list of dest devices */
10972 target_offsets
, /* migration record offsets */
10975 map_dest
->raid_level
,
10977 -1, /* source backup file descriptor */
10978 0, /* input buf offset
10979 * always 0 buf is already offseted */
10983 pr_err("Error restoring stripes\n");
10993 free(target_offsets
);
10998 /*******************************************************************************
10999 * Function: save_checkpoint_imsm
11000 * Description: Function called for current unit status update
11001 * in the migration record. It writes it to disk.
11003 * super : imsm internal array info
11004 * info : general array info
11008 * 2: failure, means no valid migration record
11009 * / no general migration in progress /
11010 ******************************************************************************/
11011 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
11013 struct intel_super
*super
= st
->sb
;
11014 unsigned long long blocks_per_unit
;
11015 unsigned long long curr_migr_unit
;
11017 if (load_imsm_migr_rec(super
) != 0) {
11018 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
11022 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
11023 if (blocks_per_unit
== 0) {
11024 dprintf("imsm: no migration in progress.\n");
11027 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
11028 /* check if array is alligned to copy area
11029 * if it is not alligned, add one to current migration unit value
11030 * this can happend on array reshape finish only
11032 if (info
->reshape_progress
% blocks_per_unit
)
11035 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
11036 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
11037 set_migr_dest_1st_member_lba(super
->migr_rec
,
11038 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
11040 if (write_imsm_migr_rec(st
) < 0) {
11041 dprintf("imsm: Cannot write migration record outside backup area\n");
11048 /*******************************************************************************
11049 * Function: recover_backup_imsm
11050 * Description: Function recovers critical data from the Migration Copy Area
11051 * while assembling an array.
11053 * super : imsm internal array info
11054 * info : general array info
11056 * 0 : success (or there is no data to recover)
11058 ******************************************************************************/
11059 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
11061 struct intel_super
*super
= st
->sb
;
11062 struct migr_record
*migr_rec
= super
->migr_rec
;
11063 struct imsm_map
*map_dest
;
11064 struct intel_dev
*id
= NULL
;
11065 unsigned long long read_offset
;
11066 unsigned long long write_offset
;
11068 int new_disks
, err
;
11071 unsigned int sector_size
= super
->sector_size
;
11072 unsigned long long curr_migr_unit
= current_migr_unit(migr_rec
);
11073 unsigned long long num_migr_units
= get_num_migr_units(migr_rec
);
11075 int skipped_disks
= 0;
11076 struct dl
*dl_disk
;
11078 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
11082 /* recover data only during assemblation */
11083 if (strncmp(buffer
, "inactive", 8) != 0)
11085 /* no data to recover */
11086 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
11088 if (curr_migr_unit
>= num_migr_units
)
11091 /* find device during reshape */
11092 for (id
= super
->devlist
; id
; id
= id
->next
)
11093 if (is_gen_migration(id
->dev
))
11098 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
11099 new_disks
= map_dest
->num_members
;
11101 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
11103 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
11104 pba_of_lba0(map_dest
)) * 512;
11106 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11107 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
11110 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
11111 if (dl_disk
->index
< 0)
11114 if (dl_disk
->fd
< 0) {
11118 if (lseek64(dl_disk
->fd
, read_offset
, SEEK_SET
) < 0) {
11119 pr_err("Cannot seek to block: %s\n",
11124 if (read(dl_disk
->fd
, buf
, unit_len
) != unit_len
) {
11125 pr_err("Cannot read copy area block: %s\n",
11130 if (lseek64(dl_disk
->fd
, write_offset
, SEEK_SET
) < 0) {
11131 pr_err("Cannot seek to block: %s\n",
11136 if (write(dl_disk
->fd
, buf
, unit_len
) != unit_len
) {
11137 pr_err("Cannot restore block: %s\n",
11144 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11148 pr_err("Cannot restore data from backup. Too many failed disks\n");
11152 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11153 /* ignore error == 2, this can mean end of reshape here
11155 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11164 static char disk_by_path
[] = "/dev/disk/by-path/";
11166 static const char *imsm_get_disk_controller_domain(const char *path
)
11168 char disk_path
[PATH_MAX
];
11172 strcpy(disk_path
, disk_by_path
);
11173 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
11174 if (stat(disk_path
, &st
) == 0) {
11175 struct sys_dev
* hba
;
11178 path
= devt_to_devpath(st
.st_rdev
);
11181 hba
= find_disk_attached_hba(-1, path
);
11182 if (hba
&& hba
->type
== SYS_DEV_SAS
)
11184 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
11186 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
11188 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
11192 dprintf("path: %s hba: %s attached: %s\n",
11193 path
, (hba
) ? hba
->path
: "NULL", drv
);
11199 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11201 static char devnm
[32];
11202 char subdev_name
[20];
11203 struct mdstat_ent
*mdstat
;
11205 sprintf(subdev_name
, "%d", subdev
);
11206 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11210 strcpy(devnm
, mdstat
->devnm
);
11211 free_mdstat(mdstat
);
11215 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11216 struct geo_params
*geo
,
11217 int *old_raid_disks
,
11220 /* currently we only support increasing the number of devices
11221 * for a container. This increases the number of device for each
11222 * member array. They must all be RAID0 or RAID5.
11225 struct mdinfo
*info
, *member
;
11226 int devices_that_can_grow
= 0;
11228 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11230 if (geo
->size
> 0 ||
11231 geo
->level
!= UnSet
||
11232 geo
->layout
!= UnSet
||
11233 geo
->chunksize
!= 0 ||
11234 geo
->raid_disks
== UnSet
) {
11235 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11239 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11240 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11244 info
= container_content_imsm(st
, NULL
);
11245 for (member
= info
; member
; member
= member
->next
) {
11248 dprintf("imsm: checking device_num: %i\n",
11249 member
->container_member
);
11251 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11252 /* we work on container for Online Capacity Expansion
11253 * only so raid_disks has to grow
11255 dprintf("imsm: for container operation raid disks increase is required\n");
11259 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11260 /* we cannot use this container with other raid level
11262 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11263 info
->array
.level
);
11266 /* check for platform support
11267 * for this raid level configuration
11269 struct intel_super
*super
= st
->sb
;
11270 if (!is_raid_level_supported(super
->orom
,
11271 member
->array
.level
,
11272 geo
->raid_disks
)) {
11273 dprintf("platform does not support raid%d with %d disk%s\n",
11276 geo
->raid_disks
> 1 ? "s" : "");
11279 /* check if component size is aligned to chunk size
11281 if (info
->component_size
%
11282 (info
->array
.chunk_size
/512)) {
11283 dprintf("Component size is not aligned to chunk size\n");
11288 if (*old_raid_disks
&&
11289 info
->array
.raid_disks
!= *old_raid_disks
)
11291 *old_raid_disks
= info
->array
.raid_disks
;
11293 /* All raid5 and raid0 volumes in container
11294 * have to be ready for Online Capacity Expansion
11295 * so they need to be assembled. We have already
11296 * checked that no recovery etc is happening.
