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
, 1, NULL
);
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
)) {
2249 char device
[PATH_MAX
];
2254 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2256 path
= devt_to_devpath(makedev(major
, minor
), 1, NULL
);
2259 if (!path_attached_to_hba(path
, hba_path
)) {
2265 /* retrieve the scsi device */
2266 if (!devt_to_devpath(makedev(major
, minor
), 1, device
)) {
2268 pr_err("failed to get device\n");
2272 if (devpath_to_char(device
, "type", buf
, sizeof(buf
), 0)) {
2276 type
= strtoul(buf
, NULL
, 10);
2278 /* if it's not a disk print the vendor and model */
2279 if (!(type
== 0 || type
== 7 || type
== 14)) {
2283 if (devpath_to_char(device
, "vendor", buf
,
2284 sizeof(buf
), 0) == 0) {
2285 strncpy(vendor
, buf
, sizeof(vendor
));
2286 vendor
[sizeof(vendor
) - 1] = '\0';
2287 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2288 while (isspace(*c
) || *c
== '\0')
2293 if (devpath_to_char(device
, "model", buf
,
2294 sizeof(buf
), 0) == 0) {
2295 strncpy(model
, buf
, sizeof(model
));
2296 model
[sizeof(model
) - 1] = '\0';
2297 c
= (char *) &model
[sizeof(model
) - 1];
2298 while (isspace(*c
) || *c
== '\0')
2302 if (vendor
[0] && model
[0])
2303 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2305 switch (type
) { /* numbers from hald/linux/device.c */
2306 case 1: sprintf(buf
, "tape"); break;
2307 case 2: sprintf(buf
, "printer"); break;
2308 case 3: sprintf(buf
, "processor"); break;
2310 case 5: sprintf(buf
, "cdrom"); break;
2311 case 6: sprintf(buf
, "scanner"); break;
2312 case 8: sprintf(buf
, "media_changer"); break;
2313 case 9: sprintf(buf
, "comm"); break;
2314 case 12: sprintf(buf
, "raid"); break;
2315 default: sprintf(buf
, "unknown");
2320 /* chop device path to 'host%d' and calculate the port number */
2321 c
= strchr(&path
[hba_len
], '/');
2324 pr_err("%s - invalid path name\n", path
+ hba_len
);
2329 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2330 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2334 *c
= '/'; /* repair the full string */
2335 pr_err("failed to determine port number for %s\n",
2342 /* mark this port as used */
2343 port_mask
&= ~(1 << port
);
2345 /* print out the device information */
2347 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2351 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2353 printf(" Port%d : - disk info unavailable -\n", port
);
2355 fd2devname(fd
, buf
);
2356 printf(" Port%d : %s", port
, buf
);
2357 if (imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2359 printf(" (%s)\n", buf
);
2374 for (i
= 0; i
< port_count
; i
++)
2375 if (port_mask
& (1 << i
))
2376 printf(" Port%d : - no device attached -\n", i
);
2382 static int print_nvme_info(struct sys_dev
*hba
)
2387 dir
= opendir("/sys/block/");
2391 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2392 char ns_path
[PATH_MAX
];
2393 char cntrl_path
[PATH_MAX
];
2397 if (!strstr(ent
->d_name
, "nvme"))
2400 fd
= open_dev(ent
->d_name
);
2404 if (!diskfd_to_devpath(fd
, 0, ns_path
) ||
2405 !diskfd_to_devpath(fd
, 1, cntrl_path
))
2408 if (!path_attached_to_hba(cntrl_path
, hba
->path
))
2411 if (!imsm_is_nvme_namespace_supported(fd
, 0))
2414 fd2devname(fd
, buf
);
2415 if (hba
->type
== SYS_DEV_VMD
)
2416 printf(" NVMe under VMD : %s", buf
);
2417 else if (hba
->type
== SYS_DEV_NVME
)
2418 printf(" NVMe Device : %s", buf
);
2420 if (!imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2422 printf(" (%s)\n", buf
);
2435 static void print_found_intel_controllers(struct sys_dev
*elem
)
2437 for (; elem
; elem
= elem
->next
) {
2438 pr_err("found Intel(R) ");
2439 if (elem
->type
== SYS_DEV_SATA
)
2440 fprintf(stderr
, "SATA ");
2441 else if (elem
->type
== SYS_DEV_SAS
)
2442 fprintf(stderr
, "SAS ");
2443 else if (elem
->type
== SYS_DEV_NVME
)
2444 fprintf(stderr
, "NVMe ");
2446 if (elem
->type
== SYS_DEV_VMD
)
2447 fprintf(stderr
, "VMD domain");
2449 fprintf(stderr
, "RAID controller");
2452 fprintf(stderr
, " at %s", elem
->pci_id
);
2453 fprintf(stderr
, ".\n");
2458 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2465 if ((dir
= opendir(hba_path
)) == NULL
)
2468 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2471 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2472 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2474 if (*port_count
== 0)
2476 else if (host
< host_base
)
2479 if (host
+ 1 > *port_count
+ host_base
)
2480 *port_count
= host
+ 1 - host_base
;
2486 static void print_imsm_capability(const struct imsm_orom
*orom
)
2488 printf(" Platform : Intel(R) ");
2489 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2490 printf("Matrix Storage Manager\n");
2491 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2492 printf("Virtual RAID on CPU\n");
2494 printf("Rapid Storage Technology%s\n",
2495 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2496 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2497 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2498 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2499 printf(" RAID Levels :%s%s%s%s%s\n",
2500 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2501 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2502 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2503 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2504 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2505 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2506 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2507 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2508 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2509 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2510 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2511 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2512 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2513 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2514 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2515 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2516 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2517 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2518 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2519 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2520 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2521 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2522 printf(" 2TB volumes :%s supported\n",
2523 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2524 printf(" 2TB disks :%s supported\n",
2525 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2526 printf(" Max Disks : %d\n", orom
->tds
);
2527 printf(" Max Volumes : %d per array, %d per %s\n",
2528 orom
->vpa
, orom
->vphba
,
2529 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2533 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2535 printf("MD_FIRMWARE_TYPE=imsm\n");
2536 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2537 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2538 orom
->hotfix_ver
, orom
->build
);
2539 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2540 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2541 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2542 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2543 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2544 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2545 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2546 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2547 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2548 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2549 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2550 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2551 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2552 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2553 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2554 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2555 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2556 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2557 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2558 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2559 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2560 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2561 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2562 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2563 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2564 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2565 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2566 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2569 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2571 /* There are two components to imsm platform support, the ahci SATA
2572 * controller and the option-rom. To find the SATA controller we
2573 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2574 * controller with the Intel vendor id is present. This approach
2575 * allows mdadm to leverage the kernel's ahci detection logic, with the
2576 * caveat that if ahci.ko is not loaded mdadm will not be able to
2577 * detect platform raid capabilities. The option-rom resides in a
2578 * platform "Adapter ROM". We scan for its signature to retrieve the
2579 * platform capabilities. If raid support is disabled in the BIOS the
2580 * option-rom capability structure will not be available.
2582 struct sys_dev
*list
, *hba
;
2587 if (enumerate_only
) {
2588 if (check_env("IMSM_NO_PLATFORM"))
2590 list
= find_intel_devices();
2593 for (hba
= list
; hba
; hba
= hba
->next
) {
2594 if (find_imsm_capability(hba
)) {
2604 list
= find_intel_devices();
2607 pr_err("no active Intel(R) RAID controller found.\n");
2609 } else if (verbose
> 0)
2610 print_found_intel_controllers(list
);
2612 for (hba
= list
; hba
; hba
= hba
->next
) {
2613 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2615 if (!find_imsm_capability(hba
)) {
2617 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2618 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2619 get_sys_dev_type(hba
->type
));
2625 if (controller_path
&& result
== 1) {
2626 pr_err("no active Intel(R) RAID controller found under %s\n",
2631 const struct orom_entry
*entry
;
2633 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2634 if (entry
->type
== SYS_DEV_VMD
) {
2635 print_imsm_capability(&entry
->orom
);
2636 printf(" 3rd party NVMe :%s supported\n",
2637 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2638 for (hba
= list
; hba
; hba
= hba
->next
) {
2639 if (hba
->type
== SYS_DEV_VMD
) {
2641 printf(" I/O Controller : %s (%s)\n",
2642 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2643 if (print_nvme_info(hba
)) {
2645 pr_err("failed to get devices attached to VMD domain.\n");
2654 print_imsm_capability(&entry
->orom
);
2655 if (entry
->type
== SYS_DEV_NVME
) {
2656 for (hba
= list
; hba
; hba
= hba
->next
) {
2657 if (hba
->type
== SYS_DEV_NVME
)
2658 print_nvme_info(hba
);
2664 struct devid_list
*devid
;
2665 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2666 hba
= device_by_id(devid
->devid
);
2670 printf(" I/O Controller : %s (%s)\n",
2671 hba
->path
, get_sys_dev_type(hba
->type
));
2672 if (hba
->type
== SYS_DEV_SATA
) {
2673 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2674 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2676 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2687 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2689 struct sys_dev
*list
, *hba
;
2692 list
= find_intel_devices();
2695 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2700 for (hba
= list
; hba
; hba
= hba
->next
) {
2701 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2703 if (!find_imsm_capability(hba
) && verbose
> 0) {
2705 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2706 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2712 const struct orom_entry
*entry
;
2714 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2715 if (entry
->type
== SYS_DEV_VMD
) {
2716 for (hba
= list
; hba
; hba
= hba
->next
)
2717 print_imsm_capability_export(&entry
->orom
);
2720 print_imsm_capability_export(&entry
->orom
);
2726 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2728 /* the imsm metadata format does not specify any host
2729 * identification information. We return -1 since we can never
2730 * confirm nor deny whether a given array is "meant" for this
2731 * host. We rely on compare_super and the 'family_num' fields to
2732 * exclude member disks that do not belong, and we rely on
2733 * mdadm.conf to specify the arrays that should be assembled.
2734 * Auto-assembly may still pick up "foreign" arrays.
2740 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2742 /* The uuid returned here is used for:
2743 * uuid to put into bitmap file (Create, Grow)
2744 * uuid for backup header when saving critical section (Grow)
2745 * comparing uuids when re-adding a device into an array
2746 * In these cases the uuid required is that of the data-array,
2747 * not the device-set.
2748 * uuid to recognise same set when adding a missing device back
2749 * to an array. This is a uuid for the device-set.
2751 * For each of these we can make do with a truncated
2752 * or hashed uuid rather than the original, as long as
2754 * In each case the uuid required is that of the data-array,
2755 * not the device-set.
2757 /* imsm does not track uuid's so we synthesis one using sha1 on
2758 * - The signature (Which is constant for all imsm array, but no matter)
2759 * - the orig_family_num of the container
2760 * - the index number of the volume
2761 * - the 'serial' number of the volume.
2762 * Hopefully these are all constant.
2764 struct intel_super
*super
= st
->sb
;
2767 struct sha1_ctx ctx
;
2768 struct imsm_dev
*dev
= NULL
;
2771 /* some mdadm versions failed to set ->orig_family_num, in which
2772 * case fall back to ->family_num. orig_family_num will be
2773 * fixed up with the first metadata update.
2775 family_num
= super
->anchor
->orig_family_num
;
2776 if (family_num
== 0)
2777 family_num
= super
->anchor
->family_num
;
2778 sha1_init_ctx(&ctx
);
2779 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2780 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2781 if (super
->current_vol
>= 0)
2782 dev
= get_imsm_dev(super
, super
->current_vol
);
2784 __u32 vol
= super
->current_vol
;
2785 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2786 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2788 sha1_finish_ctx(&ctx
, buf
);
2789 memcpy(uuid
, buf
, 4*4);
2794 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2796 __u8
*v
= get_imsm_version(mpb
);
2797 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2798 char major
[] = { 0, 0, 0 };
2799 char minor
[] = { 0 ,0, 0 };
2800 char patch
[] = { 0, 0, 0 };
2801 char *ver_parse
[] = { major
, minor
, patch
};
2805 while (*v
!= '\0' && v
< end
) {
2806 if (*v
!= '.' && j
< 2)
2807 ver_parse
[i
][j
++] = *v
;
2815 *m
= strtol(minor
, NULL
, 0);
2816 *p
= strtol(patch
, NULL
, 0);
2820 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2822 /* migr_strip_size when repairing or initializing parity */
2823 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2824 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2826 switch (get_imsm_raid_level(map
)) {
2831 return 128*1024 >> 9;
2835 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2837 /* migr_strip_size when rebuilding a degraded disk, no idea why
2838 * this is different than migr_strip_size_resync(), but it's good
2841 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2842 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2844 switch (get_imsm_raid_level(map
)) {
2847 if (map
->num_members
% map
->num_domains
== 0)
2848 return 128*1024 >> 9;
2852 return max((__u32
) 64*1024 >> 9, chunk
);
2854 return 128*1024 >> 9;
2858 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2860 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2861 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2862 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2863 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2865 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2868 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2870 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2871 int level
= get_imsm_raid_level(lo
);
2873 if (level
== 1 || level
== 10) {
2874 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2876 return hi
->num_domains
;
2878 return num_stripes_per_unit_resync(dev
);
2881 static __u8
imsm_num_data_members(struct imsm_map
*map
)
2883 /* named 'imsm_' because raid0, raid1 and raid10
2884 * counter-intuitively have the same number of data disks
2886 switch (get_imsm_raid_level(map
)) {
2888 return map
->num_members
;
2892 return map
->num_members
/2;
2894 return map
->num_members
- 1;
2896 dprintf("unsupported raid level\n");
2901 static unsigned long long calc_component_size(struct imsm_map
*map
,
2902 struct imsm_dev
*dev
)
2904 unsigned long long component_size
;
2905 unsigned long long dev_size
= imsm_dev_size(dev
);
2906 long long calc_dev_size
= 0;
2907 unsigned int member_disks
= imsm_num_data_members(map
);
2909 if (member_disks
== 0)
2912 component_size
= per_dev_array_size(map
);
2913 calc_dev_size
= component_size
* member_disks
;
2915 /* Component size is rounded to 1MB so difference between size from
2916 * metadata and size calculated from num_data_stripes equals up to
2917 * 2048 blocks per each device. If the difference is higher it means
2918 * that array size was expanded and num_data_stripes was not updated.
2920 if (llabs(calc_dev_size
- (long long)dev_size
) >
2921 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
2922 component_size
= dev_size
/ member_disks
;
2923 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
2924 component_size
/ map
->blocks_per_strip
,
2925 num_data_stripes(map
));
2928 return component_size
;
2931 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2933 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2934 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2936 switch(get_imsm_raid_level(map
)) {
2939 return chunk
* map
->num_domains
;
2941 return chunk
* map
->num_members
;
2947 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2949 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2950 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2951 __u32 strip
= block
/ chunk
;
2953 switch (get_imsm_raid_level(map
)) {
2956 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2957 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2959 return vol_stripe
* chunk
+ block
% chunk
;
2961 __u32 stripe
= strip
/ (map
->num_members
- 1);
2963 return stripe
* chunk
+ block
% chunk
;
2970 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2971 struct imsm_dev
*dev
)
2973 /* calculate the conversion factor between per member 'blocks'
2974 * (md/{resync,rebuild}_start) and imsm migration units, return
2975 * 0 for the 'not migrating' and 'unsupported migration' cases
2977 if (!dev
->vol
.migr_state
)
2980 switch (migr_type(dev
)) {
2981 case MIGR_GEN_MIGR
: {
2982 struct migr_record
*migr_rec
= super
->migr_rec
;
2983 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2988 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2989 __u32 stripes_per_unit
;
2990 __u32 blocks_per_unit
;
2999 /* yes, this is really the translation of migr_units to
3000 * per-member blocks in the 'resync' case
3002 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
3003 migr_chunk
= migr_strip_blocks_resync(dev
);
3004 disks
= imsm_num_data_members(map
);
3005 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
3006 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
3007 segment
= blocks_per_unit
/ stripe
;
3008 block_rel
= blocks_per_unit
- segment
* stripe
;
3009 parity_depth
= parity_segment_depth(dev
);
3010 block_map
= map_migr_block(dev
, block_rel
);
3011 return block_map
+ parity_depth
* segment
;
3013 case MIGR_REBUILD
: {
3014 __u32 stripes_per_unit
;
3017 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
3018 migr_chunk
= migr_strip_blocks_rebuild(dev
);
3019 return migr_chunk
* stripes_per_unit
;
3021 case MIGR_STATE_CHANGE
:
3027 static int imsm_level_to_layout(int level
)
3035 return ALGORITHM_LEFT_ASYMMETRIC
;
3042 /*******************************************************************************
3043 * Function: read_imsm_migr_rec
3044 * Description: Function reads imsm migration record from last sector of disk
3046 * fd : disk descriptor
3047 * super : metadata info
3051 ******************************************************************************/
3052 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
3055 unsigned int sector_size
= super
->sector_size
;
3056 unsigned long long dsize
;
3058 get_dev_size(fd
, NULL
, &dsize
);
3059 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
3061 pr_err("Cannot seek to anchor block: %s\n",
3065 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
3066 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3067 MIGR_REC_BUF_SECTORS
*sector_size
) {
3068 pr_err("Cannot read migr record block: %s\n",
3073 if (sector_size
== 4096)
3074 convert_from_4k_imsm_migr_rec(super
);
3080 static struct imsm_dev
*imsm_get_device_during_migration(
3081 struct intel_super
*super
)
3084 struct intel_dev
*dv
;
3086 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
3087 if (is_gen_migration(dv
->dev
))
3093 /*******************************************************************************
3094 * Function: load_imsm_migr_rec
3095 * Description: Function reads imsm migration record (it is stored at the last
3098 * super : imsm internal array info
3102 * -2 : no migration in progress
3103 ******************************************************************************/
3104 static int load_imsm_migr_rec(struct intel_super
*super
)
3110 struct imsm_dev
*dev
;
3111 struct imsm_map
*map
;
3115 /* find map under migration */
3116 dev
= imsm_get_device_during_migration(super
);
3117 /* nothing to load,no migration in progress?
3122 map
= get_imsm_map(dev
, MAP_0
);
3126 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3127 /* skip spare and failed disks
3131 /* read only from one of the first two slots
3133 slot
= get_imsm_disk_slot(map
, dl
->index
);
3134 if (slot
> 1 || slot
< 0)
3138 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3139 fd
= dev_open(nm
, O_RDONLY
);
3152 retval
= read_imsm_migr_rec(fd
, super
);
3159 /*******************************************************************************
3160 * function: imsm_create_metadata_checkpoint_update
3161 * Description: It creates update for checkpoint change.
3163 * super : imsm internal array info
3164 * u : pointer to prepared update
3167 * If length is equal to 0, input pointer u contains no update
3168 ******************************************************************************/
3169 static int imsm_create_metadata_checkpoint_update(
3170 struct intel_super
*super
,
3171 struct imsm_update_general_migration_checkpoint
**u
)
3174 int update_memory_size
= 0;
3176 dprintf("(enter)\n");
3182 /* size of all update data without anchor */
3183 update_memory_size
=
3184 sizeof(struct imsm_update_general_migration_checkpoint
);
3186 *u
= xcalloc(1, update_memory_size
);
3188 dprintf("error: cannot get memory\n");
3191 (*u
)->type
= update_general_migration_checkpoint
;
3192 (*u
)->curr_migr_unit
= current_migr_unit(super
->migr_rec
);
3193 dprintf("prepared for %llu\n", (unsigned long long)(*u
)->curr_migr_unit
);
3195 return update_memory_size
;
3198 static void imsm_update_metadata_locally(struct supertype
*st
,
3199 void *buf
, int len
);
3201 /*******************************************************************************
3202 * Function: write_imsm_migr_rec
3203 * Description: Function writes imsm migration record
3204 * (at the last sector of disk)
3206 * super : imsm internal array info
3210 ******************************************************************************/
3211 static int write_imsm_migr_rec(struct supertype
*st
)
3213 struct intel_super
*super
= st
->sb
;
3214 unsigned int sector_size
= super
->sector_size
;
3215 unsigned long long dsize
;
3219 struct imsm_update_general_migration_checkpoint
*u
;
3220 struct imsm_dev
*dev
;
3221 struct imsm_map
*map
;
3223 /* find map under migration */
3224 dev
= imsm_get_device_during_migration(super
);
3225 /* if no migration, write buffer anyway to clear migr_record
3226 * on disk based on first available device
3229 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3230 super
->current_vol
);
3232 map
= get_imsm_map(dev
, MAP_0
);
3234 if (sector_size
== 4096)
3235 convert_to_4k_imsm_migr_rec(super
);
3236 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3239 /* skip failed and spare devices */
3242 /* write to 2 first slots only */
3244 slot
= get_imsm_disk_slot(map
, sd
->index
);
3245 if (map
== NULL
|| slot
> 1 || slot
< 0)
3248 get_dev_size(sd
->fd
, NULL
, &dsize
);
3249 if (lseek64(sd
->fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*
3252 pr_err("Cannot seek to anchor block: %s\n",
3256 if ((unsigned int)write(sd
->fd
, super
->migr_rec_buf
,
3257 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3258 MIGR_REC_BUF_SECTORS
*sector_size
) {
3259 pr_err("Cannot write migr record block: %s\n",
3264 if (sector_size
== 4096)
3265 convert_from_4k_imsm_migr_rec(super
);
3266 /* update checkpoint information in metadata */
3267 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3269 dprintf("imsm: Cannot prepare update\n");
3272 /* update metadata locally */
3273 imsm_update_metadata_locally(st
, u
, len
);
3274 /* and possibly remotely */
3275 if (st
->update_tail
) {
3276 append_metadata_update(st
, u
, len
);
3277 /* during reshape we do all work inside metadata handler
3278 * manage_reshape(), so metadata update has to be triggered
3281 flush_metadata_updates(st
);
3282 st
->update_tail
= &st
->updates
;
3291 /* spare/missing disks activations are not allowe when
3292 * array/container performs reshape operation, because
3293 * all arrays in container works on the same disks set
3295 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3298 struct intel_dev
*i_dev
;
3299 struct imsm_dev
*dev
;
3301 /* check whole container
3303 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3305 if (is_gen_migration(dev
)) {
3306 /* No repair during any migration in container
3314 static unsigned long long imsm_component_size_alignment_check(int level
,
3316 unsigned int sector_size
,
3317 unsigned long long component_size
)
3319 unsigned int component_size_alignment
;
3321 /* check component size alignment
3323 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3325 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3326 level
, chunk_size
, component_size
,
3327 component_size_alignment
);
3329 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3330 dprintf("imsm: reported component size aligned from %llu ",
3332 component_size
-= component_size_alignment
;
3333 dprintf_cont("to %llu (%i).\n",
3334 component_size
, component_size_alignment
);
3337 return component_size
;
3340 /*******************************************************************************
3341 * Function: get_bitmap_header_sector
3342 * Description: Returns the sector where the bitmap header is placed.