11298 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11299 st
->container_devnm
);
11300 if (result
== NULL
) {
11301 dprintf("imsm: cannot find array\n");
11304 devices_that_can_grow
++;
11307 if (!member
&& devices_that_can_grow
)
11311 dprintf("Container operation allowed\n");
11313 dprintf("Error: %i\n", ret_val
);
11318 /* Function: get_spares_for_grow
11319 * Description: Allocates memory and creates list of spare devices
11320 * avaliable in container. Checks if spare drive size is acceptable.
11321 * Parameters: Pointer to the supertype structure
11322 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11325 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11327 struct spare_criteria sc
;
11329 get_spare_criteria_imsm(st
, &sc
);
11330 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11333 /******************************************************************************
11334 * function: imsm_create_metadata_update_for_reshape
11335 * Function creates update for whole IMSM container.
11337 ******************************************************************************/
11338 static int imsm_create_metadata_update_for_reshape(
11339 struct supertype
*st
,
11340 struct geo_params
*geo
,
11341 int old_raid_disks
,
11342 struct imsm_update_reshape
**updatep
)
11344 struct intel_super
*super
= st
->sb
;
11345 struct imsm_super
*mpb
= super
->anchor
;
11346 int update_memory_size
;
11347 struct imsm_update_reshape
*u
;
11348 struct mdinfo
*spares
;
11351 struct mdinfo
*dev
;
11353 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11355 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11357 /* size of all update data without anchor */
11358 update_memory_size
= sizeof(struct imsm_update_reshape
);
11360 /* now add space for spare disks that we need to add. */
11361 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11363 u
= xcalloc(1, update_memory_size
);
11364 u
->type
= update_reshape_container_disks
;
11365 u
->old_raid_disks
= old_raid_disks
;
11366 u
->new_raid_disks
= geo
->raid_disks
;
11368 /* now get spare disks list
11370 spares
= get_spares_for_grow(st
);
11372 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11373 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11378 /* we have got spares
11379 * update disk list in imsm_disk list table in anchor
11381 dprintf("imsm: %i spares are available.\n\n",
11382 spares
->array
.spare_disks
);
11384 dev
= spares
->devs
;
11385 for (i
= 0; i
< delta_disks
; i
++) {
11390 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11392 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11393 dl
->index
= mpb
->num_disks
;
11401 sysfs_free(spares
);
11403 dprintf("imsm: reshape update preparation :");
11404 if (i
== delta_disks
) {
11405 dprintf_cont(" OK\n");
11407 return update_memory_size
;
11410 dprintf_cont(" Error\n");
11415 /******************************************************************************
11416 * function: imsm_create_metadata_update_for_size_change()
11417 * Creates update for IMSM array for array size change.
11419 ******************************************************************************/
11420 static int imsm_create_metadata_update_for_size_change(
11421 struct supertype
*st
,
11422 struct geo_params
*geo
,
11423 struct imsm_update_size_change
**updatep
)
11425 struct intel_super
*super
= st
->sb
;
11426 int update_memory_size
;
11427 struct imsm_update_size_change
*u
;
11429 dprintf("(enter) New size = %llu\n", geo
->size
);
11431 /* size of all update data without anchor */
11432 update_memory_size
= sizeof(struct imsm_update_size_change
);
11434 u
= xcalloc(1, update_memory_size
);
11435 u
->type
= update_size_change
;
11436 u
->subdev
= super
->current_vol
;
11437 u
->new_size
= geo
->size
;
11439 dprintf("imsm: reshape update preparation : OK\n");
11442 return update_memory_size
;
11445 /******************************************************************************
11446 * function: imsm_create_metadata_update_for_migration()
11447 * Creates update for IMSM array.
11449 ******************************************************************************/
11450 static int imsm_create_metadata_update_for_migration(
11451 struct supertype
*st
,
11452 struct geo_params
*geo
,
11453 struct imsm_update_reshape_migration
**updatep
)
11455 struct intel_super
*super
= st
->sb
;
11456 int update_memory_size
;
11457 struct imsm_update_reshape_migration
*u
;
11458 struct imsm_dev
*dev
;
11459 int previous_level
= -1;
11461 dprintf("(enter) New Level = %i\n", geo
->level
);
11463 /* size of all update data without anchor */
11464 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11466 u
= xcalloc(1, update_memory_size
);
11467 u
->type
= update_reshape_migration
;
11468 u
->subdev
= super
->current_vol
;
11469 u
->new_level
= geo
->level
;
11470 u
->new_layout
= geo
->layout
;
11471 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11472 u
->new_disks
[0] = -1;
11473 u
->new_chunksize
= -1;
11475 dev
= get_imsm_dev(super
, u
->subdev
);
11477 struct imsm_map
*map
;
11479 map
= get_imsm_map(dev
, MAP_0
);
11481 int current_chunk_size
=
11482 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11484 if (geo
->chunksize
!= current_chunk_size
) {
11485 u
->new_chunksize
= geo
->chunksize
/ 1024;
11486 dprintf("imsm: chunk size change from %i to %i\n",
11487 current_chunk_size
, u
->new_chunksize
);
11489 previous_level
= map
->raid_level
;
11492 if (geo
->level
== 5 && previous_level
== 0) {
11493 struct mdinfo
*spares
= NULL
;
11495 u
->new_raid_disks
++;
11496 spares
= get_spares_for_grow(st
);
11497 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11499 sysfs_free(spares
);
11500 update_memory_size
= 0;
11501 pr_err("cannot get spare device for requested migration\n");
11504 sysfs_free(spares
);
11506 dprintf("imsm: reshape update preparation : OK\n");
11509 return update_memory_size
;
11512 static void imsm_update_metadata_locally(struct supertype
*st
,
11513 void *buf
, int len
)
11515 struct metadata_update mu
;
11520 mu
.space_list
= NULL
;
11522 if (imsm_prepare_update(st
, &mu
))
11523 imsm_process_update(st
, &mu
);
11525 while (mu
.space_list
) {
11526 void **space
= mu
.space_list
;
11527 mu
.space_list
= *space
;
11532 /***************************************************************************
11533 * Function: imsm_analyze_change
11534 * Description: Function analyze change for single volume
11535 * and validate if transition is supported
11536 * Parameters: Geometry parameters, supertype structure,
11537 * metadata change direction (apply/rollback)
11538 * Returns: Operation type code on success, -1 if fail