3344 * st : supertype information
3345 * dev_idx : index of the device with bitmap
3348 * The sector where the bitmap header is placed
3349 ******************************************************************************/
3350 static unsigned long long get_bitmap_header_sector(struct intel_super
*super
,
3353 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3354 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3356 if (!super
->sector_size
) {
3357 dprintf("sector size is not set\n");
3361 return pba_of_lba0(map
) + calc_component_size(map
, dev
) +
3362 (IMSM_BITMAP_HEADER_OFFSET
/ super
->sector_size
);
3365 /*******************************************************************************
3366 * Function: get_bitmap_sector
3367 * Description: Returns the sector where the bitmap is placed.
3369 * st : supertype information
3370 * dev_idx : index of the device with bitmap
3373 * The sector where the bitmap is placed
3374 ******************************************************************************/
3375 static unsigned long long get_bitmap_sector(struct intel_super
*super
,
3378 if (!super
->sector_size
) {
3379 dprintf("sector size is not set\n");
3383 return get_bitmap_header_sector(super
, dev_idx
) +
3384 (IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
);
3387 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3389 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3390 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3392 return pba_of_lba0(map
) +
3393 (num_data_stripes(map
) * map
->blocks_per_strip
);
3396 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3398 struct intel_super
*super
= st
->sb
;
3399 struct migr_record
*migr_rec
= super
->migr_rec
;
3400 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3401 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3402 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3403 struct imsm_map
*map_to_analyse
= map
;
3405 int map_disks
= info
->array
.raid_disks
;
3407 memset(info
, 0, sizeof(*info
));
3409 map_to_analyse
= prev_map
;
3411 dl
= super
->current_disk
;
3413 info
->container_member
= super
->current_vol
;
3414 info
->array
.raid_disks
= map
->num_members
;
3415 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3416 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3417 info
->array
.md_minor
= -1;
3418 info
->array
.ctime
= 0;
3419 info
->array
.utime
= 0;
3420 info
->array
.chunk_size
=
3421 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3422 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3423 info
->custom_array_size
= imsm_dev_size(dev
);
3424 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3426 if (is_gen_migration(dev
)) {
3427 info
->reshape_active
= 1;
3428 info
->new_level
= get_imsm_raid_level(map
);
3429 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3430 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3431 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3432 if (info
->delta_disks
) {
3433 /* this needs to be applied to every array
3436 info
->reshape_active
= CONTAINER_RESHAPE
;
3438 /* We shape information that we give to md might have to be
3439 * modify to cope with md's requirement for reshaping arrays.
3440 * For example, when reshaping a RAID0, md requires it to be
3441 * presented as a degraded RAID4.
3442 * Also if a RAID0 is migrating to a RAID5 we need to specify
3443 * the array as already being RAID5, but the 'before' layout
3444 * is a RAID4-like layout.
3446 switch (info
->array
.level
) {
3448 switch(info
->new_level
) {
3450 /* conversion is happening as RAID4 */
3451 info
->array
.level
= 4;
3452 info
->array
.raid_disks
+= 1;
3455 /* conversion is happening as RAID5 */
3456 info
->array
.level
= 5;
3457 info
->array
.layout
= ALGORITHM_PARITY_N
;
3458 info
->delta_disks
-= 1;
3461 /* FIXME error message */
3462 info
->array
.level
= UnSet
;
3468 info
->new_level
= UnSet
;
3469 info
->new_layout
= UnSet
;
3470 info
->new_chunk
= info
->array
.chunk_size
;
3471 info
->delta_disks
= 0;
3475 info
->disk
.major
= dl
->major
;
3476 info
->disk
.minor
= dl
->minor
;
3477 info
->disk
.number
= dl
->index
;
3478 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3482 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3483 info
->component_size
= calc_component_size(map
, dev
);
3484 info
->component_size
= imsm_component_size_alignment_check(
3486 info
->array
.chunk_size
,
3488 info
->component_size
);
3489 info
->bb
.supported
= 1;
3491 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3492 info
->recovery_start
= MaxSector
;
3494 if (info
->array
.level
== 5 &&
3495 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3496 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3497 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3498 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3499 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3500 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3502 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3504 } else if (info
->array
.level
<= 0) {
3505 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3507 if (dev
->rwh_policy
== RWH_BITMAP
) {
3508 info
->bitmap_offset
= get_bitmap_sector(super
, super
->current_vol
);
3509 info
->consistency_policy
= CONSISTENCY_POLICY_BITMAP
;
3511 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3515 info
->reshape_progress
= 0;
3516 info
->resync_start
= MaxSector
;
3517 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3518 !(info
->array
.state
& 1)) &&
3519 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3520 info
->resync_start
= 0;
3522 if (dev
->vol
.migr_state
) {
3523 switch (migr_type(dev
)) {
3526 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3528 __u64 units
= vol_curr_migr_unit(dev
);
3530 info
->resync_start
= blocks_per_unit
* units
;
3533 case MIGR_GEN_MIGR
: {
3534 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3536 __u64 units
= current_migr_unit(migr_rec
);
3539 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3541 (get_num_migr_units(migr_rec
)-1)) &&
3542 (super
->migr_rec
->rec_status
==
3543 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3546 info
->reshape_progress
= blocks_per_unit
* units
;
3548 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3549 (unsigned long long)units
,
3550 (unsigned long long)blocks_per_unit
,
3551 info
->reshape_progress
);
3553 used_disks
= imsm_num_data_members(prev_map
);
3554 if (used_disks
> 0) {
3555 info
->custom_array_size
= per_dev_array_size(map
) *
3560 /* we could emulate the checkpointing of
3561 * 'sync_action=check' migrations, but for now
3562 * we just immediately complete them
3565 /* this is handled by container_content_imsm() */
3566 case MIGR_STATE_CHANGE
:
3567 /* FIXME handle other migrations */
3569 /* we are not dirty, so... */
3570 info
->resync_start
= MaxSector
;
3574 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3575 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3577 info
->array
.major_version
= -1;
3578 info
->array
.minor_version
= -2;
3579 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3580 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3581 uuid_from_super_imsm(st
, info
->uuid
);
3585 for (i
=0; i
<map_disks
; i
++) {
3587 if (i
< info
->array
.raid_disks
) {
3588 struct imsm_disk
*dsk
;
3589 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3590 dsk
= get_imsm_disk(super
, j
);
3591 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3598 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3599 int failed
, int look_in_map
);
3601 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3604 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3606 if (is_gen_migration(dev
)) {
3609 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3611 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3612 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3613 if (map2
->map_state
!= map_state
) {
3614 map2
->map_state
= map_state
;
3615 super
->updates_pending
++;
3620 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3624 for (d
= super
->missing
; d
; d
= d
->next
)
3625 if (d
->index
== index
)
3630 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3632 struct intel_super
*super
= st
->sb
;
3633 struct imsm_disk
*disk
;
3634 int map_disks
= info
->array
.raid_disks
;
3635 int max_enough
= -1;
3637 struct imsm_super
*mpb
;
3639 if (super
->current_vol
>= 0) {
3640 getinfo_super_imsm_volume(st
, info
, map
);
3643 memset(info
, 0, sizeof(*info
));
3645 /* Set raid_disks to zero so that Assemble will always pull in valid
3648 info
->array
.raid_disks
= 0;
3649 info
->array
.level
= LEVEL_CONTAINER
;
3650 info
->array
.layout
= 0;
3651 info
->array
.md_minor
= -1;
3652 info
->array
.ctime
= 0; /* N/A for imsm */
3653 info
->array
.utime
= 0;
3654 info
->array
.chunk_size
= 0;
3656 info
->disk
.major
= 0;
3657 info
->disk
.minor
= 0;
3658 info
->disk
.raid_disk
= -1;
3659 info
->reshape_active
= 0;
3660 info
->array
.major_version
= -1;
3661 info
->array
.minor_version
= -2;
3662 strcpy(info
->text_version
, "imsm");
3663 info
->safe_mode_delay
= 0;
3664 info
->disk
.number
= -1;
3665 info
->disk
.state
= 0;
3667 info
->recovery_start
= MaxSector
;
3668 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3669 info
->bb
.supported
= 1;
3671 /* do we have the all the insync disks that we expect? */
3672 mpb
= super
->anchor
;
3673 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3675 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3676 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3677 int failed
, enough
, j
, missing
= 0;
3678 struct imsm_map
*map
;
3681 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3682 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3683 map
= get_imsm_map(dev
, MAP_0
);
3685 /* any newly missing disks?
3686 * (catches single-degraded vs double-degraded)
3688 for (j
= 0; j
< map
->num_members
; j
++) {
3689 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3690 __u32 idx
= ord_to_idx(ord
);
3692 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3693 info
->disk
.raid_disk
= j
;
3695 if (!(ord
& IMSM_ORD_REBUILD
) &&
3696 get_imsm_missing(super
, idx
)) {
3702 if (state
== IMSM_T_STATE_FAILED
)
3704 else if (state
== IMSM_T_STATE_DEGRADED
&&
3705 (state
!= map
->map_state
|| missing
))
3707 else /* we're normal, or already degraded */
3709 if (is_gen_migration(dev
) && missing
) {
3710 /* during general migration we need all disks
3711 * that process is running on.
3712 * No new missing disk is allowed.
3716 /* no more checks necessary
3720 /* in the missing/failed disk case check to see
3721 * if at least one array is runnable
3723 max_enough
= max(max_enough
, enough
);
3725 dprintf("enough: %d\n", max_enough
);
3726 info
->container_enough
= max_enough
;
3729 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3731 disk
= &super
->disks
->disk
;
3732 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3733 info
->component_size
= reserved
;
3734 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3735 /* we don't change info->disk.raid_disk here because
3736 * this state will be finalized in mdmon after we have
3737 * found the 'most fresh' version of the metadata
3739 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3740 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3741 0 : (1 << MD_DISK_SYNC
);
3744 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3745 * ->compare_super may have updated the 'num_raid_devs' field for spares
3747 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3748 uuid_from_super_imsm(st
, info
->uuid
);
3750 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3752 /* I don't know how to compute 'map' on imsm, so use safe default */
3755 for (i
= 0; i
< map_disks
; i
++)
3761 /* allocates memory and fills disk in mdinfo structure
3762 * for each disk in array */
3763 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3765 struct mdinfo
*mddev
;
3766 struct intel_super
*super
= st
->sb
;
3767 struct imsm_disk
*disk
;
3770 if (!super
|| !super
->disks
)
3773 mddev
= xcalloc(1, sizeof(*mddev
));
3777 tmp
= xcalloc(1, sizeof(*tmp
));
3779 tmp
->next
= mddev
->devs
;
3781 tmp
->disk
.number
= count
++;
3782 tmp
->disk
.major
= dl
->major
;
3783 tmp
->disk
.minor
= dl
->minor
;
3784 tmp
->disk
.state
= is_configured(disk
) ?
3785 (1 << MD_DISK_ACTIVE
) : 0;
3786 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3787 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3788 tmp
->disk
.raid_disk
= -1;
3794 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3795 char *update
, char *devname
, int verbose
,
3796 int uuid_set
, char *homehost
)
3798 /* For 'assemble' and 'force' we need to return non-zero if any
3799 * change was made. For others, the return value is ignored.
3800 * Update options are:
3801 * force-one : This device looks a bit old but needs to be included,
3802 * update age info appropriately.
3803 * assemble: clear any 'faulty' flag to allow this device to
3805 * force-array: Array is degraded but being forced, mark it clean
3806 * if that will be needed to assemble it.
3808 * newdev: not used ????
3809 * grow: Array has gained a new device - this is currently for
3811 * resync: mark as dirty so a resync will happen.
3812 * name: update the name - preserving the homehost
3813 * uuid: Change the uuid of the array to match watch is given
3815 * Following are not relevant for this imsm:
3816 * sparc2.2 : update from old dodgey metadata
3817 * super-minor: change the preferred_minor number
3818 * summaries: update redundant counters.
3819 * homehost: update the recorded homehost
3820 * _reshape_progress: record new reshape_progress position.
3823 struct intel_super
*super
= st
->sb
;
3824 struct imsm_super
*mpb
;
3826 /* we can only update container info */
3827 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3830 mpb
= super
->anchor
;
3832 if (strcmp(update
, "uuid") == 0) {
3833 /* We take this to mean that the family_num should be updated.
3834 * However that is much smaller than the uuid so we cannot really
3835 * allow an explicit uuid to be given. And it is hard to reliably
3837 * So if !uuid_set we know the current uuid is random and just used
3838 * the first 'int' and copy it to the other 3 positions.
3839 * Otherwise we require the 4 'int's to be the same as would be the
3840 * case if we are using a random uuid. So an explicit uuid will be
3841 * accepted as long as all for ints are the same... which shouldn't hurt
3844 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3847 if (info
->uuid
[0] != info
->uuid
[1] ||
3848 info
->uuid
[1] != info
->uuid
[2] ||
3849 info
->uuid
[2] != info
->uuid
[3])
3855 mpb
->orig_family_num
= info
->uuid
[0];
3856 } else if (strcmp(update
, "assemble") == 0)
3861 /* successful update? recompute checksum */
3863 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3868 static size_t disks_to_mpb_size(int disks
)
3872 size
= sizeof(struct imsm_super
);
3873 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3874 size
+= 2 * sizeof(struct imsm_dev
);
3875 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3876 size
+= (4 - 2) * sizeof(struct imsm_map
);
3877 /* 4 possible disk_ord_tbl's */
3878 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3879 /* maximum bbm log */
3880 size
+= sizeof(struct bbm_log
);
3885 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3886 unsigned long long data_offset
)
3888 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3891 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3894 static void free_devlist(struct intel_super
*super
)
3896 struct intel_dev
*dv
;
3898 while (super
->devlist
) {
3899 dv
= super
->devlist
->next
;
3900 free(super
->devlist
->dev
);
3901 free(super
->devlist
);
3902 super
->devlist
= dv
;
3906 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3908 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3911 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
,
3916 * 0 same, or first was empty, and second was copied
3917 * 1 second had wrong number
3919 * 3 wrong other info
3921 struct intel_super
*first
= st
->sb
;
3922 struct intel_super
*sec
= tst
->sb
;
3929 /* in platform dependent environment test if the disks
3930 * use the same Intel hba
3931 * If not on Intel hba at all, allow anything.
3933 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3934 if (first
->hba
->type
!= sec
->hba
->type
) {
3936 pr_err("HBAs of devices do not match %s != %s\n",
3937 get_sys_dev_type(first
->hba
->type
),
3938 get_sys_dev_type(sec
->hba
->type
));
3942 if (first
->orom
!= sec
->orom
) {
3944 pr_err("HBAs of devices do not match %s != %s\n",
3945 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3951 /* if an anchor does not have num_raid_devs set then it is a free
3954 if (first
->anchor
->num_raid_devs
> 0 &&
3955 sec
->anchor
->num_raid_devs
> 0) {
3956 /* Determine if these disks might ever have been
3957 * related. Further disambiguation can only take place
3958 * in load_super_imsm_all
3960 __u32 first_family
= first
->anchor
->orig_family_num
;
3961 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3963 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3964 MAX_SIGNATURE_LENGTH
) != 0)
3967 if (first_family
== 0)
3968 first_family
= first
->anchor
->family_num
;
3969 if (sec_family
== 0)
3970 sec_family
= sec
->anchor
->family_num
;
3972 if (first_family
!= sec_family
)
3977 /* if 'first' is a spare promote it to a populated mpb with sec's
3980 if (first
->anchor
->num_raid_devs
== 0 &&
3981 sec
->anchor
->num_raid_devs
> 0) {
3983 struct intel_dev
*dv
;
3984 struct imsm_dev
*dev
;
3986 /* we need to copy raid device info from sec if an allocation
3987 * fails here we don't associate the spare
3989 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3990 dv
= xmalloc(sizeof(*dv
));
3991 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3994 dv
->next
= first
->devlist
;
3995 first
->devlist
= dv
;
3997 if (i
< sec
->anchor
->num_raid_devs
) {
3998 /* allocation failure */
3999 free_devlist(first
);
4000 pr_err("imsm: failed to associate spare\n");
4003 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
4004 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
4005 first
->anchor
->family_num
= sec
->anchor
->family_num
;
4006 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
4007 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
4008 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
4014 static void fd2devname(int fd
, char *name
)
4022 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
4025 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
4027 char path
[PATH_MAX
];
4028 char *name
= fd2kname(fd
);
4033 if (strncmp(name
, "nvme", 4) != 0)
4036 if (!diskfd_to_devpath(fd
, 1, path
))
4039 return devpath_to_char(path
, "serial", buf
, buf_len
, 0);
4042 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
4044 static int imsm_read_serial(int fd
, char *devname
,
4045 __u8
*serial
, size_t serial_buf_len
)
4054 memset(buf
, 0, sizeof(buf
));
4056 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
4059 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
4061 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
4062 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
4063 fd2devname(fd
, (char *) serial
);
4069 pr_err("Failed to retrieve serial for %s\n",
4074 /* trim all whitespace and non-printable characters and convert
4077 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4080 /* ':' is reserved for use in placeholder serial
4081 * numbers for missing disks
4092 if (len
> serial_buf_len
) {
4093 /* truncate leading characters */
4094 dest
+= len
- serial_buf_len
;
4095 len
= serial_buf_len
;
4098 memset(serial
, 0, serial_buf_len
);
4099 memcpy(serial
, dest
, len
);
4104 static int serialcmp(__u8
*s1
, __u8
*s2
)
4106 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4109 static void serialcpy(__u8
*dest
, __u8
*src
)
4111 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4114 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4118 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4119 if (serialcmp(dl
->serial
, serial
) == 0)
4125 static struct imsm_disk
*
4126 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4130 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4131 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4133 if (serialcmp(disk
->serial
, serial
) == 0) {
4144 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4146 struct imsm_disk
*disk
;
4151 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4153 rv
= imsm_read_serial(fd
, devname
, serial
, MAX_RAID_SERIAL_LEN
);
4158 dl
= xcalloc(1, sizeof(*dl
));
4161 dl
->major
= major(stb
.st_rdev
);
4162 dl
->minor
= minor(stb
.st_rdev
);
4163 dl
->next
= super
->disks
;
4164 dl
->fd
= keep_fd
? fd
: -1;
4165 assert(super
->disks
== NULL
);
4167 serialcpy(dl
->serial
, serial
);
4170 fd2devname(fd
, name
);
4172 dl
->devname
= xstrdup(devname
);
4174 dl
->devname
= xstrdup(name
);
4176 /* look up this disk's index in the current anchor */
4177 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4180 /* only set index on disks that are a member of a
4181 * populated contianer, i.e. one with raid_devs
4183 if (is_failed(&dl
->disk
))
4185 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4192 /* When migrating map0 contains the 'destination' state while map1
4193 * contains the current state. When not migrating map0 contains the
4194 * current state. This routine assumes that map[0].map_state is set to
4195 * the current array state before being called.
4197 * Migration is indicated by one of the following states
4198 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4199 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4200 * map1state=unitialized)
4201 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4203 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4204 * map1state=degraded)
4205 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4208 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4209 __u8 to_state
, int migr_type
)
4211 struct imsm_map
*dest
;
4212 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4214 dev
->vol
.migr_state
= 1;
4215 set_migr_type(dev
, migr_type
);
4216 set_vol_curr_migr_unit(dev
, 0);
4217 dest
= get_imsm_map(dev
, MAP_1
);
4219 /* duplicate and then set the target end state in map[0] */
4220 memcpy(dest
, src
, sizeof_imsm_map(src
));
4221 if (migr_type
== MIGR_GEN_MIGR
) {
4225 for (i
= 0; i
< src
->num_members
; i
++) {
4226 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4227 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4231 if (migr_type
== MIGR_GEN_MIGR
)
4232 /* Clear migration record */
4233 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4235 src
->map_state
= to_state
;
4238 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4241 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4242 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4246 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4247 * completed in the last migration.