11539 ****************************************************************************/
11540 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11541 struct geo_params
*geo
,
11544 struct mdinfo info
;
11546 int check_devs
= 0;
11548 /* number of added/removed disks in operation result */
11549 int devNumChange
= 0;
11550 /* imsm compatible layout value for array geometry verification */
11551 int imsm_layout
= -1;
11553 struct imsm_dev
*dev
;
11554 struct imsm_map
*map
;
11555 struct intel_super
*super
;
11556 unsigned long long current_size
;
11557 unsigned long long free_size
;
11558 unsigned long long max_size
;
11561 getinfo_super_imsm_volume(st
, &info
, NULL
);
11562 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11563 geo
->level
!= UnSet
) {
11564 switch (info
.array
.level
) {
11566 if (geo
->level
== 5) {
11567 change
= CH_MIGRATION
;
11568 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11569 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11571 goto analyse_change_exit
;
11573 imsm_layout
= geo
->layout
;
11575 devNumChange
= 1; /* parity disk added */
11576 } else if (geo
->level
== 10) {
11577 change
= CH_TAKEOVER
;
11579 devNumChange
= 2; /* two mirrors added */
11580 imsm_layout
= 0x102; /* imsm supported layout */
11585 if (geo
->level
== 0) {
11586 change
= CH_TAKEOVER
;
11588 devNumChange
= -(geo
->raid_disks
/2);
11589 imsm_layout
= 0; /* imsm raid0 layout */
11593 if (change
== -1) {
11594 pr_err("Error. Level Migration from %d to %d not supported!\n",
11595 info
.array
.level
, geo
->level
);
11596 goto analyse_change_exit
;
11599 geo
->level
= info
.array
.level
;
11601 if (geo
->layout
!= info
.array
.layout
&&
11602 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11603 change
= CH_MIGRATION
;
11604 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11605 geo
->layout
== 5) {
11606 /* reshape 5 -> 4 */
11607 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11608 geo
->layout
== 0) {
11609 /* reshape 4 -> 5 */
11613 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11614 info
.array
.layout
, geo
->layout
);
11616 goto analyse_change_exit
;
11619 geo
->layout
= info
.array
.layout
;
11620 if (imsm_layout
== -1)
11621 imsm_layout
= info
.array
.layout
;
11624 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11625 geo
->chunksize
!= info
.array
.chunk_size
) {
11626 if (info
.array
.level
== 10) {
11627 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11629 goto analyse_change_exit
;
11630 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11631 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11632 geo
->chunksize
/1024, info
.component_size
/2);
11634 goto analyse_change_exit
;
11636 change
= CH_MIGRATION
;
11638 geo
->chunksize
= info
.array
.chunk_size
;
11641 chunk
= geo
->chunksize
/ 1024;
11644 dev
= get_imsm_dev(super
, super
->current_vol
);
11645 map
= get_imsm_map(dev
, MAP_0
);
11646 data_disks
= imsm_num_data_members(map
);
11647 /* compute current size per disk member
11649 current_size
= info
.custom_array_size
/ data_disks
;
11651 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11652 /* align component size
11654 geo
->size
= imsm_component_size_alignment_check(
11655 get_imsm_raid_level(dev
->vol
.map
),
11656 chunk
* 1024, super
->sector_size
,
11658 if (geo
->size
== 0) {
11659 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11661 goto analyse_change_exit
;
11665 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11666 if (change
!= -1) {
11667 pr_err("Error. Size change should be the only one at a time.\n");
11669 goto analyse_change_exit
;
11671 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11672 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11673 super
->current_vol
, st
->devnm
);
11674 goto analyse_change_exit
;
11676 /* check the maximum available size
11678 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11679 0, chunk
, &free_size
);
11681 /* Cannot find maximum available space
11685 max_size
= free_size
+ current_size
;
11686 /* align component size
11688 max_size
= imsm_component_size_alignment_check(
11689 get_imsm_raid_level(dev
->vol
.map
),
11690 chunk
* 1024, super
->sector_size
,
11693 if (geo
->size
== MAX_SIZE
) {
11694 /* requested size change to the maximum available size
11696 if (max_size
== 0) {
11697 pr_err("Error. Cannot find maximum available space.\n");
11699 goto analyse_change_exit
;
11701 geo
->size
= max_size
;
11704 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11705 /* accept size for rollback only
11708 /* round size due to metadata compatibility
11710 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11711 << SECT_PER_MB_SHIFT
;
11712 dprintf("Prepare update for size change to %llu\n",
11714 if (current_size
>= geo
->size
) {
11715 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11716 current_size
, geo
->size
);
11717 goto analyse_change_exit
;
11719 if (max_size
&& geo
->size
> max_size
) {
11720 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11721 max_size
, geo
->size
);
11722 goto analyse_change_exit
;
11725 geo
->size
*= data_disks
;
11726 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11727 change
= CH_ARRAY_SIZE
;
11729 if (!validate_geometry_imsm(st
,
11732 geo
->raid_disks
+ devNumChange
,
11734 geo
->size
, INVALID_SECTORS
,
11735 0, 0, info
.consistency_policy
, 1))
11739 struct intel_super
*super
= st
->sb
;
11740 struct imsm_super
*mpb
= super
->anchor
;
11742 if (mpb
->num_raid_devs
> 1) {
11743 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11749 analyse_change_exit
:
11750 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11751 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11752 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11758 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11760 struct intel_super
*super
= st
->sb
;
11761 struct imsm_update_takeover
*u
;
11763 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11765 u
->type
= update_takeover
;
11766 u
->subarray
= super
->current_vol
;
11768 /* 10->0 transition */
11769 if (geo
->level
== 0)
11770 u
->direction
= R10_TO_R0
;
11772 /* 0->10 transition */
11773 if (geo
->level
== 10)
11774 u
->direction
= R0_TO_R10
;
11776 /* update metadata locally */
11777 imsm_update_metadata_locally(st
, u
,
11778 sizeof(struct imsm_update_takeover
));
11779 /* and possibly remotely */
11780 if (st
->update_tail
)
11781 append_metadata_update(st
, u
,
11782 sizeof(struct imsm_update_takeover
));
11789 /* Flush size update if size calculated by num_data_stripes is higher than
11790 * imsm_dev_size to eliminate differences during reshape.
11791 * Mdmon will recalculate them correctly.
11792 * If subarray index is not set then check whole container.
11794 * 0 - no error occurred
11795 * 1 - error detected
11797 static int imsm_fix_size_mismatch(struct supertype
*st
, int subarray_index
)
11799 struct intel_super
*super
= st
->sb
;
11800 int tmp
= super
->current_vol
;
11804 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
11805 if (subarray_index
>= 0 && i
!= subarray_index
)
11807 super
->current_vol
= i
;
11808 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11809 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11810 unsigned int disc_count
= imsm_num_data_members(map
);
11811 struct geo_params geo
;
11812 struct imsm_update_size_change
*update
;
11813 unsigned long long calc_size
= per_dev_array_size(map
) * disc_count
;
11814 unsigned long long d_size
= imsm_dev_size(dev
);
11817 if (calc_size
== d_size
|| dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
11820 /* There is a difference, verify that imsm_dev_size is
11821 * rounded correctly and push update.
11823 if (d_size
!= round_size_to_mb(d_size
, disc_count
)) {
11824 dprintf("imsm: Size of volume %d is not rounded correctly\n",
11828 memset(&geo
, 0, sizeof(struct geo_params
));
11830 u_size
= imsm_create_metadata_update_for_size_change(st
, &geo
,
11833 dprintf("imsm: Cannot prepare size change update\n");
11836 imsm_update_metadata_locally(st
, update
, u_size
);
11837 if (st
->update_tail
) {
11838 append_metadata_update(st
, update
, u_size
);
11839 flush_metadata_updates(st
);
11840 st
->update_tail
= &st
->updates
;
11842 imsm_sync_metadata(st
);
11847 super
->current_vol
= tmp
;
11851 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11853 int layout
, int chunksize
, int raid_disks
,
11854 int delta_disks
, char *backup
, char *dev
,
11855 int direction
, int verbose
)
11858 struct geo_params geo
;
11860 dprintf("(enter)\n");
11862 memset(&geo
, 0, sizeof(struct geo_params
));
11864 geo
.dev_name
= dev
;
11865 strcpy(geo
.devnm
, st
->devnm
);
11868 geo
.layout
= layout
;
11869 geo
.chunksize
= chunksize
;
11870 geo
.raid_disks
= raid_disks
;
11871 if (delta_disks
!= UnSet
)
11872 geo
.raid_disks
+= delta_disks
;
11874 dprintf("for level : %i\n", geo
.level
);
11875 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11877 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11878 /* On container level we can only increase number of devices. */
11879 dprintf("imsm: info: Container operation\n");
11880 int old_raid_disks
= 0;
11882 if (imsm_reshape_is_allowed_on_container(
11883 st
, &geo
, &old_raid_disks
, direction
)) {
11884 struct imsm_update_reshape
*u
= NULL
;
11887 if (imsm_fix_size_mismatch(st
, -1)) {
11888 dprintf("imsm: Cannot fix size mismatch\n");
11889 goto exit_imsm_reshape_super
;
11892 len
= imsm_create_metadata_update_for_reshape(
11893 st
, &geo
, old_raid_disks
, &u
);
11896 dprintf("imsm: Cannot prepare update\n");
11897 goto exit_imsm_reshape_super
;
11901 /* update metadata locally */
11902 imsm_update_metadata_locally(st
, u
, len
);
11903 /* and possibly remotely */
11904 if (st
->update_tail
)
11905 append_metadata_update(st
, u
, len
);
11910 pr_err("(imsm) Operation is not allowed on this container\n");
11913 /* On volume level we support following operations
11914 * - takeover: raid10 -> raid0; raid0 -> raid10
11915 * - chunk size migration
11916 * - migration: raid5 -> raid0; raid0 -> raid5
11918 struct intel_super
*super
= st
->sb
;
11919 struct intel_dev
*dev
= super
->devlist
;
11921 dprintf("imsm: info: Volume operation\n");
11922 /* find requested device */
11925 imsm_find_array_devnm_by_subdev(
11926 dev
->index
, st
->container_devnm
);
11927 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11932 pr_err("Cannot find %s (%s) subarray\n",
11933 geo
.dev_name
, geo
.devnm
);
11934 goto exit_imsm_reshape_super
;
11936 super
->current_vol
= dev
->index
;
11937 change
= imsm_analyze_change(st
, &geo
, direction
);
11940 ret_val
= imsm_takeover(st
, &geo
);
11942 case CH_MIGRATION
: {
11943 struct imsm_update_reshape_migration
*u
= NULL
;
11945 imsm_create_metadata_update_for_migration(
11948 dprintf("imsm: Cannot prepare update\n");
11952 /* update metadata locally */
11953 imsm_update_metadata_locally(st
, u
, len
);
11954 /* and possibly remotely */
11955 if (st
->update_tail
)
11956 append_metadata_update(st
, u
, len
);
11961 case CH_ARRAY_SIZE
: {
11962 struct imsm_update_size_change
*u
= NULL
;
11964 imsm_create_metadata_update_for_size_change(
11967 dprintf("imsm: Cannot prepare update\n");
11971 /* update metadata locally */
11972 imsm_update_metadata_locally(st
, u
, len
);
11973 /* and possibly remotely */
11974 if (st
->update_tail
)
11975 append_metadata_update(st
, u
, len
);
11985 exit_imsm_reshape_super
:
11986 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11990 #define COMPLETED_OK 0
11991 #define COMPLETED_NONE 1
11992 #define COMPLETED_DELAYED 2
11994 static int read_completed(int fd
, unsigned long long *val
)
11999 ret
= sysfs_fd_get_str(fd
, buf
, 50);
12003 ret
= COMPLETED_OK
;
12004 if (strncmp(buf
, "none", 4) == 0) {
12005 ret
= COMPLETED_NONE
;
12006 } else if (strncmp(buf
, "delayed", 7) == 0) {
12007 ret
= COMPLETED_DELAYED
;
12010 *val
= strtoull(buf
, &ep
, 0);
12011 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
12017 /*******************************************************************************
12018 * Function: wait_for_reshape_imsm
12019 * Description: Function writes new sync_max value and waits until
12020 * reshape process reach new position
12022 * sra : general array info
12023 * ndata : number of disks in new array's layout
12026 * 1 : there is no reshape in progress,
12028 ******************************************************************************/
12029 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
12031 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
12033 unsigned long long completed
;
12034 /* to_complete : new sync_max position */
12035 unsigned long long to_complete
= sra
->reshape_progress
;
12036 unsigned long long position_to_set
= to_complete
/ ndata
;
12039 dprintf("cannot open reshape_position\n");
12044 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
12046 dprintf("cannot read reshape_position (no reshape in progres)\n");
12055 if (completed
> position_to_set
) {
12056 dprintf("wrong next position to set %llu (%llu)\n",
12057 to_complete
, position_to_set
);
12061 dprintf("Position set: %llu\n", position_to_set
);
12062 if (sysfs_set_num(sra
, NULL
, "sync_max",
12063 position_to_set
) != 0) {
12064 dprintf("cannot set reshape position to %llu\n",
12073 int timeout
= 3000;
12075 sysfs_wait(fd
, &timeout
);
12076 if (sysfs_get_str(sra
, NULL
, "sync_action",
12078 strncmp(action
, "reshape", 7) != 0) {
12079 if (strncmp(action
, "idle", 4) == 0)
12085 rc
= read_completed(fd
, &completed
);
12087 dprintf("cannot read reshape_position (in loop)\n");
12090 } else if (rc
== COMPLETED_NONE
)
12092 } while (completed
< position_to_set
);
12098 /*******************************************************************************
12099 * Function: check_degradation_change
12100 * Description: Check that array hasn't become failed.