4249 * FIXME add support for raid-level-migration
4251 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4252 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4253 /* when final map state is other than expected
4254 * merge maps (not for migration)
4258 for (i
= 0; i
< prev
->num_members
; i
++)
4259 for (j
= 0; j
< map
->num_members
; j
++)
4260 /* during online capacity expansion
4261 * disks position can be changed
4262 * if takeover is used
4264 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4265 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4266 map
->disk_ord_tbl
[j
] |=
4267 prev
->disk_ord_tbl
[i
];
4270 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4271 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4274 dev
->vol
.migr_state
= 0;
4275 set_migr_type(dev
, 0);
4276 set_vol_curr_migr_unit(dev
, 0);
4277 map
->map_state
= map_state
;
4280 static int parse_raid_devices(struct intel_super
*super
)
4283 struct imsm_dev
*dev_new
;
4284 size_t len
, len_migr
;
4286 size_t space_needed
= 0;
4287 struct imsm_super
*mpb
= super
->anchor
;
4289 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4290 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4291 struct intel_dev
*dv
;
4293 len
= sizeof_imsm_dev(dev_iter
, 0);
4294 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4296 space_needed
+= len_migr
- len
;
4298 dv
= xmalloc(sizeof(*dv
));
4299 if (max_len
< len_migr
)
4301 if (max_len
> len_migr
)
4302 space_needed
+= max_len
- len_migr
;
4303 dev_new
= xmalloc(max_len
);
4304 imsm_copy_dev(dev_new
, dev_iter
);
4307 dv
->next
= super
->devlist
;
4308 super
->devlist
= dv
;
4311 /* ensure that super->buf is large enough when all raid devices
4314 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4317 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4318 super
->sector_size
);
4319 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4322 memcpy(buf
, super
->buf
, super
->len
);
4323 memset(buf
+ super
->len
, 0, len
- super
->len
);
4329 super
->extra_space
+= space_needed
;
4334 /*******************************************************************************
4335 * Function: check_mpb_migr_compatibility
4336 * Description: Function checks for unsupported migration features:
4337 * - migration optimization area (pba_of_lba0)
4338 * - descending reshape (ascending_migr)
4340 * super : imsm metadata information
4342 * 0 : migration is compatible
4343 * -1 : migration is not compatible
4344 ******************************************************************************/
4345 int check_mpb_migr_compatibility(struct intel_super
*super
)
4347 struct imsm_map
*map0
, *map1
;
4348 struct migr_record
*migr_rec
= super
->migr_rec
;
4351 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4352 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4355 dev_iter
->vol
.migr_state
== 1 &&
4356 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4357 /* This device is migrating */
4358 map0
= get_imsm_map(dev_iter
, MAP_0
);
4359 map1
= get_imsm_map(dev_iter
, MAP_1
);
4360 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4361 /* migration optimization area was used */
4363 if (migr_rec
->ascending_migr
== 0 &&
4364 migr_rec
->dest_depth_per_unit
> 0)
4365 /* descending reshape not supported yet */
4372 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4374 /* load_imsm_mpb - read matrix metadata
4375 * allocates super->mpb to be freed by free_imsm
4377 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4379 unsigned long long dsize
;
4380 unsigned long long sectors
;
4381 unsigned int sector_size
= super
->sector_size
;
4383 struct imsm_super
*anchor
;
4386 get_dev_size(fd
, NULL
, &dsize
);
4387 if (dsize
< 2*sector_size
) {
4389 pr_err("%s: device to small for imsm\n",
4394 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4396 pr_err("Cannot seek to anchor block on %s: %s\n",
4397 devname
, strerror(errno
));
4401 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4403 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4406 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4408 pr_err("Cannot read anchor block on %s: %s\n",
4409 devname
, strerror(errno
));
4414 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4416 pr_err("no IMSM anchor on %s\n", devname
);
4421 __free_imsm(super
, 0);
4422 /* reload capability and hba */
4424 /* capability and hba must be updated with new super allocation */
4425 find_intel_hba_capability(fd
, super
, devname
);
4426 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4427 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4429 pr_err("unable to allocate %zu byte mpb buffer\n",
4434 memcpy(super
->buf
, anchor
, sector_size
);
4436 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4439 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4440 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4441 pr_err("could not allocate migr_rec buffer\n");
4445 super
->clean_migration_record_by_mdmon
= 0;
4448 check_sum
= __gen_imsm_checksum(super
->anchor
);
4449 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4451 pr_err("IMSM checksum %x != %x on %s\n",
4453 __le32_to_cpu(super
->anchor
->check_sum
),
4461 /* read the extended mpb */
4462 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4464 pr_err("Cannot seek to extended mpb on %s: %s\n",
4465 devname
, strerror(errno
));
4469 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4470 super
->len
- sector_size
) != super
->len
- sector_size
) {
4472 pr_err("Cannot read extended mpb on %s: %s\n",
4473 devname
, strerror(errno
));
4477 check_sum
= __gen_imsm_checksum(super
->anchor
);
4478 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4480 pr_err("IMSM checksum %x != %x on %s\n",
4481 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4489 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4491 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4492 static void clear_hi(struct intel_super
*super
)
4494 struct imsm_super
*mpb
= super
->anchor
;
4496 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4498 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4499 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4500 disk
->total_blocks_hi
= 0;
4502 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4503 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4506 for (n
= 0; n
< 2; ++n
) {
4507 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4510 map
->pba_of_lba0_hi
= 0;
4511 map
->blocks_per_member_hi
= 0;
4512 map
->num_data_stripes_hi
= 0;
4518 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4522 err
= load_imsm_mpb(fd
, super
, devname
);
4525 if (super
->sector_size
== 4096)
4526 convert_from_4k(super
);
4527 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4530 err
= parse_raid_devices(super
);
4533 err
= load_bbm_log(super
);
4538 static void __free_imsm_disk(struct dl
*d
)
4550 static void free_imsm_disks(struct intel_super
*super
)
4554 while (super
->disks
) {
4556 super
->disks
= d
->next
;
4557 __free_imsm_disk(d
);
4559 while (super
->disk_mgmt_list
) {
4560 d
= super
->disk_mgmt_list
;
4561 super
->disk_mgmt_list
= d
->next
;
4562 __free_imsm_disk(d
);
4564 while (super
->missing
) {
4566 super
->missing
= d
->next
;
4567 __free_imsm_disk(d
);
4572 /* free all the pieces hanging off of a super pointer */
4573 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4575 struct intel_hba
*elem
, *next
;
4581 /* unlink capability description */
4583 if (super
->migr_rec_buf
) {
4584 free(super
->migr_rec_buf
);
4585 super
->migr_rec_buf
= NULL
;
4588 free_imsm_disks(super
);
4589 free_devlist(super
);
4593 free((void *)elem
->path
);
4599 free(super
->bbm_log
);
4603 static void free_imsm(struct intel_super
*super
)
4605 __free_imsm(super
, 1);
4606 free(super
->bb
.entries
);
4610 static void free_super_imsm(struct supertype
*st
)
4612 struct intel_super
*super
= st
->sb
;
4621 static struct intel_super
*alloc_super(void)
4623 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4625 super
->current_vol
= -1;
4626 super
->create_offset
= ~((unsigned long long) 0);
4628 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4629 sizeof(struct md_bb_entry
));
4630 if (!super
->bb
.entries
) {
4639 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4641 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4643 struct sys_dev
*hba_name
;
4646 if (fd
>= 0 && test_partition(fd
)) {
4647 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4651 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4656 hba_name
= find_disk_attached_hba(fd
, NULL
);
4659 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4663 rv
= attach_hba_to_super(super
, hba_name
);
4666 struct intel_hba
*hba
= super
->hba
;
4668 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4669 " but the container is assigned to Intel(R) %s %s (",
4671 get_sys_dev_type(hba_name
->type
),
4672 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4673 hba_name
->pci_id
? : "Err!",
4674 get_sys_dev_type(super
->hba
->type
),
4675 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4678 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4680 fprintf(stderr
, ", ");
4683 fprintf(stderr
, ").\n"
4684 " Mixing devices attached to different controllers is not allowed.\n");
4688 super
->orom
= find_imsm_capability(hba_name
);
4695 /* find_missing - helper routine for load_super_imsm_all that identifies
4696 * disks that have disappeared from the system. This routine relies on
4697 * the mpb being uptodate, which it is at load time.
4699 static int find_missing(struct intel_super
*super
)
4702 struct imsm_super
*mpb
= super
->anchor
;
4704 struct imsm_disk
*disk
;
4706 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4707 disk
= __get_imsm_disk(mpb
, i
);
4708 dl
= serial_to_dl(disk
->serial
, super
);
4712 dl
= xmalloc(sizeof(*dl
));
4716 dl
->devname
= xstrdup("missing");
4718 serialcpy(dl
->serial
, disk
->serial
);
4721 dl
->next
= super
->missing
;
4722 super
->missing
= dl
;
4728 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4730 struct intel_disk
*idisk
= disk_list
;
4733 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4735 idisk
= idisk
->next
;
4741 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4742 struct intel_super
*super
,
4743 struct intel_disk
**disk_list
)
4745 struct imsm_disk
*d
= &super
->disks
->disk
;
4746 struct imsm_super
*mpb
= super
->anchor
;
4749 for (i
= 0; i
< tbl_size
; i
++) {
4750 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4751 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4753 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4754 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4755 dprintf("mpb from %d:%d matches %d:%d\n",
4756 super
->disks
->major
,
4757 super
->disks
->minor
,
4758 table
[i
]->disks
->major
,
4759 table
[i
]->disks
->minor
);
4763 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4764 is_configured(d
) == is_configured(tbl_d
)) &&
4765 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4766 /* current version of the mpb is a
4767 * better candidate than the one in
4768 * super_table, but copy over "cross
4769 * generational" status
4771 struct intel_disk
*idisk
;
4773 dprintf("mpb from %d:%d replaces %d:%d\n",
4774 super
->disks
->major
,
4775 super
->disks
->minor
,
4776 table
[i
]->disks
->major
,
4777 table
[i
]->disks
->minor
);
4779 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4780 if (idisk
&& is_failed(&idisk
->disk
))
4781 tbl_d
->status
|= FAILED_DISK
;
4784 struct intel_disk
*idisk
;
4785 struct imsm_disk
*disk
;
4787 /* tbl_mpb is more up to date, but copy
4788 * over cross generational status before
4791 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4792 if (disk
&& is_failed(disk
))
4793 d
->status
|= FAILED_DISK
;
4795 idisk
= disk_list_get(d
->serial
, *disk_list
);
4798 if (disk
&& is_configured(disk
))
4799 idisk
->disk
.status
|= CONFIGURED_DISK
;
4802 dprintf("mpb from %d:%d prefer %d:%d\n",
4803 super
->disks
->major
,
4804 super
->disks
->minor
,
4805 table
[i
]->disks
->major
,
4806 table
[i
]->disks
->minor
);
4814 table
[tbl_size
++] = super
;
4818 /* update/extend the merged list of imsm_disk records */
4819 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4820 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4821 struct intel_disk
*idisk
;
4823 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4825 idisk
->disk
.status
|= disk
->status
;
4826 if (is_configured(&idisk
->disk
) ||
4827 is_failed(&idisk
->disk
))
4828 idisk
->disk
.status
&= ~(SPARE_DISK
);
4830 idisk
= xcalloc(1, sizeof(*idisk
));
4831 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4832 idisk
->disk
= *disk
;
4833 idisk
->next
= *disk_list
;
4837 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4844 static struct intel_super
*
4845 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4848 struct imsm_super
*mpb
= super
->anchor
;
4852 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4853 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4854 struct intel_disk
*idisk
;
4856 idisk
= disk_list_get(disk
->serial
, disk_list
);
4858 if (idisk
->owner
== owner
||
4859 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4862 dprintf("'%.16s' owner %d != %d\n",
4863 disk
->serial
, idisk
->owner
,
4866 dprintf("unknown disk %x [%d]: %.16s\n",
4867 __le32_to_cpu(mpb
->family_num
), i
,
4873 if (ok_count
== mpb
->num_disks
)
4878 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4880 struct intel_super
*s
;
4882 for (s
= super_list
; s
; s
= s
->next
) {
4883 if (family_num
!= s
->anchor
->family_num
)
4885 pr_err("Conflict, offlining family %#x on '%s'\n",
4886 __le32_to_cpu(family_num
), s
->disks
->devname
);
4890 static struct intel_super
*
4891 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4893 struct intel_super
*super_table
[len
];
4894 struct intel_disk
*disk_list
= NULL
;
4895 struct intel_super
*champion
, *spare
;
4896 struct intel_super
*s
, **del
;
4901 memset(super_table
, 0, sizeof(super_table
));
4902 for (s
= *super_list
; s
; s
= s
->next
)
4903 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4905 for (i
= 0; i
< tbl_size
; i
++) {
4906 struct imsm_disk
*d
;
4907 struct intel_disk
*idisk
;
4908 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4911 d
= &s
->disks
->disk
;
4913 /* 'd' must appear in merged disk list for its
4914 * configuration to be valid
4916 idisk
= disk_list_get(d
->serial
, disk_list
);
4917 if (idisk
&& idisk
->owner
== i
)
4918 s
= validate_members(s
, disk_list
, i
);
4923 dprintf("marking family: %#x from %d:%d offline\n",
4925 super_table
[i
]->disks
->major
,
4926 super_table
[i
]->disks
->minor
);
4930 /* This is where the mdadm implementation differs from the Windows
4931 * driver which has no strict concept of a container. We can only
4932 * assemble one family from a container, so when returning a prodigal
4933 * array member to this system the code will not be able to disambiguate
4934 * the container contents that should be assembled ("foreign" versus
4935 * "local"). It requires user intervention to set the orig_family_num
4936 * to a new value to establish a new container. The Windows driver in
4937 * this situation fixes up the volume name in place and manages the
4938 * foreign array as an independent entity.
4943 for (i
= 0; i
< tbl_size
; i
++) {
4944 struct intel_super
*tbl_ent
= super_table
[i
];
4950 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4955 if (s
&& !is_spare
) {
4956 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4958 } else if (!s
&& !is_spare
)
4971 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4972 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4974 /* collect all dl's onto 'champion', and update them to
4975 * champion's version of the status
4977 for (s
= *super_list
; s
; s
= s
->next
) {
4978 struct imsm_super
*mpb
= champion
->anchor
;
4979 struct dl
*dl
= s
->disks
;
4984 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4986 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4987 struct imsm_disk
*disk
;
4989 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4992 /* only set index on disks that are a member of
4993 * a populated contianer, i.e. one with
4996 if (is_failed(&dl
->disk
))
4998 else if (is_spare(&dl
->disk
))
5004 if (i
>= mpb
->num_disks
) {
5005 struct intel_disk
*idisk
;
5007 idisk
= disk_list_get(dl
->serial
, disk_list
);
5008 if (idisk
&& is_spare(&idisk
->disk
) &&
5009 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
5017 dl
->next
= champion
->disks
;
5018 champion
->disks
= dl
;
5022 /* delete 'champion' from super_list */
5023 for (del
= super_list
; *del
; ) {
5024 if (*del
== champion
) {
5025 *del
= (*del
)->next
;
5028 del
= &(*del
)->next
;
5030 champion
->next
= NULL
;
5034 struct intel_disk
*idisk
= disk_list
;
5036 disk_list
= disk_list
->next
;
5044 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
5045 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5046 int major
, int minor
, int keep_fd
);
5048 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5049 int *max
, int keep_fd
);
5051 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
5052 char *devname
, struct md_list
*devlist
,
5055 struct intel_super
*super_list
= NULL
;
5056 struct intel_super
*super
= NULL
;
5061 /* 'fd' is an opened container */
5062 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
5064 /* get super block from devlist devices */
5065 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
5068 /* all mpbs enter, maybe one leaves */
5069 super
= imsm_thunderdome(&super_list
, i
);
5075 if (find_missing(super
) != 0) {
5081 /* load migration record */
5082 err
= load_imsm_migr_rec(super
);
5084 /* migration is in progress,
5085 * but migr_rec cannot be loaded,
5091 /* Check migration compatibility */
5092 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5093 pr_err("Unsupported migration detected");
5095 fprintf(stderr
, " on %s\n", devname
);
5097 fprintf(stderr
, " (IMSM).\n");
5106 while (super_list
) {
5107 struct intel_super
*s
= super_list
;
5109 super_list
= super_list
->next
;
5118 strcpy(st
->container_devnm
, fd2devnm(fd
));
5120 st
->container_devnm
[0] = 0;
5121 if (err
== 0 && st
->ss
== NULL
) {
5122 st
->ss
= &super_imsm
;
5123 st
->minor_version
= 0;
5124 st
->max_devs
= IMSM_MAX_DEVICES
;
5130 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5131 int *max
, int keep_fd
)
5133 struct md_list
*tmpdev
;
5137 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5138 if (tmpdev
->used
!= 1)
5140 if (tmpdev
->container
== 1) {
5142 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5144 pr_err("cannot open device %s: %s\n",
5145 tmpdev
->devname
, strerror(errno
));
5149 err
= get_sra_super_block(fd
, super_list
,
5150 tmpdev
->devname
, &lmax
,
5159 int major
= major(tmpdev
->st_rdev
);
5160 int minor
= minor(tmpdev
->st_rdev
);
5161 err
= get_super_block(super_list
,
5178 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5179 int major
, int minor
, int keep_fd
)
5181 struct intel_super
*s
;
5193 sprintf(nm
, "%d:%d", major
, minor
);
5194 dfd
= dev_open(nm
, O_RDWR
);
5200 if (!get_dev_sector_size(dfd
, NULL
, &s
->sector_size
)) {
5204 find_intel_hba_capability(dfd
, s
, devname
);
5205 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5207 /* retry the load if we might have raced against mdmon */
5208 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5209 for (retry
= 0; retry
< 3; retry
++) {
5211 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5217 s
->next
= *super_list
;
5225 if (dfd
>= 0 && !keep_fd
)
5232 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5239 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5243 if (sra
->array
.major_version
!= -1 ||
5244 sra
->array
.minor_version
!= -2 ||
5245 strcmp(sra
->text_version
, "imsm") != 0) {
5250 devnm
= fd2devnm(fd
);
5251 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5252 if (get_super_block(super_list
, devnm
, devname
,
5253 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5264 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5266 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5269 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5271 struct intel_super
*super
;
5275 if (test_partition(fd
))
5276 /* IMSM not allowed on partitions */
5279 free_super_imsm(st
);
5281 super
= alloc_super();
5282 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
))
5286 /* Load hba and capabilities if they exist.
5287 * But do not preclude loading metadata in case capabilities or hba are
5288 * non-compliant and ignore_hw_compat is set.
5290 rv
= find_intel_hba_capability(fd
, super
, devname
);
5291 /* no orom/efi or non-intel hba of the disk */
5292 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5294 pr_err("No OROM/EFI properties for %s\n", devname
);
5298 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5300 /* retry the load if we might have raced against mdmon */
5302 struct mdstat_ent
*mdstat
= NULL
;
5303 char *name
= fd2kname(fd
);
5306 mdstat
= mdstat_by_component(name
);
5308 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5309 for (retry
= 0; retry
< 3; retry
++) {
5311 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5317 free_mdstat(mdstat
);
5322 pr_err("Failed to load all information sections on %s\n", devname
);
5328 if (st
->ss
== NULL
) {
5329 st
->ss
= &super_imsm
;
5330 st
->minor_version
= 0;
5331 st
->max_devs
= IMSM_MAX_DEVICES
;
5334 /* load migration record */
5335 if (load_imsm_migr_rec(super
) == 0) {
5336 /* Check for unsupported migration features */
5337 if (check_mpb_migr_compatibility(super
) != 0) {
5338 pr_err("Unsupported migration detected");
5340 fprintf(stderr
, " on %s\n", devname
);
5342 fprintf(stderr
, " (IMSM).\n");
5350 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5352 if (info
->level
== 1)
5354 return info
->chunk_size
>> 9;
5357 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5358 unsigned long long size
)
5360 if (info
->level
== 1)
5363 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5366 static void imsm_update_version_info(struct intel_super
*super
)
5368 /* update the version and attributes */
5369 struct imsm_super
*mpb
= super
->anchor
;
5371 struct imsm_dev
*dev
;
5372 struct imsm_map
*map
;
5375 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5376 dev
= get_imsm_dev(super
, i
);
5377 map
= get_imsm_map(dev
, MAP_0
);
5378 if (__le32_to_cpu(dev
->size_high
) > 0)
5379 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5381 /* FIXME detect when an array spans a port multiplier */
5383 mpb
->attributes
|= MPB_ATTRIB_PM
;
5386 if (mpb
->num_raid_devs
> 1 ||
5387 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5388 version
= MPB_VERSION_ATTRIBS
;
5389 switch (get_imsm_raid_level(map
)) {
5390 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5391 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5392 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5393 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5396 if (map
->num_members
>= 5)
5397 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5398 else if (dev
->status
== DEV_CLONE_N_GO
)
5399 version
= MPB_VERSION_CNG
;
5400 else if (get_imsm_raid_level(map
) == 5)
5401 version
= MPB_VERSION_RAID5
;
5402 else if (map
->num_members
>= 3)
5403 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5404 else if (get_imsm_raid_level(map
) == 1)
5405 version
= MPB_VERSION_RAID1
;
5407 version
= MPB_VERSION_RAID0
;
5409 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5413 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5415 struct imsm_super
*mpb
= super
->anchor
;
5416 char *reason
= NULL
;
5418 size_t len
= strlen(name
);
5422 while (isspace(start
[len
- 1]))
5424 while (*start
&& isspace(*start
))
5426 memmove(name
, start
, len
+ 1);
5429 if (len
> MAX_RAID_SERIAL_LEN
)
5430 reason
= "must be 16 characters or less";
5432 reason
= "must be a non-empty string";
5434 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5435 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5437 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5438 reason
= "already exists";
5443 if (reason
&& !quiet
)
5444 pr_err("imsm volume name %s\n", reason
);
5449 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5450 struct shape
*s
, char *name
,
5451 char *homehost
, int *uuid
,
5452 long long data_offset
)
5454 /* We are creating a volume inside a pre-existing container.
5455 * so st->sb is already set.
5457 struct intel_super
*super
= st
->sb
;
5458 unsigned int sector_size
= super
->sector_size
;
5459 struct imsm_super
*mpb
= super
->anchor
;
5460 struct intel_dev
*dv
;
5461 struct imsm_dev
*dev
;
5462 struct imsm_vol
*vol
;
5463 struct imsm_map
*map
;
5464 int idx
= mpb
->num_raid_devs
;
5467 unsigned long long array_blocks
;
5468 size_t size_old
, size_new
;
5469 unsigned long long num_data_stripes
;
5470 unsigned int data_disks
;
5471 unsigned long long size_per_member
;
5473 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5474 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5478 /* ensure the mpb is large enough for the new data */
5479 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5480 size_new
= disks_to_mpb_size(info
->nr_disks
);
5481 if (size_new
> size_old
) {
5483 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5485 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5486 pr_err("could not allocate new mpb\n");
5489 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5490 MIGR_REC_BUF_SECTORS
*
5491 MAX_SECTOR_SIZE
) != 0) {
5492 pr_err("could not allocate migr_rec buffer\n");
5498 memcpy(mpb_new
, mpb
, size_old
);
5501 super
->anchor
= mpb_new
;
5502 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5503 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5504 super
->len
= size_round
;
5506 super
->current_vol
= idx
;
5508 /* handle 'failed_disks' by either:
5509 * a) create dummy disk entries in the table if this the first
5510 * volume in the array. We add them here as this is the only
5511 * opportunity to add them. add_to_super_imsm_volume()
5512 * handles the non-failed disks and continues incrementing
5514 * b) validate that 'failed_disks' matches the current number
5515 * of missing disks if the container is populated
5517 if (super
->current_vol
== 0) {
5519 for (i
= 0; i
< info
->failed_disks
; i
++) {
5520 struct imsm_disk
*disk
;
5523 disk
= __get_imsm_disk(mpb
, i
);
5524 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5525 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5526 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5527 "missing:%d", (__u8
)i
);
5529 find_missing(super
);
5534 for (d
= super
->missing
; d
; d
= d
->next
)
5536 if (info
->failed_disks
> missing
) {
5537 pr_err("unable to add 'missing' disk to container\n");
5542 if (!check_name(super
, name
, 0))
5544 dv
= xmalloc(sizeof(*dv
));
5545 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5547 * Explicitly allow truncating to not confuse gcc's
5548 * -Werror=stringop-truncation
5550 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5551 memcpy(dev
->volume
, name
, namelen
);
5552 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5553 info
->layout
, info
->chunk_size
,
5554 s
->size
* BLOCKS_PER_KB
);
5555 data_disks
= get_data_disks(info
->level
, info
->layout
,
5557 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5558 size_per_member
= array_blocks
/ data_disks
;
5560 set_imsm_dev_size(dev
, array_blocks
);
5561 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5563 vol
->migr_state
= 0;
5564 set_migr_type(dev
, MIGR_INIT
);
5565 vol
->dirty
= !info
->state
;
5566 set_vol_curr_migr_unit(dev
, 0);
5567 map
= get_imsm_map(dev
, MAP_0
);
5568 set_pba_of_lba0(map
, super
->create_offset
);
5569 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5570 map
->failed_disk_num
= ~0;
5571 if (info
->level
> 0)
5572 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5573 : IMSM_T_STATE_UNINITIALIZED
);
5575 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5576 IMSM_T_STATE_NORMAL
;
5579 if (info
->level
== 1 && info
->raid_disks
> 2) {
5582 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5586 map
->raid_level
= info
->level
;
5587 if (info
->level
== 10) {
5588 map
->raid_level
= 1;
5589 map
->num_domains
= info
->raid_disks
/ 2;
5590 } else if (info
->level
== 1)
5591 map
->num_domains
= info
->raid_disks
;
5593 map
->num_domains
= 1;
5595 /* info->size is only int so use the 'size' parameter instead */
5596 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5597 num_data_stripes
/= map
->num_domains
;
5598 set_num_data_stripes(map
, num_data_stripes
);
5600 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5601 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5605 map
->num_members
= info
->raid_disks
;
5606 for (i
= 0; i
< map
->num_members
; i
++) {
5607 /* initialized in add_to_super */
5608 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5610 mpb
->num_raid_devs
++;
5611 mpb
->num_raid_devs_created
++;
5612 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5614 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5615 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5616 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5617 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5621 pr_err("imsm does not support consistency policy %s\n",
5622 map_num(consistency_policies
, s
->consistency_policy
));
5627 dv
->index
= super
->current_vol
;
5628 dv
->next
= super
->devlist
;
5629 super
->devlist
= dv
;
5631 imsm_update_version_info(super
);
5636 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5637 struct shape
*s
, char *name
,
5638 char *homehost
, int *uuid
,
5639 unsigned long long data_offset
)
5641 /* This is primarily called by Create when creating a new array.
5642 * We will then get add_to_super called for each component, and then
5643 * write_init_super called to write it out to each device.
5644 * For IMSM, Create can create on fresh devices or on a pre-existing
5646 * To create on a pre-existing array a different method will be called.
5647 * This one is just for fresh drives.
5649 struct intel_super
*super
;
5650 struct imsm_super
*mpb
;
5654 if (data_offset
!= INVALID_SECTORS
) {
5655 pr_err("data-offset not supported by imsm\n");
5660 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5664 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5666 mpb_size
= MAX_SECTOR_SIZE
;
5668 super
= alloc_super();
5670 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5675 pr_err("could not allocate superblock\n");
5678 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5679 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5680 pr_err("could not allocate migr_rec buffer\n");
5685 memset(super
->buf
, 0, mpb_size
);
5687 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5691 /* zeroing superblock */
5695 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5697 version
= (char *) mpb
->sig
;
5698 strcpy(version
, MPB_SIGNATURE
);
5699 version
+= strlen(MPB_SIGNATURE
);
5700 strcpy(version
, MPB_VERSION_RAID0
);
5705 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5707 unsigned int member_sector_size
;
5710 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5714 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5716 if (member_sector_size
!= super
->sector_size
)
5721 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5722 int fd
, char *devname
)
5724 struct intel_super
*super
= st
->sb
;
5725 struct imsm_super
*mpb
= super
->anchor
;
5726 struct imsm_disk
*_disk
;
5727 struct imsm_dev
*dev
;
5728 struct imsm_map
*map
;
5732 dev
= get_imsm_dev(super
, super
->current_vol
);
5733 map
= get_imsm_map(dev
, MAP_0
);
5735 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5736 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5742 /* we're doing autolayout so grab the pre-marked (in
5743 * validate_geometry) raid_disk
5745 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5746 if (dl
->raiddisk
== dk
->raid_disk
)
5749 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5750 if (dl
->major
== dk
->major
&&
5751 dl
->minor
== dk
->minor
)
5756 pr_err("%s is not a member of the same container\n", devname
);
5760 if (mpb
->num_disks
== 0)
5761 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5762 &super
->sector_size
))
5765 if (!drive_validate_sector_size(super
, dl
)) {
5766 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5770 /* add a pristine spare to the metadata */
5771 if (dl
->index
< 0) {
5772 dl
->index
= super
->anchor
->num_disks
;
5773 super
->anchor
->num_disks
++;
5775 /* Check the device has not already been added */
5776 slot
= get_imsm_disk_slot(map
, dl
->index
);
5778 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5779 pr_err("%s has been included in this array twice\n",
5783 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5784 dl
->disk
.status
= CONFIGURED_DISK
;
5786 /* update size of 'missing' disks to be at least as large as the
5787 * largest acitve member (we only have dummy missing disks when
5788 * creating the first volume)
5790 if (super
->current_vol
== 0) {
5791 for (df
= super
->missing
; df
; df
= df
->next
) {
5792 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5793 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5794 _disk
= __get_imsm_disk(mpb
, df
->index
);
5799 /* refresh unset/failed slots to point to valid 'missing' entries */
5800 for (df
= super
->missing
; df
; df
= df
->next
)
5801 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5802 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5804 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5806 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5807 if (is_gen_migration(dev
)) {
5808 struct imsm_map
*map2
= get_imsm_map(dev
,
5810 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5811 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5812 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5815 if ((unsigned)df
->index
==
5817 set_imsm_ord_tbl_ent(map2
,
5823 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5827 /* if we are creating the first raid device update the family number */
5828 if (super
->current_vol
== 0) {
5830 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5832 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5833 if (!_dev
|| !_disk
) {
5834 pr_err("BUG mpb setup error\n");
5840 sum
+= __gen_imsm_checksum(mpb
);
5841 mpb
->family_num
= __cpu_to_le32(sum
);
5842 mpb
->orig_family_num
= mpb
->family_num
;
5843 mpb
->creation_time
= __cpu_to_le64((__u64
)time(NULL
));
5845 super
->current_disk
= dl
;
5850 * Function marks disk as spare and restores disk serial
5851 * in case it was previously marked as failed by takeover operation
5853 * -1 : critical error
5854 * 0 : disk is marked as spare but serial is not set
5857 int mark_spare(struct dl
*disk
)
5859 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5866 if (!imsm_read_serial(disk
->fd
, NULL
, serial
, MAX_RAID_SERIAL_LEN
)) {
5867 /* Restore disk serial number, because takeover marks disk
5868 * as failed and adds to serial ':0' before it becomes
5871 serialcpy(disk
->serial
, serial
);
5872 serialcpy(disk
->disk
.serial
, serial
);
5875 disk
->disk
.status
= SPARE_DISK
;
5882 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
);
5884 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5885 int fd
, char *devname
,
5886 unsigned long long data_offset
)
5888 struct intel_super
*super
= st
->sb
;
5890 unsigned long long size
;
5891 unsigned int member_sector_size
;
5896 /* If we are on an RAID enabled platform check that the disk is
5897 * attached to the raid controller.
5898 * We do not need to test disks attachment for container based additions,
5899 * they shall be already tested when container was created/assembled.
5901 rv
= find_intel_hba_capability(fd
, super
, devname
);
5902 /* no orom/efi or non-intel hba of the disk */
5904 dprintf("capability: %p fd: %d ret: %d\n",
5905 super
->orom
, fd
, rv
);
5909 if (super
->current_vol
>= 0)
5910 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5913 dd
= xcalloc(sizeof(*dd
), 1);
5914 dd
->major
= major(stb
.st_rdev
);
5915 dd
->minor
= minor(stb
.st_rdev
);
5916 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5919 dd
->action
= DISK_ADD
;
5920 rv
= imsm_read_serial(fd
, devname
, dd
->serial
, MAX_RAID_SERIAL_LEN
);
5922 pr_err("failed to retrieve scsi serial, aborting\n");
5929 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5930 (super
->hba
->type
== SYS_DEV_VMD
))) {
5932 char cntrl_path
[PATH_MAX
];
5934 char pci_dev_path
[PATH_MAX
];
5936 if (!diskfd_to_devpath(fd
, 2, pci_dev_path
) ||
5937 !diskfd_to_devpath(fd
, 1, cntrl_path
)) {
5938 pr_err("failed to get dev paths, aborting\n");
5946 cntrl_name
= basename(cntrl_path
);
5947 if (is_multipath_nvme(fd
))
5948 pr_err("%s controller supports Multi-Path I/O, Intel (R) VROC does not support multipathing\n",
5951 if (devpath_to_vendor(pci_dev_path
) == 0x8086) {
5953 * If Intel's NVMe drive has serial ended with
5954 * "-A","-B","-1" or "-2" it means that this is "x8"
5955 * device (double drive on single PCIe card).
5956 * User should be warned about potential data loss.
5958 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5959 /* Skip empty character at the end */
5960 if (dd
->serial
[i
] == 0)
5963 if (((dd
->serial
[i
] == 'A') ||
5964 (dd
->serial
[i
] == 'B') ||
5965 (dd
->serial
[i
] == '1') ||
5966 (dd
->serial
[i
] == '2')) &&
5967 (dd
->serial
[i
-1] == '-'))
5968 pr_err("\tThe action you are about to take may put your data at risk.\n"
5969 "\tPlease note that x8 devices may consist of two separate x4 devices "
5970 "located on a single PCIe port.\n"
5971 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5974 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5975 !imsm_orom_has_tpv_support(super
->orom
)) {
5976 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5977 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
5984 get_dev_size(fd
, NULL
, &size
);
5985 if (!get_dev_sector_size(fd
, NULL
, &member_sector_size
))
5988 if (super
->sector_size
== 0) {
5989 /* this a first device, so sector_size is not set yet */
5990 super
->sector_size
= member_sector_size
;
5993 /* clear migr_rec when adding disk to container */
5994 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5995 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5997 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5998 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5999 MIGR_REC_BUF_SECTORS
*member_sector_size
)
6000 perror("Write migr_rec failed");
6004 serialcpy(dd
->disk
.serial
, dd
->serial
);
6005 set_total_blocks(&dd
->disk
, size
);
6006 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
6007 struct imsm_super
*mpb
= super
->anchor
;
6008 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6011 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
6012 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
6014 dd
->disk
.scsi_id
= __cpu_to_le32(0);
6016 if (st
->update_tail
) {
6017 dd
->next
= super
->disk_mgmt_list
;
6018 super
->disk_mgmt_list
= dd
;
6020 /* this is called outside of mdmon
6021 * write initial spare metadata
6022 * mdmon will overwrite it.
6024 dd
->next
= super
->disks
;
6026 write_super_imsm_spare(super
, dd
);
6032 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
6034 struct intel_super
*super
= st
->sb
;
6037 /* remove from super works only in mdmon - for communication
6038 * manager - monitor. Check if communication memory buffer
6041 if (!st
->update_tail
) {
6042 pr_err("shall be used in mdmon context only\n");
6045 dd
= xcalloc(1, sizeof(*dd
));
6046 dd
->major
= dk
->major
;
6047 dd
->minor
= dk
->minor
;
6050 dd
->action
= DISK_REMOVE
;
6052 dd
->next
= super
->disk_mgmt_list
;
6053 super
->disk_mgmt_list
= dd
;
6058 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
6061 char buf
[MAX_SECTOR_SIZE
];
6062 struct imsm_super anchor
;
6063 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
6066 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
)
6068 struct imsm_super
*mpb
= super
->anchor
;
6069 struct imsm_super
*spare
= &spare_record
.anchor
;
6075 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
6076 spare
->generation_num
= __cpu_to_le32(1UL);
6077 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
6078 spare
->num_disks
= 1;
6079 spare
->num_raid_devs
= 0;
6080 spare
->cache_size
= mpb
->cache_size
;
6081 spare
->pwr_cycle_count
= __cpu_to_le32(1);
6083 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
6084 MPB_SIGNATURE MPB_VERSION_RAID0
);
6086 spare
->disk
[0] = d
->disk
;
6087 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
6088 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6090 if (super
->sector_size
== 4096)
6091 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6093 sum
= __gen_imsm_checksum(spare
);
6094 spare
->family_num
= __cpu_to_le32(sum
);
6095 spare
->orig_family_num
= 0;
6096 sum
= __gen_imsm_checksum(spare
);
6097 spare
->check_sum
= __cpu_to_le32(sum
);
6099 if (store_imsm_mpb(d
->fd
, spare
)) {
6100 pr_err("failed for device %d:%d %s\n",
6101 d
->major
, d
->minor
, strerror(errno
));
6107 /* spare records have their own family number and do not have any defined raid
6110 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
6114 for (d
= super
->disks
; d
; d
= d
->next
) {
6118 if (write_super_imsm_spare(super
, d
))
6130 static int write_super_imsm(struct supertype
*st
, int doclose
)
6132 struct intel_super
*super
= st
->sb
;
6133 unsigned int sector_size
= super
->sector_size
;
6134 struct imsm_super
*mpb
= super
->anchor
;
6140 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6142 int clear_migration_record
= 1;
6145 /* 'generation' is incremented everytime the metadata is written */
6146 generation
= __le32_to_cpu(mpb
->generation_num
);
6148 mpb
->generation_num
= __cpu_to_le32(generation
);
6150 /* fix up cases where previous mdadm releases failed to set
6153 if (mpb
->orig_family_num
== 0)
6154 mpb
->orig_family_num
= mpb
->family_num
;
6156 for (d
= super
->disks
; d
; d
= d
->next
) {
6160 mpb
->disk
[d
->index
] = d
->disk
;
6164 for (d
= super
->missing
; d
; d
= d
->next
) {
6165 mpb
->disk
[d
->index
] = d
->disk
;
6168 mpb
->num_disks
= num_disks
;
6169 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6171 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6172 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6173 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6175 imsm_copy_dev(dev
, dev2
);
6176 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6178 if (is_gen_migration(dev2
))
6179 clear_migration_record
= 0;
6182 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6185 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6186 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6188 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6190 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6191 mpb_size
+= bbm_log_size
;
6192 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6195 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6198 /* recalculate checksum */
6199 sum
= __gen_imsm_checksum(mpb
);
6200 mpb
->check_sum
= __cpu_to_le32(sum
);
6202 if (super
->clean_migration_record_by_mdmon
) {
6203 clear_migration_record
= 1;
6204 super
->clean_migration_record_by_mdmon
= 0;
6206 if (clear_migration_record
)
6207 memset(super
->migr_rec_buf
, 0,
6208 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6210 if (sector_size
== 4096)
6211 convert_to_4k(super
);
6213 /* write the mpb for disks that compose raid devices */
6214 for (d
= super
->disks
; d
; d
= d
->next
) {
6215 if (d
->index
< 0 || is_failed(&d
->disk
))
6218 if (clear_migration_record
) {
6219 unsigned long long dsize
;
6221 get_dev_size(d
->fd
, NULL
, &dsize
);
6222 if (lseek64(d
->fd
, dsize
- sector_size
,
6224 if ((unsigned int)write(d
->fd
,
6225 super
->migr_rec_buf
,
6226 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6227 MIGR_REC_BUF_SECTORS
*sector_size
)
6228 perror("Write migr_rec failed");
6232 if (store_imsm_mpb(d
->fd
, mpb
))
6234 "failed for device %d:%d (fd: %d)%s\n",
6236 d
->fd
, strerror(errno
));
6245 return write_super_imsm_spares(super
, doclose
);
6250 static int create_array(struct supertype
*st
, int dev_idx
)
6253 struct imsm_update_create_array
*u
;
6254 struct intel_super
*super
= st
->sb
;
6255 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6256 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6257 struct disk_info
*inf
;
6258 struct imsm_disk
*disk
;
6261 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6262 sizeof(*inf
) * map
->num_members
;
6264 u
->type
= update_create_array
;
6265 u
->dev_idx
= dev_idx
;
6266 imsm_copy_dev(&u
->dev
, dev
);
6267 inf
= get_disk_info(u
);
6268 for (i
= 0; i
< map
->num_members
; i
++) {
6269 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6271 disk
= get_imsm_disk(super
, idx
);
6273 disk
= get_imsm_missing(super
, idx
);
6274 serialcpy(inf
[i
].serial
, disk
->serial
);
6276 append_metadata_update(st
, u
, len
);
6281 static int mgmt_disk(struct supertype
*st
)
6283 struct intel_super
*super
= st
->sb
;
6285 struct imsm_update_add_remove_disk
*u
;
6287 if (!super
->disk_mgmt_list
)
6292 u
->type
= update_add_remove_disk
;
6293 append_metadata_update(st
, u
, len
);
6298 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6300 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6302 struct ppl_header
*ppl_hdr
= buf
;
6305 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6307 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6309 perror("Failed to seek to PPL header location");
6313 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6315 perror("Write PPL header failed");
6324 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6326 struct intel_super
*super
= st
->sb
;
6328 struct ppl_header
*ppl_hdr
;
6331 /* first clear entire ppl space */
6332 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6336 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6338 pr_err("Failed to allocate PPL header buffer\n");
6342 memset(buf
, 0, PPL_HEADER_SIZE
);
6344 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6345 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6347 if (info
->mismatch_cnt
) {
6349 * We are overwriting an invalid ppl. Make one entry with wrong
6350 * checksum to prevent the kernel from skipping resync.
6352 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6353 ppl_hdr
->entries
[0].checksum
= ~0;
6356 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6362 static int is_rebuilding(struct imsm_dev
*dev
);
6364 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6365 struct mdinfo
*disk
)
6367 struct intel_super
*super
= st
->sb
;
6369 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6371 struct ppl_header
*ppl_hdr
= NULL
;
6373 struct imsm_dev
*dev
;
6376 unsigned long long ppl_offset
= 0;
6377 unsigned long long prev_gen_num
= 0;
6379 if (disk
->disk
.raid_disk
< 0)
6382 dev
= get_imsm_dev(super
, info
->container_member
);
6383 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6384 d
= get_imsm_dl_disk(super
, idx
);
6386 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6389 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6390 pr_err("Failed to allocate PPL header buffer\n");
6396 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6399 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6401 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6403 perror("Failed to seek to PPL header location");
6408 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6409 perror("Read PPL header failed");
6416 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6417 ppl_hdr
->checksum
= 0;
6419 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6420 dprintf("Wrong PPL header checksum on %s\n",
6425 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6426 /* previous was newest, it was already checked */
6430 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6431 super
->anchor
->orig_family_num
)) {
6432 dprintf("Wrong PPL header signature on %s\n",
6439 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6441 ppl_offset
+= PPL_HEADER_SIZE
;
6442 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6444 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6447 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6459 * Update metadata to use mutliple PPLs area (1MB).
6460 * This is done once for all RAID members
6462 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6463 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6465 struct mdinfo
*member_dev
;
6467 sprintf(subarray
, "%d", info
->container_member
);
6469 if (mdmon_running(st
->container_devnm
))
6470 st
->update_tail
= &st
->updates
;
6472 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6473 pr_err("Failed to update subarray %s\n",
6476 if (st
->update_tail
)
6477 flush_metadata_updates(st
);
6479 st
->ss
->sync_metadata(st
);
6480 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6481 for (member_dev
= info
->devs
; member_dev
;
6482 member_dev
= member_dev
->next
)
6483 member_dev
->ppl_size
=
6484 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6489 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6491 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6492 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6493 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6494 (is_rebuilding(dev
) &&
6495 vol_curr_migr_unit(dev
) == 0 &&
6496 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6497 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6499 info
->mismatch_cnt
++;
6500 } else if (ret
== 0 &&
6501 ppl_hdr
->entries_count
== 0 &&
6502 is_rebuilding(dev
) &&
6503 info
->resync_start
== 0) {
6505 * The header has no entries - add a single empty entry and
6506 * rewrite the header to prevent the kernel from going into
6507 * resync after an interrupted rebuild.
6509 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6510 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6518 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6520 struct intel_super
*super
= st
->sb
;
6524 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6525 info
->array
.level
!= 5)
6528 for (d
= super
->disks
; d
; d
= d
->next
) {
6529 if (d
->index
< 0 || is_failed(&d
->disk
))
6532 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6540 /*******************************************************************************
6541 * Function: write_init_bitmap_imsm_vol
6542 * Description: Write a bitmap header and prepares the area for the bitmap.
6544 * st : supertype information
6545 * vol_idx : the volume index to use
6550 ******************************************************************************/
6551 static int write_init_bitmap_imsm_vol(struct supertype
*st
, int vol_idx
)
6553 struct intel_super
*super
= st
->sb
;
6554 int prev_current_vol
= super
->current_vol
;
6558 super
->current_vol
= vol_idx
;
6559 for (d
= super
->disks
; d
; d
= d
->next
) {
6560 if (d
->index
< 0 || is_failed(&d
->disk
))
6562 ret
= st
->ss
->write_bitmap(st
, d
->fd
, NoUpdate
);
6566 super
->current_vol
= prev_current_vol
;
6570 /*******************************************************************************
6571 * Function: write_init_bitmap_imsm_all
6572 * Description: Write a bitmap header and prepares the area for the bitmap.