12102 * info : for sysfs access
12103 * sources : source disks descriptors
12104 * degraded: previous degradation level
12106 * degradation level
12107 ******************************************************************************/
12108 int check_degradation_change(struct mdinfo
*info
,
12112 unsigned long long new_degraded
;
12115 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
12116 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
12117 /* check each device to ensure it is still working */
12120 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
12121 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
12123 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
12126 if (sysfs_get_str(info
,
12127 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
12128 strstr(sbuf
, "faulty") ||
12129 strstr(sbuf
, "in_sync") == NULL
) {
12130 /* this device is dead */
12131 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
12132 if (sd
->disk
.raid_disk
>= 0 &&
12133 sources
[sd
->disk
.raid_disk
] >= 0) {
12135 sd
->disk
.raid_disk
]);
12136 sources
[sd
->disk
.raid_disk
] =
12145 return new_degraded
;
12148 /*******************************************************************************
12149 * Function: imsm_manage_reshape
12150 * Description: Function finds array under reshape and it manages reshape
12151 * process. It creates stripes backups (if required) and sets
12154 * afd : Backup handle (nattive) - not used
12155 * sra : general array info
12156 * reshape : reshape parameters - not used
12157 * st : supertype structure
12158 * blocks : size of critical section [blocks]
12159 * fds : table of source device descriptor
12160 * offsets : start of array (offest per devices)
12162 * destfd : table of destination device descriptor
12163 * destoffsets : table of destination offsets (per device)
12165 * 1 : success, reshape is done
12167 ******************************************************************************/
12168 static int imsm_manage_reshape(
12169 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
12170 struct supertype
*st
, unsigned long backup_blocks
,
12171 int *fds
, unsigned long long *offsets
,
12172 int dests
, int *destfd
, unsigned long long *destoffsets
)
12175 struct intel_super
*super
= st
->sb
;
12176 struct intel_dev
*dv
;
12177 unsigned int sector_size
= super
->sector_size
;
12178 struct imsm_dev
*dev
= NULL
;
12179 struct imsm_map
*map_src
, *map_dest
;
12180 int migr_vol_qan
= 0;
12181 int ndata
, odata
; /* [bytes] */
12182 int chunk
; /* [bytes] */
12183 struct migr_record
*migr_rec
;
12185 unsigned int buf_size
; /* [bytes] */
12186 unsigned long long max_position
; /* array size [bytes] */
12187 unsigned long long next_step
; /* [blocks]/[bytes] */
12188 unsigned long long old_data_stripe_length
;
12189 unsigned long long start_src
; /* [bytes] */
12190 unsigned long long start
; /* [bytes] */
12191 unsigned long long start_buf_shift
; /* [bytes] */
12193 int source_layout
= 0;
12194 int subarray_index
= -1;
12199 if (!fds
|| !offsets
)
12202 /* Find volume during the reshape */
12203 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
12204 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12205 dv
->dev
->vol
.migr_state
== 1) {
12208 subarray_index
= dv
->index
;
12211 /* Only one volume can migrate at the same time */
12212 if (migr_vol_qan
!= 1) {
12213 pr_err("%s", migr_vol_qan
?
12214 "Number of migrating volumes greater than 1\n" :
12215 "There is no volume during migrationg\n");
12219 map_dest
= get_imsm_map(dev
, MAP_0
);
12220 map_src
= get_imsm_map(dev
, MAP_1
);
12221 if (map_src
== NULL
)
12224 ndata
= imsm_num_data_members(map_dest
);
12225 odata
= imsm_num_data_members(map_src
);
12227 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12228 old_data_stripe_length
= odata
* chunk
;
12230 migr_rec
= super
->migr_rec
;
12232 /* initialize migration record for start condition */
12233 if (sra
->reshape_progress
== 0)
12234 init_migr_record_imsm(st
, dev
, sra
);
12236 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12237 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12240 /* Save checkpoint to update migration record for current
12241 * reshape position (in md). It can be farther than current
12242 * reshape position in metadata.
12244 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12245 /* ignore error == 2, this can mean end of reshape here
12247 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12252 /* size for data */
12253 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12254 /* extend buffer size for parity disk */
12255 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12256 /* add space for stripe alignment */
12257 buf_size
+= old_data_stripe_length
;
12258 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12259 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12263 max_position
= sra
->component_size
* ndata
;
12264 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12266 while (current_migr_unit(migr_rec
) <
12267 get_num_migr_units(migr_rec
)) {
12268 /* current reshape position [blocks] */
12269 unsigned long long current_position
=
12270 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12271 * current_migr_unit(migr_rec
);
12272 unsigned long long border
;
12274 /* Check that array hasn't become failed.