6573 * Operation is executed for volumes with CONSISTENCY_POLICY_BITMAP.
6575 * st : supertype information
6576 * info : info about the volume where the bitmap should be written
6577 * vol_idx : the volume index to use
6582 ******************************************************************************/
6583 static int write_init_bitmap_imsm_all(struct supertype
*st
, struct mdinfo
*info
,
6588 if (info
&& (info
->consistency_policy
== CONSISTENCY_POLICY_BITMAP
))
6589 ret
= write_init_bitmap_imsm_vol(st
, vol_idx
);
6594 static int write_init_super_imsm(struct supertype
*st
)
6596 struct intel_super
*super
= st
->sb
;
6597 int current_vol
= super
->current_vol
;
6601 getinfo_super_imsm(st
, &info
, NULL
);
6603 /* we are done with current_vol reset it to point st at the container */
6604 super
->current_vol
= -1;
6606 if (st
->update_tail
) {
6607 /* queue the recently created array / added disk
6608 * as a metadata update */
6610 /* determine if we are creating a volume or adding a disk */
6611 if (current_vol
< 0) {
6612 /* in the mgmt (add/remove) disk case we are running
6613 * in mdmon context, so don't close fd's
6617 /* adding the second volume to the array */
6618 rv
= write_init_ppl_imsm_all(st
, &info
);
6620 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6622 rv
= create_array(st
, current_vol
);
6626 for (d
= super
->disks
; d
; d
= d
->next
)
6627 Kill(d
->devname
, NULL
, 0, -1, 1);
6628 if (current_vol
>= 0) {
6629 rv
= write_init_ppl_imsm_all(st
, &info
);
6631 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6635 rv
= write_super_imsm(st
, 1);
6641 static int store_super_imsm(struct supertype
*st
, int fd
)
6643 struct intel_super
*super
= st
->sb
;
6644 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6649 if (super
->sector_size
== 4096)
6650 convert_to_4k(super
);
6651 return store_imsm_mpb(fd
, mpb
);
6654 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6655 int layout
, int raiddisks
, int chunk
,
6656 unsigned long long size
,
6657 unsigned long long data_offset
,
6659 unsigned long long *freesize
,
6663 unsigned long long ldsize
;
6664 struct intel_super
*super
= NULL
;
6667 if (level
!= LEVEL_CONTAINER
)
6672 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6675 pr_err("imsm: Cannot open %s: %s\n",
6676 dev
, strerror(errno
));
6679 if (!get_dev_size(fd
, dev
, &ldsize
))
6682 /* capabilities retrieve could be possible
6683 * note that there is no fd for the disks in array.
6685 super
= alloc_super();
6689 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
))
6692 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6696 fd2devname(fd
, str
);
6697 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6698 fd
, str
, super
->orom
, rv
, raiddisks
);
6700 /* no orom/efi or non-intel hba of the disk */
6705 if (raiddisks
> super
->orom
->tds
) {
6707 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6708 raiddisks
, super
->orom
->tds
);
6711 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6712 (ldsize
>> 9) >> 32 > 0) {
6714 pr_err("%s exceeds maximum platform supported size\n", dev
);
6718 if (super
->hba
->type
== SYS_DEV_VMD
||
6719 super
->hba
->type
== SYS_DEV_NVME
) {
6720 if (!imsm_is_nvme_namespace_supported(fd
, 1)) {
6722 pr_err("NVMe namespace %s is not supported by IMSM\n",
6729 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6739 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6741 const unsigned long long base_start
= e
[*idx
].start
;
6742 unsigned long long end
= base_start
+ e
[*idx
].size
;
6745 if (base_start
== end
)
6749 for (i
= *idx
; i
< num_extents
; i
++) {
6750 /* extend overlapping extents */
6751 if (e
[i
].start
>= base_start
&&
6752 e
[i
].start
<= end
) {
6755 if (e
[i
].start
+ e
[i
].size
> end
)
6756 end
= e
[i
].start
+ e
[i
].size
;
6757 } else if (e
[i
].start
> end
) {
6763 return end
- base_start
;
6766 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6768 /* build a composite disk with all known extents and generate a new
6769 * 'maxsize' given the "all disks in an array must share a common start
6770 * offset" constraint
6772 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6776 unsigned long long pos
;
6777 unsigned long long start
= 0;
6778 unsigned long long maxsize
;
6779 unsigned long reserve
;
6781 /* coalesce and sort all extents. also, check to see if we need to
6782 * reserve space between member arrays
6785 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6788 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6791 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6796 while (i
< sum_extents
) {
6797 e
[j
].start
= e
[i
].start
;
6798 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6800 if (e
[j
-1].size
== 0)
6809 unsigned long long esize
;
6811 esize
= e
[i
].start
- pos
;
6812 if (esize
>= maxsize
) {
6817 pos
= e
[i
].start
+ e
[i
].size
;
6819 } while (e
[i
-1].size
);
6825 /* FIXME assumes volume at offset 0 is the first volume in a
6828 if (start_extent
> 0)
6829 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6833 if (maxsize
< reserve
)
6836 super
->create_offset
= ~((unsigned long long) 0);
6837 if (start
+ reserve
> super
->create_offset
)
6838 return 0; /* start overflows create_offset */
6839 super
->create_offset
= start
+ reserve
;
6841 return maxsize
- reserve
;
6844 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6846 if (level
< 0 || level
== 6 || level
== 4)
6849 /* if we have an orom prevent invalid raid levels */
6852 case 0: return imsm_orom_has_raid0(orom
);
6855 return imsm_orom_has_raid1e(orom
);
6856 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6857 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6858 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6861 return 1; /* not on an Intel RAID platform so anything goes */
6867 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6868 int dpa
, int verbose
)
6870 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6871 struct mdstat_ent
*memb
;
6877 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6878 if (memb
->metadata_version
&&
6879 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6880 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6881 !is_subarray(memb
->metadata_version
+9) &&
6883 struct dev_member
*dev
= memb
->members
;
6885 while(dev
&& (fd
< 0)) {
6886 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6887 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6889 fd
= open(path
, O_RDONLY
, 0);
6890 if (num
<= 0 || fd
< 0) {
6891 pr_vrb("Cannot open %s: %s\n",
6892 dev
->name
, strerror(errno
));
6898 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6899 struct mdstat_ent
*vol
;
6900 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6901 if (vol
->active
> 0 &&
6902 vol
->metadata_version
&&
6903 is_container_member(vol
, memb
->devnm
)) {
6908 if (*devlist
&& (found
< dpa
)) {
6909 dv
= xcalloc(1, sizeof(*dv
));
6910 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6911 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6914 dv
->next
= *devlist
;
6922 free_mdstat(mdstat
);
6927 static struct md_list
*
6928 get_loop_devices(void)
6931 struct md_list
*devlist
= NULL
;
6934 for(i
= 0; i
< 12; i
++) {
6935 dv
= xcalloc(1, sizeof(*dv
));
6936 dv
->devname
= xmalloc(40);
6937 sprintf(dv
->devname
, "/dev/loop%d", i
);
6945 static struct md_list
*
6946 get_devices(const char *hba_path
)
6948 struct md_list
*devlist
= NULL
;
6955 devlist
= get_loop_devices();
6958 /* scroll through /sys/dev/block looking for devices attached to
6961 dir
= opendir("/sys/dev/block");
6962 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6967 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6969 path
= devt_to_devpath(makedev(major
, minor
), 1, NULL
);
6972 if (!path_attached_to_hba(path
, hba_path
)) {
6979 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6981 fd2devname(fd
, buf
);
6984 pr_err("cannot open device: %s\n",
6989 dv
= xcalloc(1, sizeof(*dv
));
6990 dv
->devname
= xstrdup(buf
);
6997 devlist
= devlist
->next
;
7007 count_volumes_list(struct md_list
*devlist
, char *homehost
,
7008 int verbose
, int *found
)
7010 struct md_list
*tmpdev
;
7012 struct supertype
*st
;
7014 /* first walk the list of devices to find a consistent set
7015 * that match the criterea, if that is possible.
7016 * We flag the ones we like with 'used'.
7019 st
= match_metadata_desc_imsm("imsm");
7021 pr_vrb("cannot allocate memory for imsm supertype\n");
7025 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7026 char *devname
= tmpdev
->devname
;
7028 struct supertype
*tst
;
7030 if (tmpdev
->used
> 1)
7032 tst
= dup_super(st
);
7034 pr_vrb("cannot allocate memory for imsm supertype\n");
7037 tmpdev
->container
= 0;
7038 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
7040 dprintf("cannot open device %s: %s\n",
7041 devname
, strerror(errno
));
7043 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
7045 } else if (must_be_container(dfd
)) {
7046 struct supertype
*cst
;
7047 cst
= super_by_fd(dfd
, NULL
);
7049 dprintf("cannot recognize container type %s\n",
7052 } else if (tst
->ss
!= st
->ss
) {
7053 dprintf("non-imsm container - ignore it: %s\n",
7056 } else if (!tst
->ss
->load_container
||
7057 tst
->ss
->load_container(tst
, dfd
, NULL
))
7060 tmpdev
->container
= 1;
7063 cst
->ss
->free_super(cst
);
7065 tmpdev
->st_rdev
= rdev
;
7066 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
7067 dprintf("no RAID superblock on %s\n",
7070 } else if (tst
->ss
->compare_super
== NULL
) {
7071 dprintf("Cannot assemble %s metadata on %s\n",
7072 tst
->ss
->name
, devname
);
7078 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
7079 /* Ignore unrecognised devices during auto-assembly */
7084 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
7086 if (st
->minor_version
== -1)
7087 st
->minor_version
= tst
->minor_version
;
7089 if (memcmp(info
.uuid
, uuid_zero
,
7090 sizeof(int[4])) == 0) {
7091 /* this is a floating spare. It cannot define
7092 * an array unless there are no more arrays of
7093 * this type to be found. It can be included
7094 * in an array of this type though.
7100 if (st
->ss
!= tst
->ss
||
7101 st
->minor_version
!= tst
->minor_version
||
7102 st
->ss
->compare_super(st
, tst
, 1) != 0) {
7103 /* Some mismatch. If exactly one array matches this host,
7104 * we can resolve on that one.
7105 * Or, if we are auto assembling, we just ignore the second
7108 dprintf("superblock on %s doesn't match others - assembly aborted\n",
7114 dprintf("found: devname: %s\n", devname
);
7118 tst
->ss
->free_super(tst
);
7122 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
7123 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
7124 for (iter
= head
; iter
; iter
= iter
->next
) {
7125 dprintf("content->text_version: %s vol\n",
7126 iter
->text_version
);
7127 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
7128 /* do not assemble arrays with unsupported
7130 dprintf("Cannot activate member %s.\n",
7131 iter
->text_version
);
7138 dprintf("No valid super block on device list: err: %d %p\n",
7142 dprintf("no more devices to examine\n");
7145 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7146 if (tmpdev
->used
== 1 && tmpdev
->found
) {
7148 if (count
< tmpdev
->found
)
7151 count
-= tmpdev
->found
;
7154 if (tmpdev
->used
== 1)
7159 st
->ss
->free_super(st
);
7163 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
7166 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
7168 const struct orom_entry
*entry
;
7169 struct devid_list
*dv
, *devid_list
;
7174 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7175 if (strstr(idev
->path
, hba_path
))
7179 if (!idev
|| !idev
->dev_id
)
7182 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7184 if (!entry
|| !entry
->devid_list
)
7187 devid_list
= entry
->devid_list
;
7188 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7189 struct md_list
*devlist
;
7190 struct sys_dev
*device
= NULL
;
7195 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7197 device
= device_by_id(dv
->devid
);
7200 hpath
= device
->path
;
7204 devlist
= get_devices(hpath
);
7205 /* if no intel devices return zero volumes */
7206 if (devlist
== NULL
)
7209 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7211 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7212 if (devlist
== NULL
)
7216 count
+= count_volumes_list(devlist
,
7220 dprintf("found %d count: %d\n", found
, count
);
7223 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7226 struct md_list
*dv
= devlist
;
7227 devlist
= devlist
->next
;
7235 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7239 if (hba
->type
== SYS_DEV_VMD
) {
7240 struct sys_dev
*dev
;
7243 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7244 if (dev
->type
== SYS_DEV_VMD
)
7245 count
+= __count_volumes(dev
->path
, dpa
,
7250 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7253 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7255 /* up to 512 if the plaform supports it, otherwise the platform max.
7256 * 128 if no platform detected
7258 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7260 return min(512, (1 << fs
));
7264 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7265 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7267 /* check/set platform and metadata limits/defaults */
7268 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7269 pr_vrb("platform supports a maximum of %d disks per array\n",
7274 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7275 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7276 pr_vrb("platform does not support raid%d with %d disk%s\n",
7277 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7281 if (*chunk
== 0 || *chunk
== UnSet
)
7282 *chunk
= imsm_default_chunk(super
->orom
);
7284 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7285 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7289 if (layout
!= imsm_level_to_layout(level
)) {
7291 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7292 else if (level
== 10)
7293 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7295 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7300 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7301 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7302 pr_vrb("platform does not support a volume size over 2TB\n");
7309 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7310 * FIX ME add ahci details
7312 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7313 int layout
, int raiddisks
, int *chunk
,
7314 unsigned long long size
,
7315 unsigned long long data_offset
,
7317 unsigned long long *freesize
,
7321 struct intel_super
*super
= st
->sb
;
7322 struct imsm_super
*mpb
;
7324 unsigned long long pos
= 0;
7325 unsigned long long maxsize
;
7329 /* We must have the container info already read in. */
7333 mpb
= super
->anchor
;
7335 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7336 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7340 /* General test: make sure there is space for
7341 * 'raiddisks' device extents of size 'size' at a given
7344 unsigned long long minsize
= size
;
7345 unsigned long long start_offset
= MaxSector
;
7348 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7349 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7354 e
= get_extents(super
, dl
, 0);
7357 unsigned long long esize
;
7358 esize
= e
[i
].start
- pos
;
7359 if (esize
>= minsize
)
7361 if (found
&& start_offset
== MaxSector
) {
7364 } else if (found
&& pos
!= start_offset
) {
7368 pos
= e
[i
].start
+ e
[i
].size
;
7370 } while (e
[i
-1].size
);
7375 if (dcnt
< raiddisks
) {
7377 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7384 /* This device must be a member of the set */
7385 if (!stat_is_blkdev(dev
, &rdev
))
7387 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7388 if (dl
->major
== (int)major(rdev
) &&
7389 dl
->minor
== (int)minor(rdev
))
7394 pr_err("%s is not in the same imsm set\n", dev
);
7396 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7397 /* If a volume is present then the current creation attempt
7398 * cannot incorporate new spares because the orom may not
7399 * understand this configuration (all member disks must be
7400 * members of each array in the container).
7402 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7403 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7405 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7406 mpb
->num_disks
!= raiddisks
) {
7407 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7411 /* retrieve the largest free space block */
7412 e
= get_extents(super
, dl
, 0);
7417 unsigned long long esize
;
7419 esize
= e
[i
].start
- pos
;
7420 if (esize
>= maxsize
)
7422 pos
= e
[i
].start
+ e
[i
].size
;
7424 } while (e
[i
-1].size
);
7429 pr_err("unable to determine free space for: %s\n",
7433 if (maxsize
< size
) {
7435 pr_err("%s not enough space (%llu < %llu)\n",
7436 dev
, maxsize
, size
);
7440 /* count total number of extents for merge */
7442 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7444 i
+= dl
->extent_cnt
;
7446 maxsize
= merge_extents(super
, i
);
7448 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7449 pr_err("attempting to create a second volume with size less then remaining space.\n");
7451 if (maxsize
< size
|| maxsize
== 0) {
7454 pr_err("no free space left on device. Aborting...\n");
7456 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7462 *freesize
= maxsize
;
7465 int count
= count_volumes(super
->hba
,
7466 super
->orom
->dpa
, verbose
);
7467 if (super
->orom
->vphba
<= count
) {
7468 pr_vrb("platform does not support more than %d raid volumes.\n",
7469 super
->orom
->vphba
);
7476 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7477 unsigned long long size
, int chunk
,
7478 unsigned long long *freesize
)
7480 struct intel_super
*super
= st
->sb
;
7481 struct imsm_super
*mpb
= super
->anchor
;
7486 unsigned long long maxsize
;
7487 unsigned long long minsize
;
7491 /* find the largest common start free region of the possible disks */
7495 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7501 /* don't activate new spares if we are orom constrained
7502 * and there is already a volume active in the container
7504 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7507 e
= get_extents(super
, dl
, 0);
7510 for (i
= 1; e
[i
-1].size
; i
++)
7518 maxsize
= merge_extents(super
, extent_cnt
);
7522 minsize
= chunk
* 2;
7524 if (cnt
< raiddisks
||
7525 (super
->orom
&& used
&& used
!= raiddisks
) ||
7526 maxsize
< minsize
||
7528 pr_err("not enough devices with space to create array.\n");
7529 return 0; /* No enough free spaces large enough */
7540 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7541 pr_err("attempting to create a second volume with size less then remaining space.\n");
7543 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7545 dl
->raiddisk
= cnt
++;
7549 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7554 static int reserve_space(struct supertype
*st
, int raiddisks
,
7555 unsigned long long size
, int chunk
,
7556 unsigned long long *freesize
)
7558 struct intel_super
*super
= st
->sb
;
7563 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7566 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7568 dl
->raiddisk
= cnt
++;
7575 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7576 int raiddisks
, int *chunk
, unsigned long long size
,
7577 unsigned long long data_offset
,
7578 char *dev
, unsigned long long *freesize
,
7579 int consistency_policy
, int verbose
)
7586 * if given unused devices create a container
7587 * if given given devices in a container create a member volume
7589 if (level
== LEVEL_CONTAINER
) {
7590 /* Must be a fresh device to add to a container */
7591 return validate_geometry_imsm_container(st
, level
, layout
,
7600 * Size is given in sectors.
7602 if (size
&& (size
< 2048)) {
7603 pr_err("Given size must be greater than 1M.\n");
7604 /* Depends on algorithm in Create.c :
7605 * if container was given (dev == NULL) return -1,
7606 * if block device was given ( dev != NULL) return 0.
7608 return dev
? -1 : 0;
7613 struct intel_super
*super
= st
->sb
;
7614 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7615 raiddisks
, chunk
, size
,
7618 /* we are being asked to automatically layout a
7619 * new volume based on the current contents of
7620 * the container. If the the parameters can be
7621 * satisfied reserve_space will record the disks,
7622 * start offset, and size of the volume to be
7623 * created. add_to_super and getinfo_super
7624 * detect when autolayout is in progress.