12276 degraded
= check_degradation_change(sra
, fds
, degraded
);
12277 if (degraded
> 1) {
12278 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12282 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12284 if ((current_position
+ next_step
) > max_position
)
12285 next_step
= max_position
- current_position
;
12287 start
= current_position
* 512;
12289 /* align reading start to old geometry */
12290 start_buf_shift
= start
% old_data_stripe_length
;
12291 start_src
= start
- start_buf_shift
;
12293 border
= (start_src
/ odata
) - (start
/ ndata
);
12295 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12296 /* save critical stripes to buf
12297 * start - start address of current unit
12298 * to backup [bytes]
12299 * start_src - start address of current unit
12300 * to backup alligned to source array
12303 unsigned long long next_step_filler
;
12304 unsigned long long copy_length
= next_step
* 512;
12306 /* allign copy area length to stripe in old geometry */
12307 next_step_filler
= ((copy_length
+ start_buf_shift
)
12308 % old_data_stripe_length
);
12309 if (next_step_filler
)
12310 next_step_filler
= (old_data_stripe_length
12311 - next_step_filler
);
12312 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12313 start
, start_src
, copy_length
,
12314 start_buf_shift
, next_step_filler
);
12316 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12317 chunk
, map_src
->raid_level
,
12318 source_layout
, 0, NULL
, start_src
,
12320 next_step_filler
+ start_buf_shift
,
12322 dprintf("imsm: Cannot save stripes to buffer\n");
12325 /* Convert data to destination format and store it
12326 * in backup general migration area
12328 if (save_backup_imsm(st
, dev
, sra
,
12329 buf
+ start_buf_shift
, copy_length
)) {
12330 dprintf("imsm: Cannot save stripes to target devices\n");
12333 if (save_checkpoint_imsm(st
, sra
,
12334 UNIT_SRC_IN_CP_AREA
)) {
12335 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12339 /* set next step to use whole border area */
12340 border
/= next_step
;
12342 next_step
*= border
;
12344 /* When data backed up, checkpoint stored,
12345 * kick the kernel to reshape unit of data
12347 next_step
= next_step
+ sra
->reshape_progress
;
12348 /* limit next step to array max position */
12349 if (next_step
> max_position
)
12350 next_step
= max_position
;
12351 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12352 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12353 sra
->reshape_progress
= next_step
;
12355 /* wait until reshape finish */
12356 if (wait_for_reshape_imsm(sra
, ndata
)) {
12357 dprintf("wait_for_reshape_imsm returned error!\n");
12363 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12364 /* ignore error == 2, this can mean end of reshape here
12366 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12372 /* clear migr_rec on disks after successful migration */
12375 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12376 for (d
= super
->disks
; d
; d
= d
->next
) {
12377 if (d
->index
< 0 || is_failed(&d
->disk
))
12379 unsigned long long dsize
;
12381 get_dev_size(d
->fd
, NULL
, &dsize
);
12382 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12384 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12385 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12386 MIGR_REC_BUF_SECTORS
*sector_size
)
12387 perror("Write migr_rec failed");
12391 /* return '1' if done */
12394 /* After the reshape eliminate size mismatch in metadata.
12395 * Don't update md/component_size here, volume hasn't
12396 * to take whole space. It is allowed by kernel.
12397 * md/component_size will be set propoperly after next assembly.
12399 imsm_fix_size_mismatch(st
, subarray_index
);
12403 /* See Grow.c: abort_reshape() for further explanation */
12404 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12405 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12406 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12411 /*******************************************************************************
12412 * Function: calculate_bitmap_min_chunksize
12413 * Description: Calculates the minimal valid bitmap chunk size
12415 * max_bits : indicate how many bits can be used for the bitmap
12416 * data_area_size : the size of the data area covered by the bitmap
12419 * The bitmap chunk size
12420 ******************************************************************************/
12421 static unsigned long long
12422 calculate_bitmap_min_chunksize(unsigned long long max_bits
,
12423 unsigned long long data_area_size
)
12425 unsigned long long min_chunk
=
12426 4096; /* sub-page chunks don't work yet.. */
12427 unsigned long long bits
= data_area_size
/ min_chunk
+ 1;
12429 while (bits
> max_bits
) {
12431 bits
= (bits
+ 1) / 2;
12436 /*******************************************************************************
12437 * Function: calculate_bitmap_chunksize
12438 * Description: Calculates the bitmap chunk size for the given device
12440 * st : supertype information
12441 * dev : device for the bitmap
12444 * The bitmap chunk size
12445 ******************************************************************************/
12446 static unsigned long long calculate_bitmap_chunksize(struct supertype
*st
,
12447 struct imsm_dev
*dev
)
12449 struct intel_super
*super
= st
->sb
;
12450 unsigned long long min_chunksize
;
12451 unsigned long long result
= IMSM_DEFAULT_BITMAP_CHUNKSIZE
;
12452 size_t dev_size
= imsm_dev_size(dev
);
12454 min_chunksize
= calculate_bitmap_min_chunksize(
12455 IMSM_BITMAP_AREA_SIZE
* super
->sector_size
, dev_size
);
12457 if (result
< min_chunksize
)
12458 result
= min_chunksize
;
12463 /*******************************************************************************
12464 * Function: init_bitmap_header
12465 * Description: Initialize the bitmap header structure
12467 * st : supertype information
12468 * bms : bitmap header struct to initialize
12469 * dev : device for the bitmap
12474 ******************************************************************************/
12475 static int init_bitmap_header(struct supertype
*st
, struct bitmap_super_s
*bms
,
12476 struct imsm_dev
*dev
)
12483 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
12484 bms
->version
= __cpu_to_le32(BITMAP_MAJOR_HI
);
12485 bms
->daemon_sleep
= __cpu_to_le32(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
);
12486 bms
->sync_size
= __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
);
12487 bms
->write_behind
= __cpu_to_le32(0);
12489 uuid_from_super_imsm(st
, vol_uuid
);
12490 memcpy(bms
->uuid
, vol_uuid
, 16);
12492 bms
->chunksize
= calculate_bitmap_chunksize(st
, dev
);
12497 /*******************************************************************************
12498 * Function: validate_internal_bitmap_for_drive
12499 * Description: Verify if the bitmap header for a given drive.
12501 * st : supertype information
12502 * offset : The offset from the beginning of the drive where to look for
12503 * the bitmap header.