7626 /* assuming that freesize is always given when array is
7628 if (super
->orom
&& freesize
) {
7630 count
= count_volumes(super
->hba
,
7631 super
->orom
->dpa
, verbose
);
7632 if (super
->orom
->vphba
<= count
) {
7633 pr_vrb("platform does not support more than %d raid volumes.\n",
7634 super
->orom
->vphba
);
7639 return reserve_space(st
, raiddisks
, size
,
7645 /* creating in a given container */
7646 return validate_geometry_imsm_volume(st
, level
, layout
,
7647 raiddisks
, chunk
, size
,
7649 dev
, freesize
, verbose
);
7652 /* This device needs to be a device in an 'imsm' container */
7653 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7656 pr_err("Cannot create this array on device %s\n",
7661 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7663 pr_err("Cannot open %s: %s\n",
7664 dev
, strerror(errno
));
7667 /* Well, it is in use by someone, maybe an 'imsm' container. */
7668 cfd
= open_container(fd
);
7672 pr_err("Cannot use %s: It is busy\n",
7676 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7677 if (sra
&& sra
->array
.major_version
== -1 &&
7678 strcmp(sra
->text_version
, "imsm") == 0)
7682 /* This is a member of a imsm container. Load the container
7683 * and try to create a volume
7685 struct intel_super
*super
;
7687 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7689 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7691 return validate_geometry_imsm_volume(st
, level
, layout
,
7693 size
, data_offset
, dev
,
7700 pr_err("failed container membership check\n");
7706 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7708 struct intel_super
*super
= st
->sb
;
7710 if (level
&& *level
== UnSet
)
7711 *level
= LEVEL_CONTAINER
;
7713 if (level
&& layout
&& *layout
== UnSet
)
7714 *layout
= imsm_level_to_layout(*level
);
7716 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7717 *chunk
= imsm_default_chunk(super
->orom
);
7720 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7722 static int kill_subarray_imsm(struct supertype
*st
, char *subarray_id
)
7724 /* remove the subarray currently referenced by subarray_id */
7726 struct intel_dev
**dp
;
7727 struct intel_super
*super
= st
->sb
;
7728 __u8 current_vol
= strtoul(subarray_id
, NULL
, 10);
7729 struct imsm_super
*mpb
= super
->anchor
;
7731 if (mpb
->num_raid_devs
== 0)
7734 /* block deletions that would change the uuid of active subarrays
7736 * FIXME when immutable ids are available, but note that we'll
7737 * also need to fixup the invalidated/active subarray indexes in
7740 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7743 if (i
< current_vol
)
7745 sprintf(subarray
, "%u", i
);
7746 if (is_subarray_active(subarray
, st
->devnm
)) {
7747 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7754 if (st
->update_tail
) {
7755 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7757 u
->type
= update_kill_array
;
7758 u
->dev_idx
= current_vol
;
7759 append_metadata_update(st
, u
, sizeof(*u
));
7764 for (dp
= &super
->devlist
; *dp
;)
7765 if ((*dp
)->index
== current_vol
) {
7768 handle_missing(super
, (*dp
)->dev
);
7769 if ((*dp
)->index
> current_vol
)
7774 /* no more raid devices, all active components are now spares,
7775 * but of course failed are still failed
7777 if (--mpb
->num_raid_devs
== 0) {
7780 for (d
= super
->disks
; d
; d
= d
->next
)
7785 super
->updates_pending
++;
7790 static int get_rwh_policy_from_update(char *update
)
7792 if (strcmp(update
, "ppl") == 0)
7793 return RWH_MULTIPLE_DISTRIBUTED
;
7794 else if (strcmp(update
, "no-ppl") == 0)
7795 return RWH_MULTIPLE_OFF
;
7796 else if (strcmp(update
, "bitmap") == 0)
7798 else if (strcmp(update
, "no-bitmap") == 0)
7803 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7804 char *update
, struct mddev_ident
*ident
)
7806 /* update the subarray currently referenced by ->current_vol */
7807 struct intel_super
*super
= st
->sb
;
7808 struct imsm_super
*mpb
= super
->anchor
;
7810 if (strcmp(update
, "name") == 0) {
7811 char *name
= ident
->name
;
7815 if (is_subarray_active(subarray
, st
->devnm
)) {
7816 pr_err("Unable to update name of active subarray\n");
7820 if (!check_name(super
, name
, 0))
7823 vol
= strtoul(subarray
, &ep
, 10);
7824 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7827 if (st
->update_tail
) {
7828 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7830 u
->type
= update_rename_array
;
7832 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7833 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7834 append_metadata_update(st
, u
, sizeof(*u
));
7836 struct imsm_dev
*dev
;
7839 dev
= get_imsm_dev(super
, vol
);
7840 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7841 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7842 memcpy(dev
->volume
, name
, namelen
);
7843 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7844 dev
= get_imsm_dev(super
, i
);
7845 handle_missing(super
, dev
);
7847 super
->updates_pending
++;
7849 } else if (get_rwh_policy_from_update(update
) != -1) {
7852 int vol
= strtoul(subarray
, &ep
, 10);
7854 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7857 new_policy
= get_rwh_policy_from_update(update
);
7859 if (st
->update_tail
) {
7860 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7862 u
->type
= update_rwh_policy
;
7864 u
->new_policy
= new_policy
;
7865 append_metadata_update(st
, u
, sizeof(*u
));
7867 struct imsm_dev
*dev
;
7869 dev
= get_imsm_dev(super
, vol
);
7870 dev
->rwh_policy
= new_policy
;
7871 super
->updates_pending
++;
7873 if (new_policy
== RWH_BITMAP
)
7874 return write_init_bitmap_imsm_vol(st
, vol
);
7881 static int is_gen_migration(struct imsm_dev
*dev
)
7886 if (!dev
->vol
.migr_state
)
7889 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7895 static int is_rebuilding(struct imsm_dev
*dev
)
7897 struct imsm_map
*migr_map
;
7899 if (!dev
->vol
.migr_state
)
7902 if (migr_type(dev
) != MIGR_REBUILD
)
7905 migr_map
= get_imsm_map(dev
, MAP_1
);
7907 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7913 static int is_initializing(struct imsm_dev
*dev
)
7915 struct imsm_map
*migr_map
;
7917 if (!dev
->vol
.migr_state
)
7920 if (migr_type(dev
) != MIGR_INIT
)
7923 migr_map
= get_imsm_map(dev
, MAP_1
);
7925 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7931 static void update_recovery_start(struct intel_super
*super
,
7932 struct imsm_dev
*dev
,
7933 struct mdinfo
*array
)
7935 struct mdinfo
*rebuild
= NULL
;
7939 if (!is_rebuilding(dev
))
7942 /* Find the rebuild target, but punt on the dual rebuild case */
7943 for (d
= array
->devs
; d
; d
= d
->next
)
7944 if (d
->recovery_start
== 0) {
7951 /* (?) none of the disks are marked with
7952 * IMSM_ORD_REBUILD, so assume they are missing and the
7953 * disk_ord_tbl was not correctly updated
7955 dprintf("failed to locate out-of-sync disk\n");
7959 units
= vol_curr_migr_unit(dev
);
7960 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7963 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7965 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7967 /* Given a container loaded by load_super_imsm_all,
7968 * extract information about all the arrays into
7970 * If 'subarray' is given, just extract info about that array.
7972 * For each imsm_dev create an mdinfo, fill it in,
7973 * then look for matching devices in super->disks
7974 * and create appropriate device mdinfo.
7976 struct intel_super
*super
= st
->sb
;
7977 struct imsm_super
*mpb
= super
->anchor
;
7978 struct mdinfo
*rest
= NULL
;
7982 int spare_disks
= 0;
7983 int current_vol
= super
->current_vol
;
7985 /* do not assemble arrays when not all attributes are supported */
7986 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7988 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7991 /* count spare devices, not used in maps
7993 for (d
= super
->disks
; d
; d
= d
->next
)
7997 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7998 struct imsm_dev
*dev
;
7999 struct imsm_map
*map
;
8000 struct imsm_map
*map2
;
8001 struct mdinfo
*this;
8008 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
8011 dev
= get_imsm_dev(super
, i
);
8012 map
= get_imsm_map(dev
, MAP_0
);
8013 map2
= get_imsm_map(dev
, MAP_1
);
8014 level
= get_imsm_raid_level(map
);
8016 /* do not publish arrays that are in the middle of an
8017 * unsupported migration
8019 if (dev
->vol
.migr_state
&&
8020 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
8021 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
8025 /* do not publish arrays that are not support by controller's
8029 this = xmalloc(sizeof(*this));
8031 super
->current_vol
= i
;
8032 getinfo_super_imsm_volume(st
, this, NULL
);
8034 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
8035 /* mdadm does not support all metadata features- set the bit in all arrays state */
8036 if (!validate_geometry_imsm_orom(super
,
8037 level
, /* RAID level */
8038 imsm_level_to_layout(level
),
8039 map
->num_members
, /* raid disks */
8040 &chunk
, imsm_dev_size(dev
),
8042 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
8044 this->array
.state
|=
8045 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8046 (1<<MD_SB_BLOCK_VOLUME
);
8049 /* if array has bad blocks, set suitable bit in all arrays state */
8051 this->array
.state
|=
8052 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8053 (1<<MD_SB_BLOCK_VOLUME
);
8055 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
8056 unsigned long long recovery_start
;
8057 struct mdinfo
*info_d
;
8065 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
8066 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
8067 for (d
= super
->disks
; d
; d
= d
->next
)
8068 if (d
->index
== idx
)
8071 recovery_start
= MaxSector
;
8074 if (d
&& is_failed(&d
->disk
))
8076 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
8078 if (!(ord
& IMSM_ORD_REBUILD
))
8079 this->array
.working_disks
++;
8081 * if we skip some disks the array will be assmebled degraded;
8082 * reset resync start to avoid a dirty-degraded
8083 * situation when performing the intial sync
8088 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
8089 if ((!able_to_resync(level
, missing
) ||
8090 recovery_start
== 0))
8091 this->resync_start
= MaxSector
;
8094 * FIXME handle dirty degraded
8101 info_d
= xcalloc(1, sizeof(*info_d
));
8102 info_d
->next
= this->devs
;
8103 this->devs
= info_d
;
8105 info_d
->disk
.number
= d
->index
;
8106 info_d
->disk
.major
= d
->major
;
8107 info_d
->disk
.minor
= d
->minor
;
8108 info_d
->disk
.raid_disk
= slot
;
8109 info_d
->recovery_start
= recovery_start
;
8111 if (slot
< map2
->num_members
)
8112 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8114 this->array
.spare_disks
++;
8116 if (slot
< map
->num_members
)
8117 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8119 this->array
.spare_disks
++;
8122 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
8123 info_d
->data_offset
= pba_of_lba0(map
);
8124 info_d
->component_size
= calc_component_size(map
, dev
);
8126 if (map
->raid_level
== 5) {
8127 info_d
->ppl_sector
= this->ppl_sector
;
8128 info_d
->ppl_size
= this->ppl_size
;
8129 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
8130 recovery_start
== 0)
8131 this->resync_start
= 0;
8134 info_d
->bb
.supported
= 1;
8135 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
8136 info_d
->data_offset
,
8137 info_d
->component_size
,
8140 /* now that the disk list is up-to-date fixup recovery_start */
8141 update_recovery_start(super
, dev
, this);
8142 this->array
.spare_disks
+= spare_disks
;
8144 /* check for reshape */
8145 if (this->reshape_active
== 1)
8146 recover_backup_imsm(st
, this);
8150 super
->current_vol
= current_vol
;
8154 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
8155 int failed
, int look_in_map
)
8157 struct imsm_map
*map
;
8159 map
= get_imsm_map(dev
, look_in_map
);
8162 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
8163 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
8165 switch (get_imsm_raid_level(map
)) {
8167 return IMSM_T_STATE_FAILED
;
8170 if (failed
< map
->num_members
)
8171 return IMSM_T_STATE_DEGRADED
;
8173 return IMSM_T_STATE_FAILED
;
8178 * check to see if any mirrors have failed, otherwise we
8179 * are degraded. Even numbered slots are mirrored on
8183 /* gcc -Os complains that this is unused */
8184 int insync
= insync
;
8186 for (i
= 0; i
< map
->num_members
; i
++) {
8187 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8188 int idx
= ord_to_idx(ord
);
8189 struct imsm_disk
*disk
;
8191 /* reset the potential in-sync count on even-numbered
8192 * slots. num_copies is always 2 for imsm raid10
8197 disk
= get_imsm_disk(super
, idx
);
8198 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8201 /* no in-sync disks left in this mirror the
8205 return IMSM_T_STATE_FAILED
;
8208 return IMSM_T_STATE_DEGRADED
;
8212 return IMSM_T_STATE_DEGRADED
;
8214 return IMSM_T_STATE_FAILED
;
8220 return map
->map_state
;
8223 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8228 struct imsm_disk
*disk
;
8229 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8230 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8231 struct imsm_map
*map_for_loop
;
8236 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8237 * disks that are being rebuilt. New failures are recorded to
8238 * map[0]. So we look through all the disks we started with and
8239 * see if any failures are still present, or if any new ones
8243 if (prev
&& (map
->num_members
< prev
->num_members
))
8244 map_for_loop
= prev
;
8246 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8248 /* when MAP_X is passed both maps failures are counted
8251 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8252 i
< prev
->num_members
) {
8253 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8254 idx_1
= ord_to_idx(ord
);
8256 disk
= get_imsm_disk(super
, idx_1
);
8257 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8260 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8261 i
< map
->num_members
) {
8262 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8263 idx
= ord_to_idx(ord
);
8266 disk
= get_imsm_disk(super
, idx
);
8267 if (!disk
|| is_failed(disk
) ||
8268 ord
& IMSM_ORD_REBUILD
)
8277 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8280 struct intel_super
*super
= c
->sb
;
8281 struct imsm_super
*mpb
= super
->anchor
;
8282 struct imsm_update_prealloc_bb_mem u
;
8284 if (atoi(inst
) >= mpb
->num_raid_devs
) {
8285 pr_err("subarry index %d, out of range\n", atoi(inst
));
8289 dprintf("imsm: open_new %s\n", inst
);
8290 a
->info
.container_member
= atoi(inst
);
8292 u
.type
= update_prealloc_badblocks_mem
;
8293 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8298 static int is_resyncing(struct imsm_dev
*dev
)
8300 struct imsm_map
*migr_map
;
8302 if (!dev
->vol
.migr_state
)
8305 if (migr_type(dev
) == MIGR_INIT
||
8306 migr_type(dev
) == MIGR_REPAIR
)
8309 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8312 migr_map
= get_imsm_map(dev
, MAP_1
);
8314 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8315 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8321 /* return true if we recorded new information */
8322 static int mark_failure(struct intel_super
*super
,
8323 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8327 struct imsm_map
*map
;
8328 char buf
[MAX_RAID_SERIAL_LEN
+3];
8329 unsigned int len
, shift
= 0;
8331 /* new failures are always set in map[0] */
8332 map
= get_imsm_map(dev
, MAP_0
);
8334 slot
= get_imsm_disk_slot(map
, idx
);
8338 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8339 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8342 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8343 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8345 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8346 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8347 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8349 disk
->status
|= FAILED_DISK
;
8350 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8351 /* mark failures in second map if second map exists and this disk
8353 * This is valid for migration, initialization and rebuild
8355 if (dev
->vol
.migr_state
) {
8356 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8357 int slot2
= get_imsm_disk_slot(map2
, idx
);
8359 if (slot2
< map2
->num_members
&& slot2
>= 0)
8360 set_imsm_ord_tbl_ent(map2
, slot2
,
8361 idx
| IMSM_ORD_REBUILD
);
8363 if (map
->failed_disk_num
== 0xff ||
8364 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8365 map
->failed_disk_num
= slot
;
8367 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8372 static void mark_missing(struct intel_super
*super
,
8373 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8375 mark_failure(super
, dev
, disk
, idx
);
8377 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8380 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8381 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8384 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8388 if (!super
->missing
)
8391 /* When orom adds replacement for missing disk it does
8392 * not remove entry of missing disk, but just updates map with
8393 * new added disk. So it is not enough just to test if there is
8394 * any missing disk, we have to look if there are any failed disks
8395 * in map to stop migration */
8397 dprintf("imsm: mark missing\n");
8398 /* end process for initialization and rebuild only
8400 if (is_gen_migration(dev
) == 0) {
8401 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8405 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8406 struct imsm_map
*map1
;
8407 int i
, ord
, ord_map1
;
8410 for (i
= 0; i
< map
->num_members
; i
++) {
8411 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8412 if (!(ord
& IMSM_ORD_REBUILD
))
8415 map1
= get_imsm_map(dev
, MAP_1
);
8419 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8420 if (ord_map1
& IMSM_ORD_REBUILD
)
8425 map_state
= imsm_check_degraded(super
, dev
,
8427 end_migration(dev
, super
, map_state
);
8431 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8432 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8433 super
->updates_pending
++;
8436 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8439 unsigned long long array_blocks
;
8440 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8441 int used_disks
= imsm_num_data_members(map
);
8443 if (used_disks
== 0) {
8444 /* when problems occures
8445 * return current array_blocks value
8447 array_blocks
= imsm_dev_size(dev
);
8449 return array_blocks
;
8452 /* set array size in metadata
8455 /* OLCE size change is caused by added disks
8457 array_blocks
= per_dev_array_size(map
) * used_disks
;
8459 /* Online Volume Size Change
8460 * Using available free space
8462 array_blocks
= new_size
;
8464 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8465 set_imsm_dev_size(dev
, array_blocks
);
8467 return array_blocks
;
8470 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8472 static void imsm_progress_container_reshape(struct intel_super
*super
)
8474 /* if no device has a migr_state, but some device has a
8475 * different number of members than the previous device, start
8476 * changing the number of devices in this device to match
8479 struct imsm_super
*mpb
= super
->anchor
;
8480 int prev_disks
= -1;
8484 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8485 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8486 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8487 struct imsm_map
*map2
;
8488 int prev_num_members
;
8490 if (dev
->vol
.migr_state
)
8493 if (prev_disks
== -1)
8494 prev_disks
= map
->num_members
;
8495 if (prev_disks
== map
->num_members
)
8498 /* OK, this array needs to enter reshape mode.
8499 * i.e it needs a migr_state
8502 copy_map_size
= sizeof_imsm_map(map
);
8503 prev_num_members
= map
->num_members
;
8504 map
->num_members
= prev_disks
;
8505 dev
->vol
.migr_state
= 1;
8506 set_vol_curr_migr_unit(dev
, 0);
8507 set_migr_type(dev
, MIGR_GEN_MIGR
);
8508 for (i
= prev_num_members
;
8509 i
< map
->num_members
; i
++)
8510 set_imsm_ord_tbl_ent(map
, i
, i
);
8511 map2
= get_imsm_map(dev
, MAP_1
);
8512 /* Copy the current map */
8513 memcpy(map2
, map
, copy_map_size
);
8514 map2
->num_members
= prev_num_members
;
8516 imsm_set_array_size(dev
, -1);
8517 super
->clean_migration_record_by_mdmon
= 1;
8518 super
->updates_pending
++;
8522 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8523 * states are handled in imsm_set_disk() with one exception, when a
8524 * resync is stopped due to a new failure this routine will set the
8525 * 'degraded' state for the array.
8527 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8529 int inst
= a
->info
.container_member
;
8530 struct intel_super
*super
= a
->container
->sb
;
8531 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8532 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8533 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8534 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8535 __u32 blocks_per_unit
;
8537 if (dev
->vol
.migr_state
&&
8538 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8539 /* array state change is blocked due to reshape action
8541 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8542 * - finish the reshape (if last_checkpoint is big and action != reshape)
8543 * - update vol_curr_migr_unit
8545 if (a
->curr_action
== reshape
) {
8546 /* still reshaping, maybe update vol_curr_migr_unit */
8547 goto mark_checkpoint
;
8549 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8550 /* for some reason we aborted the reshape.
8552 * disable automatic metadata rollback
8553 * user action is required to recover process
8556 struct imsm_map
*map2
=
8557 get_imsm_map(dev
, MAP_1
);
8558 dev
->vol
.migr_state
= 0;
8559 set_migr_type(dev
, 0);
8560 set_vol_curr_migr_unit(dev
, 0);
8562 sizeof_imsm_map(map2
));
8563 super
->updates_pending
++;
8566 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8567 unsigned long long array_blocks
;
8571 used_disks
= imsm_num_data_members(map
);
8572 if (used_disks
> 0) {
8574 per_dev_array_size(map
) *
8577 round_size_to_mb(array_blocks
,
8579 a
->info
.custom_array_size
= array_blocks
;
8580 /* encourage manager to update array
8584 a
->check_reshape
= 1;
8586 /* finalize online capacity expansion/reshape */
8587 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8589 mdi
->disk
.raid_disk
,
8592 imsm_progress_container_reshape(super
);
8597 /* before we activate this array handle any missing disks */
8598 if (consistent
== 2)
8599 handle_missing(super
, dev
);
8601 if (consistent
== 2 &&
8602 (!is_resync_complete(&a
->info
) ||
8603 map_state
!= IMSM_T_STATE_NORMAL
||
8604 dev
->vol
.migr_state
))
8607 if (is_resync_complete(&a
->info
)) {
8608 /* complete intialization / resync,
8609 * recovery and interrupted recovery is completed in
8612 if (is_resyncing(dev
)) {
8613 dprintf("imsm: mark resync done\n");
8614 end_migration(dev
, super
, map_state
);
8615 super
->updates_pending
++;
8616 a
->last_checkpoint
= 0;
8618 } else if ((!is_resyncing(dev
) && !failed
) &&
8619 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8620 /* mark the start of the init process if nothing is failed */
8621 dprintf("imsm: mark resync start\n");
8622 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8623 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8625 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8626 super
->updates_pending
++;
8630 /* skip checkpointing for general migration,
8631 * it is controlled in mdadm
8633 if (is_gen_migration(dev
))
8634 goto skip_mark_checkpoint
;
8636 /* check if we can update vol_curr_migr_unit from resync_start,
8639 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8640 if (blocks_per_unit
) {
8641 set_vol_curr_migr_unit(dev
,
8642 a
->last_checkpoint
/ blocks_per_unit
);
8643 dprintf("imsm: mark checkpoint (%llu)\n",
8644 vol_curr_migr_unit(dev
));
8645 super
->updates_pending
++;
8648 skip_mark_checkpoint
:
8649 /* mark dirty / clean */
8650 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8651 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8652 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8654 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8656 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8657 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8658 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8659 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8661 super
->updates_pending
++;
8667 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8669 int inst
= a
->info
.container_member
;
8670 struct intel_super
*super
= a
->container
->sb
;
8671 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8672 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8674 if (slot
> map
->num_members
) {
8675 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8676 slot
, map
->num_members
- 1);
8683 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8686 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8688 int inst
= a
->info
.container_member
;
8689 struct intel_super
*super
= a
->container
->sb
;
8690 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8691 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8692 struct imsm_disk
*disk
;
8694 int recovery_not_finished
= 0;
8698 int rebuild_done
= 0;
8701 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8705 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8706 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8708 /* check for new failures */
8709 if (disk
&& (state
& DS_FAULTY
)) {
8710 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8711 super
->updates_pending
++;
8714 /* check if in_sync */
8715 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8716 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8718 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8720 super
->updates_pending
++;
8723 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8724 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8726 /* check if recovery complete, newly degraded, or failed */
8727 dprintf("imsm: Detected transition to state ");
8728 switch (map_state
) {
8729 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8730 dprintf("normal: ");
8731 if (is_rebuilding(dev
)) {
8732 dprintf_cont("while rebuilding");
8733 /* check if recovery is really finished */
8734 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8735 if (mdi
->recovery_start
!= MaxSector
) {
8736 recovery_not_finished
= 1;
8739 if (recovery_not_finished
) {
8741 dprintf("Rebuild has not finished yet, state not changed");
8742 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8743 a
->last_checkpoint
= mdi
->recovery_start
;
8744 super
->updates_pending
++;
8748 end_migration(dev
, super
, map_state
);
8749 map
->failed_disk_num
= ~0;
8750 super
->updates_pending
++;
8751 a
->last_checkpoint
= 0;
8754 if (is_gen_migration(dev
)) {
8755 dprintf_cont("while general migration");
8756 if (a
->last_checkpoint
>= a
->info
.component_size
)
8757 end_migration(dev
, super
, map_state
);
8759 map
->map_state
= map_state
;
8760 map
->failed_disk_num
= ~0;
8761 super
->updates_pending
++;
8765 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8766 dprintf_cont("degraded: ");
8767 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8768 dprintf_cont("mark degraded");
8769 map
->map_state
= map_state
;
8770 super
->updates_pending
++;
8771 a
->last_checkpoint
= 0;
8774 if (is_rebuilding(dev
)) {
8775 dprintf_cont("while rebuilding ");
8776 if (state
& DS_FAULTY
) {
8777 dprintf_cont("removing failed drive ");
8778 if (n
== map
->failed_disk_num
) {
8779 dprintf_cont("end migration");
8780 end_migration(dev
, super
, map_state
);
8781 a
->last_checkpoint
= 0;
8783 dprintf_cont("fail detected during rebuild, changing map state");
8784 map
->map_state
= map_state
;
8786 super
->updates_pending
++;
8792 /* check if recovery is really finished */
8793 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8794 if (mdi
->recovery_start
!= MaxSector
) {
8795 recovery_not_finished
= 1;
8798 if (recovery_not_finished
) {
8800 dprintf_cont("Rebuild has not finished yet");
8801 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8802 a
->last_checkpoint
=
8803 mdi
->recovery_start
;
8804 super
->updates_pending
++;
8809 dprintf_cont(" Rebuild done, still degraded");
8810 end_migration(dev
, super
, map_state
);
8811 a
->last_checkpoint
= 0;
8812 super
->updates_pending
++;
8814 for (i
= 0; i
< map
->num_members
; i
++) {
8815 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8817 if (idx
& IMSM_ORD_REBUILD
)
8818 map
->failed_disk_num
= i
;
8820 super
->updates_pending
++;
8823 if (is_gen_migration(dev
)) {
8824 dprintf_cont("while general migration");
8825 if (a
->last_checkpoint
>= a
->info
.component_size
)
8826 end_migration(dev
, super
, map_state
);
8828 map
->map_state
= map_state
;
8829 manage_second_map(super
, dev
);
8831 super
->updates_pending
++;
8834 if (is_initializing(dev
)) {
8835 dprintf_cont("while initialization.");
8836 map
->map_state
= map_state
;
8837 super
->updates_pending
++;
8841 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8842 dprintf_cont("failed: ");
8843 if (is_gen_migration(dev
)) {
8844 dprintf_cont("while general migration");
8845 map
->map_state
= map_state
;
8846 super
->updates_pending
++;
8849 if (map
->map_state
!= map_state
) {
8850 dprintf_cont("mark failed");
8851 end_migration(dev
, super
, map_state
);
8852 super
->updates_pending
++;
8853 a
->last_checkpoint
= 0;
8858 dprintf_cont("state %i\n", map_state
);
8863 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8866 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8867 unsigned long long dsize
;
8868 unsigned long long sectors
;
8869 unsigned int sector_size
;
8871 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
8873 get_dev_size(fd
, NULL
, &dsize
);
8875 if (mpb_size
> sector_size
) {
8876 /* -1 to account for anchor */
8877 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8879 /* write the extended mpb to the sectors preceeding the anchor */
8880 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8884 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8885 sector_size
* sectors
) != sector_size
* sectors
)
8889 /* first block is stored on second to last sector of the disk */
8890 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8893 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8899 static void imsm_sync_metadata(struct supertype
*container
)
8901 struct intel_super
*super
= container
->sb
;
8903 dprintf("sync metadata: %d\n", super
->updates_pending
);
8904 if (!super
->updates_pending
)
8907 write_super_imsm(container
, 0);
8909 super
->updates_pending
= 0;
8912 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8914 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8915 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8918 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8922 if (dl
&& is_failed(&dl
->disk
))
8926 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8931 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8932 struct active_array
*a
, int activate_new
,
8933 struct mdinfo
*additional_test_list
)
8935 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8936 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8937 struct imsm_super
*mpb
= super
->anchor
;
8938 struct imsm_map
*map
;
8939 unsigned long long pos
;
8944 __u32 array_start
= 0;
8945 __u32 array_end
= 0;
8947 struct mdinfo
*test_list
;
8949 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8950 /* If in this array, skip */
8951 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8952 if (d
->state_fd
>= 0 &&
8953 d
->disk
.major
== dl
->major
&&
8954 d
->disk
.minor
== dl
->minor
) {
8955 dprintf("%x:%x already in array\n",
8956 dl
->major
, dl
->minor
);
8961 test_list
= additional_test_list
;
8963 if (test_list
->disk
.major
== dl
->major
&&
8964 test_list
->disk
.minor
== dl
->minor
) {
8965 dprintf("%x:%x already in additional test list\n",
8966 dl
->major
, dl
->minor
);
8969 test_list
= test_list
->next
;
8974 /* skip in use or failed drives */
8975 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8977 dprintf("%x:%x status (failed: %d index: %d)\n",
8978 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8982 /* skip pure spares when we are looking for partially
8983 * assimilated drives
8985 if (dl
->index
== -1 && !activate_new
)
8988 if (!drive_validate_sector_size(super
, dl
))
8991 /* Does this unused device have the requisite free space?