12504 * d : the drive info
12509 ******************************************************************************/
12510 static int validate_internal_bitmap_for_drive(struct supertype
*st
,
12511 unsigned long long offset
,
12514 struct intel_super
*super
= st
->sb
;
12517 bitmap_super_t
*bms
;
12525 if (posix_memalign(&read_buf
, MAX_SECTOR_SIZE
, IMSM_BITMAP_HEADER_SIZE
))
12530 fd
= open(d
->devname
, O_RDONLY
, 0);
12532 dprintf("cannot open the device %s\n", d
->devname
);
12537 if (lseek64(fd
, offset
* super
->sector_size
, SEEK_SET
) < 0)
12539 if (read(fd
, read_buf
, IMSM_BITMAP_HEADER_SIZE
) !=
12540 IMSM_BITMAP_HEADER_SIZE
)
12543 uuid_from_super_imsm(st
, vol_uuid
);
12546 if ((bms
->magic
!= __cpu_to_le32(BITMAP_MAGIC
)) ||
12547 (bms
->version
!= __cpu_to_le32(BITMAP_MAJOR_HI
)) ||
12548 (!same_uuid((int *)bms
->uuid
, vol_uuid
, st
->ss
->swapuuid
))) {
12549 dprintf("wrong bitmap header detected\n");
12555 if ((d
->fd
< 0) && (fd
>= 0))
12563 /*******************************************************************************
12564 * Function: validate_internal_bitmap_imsm
12565 * Description: Verify if the bitmap header is in place and with proper data.
12567 * st : supertype information
12570 * 0 : success or device w/o RWH_BITMAP
12572 ******************************************************************************/
12573 static int validate_internal_bitmap_imsm(struct supertype
*st
)
12575 struct intel_super
*super
= st
->sb
;
12576 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
12577 unsigned long long offset
;
12583 if (dev
->rwh_policy
!= RWH_BITMAP
)
12586 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12587 for (d
= super
->disks
; d
; d
= d
->next
) {
12588 if (d
->index
< 0 || is_failed(&d
->disk
))
12591 if (validate_internal_bitmap_for_drive(st
, offset
, d
)) {
12592 pr_err("imsm: bitmap validation failed\n");
12599 /*******************************************************************************
12600 * Function: add_internal_bitmap_imsm
12601 * Description: Mark the volume to use the bitmap and updates the chunk size value.
12603 * st : supertype information
12604 * chunkp : bitmap chunk size
12605 * delay : not used for imsm
12606 * write_behind : not used for imsm
12607 * size : not used for imsm
12608 * may_change : not used for imsm
12609 * amajor : not used for imsm
12614 ******************************************************************************/
12615 static int add_internal_bitmap_imsm(struct supertype
*st
, int *chunkp
,
12616 int delay
, int write_behind
,
12617 unsigned long long size
, int may_change
,
12620 struct intel_super
*super
= st
->sb
;
12621 int vol_idx
= super
->current_vol
;
12622 struct imsm_dev
*dev
;
12624 if (!super
->devlist
|| vol_idx
== -1 || !chunkp
)
12627 dev
= get_imsm_dev(super
, vol_idx
);
12630 dprintf("cannot find the device for volume index %d\n",
12634 dev
->rwh_policy
= RWH_BITMAP
;
12636 *chunkp
= calculate_bitmap_chunksize(st
, dev
);
12641 /*******************************************************************************
12642 * Function: locate_bitmap_imsm
12643 * Description: Seek 'fd' to start of write-intent-bitmap.
12645 * st : supertype information
12646 * fd : file descriptor for the device
12647 * node_num : not used for imsm
12652 ******************************************************************************/
12653 static int locate_bitmap_imsm(struct supertype
*st
, int fd
, int node_num
)
12655 struct intel_super
*super
= st
->sb
;
12656 unsigned long long offset
;
12657 int vol_idx
= super
->current_vol
;
12659 if (!super
->devlist
|| vol_idx
== -1)
12662 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12663 dprintf("bitmap header offset is %llu\n", offset
);
12665 lseek64(fd
, offset
<< 9, 0);
12670 /*******************************************************************************
12671 * Function: write_init_bitmap_imsm
12672 * Description: Write a bitmap header and prepares the area for the bitmap.
12674 * st : supertype information
12675 * fd : file descriptor for the device
12676 * update : not used for imsm
12681 ******************************************************************************/
12682 static int write_init_bitmap_imsm(struct supertype
*st
, int fd
,
12683 enum bitmap_update update
)
12685 struct intel_super
*super
= st
->sb
;
12686 int vol_idx
= super
->current_vol
;
12688 unsigned long long offset
;
12689 bitmap_super_t bms
= { 0 };
12690 size_t written
= 0;
12695 if (!super
->devlist
|| !super
->sector_size
|| vol_idx
== -1)
12698 struct imsm_dev
*dev
= get_imsm_dev(super
, vol_idx
);
12700 /* first clear the space for bitmap header */
12701 unsigned long long bitmap_area_start
=
12702 get_bitmap_header_sector(super
, vol_idx
);
12704 dprintf("zeroing area start (%llu) and size (%u)\n", bitmap_area_start
,
12705 IMSM_BITMAP_AND_HEADER_SIZE
/ super
->sector_size
);
12706 if (zero_disk_range(fd
, bitmap_area_start
,
12707 IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
)) {
12708 pr_err("imsm: cannot zeroing the space for the bitmap\n");
12712 /* The bitmap area should be filled with "1"s to perform initial
12715 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
))
12717 memset(buf
, 0xFF, MAX_SECTOR_SIZE
);
12718 offset
= get_bitmap_sector(super
, vol_idx
);
12719 lseek64(fd
, offset
<< 9, 0);
12720 while (written
< IMSM_BITMAP_AREA_SIZE
) {
12721 to_write
= IMSM_BITMAP_AREA_SIZE
- written
;
12722 if (to_write
> MAX_SECTOR_SIZE
)
12723 to_write
= MAX_SECTOR_SIZE
;
12724 rv_num
= write(fd
, buf
, MAX_SECTOR_SIZE
);
12725 if (rv_num
!= MAX_SECTOR_SIZE
) {
12727 dprintf("cannot initialize bitmap area\n");
12733 /* write a bitmap header */
12734 init_bitmap_header(st
, &bms
, dev
);
12735 memset(buf
, 0, MAX_SECTOR_SIZE
);
12736 memcpy(buf
, &bms
, sizeof(bitmap_super_t
));
12737 if (locate_bitmap_imsm(st
, fd
, 0)) {
12739 dprintf("cannot locate the bitmap\n");
12742 if (write(fd
, buf
, MAX_SECTOR_SIZE
) != MAX_SECTOR_SIZE
) {
12744 dprintf("cannot write the bitmap header\n");
12755 /*******************************************************************************
12756 * Function: is_vol_to_setup_bitmap
12757 * Description: Checks if a bitmap should be activated on the dev.
12759 * info : info about the volume to setup the bitmap
12760 * dev : the device to check against bitmap creation
12763 * 0 : bitmap should be set up on the device
12765 ******************************************************************************/
12766 static int is_vol_to_setup_bitmap(struct mdinfo
*info
, struct imsm_dev
*dev
)
12771 if ((strcmp((char *)dev
->volume
, info
->name
) == 0) &&
12772 (dev
->rwh_policy
== RWH_BITMAP
))
12778 /*******************************************************************************
12779 * Function: set_bitmap_sysfs
12780 * Description: Set the sysfs atributes of a given volume to activate the bitmap.