8992 * It needs to be able to cover all member volumes
8994 ex
= get_extents(super
, dl
, 1);
8996 dprintf("cannot get extents\n");
8999 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9000 dev
= get_imsm_dev(super
, i
);
9001 map
= get_imsm_map(dev
, MAP_0
);
9003 /* check if this disk is already a member of
9006 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
9012 array_start
= pba_of_lba0(map
);
9013 array_end
= array_start
+
9014 per_dev_array_size(map
) - 1;
9017 /* check that we can start at pba_of_lba0 with
9018 * num_data_stripes*blocks_per_stripe of space
9020 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
9024 pos
= ex
[j
].start
+ ex
[j
].size
;
9026 } while (ex
[j
-1].size
);
9033 if (i
< mpb
->num_raid_devs
) {
9034 dprintf("%x:%x does not have %u to %u available\n",
9035 dl
->major
, dl
->minor
, array_start
, array_end
);
9045 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
9047 struct imsm_dev
*dev2
;
9048 struct imsm_map
*map
;
9054 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
9056 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
9057 if (state
== IMSM_T_STATE_FAILED
) {
9058 map
= get_imsm_map(dev2
, MAP_0
);
9061 for (slot
= 0; slot
< map
->num_members
; slot
++) {
9063 * Check if failed disks are deleted from intel
9064 * disk list or are marked to be deleted
9066 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
9067 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
9069 * Do not rebuild the array if failed disks
9070 * from failed sub-array are not removed from
9074 is_failed(&idisk
->disk
) &&
9075 (idisk
->action
!= DISK_REMOVE
))
9083 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
9084 struct metadata_update
**updates
)
9087 * Find a device with unused free space and use it to replace a
9088 * failed/vacant region in an array. We replace failed regions one a
9089 * array at a time. The result is that a new spare disk will be added
9090 * to the first failed array and after the monitor has finished
9091 * propagating failures the remainder will be consumed.
9093 * FIXME add a capability for mdmon to request spares from another
9097 struct intel_super
*super
= a
->container
->sb
;
9098 int inst
= a
->info
.container_member
;
9099 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
9100 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9101 int failed
= a
->info
.array
.raid_disks
;
9102 struct mdinfo
*rv
= NULL
;
9105 struct metadata_update
*mu
;
9107 struct imsm_update_activate_spare
*u
;
9112 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
9113 if ((d
->curr_state
& DS_FAULTY
) &&
9115 /* wait for Removal to happen */
9117 if (d
->state_fd
>= 0)
9121 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
9122 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
9124 if (imsm_reshape_blocks_arrays_changes(super
))
9127 /* Cannot activate another spare if rebuild is in progress already
9129 if (is_rebuilding(dev
)) {
9130 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
9134 if (a
->info
.array
.level
== 4)
9135 /* No repair for takeovered array
9136 * imsm doesn't support raid4
9140 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
9141 IMSM_T_STATE_DEGRADED
)
9144 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
9145 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
9150 * If there are any failed disks check state of the other volume.
9151 * Block rebuild if the another one is failed until failed disks
9152 * are removed from container.
9155 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
9156 MAX_RAID_SERIAL_LEN
, dev
->volume
);
9157 /* check if states of the other volumes allow for rebuild */
9158 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
9160 allowed
= imsm_rebuild_allowed(a
->container
,
9168 /* For each slot, if it is not working, find a spare */
9169 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
9170 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9171 if (d
->disk
.raid_disk
== i
)
9173 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
9174 if (d
&& (d
->state_fd
>= 0))
9178 * OK, this device needs recovery. Try to re-add the
9179 * previous occupant of this slot, if this fails see if
9180 * we can continue the assimilation of a spare that was
9181 * partially assimilated, finally try to activate a new
9184 dl
= imsm_readd(super
, i
, a
);
9186 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9188 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9192 /* found a usable disk with enough space */
9193 di
= xcalloc(1, sizeof(*di
));
9195 /* dl->index will be -1 in the case we are activating a
9196 * pristine spare. imsm_process_update() will create a
9197 * new index in this case. Once a disk is found to be
9198 * failed in all member arrays it is kicked from the
9201 di
->disk
.number
= dl
->index
;
9203 /* (ab)use di->devs to store a pointer to the device
9206 di
->devs
= (struct mdinfo
*) dl
;
9208 di
->disk
.raid_disk
= i
;
9209 di
->disk
.major
= dl
->major
;
9210 di
->disk
.minor
= dl
->minor
;
9212 di
->recovery_start
= 0;
9213 di
->data_offset
= pba_of_lba0(map
);
9214 di
->component_size
= a
->info
.component_size
;
9215 di
->container_member
= inst
;
9216 di
->bb
.supported
= 1;
9217 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9218 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9219 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9221 super
->random
= random32();
9225 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9226 i
, di
->data_offset
);
9230 /* No spares found */
9232 /* Now 'rv' has a list of devices to return.
9233 * Create a metadata_update record to update the
9234 * disk_ord_tbl for the array
9236 mu
= xmalloc(sizeof(*mu
));
9237 mu
->buf
= xcalloc(num_spares
,
9238 sizeof(struct imsm_update_activate_spare
));
9240 mu
->space_list
= NULL
;
9241 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9242 mu
->next
= *updates
;
9243 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9245 for (di
= rv
; di
; di
= di
->next
) {
9246 u
->type
= update_activate_spare
;
9247 u
->dl
= (struct dl
*) di
->devs
;
9249 u
->slot
= di
->disk
.raid_disk
;
9260 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9262 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9263 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9264 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9265 struct disk_info
*inf
= get_disk_info(u
);
9266 struct imsm_disk
*disk
;
9270 for (i
= 0; i
< map
->num_members
; i
++) {
9271 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9272 for (j
= 0; j
< new_map
->num_members
; j
++)
9273 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9280 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9284 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9285 if (dl
->major
== major
&& dl
->minor
== minor
)
9290 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9296 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9297 if (dl
->major
== major
&& dl
->minor
== minor
) {
9300 prev
->next
= dl
->next
;
9302 super
->disks
= dl
->next
;
9304 __free_imsm_disk(dl
);
9305 dprintf("removed %x:%x\n", major
, minor
);
9313 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9315 static int add_remove_disk_update(struct intel_super
*super
)
9317 int check_degraded
= 0;
9320 /* add/remove some spares to/from the metadata/contrainer */
9321 while (super
->disk_mgmt_list
) {
9322 struct dl
*disk_cfg
;
9324 disk_cfg
= super
->disk_mgmt_list
;
9325 super
->disk_mgmt_list
= disk_cfg
->next
;
9326 disk_cfg
->next
= NULL
;
9328 if (disk_cfg
->action
== DISK_ADD
) {
9329 disk_cfg
->next
= super
->disks
;
9330 super
->disks
= disk_cfg
;
9332 dprintf("added %x:%x\n",
9333 disk_cfg
->major
, disk_cfg
->minor
);
9334 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9335 dprintf("Disk remove action processed: %x.%x\n",
9336 disk_cfg
->major
, disk_cfg
->minor
);
9337 disk
= get_disk_super(super
,
9341 /* store action status */
9342 disk
->action
= DISK_REMOVE
;
9343 /* remove spare disks only */
9344 if (disk
->index
== -1) {
9345 remove_disk_super(super
,
9349 disk_cfg
->fd
= disk
->fd
;
9353 /* release allocate disk structure */
9354 __free_imsm_disk(disk_cfg
);
9357 return check_degraded
;
9360 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9361 struct intel_super
*super
,
9364 struct intel_dev
*id
;
9365 void **tofree
= NULL
;
9368 dprintf("(enter)\n");
9369 if (u
->subdev
< 0 || u
->subdev
> 1) {
9370 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9373 if (space_list
== NULL
|| *space_list
== NULL
) {
9374 dprintf("imsm: Error: Memory is not allocated\n");
9378 for (id
= super
->devlist
; id
; id
= id
->next
) {
9379 if (id
->index
== (unsigned)u
->subdev
) {
9380 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9381 struct imsm_map
*map
;
9382 struct imsm_dev
*new_dev
=
9383 (struct imsm_dev
*)*space_list
;
9384 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9386 struct dl
*new_disk
;
9388 if (new_dev
== NULL
)
9390 *space_list
= **space_list
;
9391 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9392 map
= get_imsm_map(new_dev
, MAP_0
);
9394 dprintf("imsm: Error: migration in progress");
9398 to_state
= map
->map_state
;
9399 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9401 /* this should not happen */
9402 if (u
->new_disks
[0] < 0) {
9403 map
->failed_disk_num
=
9404 map
->num_members
- 1;
9405 to_state
= IMSM_T_STATE_DEGRADED
;
9407 to_state
= IMSM_T_STATE_NORMAL
;
9409 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9410 if (u
->new_level
> -1)
9411 map
->raid_level
= u
->new_level
;
9412 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9413 if ((u
->new_level
== 5) &&
9414 (migr_map
->raid_level
== 0)) {
9415 int ord
= map
->num_members
- 1;
9416 migr_map
->num_members
--;
9417 if (u
->new_disks
[0] < 0)
9418 ord
|= IMSM_ORD_REBUILD
;
9419 set_imsm_ord_tbl_ent(map
,
9420 map
->num_members
- 1,
9424 tofree
= (void **)dev
;
9426 /* update chunk size
9428 if (u
->new_chunksize
> 0) {
9429 unsigned long long num_data_stripes
;
9430 struct imsm_map
*dest_map
=
9431 get_imsm_map(dev
, MAP_0
);
9433 imsm_num_data_members(dest_map
);
9435 if (used_disks
== 0)
9438 map
->blocks_per_strip
=
9439 __cpu_to_le16(u
->new_chunksize
* 2);
9441 imsm_dev_size(dev
) / used_disks
;
9442 num_data_stripes
/= map
->blocks_per_strip
;
9443 num_data_stripes
/= map
->num_domains
;
9444 set_num_data_stripes(map
, num_data_stripes
);
9447 /* ensure blocks_per_member has valid value
9449 set_blocks_per_member(map
,
9450 per_dev_array_size(map
) +
9451 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9455 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9456 migr_map
->raid_level
== map
->raid_level
)
9459 if (u
->new_disks
[0] >= 0) {
9462 new_disk
= get_disk_super(super
,
9463 major(u
->new_disks
[0]),
9464 minor(u
->new_disks
[0]));
9465 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9466 major(u
->new_disks
[0]),
9467 minor(u
->new_disks
[0]),
9468 new_disk
, new_disk
->index
);
9469 if (new_disk
== NULL
)
9470 goto error_disk_add
;
9472 new_disk
->index
= map
->num_members
- 1;
9473 /* slot to fill in autolayout
9475 new_disk
->raiddisk
= new_disk
->index
;
9476 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9477 new_disk
->disk
.status
&= ~SPARE_DISK
;
9479 goto error_disk_add
;
9482 *tofree
= *space_list
;
9483 /* calculate new size
9485 imsm_set_array_size(new_dev
, -1);
9492 *space_list
= tofree
;
9496 dprintf("Error: imsm: Cannot find disk.\n");
9500 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9501 struct intel_super
*super
)
9503 struct intel_dev
*id
;
9506 dprintf("(enter)\n");
9507 if (u
->subdev
< 0 || u
->subdev
> 1) {
9508 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9512 for (id
= super
->devlist
; id
; id
= id
->next
) {
9513 if (id
->index
== (unsigned)u
->subdev
) {
9514 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9515 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9516 int used_disks
= imsm_num_data_members(map
);
9517 unsigned long long blocks_per_member
;
9518 unsigned long long num_data_stripes
;
9519 unsigned long long new_size_per_disk
;
9521 if (used_disks
== 0)
9524 /* calculate new size
9526 new_size_per_disk
= u
->new_size
/ used_disks
;
9527 blocks_per_member
= new_size_per_disk
+
9528 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9529 num_data_stripes
= new_size_per_disk
/
9530 map
->blocks_per_strip
;
9531 num_data_stripes
/= map
->num_domains
;
9532 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9533 u
->new_size
, new_size_per_disk
,
9535 set_blocks_per_member(map
, blocks_per_member
);
9536 set_num_data_stripes(map
, num_data_stripes
);
9537 imsm_set_array_size(dev
, u
->new_size
);
9547 static int prepare_spare_to_activate(struct supertype
*st
,
9548 struct imsm_update_activate_spare
*u
)
9550 struct intel_super
*super
= st
->sb
;
9551 int prev_current_vol
= super
->current_vol
;
9552 struct active_array
*a
;
9555 for (a
= st
->arrays
; a
; a
= a
->next
)
9557 * Additional initialization (adding bitmap header, filling
9558 * the bitmap area with '1's to force initial rebuild for a whole
9559 * data-area) is required when adding the spare to the volume
9560 * with write-intent bitmap.
9562 if (a
->info
.container_member
== u
->array
&&
9563 a
->info
.consistency_policy
== CONSISTENCY_POLICY_BITMAP
) {
9566 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9572 super
->current_vol
= u
->array
;
9573 if (st
->ss
->write_bitmap(st
, dl
->fd
, NoUpdate
))
9575 super
->current_vol
= prev_current_vol
;
9580 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9581 struct intel_super
*super
,
9582 struct active_array
*active_array
)
9584 struct imsm_super
*mpb
= super
->anchor
;
9585 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9586 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9587 struct imsm_map
*migr_map
;
9588 struct active_array
*a
;
9589 struct imsm_disk
*disk
;
9596 int second_map_created
= 0;
9598 for (; u
; u
= u
->next
) {
9599 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9604 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9609 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9614 /* count failures (excluding rebuilds and the victim)
9615 * to determine map[0] state
9618 for (i
= 0; i
< map
->num_members
; i
++) {
9621 disk
= get_imsm_disk(super
,
9622 get_imsm_disk_idx(dev
, i
, MAP_X
));
9623 if (!disk
|| is_failed(disk
))
9627 /* adding a pristine spare, assign a new index */
9628 if (dl
->index
< 0) {
9629 dl
->index
= super
->anchor
->num_disks
;
9630 super
->anchor
->num_disks
++;
9633 disk
->status
|= CONFIGURED_DISK
;
9634 disk
->status
&= ~SPARE_DISK
;
9637 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9638 if (!second_map_created
) {
9639 second_map_created
= 1;
9640 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9641 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9643 map
->map_state
= to_state
;
9644 migr_map
= get_imsm_map(dev
, MAP_1
);
9645 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9646 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9647 dl
->index
| IMSM_ORD_REBUILD
);
9649 /* update the family_num to mark a new container
9650 * generation, being careful to record the existing
9651 * family_num in orig_family_num to clean up after
9652 * earlier mdadm versions that neglected to set it.
9654 if (mpb
->orig_family_num
== 0)
9655 mpb
->orig_family_num
= mpb
->family_num
;
9656 mpb
->family_num
+= super
->random
;
9658 /* count arrays using the victim in the metadata */
9660 for (a
= active_array
; a
; a
= a
->next
) {
9661 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9662 map
= get_imsm_map(dev
, MAP_0
);
9664 if (get_imsm_disk_slot(map
, victim
) >= 0)
9668 /* delete the victim if it is no longer being
9674 /* We know that 'manager' isn't touching anything,
9675 * so it is safe to delete
9677 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9678 if ((*dlp
)->index
== victim
)
9681 /* victim may be on the missing list */
9683 for (dlp
= &super
->missing
; *dlp
;
9684 dlp
= &(*dlp
)->next
)
9685 if ((*dlp
)->index
== victim
)
9687 imsm_delete(super
, dlp
, victim
);
9694 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9695 struct intel_super
*super
,
9698 struct dl
*new_disk
;
9699 struct intel_dev
*id
;
9701 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9702 int disk_count
= u
->old_raid_disks
;
9703 void **tofree
= NULL
;
9704 int devices_to_reshape
= 1;
9705 struct imsm_super
*mpb
= super
->anchor
;
9707 unsigned int dev_id
;
9709 dprintf("(enter)\n");
9711 /* enable spares to use in array */
9712 for (i
= 0; i
< delta_disks
; i
++) {
9713 new_disk
= get_disk_super(super
,
9714 major(u
->new_disks
[i
]),
9715 minor(u
->new_disks
[i
]));
9716 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9717 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9718 new_disk
, new_disk
->index
);
9719 if (new_disk
== NULL
||
9720 (new_disk
->index
>= 0 &&
9721 new_disk
->index
< u
->old_raid_disks
))
9722 goto update_reshape_exit
;
9723 new_disk
->index
= disk_count
++;
9724 /* slot to fill in autolayout
9726 new_disk
->raiddisk
= new_disk
->index
;
9727 new_disk
->disk
.status
|=
9729 new_disk
->disk
.status
&= ~SPARE_DISK
;
9732 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9733 mpb
->num_raid_devs
);
9734 /* manage changes in volume
9736 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9737 void **sp
= *space_list
;
9738 struct imsm_dev
*newdev
;
9739 struct imsm_map
*newmap
, *oldmap
;
9741 for (id
= super
->devlist
; id
; id
= id
->next
) {
9742 if (id
->index
== dev_id
)
9751 /* Copy the dev, but not (all of) the map */
9752 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9753 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9754 newmap
= get_imsm_map(newdev
, MAP_0
);
9755 /* Copy the current map */
9756 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9757 /* update one device only
9759 if (devices_to_reshape
) {
9760 dprintf("imsm: modifying subdev: %i\n",
9762 devices_to_reshape
--;
9763 newdev
->vol
.migr_state
= 1;
9764 set_vol_curr_migr_unit(newdev
, 0);
9765 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9766 newmap
->num_members
= u
->new_raid_disks
;
9767 for (i
= 0; i
< delta_disks
; i
++) {
9768 set_imsm_ord_tbl_ent(newmap
,
9769 u
->old_raid_disks
+ i
,
9770 u
->old_raid_disks
+ i
);
9772 /* New map is correct, now need to save old map
9774 newmap
= get_imsm_map(newdev
, MAP_1
);
9775 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9777 imsm_set_array_size(newdev
, -1);
9780 sp
= (void **)id
->dev
;
9785 /* Clear migration record */
9786 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9789 *space_list
= tofree
;
9792 update_reshape_exit
:
9797 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9798 struct intel_super
*super
,
9801 struct imsm_dev
*dev
= NULL
;
9802 struct intel_dev
*dv
;
9803 struct imsm_dev
*dev_new
;
9804 struct imsm_map
*map
;
9808 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9809 if (dv
->index
== (unsigned int)u
->subarray
) {
9817 map
= get_imsm_map(dev
, MAP_0
);
9819 if (u
->direction
== R10_TO_R0
) {
9820 unsigned long long num_data_stripes
;
9822 /* Number of failed disks must be half of initial disk number */
9823 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9824 (map
->num_members
/ 2))
9827 /* iterate through devices to mark removed disks as spare */
9828 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9829 if (dm
->disk
.status
& FAILED_DISK
) {
9830 int idx
= dm
->index
;
9831 /* update indexes on the disk list */
9832 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9833 the index values will end up being correct.... NB */
9834 for (du
= super
->disks
; du
; du
= du
->next
)
9835 if (du
->index
> idx
)
9837 /* mark as spare disk */
9842 map
->num_members
= map
->num_members
/ 2;
9843 map
->map_state
= IMSM_T_STATE_NORMAL
;
9844 map
->num_domains
= 1;
9845 map
->raid_level
= 0;
9846 map
->failed_disk_num
= -1;
9847 num_data_stripes
= imsm_dev_size(dev
) / 2;
9848 num_data_stripes
/= map
->blocks_per_strip
;
9849 set_num_data_stripes(map
, num_data_stripes
);
9852 if (u
->direction
== R0_TO_R10
) {
9854 unsigned long long num_data_stripes
;
9856 /* update slots in current disk list */
9857 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9861 /* create new *missing* disks */
9862 for (i
= 0; i
< map
->num_members
; i
++) {
9863 space
= *space_list
;
9866 *space_list
= *space
;
9868 memcpy(du
, super
->disks
, sizeof(*du
));
9872 du
->index
= (i
* 2) + 1;
9873 sprintf((char *)du
->disk
.serial
,
9874 " MISSING_%d", du
->index
);
9875 sprintf((char *)du
->serial
,
9876 "MISSING_%d", du
->index
);
9877 du
->next
= super
->missing
;
9878 super
->missing
= du
;
9880 /* create new dev and map */
9881 space
= *space_list
;
9884 *space_list
= *space
;
9885 dev_new
= (void *)space
;
9886 memcpy(dev_new
, dev
, sizeof(*dev
));
9887 /* update new map */
9888 map
= get_imsm_map(dev_new
, MAP_0
);
9889 map
->num_members
= map
->num_members
* 2;
9890 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9891 map
->num_domains
= 2;
9892 map
->raid_level
= 1;
9893 num_data_stripes
= imsm_dev_size(dev
) / 2;
9894 num_data_stripes
/= map
->blocks_per_strip
;
9895 num_data_stripes
/= map
->num_domains
;
9896 set_num_data_stripes(map
, num_data_stripes
);
9898 /* replace dev<->dev_new */
9901 /* update disk order table */
9902 for (du
= super
->disks
; du
; du
= du
->next
)
9904 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9905 for (du
= super
->missing
; du
; du
= du
->next
)
9906 if (du
->index
>= 0) {
9907 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9908 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9914 static void imsm_process_update(struct supertype
*st
,
9915 struct metadata_update
*update
)
9918 * crack open the metadata_update envelope to find the update record
9919 * update can be one of:
9920 * update_reshape_container_disks - all the arrays in the container
9921 * are being reshaped to have more devices. We need to mark
9922 * the arrays for general migration and convert selected spares
9923 * into active devices.