12782 * info : info about the volume where the bitmap should be setup
12783 * chunksize : bitmap chunk size
12784 * location : location of the bitmap
12789 ******************************************************************************/
12790 static int set_bitmap_sysfs(struct mdinfo
*info
, unsigned long long chunksize
,
12793 /* The bitmap/metadata is set to external to allow changing of value for
12794 * bitmap/location. When external is used, the kernel will treat an offset
12795 * related to the device's first lba (in opposition to the "internal" case
12796 * when this value is related to the beginning of the superblock).
12798 if (sysfs_set_str(info
, NULL
, "bitmap/metadata", "external")) {
12799 dprintf("failed to set bitmap/metadata\n");
12803 /* It can only be changed when no bitmap is active.
12804 * Should be bigger than 512 and must be power of 2.
12805 * It is expecting the value in bytes.
12807 if (sysfs_set_num(info
, NULL
, "bitmap/chunksize",
12808 __cpu_to_le32(chunksize
))) {
12809 dprintf("failed to set bitmap/chunksize\n");
12813 /* It is expecting the value in sectors. */
12814 if (sysfs_set_num(info
, NULL
, "bitmap/space",
12815 __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
))) {
12816 dprintf("failed to set bitmap/space\n");
12820 /* Determines the delay between the bitmap updates.
12821 * It is expecting the value in seconds.
12823 if (sysfs_set_num(info
, NULL
, "bitmap/time_base",
12824 __cpu_to_le64(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
))) {
12825 dprintf("failed to set bitmap/time_base\n");
12829 /* It is expecting the value in sectors with a sign at the beginning. */
12830 if (sysfs_set_str(info
, NULL
, "bitmap/location", location
)) {
12831 dprintf("failed to set bitmap/location\n");
12838 /*******************************************************************************
12839 * Function: set_bitmap_imsm
12840 * Description: Setup the bitmap for the given volume
12842 * st : supertype information
12843 * info : info about the volume where the bitmap should be setup
12848 ******************************************************************************/
12849 static int set_bitmap_imsm(struct supertype
*st
, struct mdinfo
*info
)
12851 struct intel_super
*super
= st
->sb
;
12852 int prev_current_vol
= super
->current_vol
;
12853 struct imsm_dev
*dev
;
12855 char location
[16] = "";
12856 unsigned long long chunksize
;
12857 struct intel_dev
*dev_it
;
12859 for (dev_it
= super
->devlist
; dev_it
; dev_it
= dev_it
->next
) {
12860 super
->current_vol
= dev_it
->index
;
12861 dev
= get_imsm_dev(super
, super
->current_vol
);
12863 if (is_vol_to_setup_bitmap(info
, dev
)) {
12864 if (validate_internal_bitmap_imsm(st
)) {
12865 dprintf("bitmap header validation failed\n");
12869 chunksize
= calculate_bitmap_chunksize(st
, dev
);
12870 dprintf("chunk size is %llu\n", chunksize
);
12872 snprintf(location
, sizeof(location
), "+%llu",
12873 get_bitmap_sector(super
, super
->current_vol
));
12874 dprintf("bitmap offset is %s\n", location
);
12876 if (set_bitmap_sysfs(info
, chunksize
, location
)) {
12877 dprintf("cannot setup the bitmap\n");
12884 super
->current_vol
= prev_current_vol
;
12888 struct superswitch super_imsm
= {
12889 .examine_super
= examine_super_imsm
,
12890 .brief_examine_super
= brief_examine_super_imsm
,
12891 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12892 .export_examine_super
= export_examine_super_imsm
,
12893 .detail_super
= detail_super_imsm
,
12894 .brief_detail_super
= brief_detail_super_imsm
,
12895 .write_init_super
= write_init_super_imsm
,
12896 .validate_geometry
= validate_geometry_imsm
,
12897 .add_to_super
= add_to_super_imsm
,
12898 .remove_from_super
= remove_from_super_imsm
,
12899 .detail_platform
= detail_platform_imsm
,
12900 .export_detail_platform
= export_detail_platform_imsm
,
12901 .kill_subarray
= kill_subarray_imsm
,
12902 .update_subarray
= update_subarray_imsm
,
12903 .load_container
= load_container_imsm
,
12904 .default_geometry
= default_geometry_imsm
,
12905 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12906 .reshape_super
= imsm_reshape_super
,
12907 .manage_reshape
= imsm_manage_reshape
,
12908 .recover_backup
= recover_backup_imsm
,
12909 .examine_badblocks
= examine_badblocks_imsm
,
12910 .match_home
= match_home_imsm
,
12911 .uuid_from_super
= uuid_from_super_imsm
,
12912 .getinfo_super
= getinfo_super_imsm
,
12913 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12914 .update_super
= update_super_imsm
,
12916 .avail_size
= avail_size_imsm
,
12917 .get_spare_criteria
= get_spare_criteria_imsm
,
12919 .compare_super
= compare_super_imsm
,
12921 .load_super
= load_super_imsm
,
12922 .init_super
= init_super_imsm
,
12923 .store_super
= store_super_imsm
,
12924 .free_super
= free_super_imsm
,
12925 .match_metadata_desc
= match_metadata_desc_imsm
,
12926 .container_content
= container_content_imsm
,
12927 .validate_container
= validate_container_imsm
,
12929 .add_internal_bitmap
= add_internal_bitmap_imsm
,
12930 .locate_bitmap
= locate_bitmap_imsm
,
12931 .write_bitmap
= write_init_bitmap_imsm
,
12932 .set_bitmap
= set_bitmap_imsm
,
12934 .write_init_ppl
= write_init_ppl_imsm
,
12935 .validate_ppl
= validate_ppl_imsm
,
12941 .open_new
= imsm_open_new
,
12942 .set_array_state
= imsm_set_array_state
,
12943 .set_disk
= imsm_set_disk
,
12944 .sync_metadata
= imsm_sync_metadata
,
12945 .activate_spare
= imsm_activate_spare
,
12946 .process_update
= imsm_process_update
,
12947 .prepare_update
= imsm_prepare_update
,
12948 .record_bad_block
= imsm_record_badblock
,
12949 .clear_bad_block
= imsm_clear_badblock
,
12950 .get_bad_blocks
= imsm_get_badblocks
,