9924 * update_activate_spare - a spare device has replaced a failed
9925 * device in an array, update the disk_ord_tbl. If this disk is
9926 * present in all member arrays then also clear the SPARE_DISK
9928 * update_create_array
9930 * update_rename_array
9931 * update_add_remove_disk
9933 struct intel_super
*super
= st
->sb
;
9934 struct imsm_super
*mpb
;
9935 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9937 /* update requires a larger buf but the allocation failed */
9938 if (super
->next_len
&& !super
->next_buf
) {
9939 super
->next_len
= 0;
9943 if (super
->next_buf
) {
9944 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9946 super
->len
= super
->next_len
;
9947 super
->buf
= super
->next_buf
;
9949 super
->next_len
= 0;
9950 super
->next_buf
= NULL
;
9953 mpb
= super
->anchor
;
9956 case update_general_migration_checkpoint
: {
9957 struct intel_dev
*id
;
9958 struct imsm_update_general_migration_checkpoint
*u
=
9959 (void *)update
->buf
;
9961 dprintf("called for update_general_migration_checkpoint\n");
9963 /* find device under general migration */
9964 for (id
= super
->devlist
; id
; id
= id
->next
) {
9965 if (is_gen_migration(id
->dev
)) {
9966 set_vol_curr_migr_unit(id
->dev
,
9968 super
->updates_pending
++;
9973 case update_takeover
: {
9974 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9975 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9976 imsm_update_version_info(super
);
9977 super
->updates_pending
++;
9982 case update_reshape_container_disks
: {
9983 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9984 if (apply_reshape_container_disks_update(
9985 u
, super
, &update
->space_list
))
9986 super
->updates_pending
++;
9989 case update_reshape_migration
: {
9990 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9991 if (apply_reshape_migration_update(
9992 u
, super
, &update
->space_list
))
9993 super
->updates_pending
++;
9996 case update_size_change
: {
9997 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9998 if (apply_size_change_update(u
, super
))
9999 super
->updates_pending
++;
10002 case update_activate_spare
: {
10003 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
10005 if (prepare_spare_to_activate(st
, u
) &&
10006 apply_update_activate_spare(u
, super
, st
->arrays
))
10007 super
->updates_pending
++;
10010 case update_create_array
: {
10011 /* someone wants to create a new array, we need to be aware of
10012 * a few races/collisions:
10013 * 1/ 'Create' called by two separate instances of mdadm
10014 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
10015 * devices that have since been assimilated via
10017 * In the event this update can not be carried out mdadm will
10018 * (FIX ME) notice that its update did not take hold.
10020 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10021 struct intel_dev
*dv
;
10022 struct imsm_dev
*dev
;
10023 struct imsm_map
*map
, *new_map
;
10024 unsigned long long start
, end
;
10025 unsigned long long new_start
, new_end
;
10027 struct disk_info
*inf
;
10030 /* handle racing creates: first come first serve */
10031 if (u
->dev_idx
< mpb
->num_raid_devs
) {
10032 dprintf("subarray %d already defined\n", u
->dev_idx
);
10036 /* check update is next in sequence */
10037 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
10038 dprintf("can not create array %d expected index %d\n",
10039 u
->dev_idx
, mpb
->num_raid_devs
);
10043 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
10044 new_start
= pba_of_lba0(new_map
);
10045 new_end
= new_start
+ per_dev_array_size(new_map
);
10046 inf
= get_disk_info(u
);
10048 /* handle activate_spare versus create race:
10049 * check to make sure that overlapping arrays do not include
10050 * overalpping disks
10052 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10053 dev
= get_imsm_dev(super
, i
);
10054 map
= get_imsm_map(dev
, MAP_0
);
10055 start
= pba_of_lba0(map
);
10056 end
= start
+ per_dev_array_size(map
);
10057 if ((new_start
>= start
&& new_start
<= end
) ||
10058 (start
>= new_start
&& start
<= new_end
))
10063 if (disks_overlap(super
, i
, u
)) {
10064 dprintf("arrays overlap\n");
10069 /* check that prepare update was successful */
10070 if (!update
->space
) {
10071 dprintf("prepare update failed\n");
10075 /* check that all disks are still active before committing
10076 * changes. FIXME: could we instead handle this by creating a
10077 * degraded array? That's probably not what the user expects,
10078 * so better to drop this update on the floor.
10080 for (i
= 0; i
< new_map
->num_members
; i
++) {
10081 dl
= serial_to_dl(inf
[i
].serial
, super
);
10083 dprintf("disk disappeared\n");
10088 super
->updates_pending
++;
10090 /* convert spares to members and fixup ord_tbl */
10091 for (i
= 0; i
< new_map
->num_members
; i
++) {
10092 dl
= serial_to_dl(inf
[i
].serial
, super
);
10093 if (dl
->index
== -1) {
10094 dl
->index
= mpb
->num_disks
;
10096 dl
->disk
.status
|= CONFIGURED_DISK
;
10097 dl
->disk
.status
&= ~SPARE_DISK
;
10099 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
10102 dv
= update
->space
;
10104 update
->space
= NULL
;
10105 imsm_copy_dev(dev
, &u
->dev
);
10106 dv
->index
= u
->dev_idx
;
10107 dv
->next
= super
->devlist
;
10108 super
->devlist
= dv
;
10109 mpb
->num_raid_devs
++;
10111 imsm_update_version_info(super
);
10114 /* mdmon knows how to release update->space, but not
10115 * ((struct intel_dev *) update->space)->dev
10117 if (update
->space
) {
10118 dv
= update
->space
;
10123 case update_kill_array
: {
10124 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
10125 int victim
= u
->dev_idx
;
10126 struct active_array
*a
;
10127 struct intel_dev
**dp
;
10128 struct imsm_dev
*dev
;
10130 /* sanity check that we are not affecting the uuid of
10131 * active arrays, or deleting an active array
10133 * FIXME when immutable ids are available, but note that
10134 * we'll also need to fixup the invalidated/active
10135 * subarray indexes in mdstat
10137 for (a
= st
->arrays
; a
; a
= a
->next
)
10138 if (a
->info
.container_member
>= victim
)
10140 /* by definition if mdmon is running at least one array
10141 * is active in the container, so checking
10142 * mpb->num_raid_devs is just extra paranoia
10144 dev
= get_imsm_dev(super
, victim
);
10145 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
10146 dprintf("failed to delete subarray-%d\n", victim
);
10150 for (dp
= &super
->devlist
; *dp
;)
10151 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
10154 if ((*dp
)->index
> (unsigned)victim
)
10158 mpb
->num_raid_devs
--;
10159 super
->updates_pending
++;
10162 case update_rename_array
: {
10163 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
10164 char name
[MAX_RAID_SERIAL_LEN
+1];
10165 int target
= u
->dev_idx
;
10166 struct active_array
*a
;
10167 struct imsm_dev
*dev
;
10169 /* sanity check that we are not affecting the uuid of
10172 memset(name
, 0, sizeof(name
));
10173 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
10174 name
[MAX_RAID_SERIAL_LEN
] = '\0';
10175 for (a
= st
->arrays
; a
; a
= a
->next
)
10176 if (a
->info
.container_member
== target
)
10178 dev
= get_imsm_dev(super
, u
->dev_idx
);
10179 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
10180 dprintf("failed to rename subarray-%d\n", target
);
10184 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
10185 super
->updates_pending
++;
10188 case update_add_remove_disk
: {
10189 /* we may be able to repair some arrays if disks are
10190 * being added, check the status of add_remove_disk
10191 * if discs has been added.
10193 if (add_remove_disk_update(super
)) {
10194 struct active_array
*a
;
10196 super
->updates_pending
++;
10197 for (a
= st
->arrays
; a
; a
= a
->next
)
10198 a
->check_degraded
= 1;
10202 case update_prealloc_badblocks_mem
:
10204 case update_rwh_policy
: {
10205 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
10206 int target
= u
->dev_idx
;
10207 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
10209 dprintf("could not find subarray-%d\n", target
);
10213 if (dev
->rwh_policy
!= u
->new_policy
) {
10214 dev
->rwh_policy
= u
->new_policy
;
10215 super
->updates_pending
++;
10220 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10224 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10226 static int imsm_prepare_update(struct supertype
*st
,
10227 struct metadata_update
*update
)
10230 * Allocate space to hold new disk entries, raid-device entries or a new
10231 * mpb if necessary. The manager synchronously waits for updates to
10232 * complete in the monitor, so new mpb buffers allocated here can be
10233 * integrated by the monitor thread without worrying about live pointers
10234 * in the manager thread.
10236 enum imsm_update_type type
;
10237 struct intel_super
*super
= st
->sb
;
10238 unsigned int sector_size
= super
->sector_size
;
10239 struct imsm_super
*mpb
= super
->anchor
;
10243 if (update
->len
< (int)sizeof(type
))
10246 type
= *(enum imsm_update_type
*) update
->buf
;
10249 case update_general_migration_checkpoint
:
10250 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10252 dprintf("called for update_general_migration_checkpoint\n");
10254 case update_takeover
: {
10255 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10256 if (update
->len
< (int)sizeof(*u
))
10258 if (u
->direction
== R0_TO_R10
) {
10259 void **tail
= (void **)&update
->space_list
;
10260 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10261 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10262 int num_members
= map
->num_members
;
10265 /* allocate memory for added disks */
10266 for (i
= 0; i
< num_members
; i
++) {
10267 size
= sizeof(struct dl
);
10268 space
= xmalloc(size
);
10273 /* allocate memory for new device */
10274 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10275 (num_members
* sizeof(__u32
));
10276 space
= xmalloc(size
);
10280 len
= disks_to_mpb_size(num_members
* 2);
10285 case update_reshape_container_disks
: {
10286 /* Every raid device in the container is about to
10287 * gain some more devices, and we will enter a
10289 * So each 'imsm_map' will be bigger, and the imsm_vol
10290 * will now hold 2 of them.
10291 * Thus we need new 'struct imsm_dev' allocations sized
10292 * as sizeof_imsm_dev but with more devices in both maps.
10294 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10295 struct intel_dev
*dl
;
10296 void **space_tail
= (void**)&update
->space_list
;
10298 if (update
->len
< (int)sizeof(*u
))
10301 dprintf("for update_reshape\n");
10303 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10304 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10306 if (u
->new_raid_disks
> u
->old_raid_disks
)
10307 size
+= sizeof(__u32
)*2*
10308 (u
->new_raid_disks
- u
->old_raid_disks
);
10312 *space_tail
= NULL
;
10315 len
= disks_to_mpb_size(u
->new_raid_disks
);
10316 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10319 case update_reshape_migration
: {
10320 /* for migration level 0->5 we need to add disks
10321 * so the same as for container operation we will copy
10322 * device to the bigger location.
10323 * in memory prepared device and new disk area are prepared
10324 * for usage in process update
10326 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10327 struct intel_dev
*id
;
10328 void **space_tail
= (void **)&update
->space_list
;
10331 int current_level
= -1;
10333 if (update
->len
< (int)sizeof(*u
))
10336 dprintf("for update_reshape\n");
10338 /* add space for bigger array in update
10340 for (id
= super
->devlist
; id
; id
= id
->next
) {
10341 if (id
->index
== (unsigned)u
->subdev
) {
10342 size
= sizeof_imsm_dev(id
->dev
, 1);
10343 if (u
->new_raid_disks
> u
->old_raid_disks
)
10344 size
+= sizeof(__u32
)*2*
10345 (u
->new_raid_disks
- u
->old_raid_disks
);
10349 *space_tail
= NULL
;
10353 if (update
->space_list
== NULL
)
10356 /* add space for disk in update
10358 size
= sizeof(struct dl
);
10362 *space_tail
= NULL
;
10364 /* add spare device to update
10366 for (id
= super
->devlist
; id
; id
= id
->next
)
10367 if (id
->index
== (unsigned)u
->subdev
) {
10368 struct imsm_dev
*dev
;
10369 struct imsm_map
*map
;
10371 dev
= get_imsm_dev(super
, u
->subdev
);
10372 map
= get_imsm_map(dev
, MAP_0
);
10373 current_level
= map
->raid_level
;
10376 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10377 struct mdinfo
*spares
;
10379 spares
= get_spares_for_grow(st
);
10382 struct mdinfo
*dev
;
10384 dev
= spares
->devs
;
10387 makedev(dev
->disk
.major
,
10389 dl
= get_disk_super(super
,
10392 dl
->index
= u
->old_raid_disks
;
10395 sysfs_free(spares
);
10398 len
= disks_to_mpb_size(u
->new_raid_disks
);
10399 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10402 case update_size_change
: {
10403 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10407 case update_activate_spare
: {
10408 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10412 case update_create_array
: {
10413 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10414 struct intel_dev
*dv
;
10415 struct imsm_dev
*dev
= &u
->dev
;
10416 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10418 struct disk_info
*inf
;
10422 if (update
->len
< (int)sizeof(*u
))
10425 inf
= get_disk_info(u
);
10426 len
= sizeof_imsm_dev(dev
, 1);
10427 /* allocate a new super->devlist entry */
10428 dv
= xmalloc(sizeof(*dv
));
10429 dv
->dev
= xmalloc(len
);
10430 update
->space
= dv
;
10432 /* count how many spares will be converted to members */
10433 for (i
= 0; i
< map
->num_members
; i
++) {
10434 dl
= serial_to_dl(inf
[i
].serial
, super
);
10436 /* hmm maybe it failed?, nothing we can do about
10441 if (count_memberships(dl
, super
) == 0)
10444 len
+= activate
* sizeof(struct imsm_disk
);
10447 case update_kill_array
: {
10448 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10452 case update_rename_array
: {
10453 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10457 case update_add_remove_disk
:
10458 /* no update->len needed */
10460 case update_prealloc_badblocks_mem
:
10461 super
->extra_space
+= sizeof(struct bbm_log
) -
10462 get_imsm_bbm_log_size(super
->bbm_log
);
10464 case update_rwh_policy
: {
10465 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10473 /* check if we need a larger metadata buffer */
10474 if (super
->next_buf
)
10475 buf_len
= super
->next_len
;
10477 buf_len
= super
->len
;
10479 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10480 /* ok we need a larger buf than what is currently allocated
10481 * if this allocation fails process_update will notice that
10482 * ->next_len is set and ->next_buf is NULL
10484 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10485 super
->extra_space
+ len
, sector_size
);
10486 if (super
->next_buf
)
10487 free(super
->next_buf
);
10489 super
->next_len
= buf_len
;
10490 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10491 memset(super
->next_buf
, 0, buf_len
);
10493 super
->next_buf
= NULL
;
10498 /* must be called while manager is quiesced */
10499 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10501 struct imsm_super
*mpb
= super
->anchor
;
10503 struct imsm_dev
*dev
;
10504 struct imsm_map
*map
;
10505 unsigned int i
, j
, num_members
;
10506 __u32 ord
, ord_map0
;
10507 struct bbm_log
*log
= super
->bbm_log
;
10509 dprintf("deleting device[%d] from imsm_super\n", index
);
10511 /* shift all indexes down one */
10512 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10513 if (iter
->index
> (int)index
)
10515 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10516 if (iter
->index
> (int)index
)
10519 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10520 dev
= get_imsm_dev(super
, i
);
10521 map
= get_imsm_map(dev
, MAP_0
);
10522 num_members
= map
->num_members
;
10523 for (j
= 0; j
< num_members
; j
++) {
10524 /* update ord entries being careful not to propagate
10525 * ord-flags to the first map
10527 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10528 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10530 if (ord_to_idx(ord
) <= index
)
10533 map
= get_imsm_map(dev
, MAP_0
);
10534 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10535 map
= get_imsm_map(dev
, MAP_1
);
10537 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10541 for (i
= 0; i
< log
->entry_count
; i
++) {
10542 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10544 if (entry
->disk_ordinal
<= index
)
10546 entry
->disk_ordinal
--;
10550 super
->updates_pending
++;
10552 struct dl
*dl
= *dlp
;
10554 *dlp
= (*dlp
)->next
;
10555 __free_imsm_disk(dl
);
10559 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10560 struct intel_super
*super
,
10561 struct imsm_dev
*dev
)
10567 struct imsm_map
*map
;
10570 ret_val
= raid_disks
/2;
10571 /* check map if all disks pairs not failed
10574 map
= get_imsm_map(dev
, MAP_0
);
10575 for (i
= 0; i
< ret_val
; i
++) {
10576 int degradation
= 0;
10577 if (get_imsm_disk(super
, i
) == NULL
)
10579 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10581 if (degradation
== 2)
10584 map
= get_imsm_map(dev
, MAP_1
);
10585 /* if there is no second map
10586 * result can be returned
10590 /* check degradation in second map
10592 for (i
= 0; i
< ret_val
; i
++) {
10593 int degradation
= 0;
10594 if (get_imsm_disk(super
, i
) == NULL
)
10596 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10598 if (degradation
== 2)
10612 /*******************************************************************************
10613 * Function: validate_container_imsm
10614 * Description: This routine validates container after assemble,
10615 * eg. if devices in container are under the same controller.
10618 * info : linked list with info about devices used in array
10622 ******************************************************************************/
10623 int validate_container_imsm(struct mdinfo
*info
)
10625 if (check_env("IMSM_NO_PLATFORM"))
10628 struct sys_dev
*idev
;
10629 struct sys_dev
*hba
= NULL
;
10630 struct sys_dev
*intel_devices
= find_intel_devices();
10631 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10632 info
->disk
.minor
), 1, NULL
);
10634 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10635 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10644 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10645 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10649 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10650 struct mdinfo
*dev
;
10652 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10653 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
,
10654 dev
->disk
.minor
), 1, NULL
);
10656 struct sys_dev
*hba2
= NULL
;
10657 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10658 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10666 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10667 get_orom_by_device_id(hba2
->dev_id
);
10669 if (hba2
&& hba
->type
!= hba2
->type
) {
10670 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10671 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10675 if (orom
!= orom2
) {
10676 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10677 " This operation is not supported and can lead to data loss.\n");
10682 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10683 " This operation is not supported and can lead to data loss.\n");
10691 /*******************************************************************************
10692 * Function: imsm_record_badblock
10693 * Description: This routine stores new bad block record in BBM log
10696 * a : array containing a bad block
10697 * slot : disk number containing a bad block
10698 * sector : bad block sector
10699 * length : bad block sectors range
10703 ******************************************************************************/
10704 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10705 unsigned long long sector
, int length
)
10707 struct intel_super
*super
= a
->container
->sb
;
10711 ord
= imsm_disk_slot_to_ord(a
, slot
);
10715 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10718 super
->updates_pending
++;
10722 /*******************************************************************************
10723 * Function: imsm_clear_badblock
10724 * Description: This routine clears bad block record from BBM log
10727 * a : array containing a bad block
10728 * slot : disk number containing a bad block
10729 * sector : bad block sector
10730 * length : bad block sectors range
10734 ******************************************************************************/
10735 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10736 unsigned long long sector
, int length
)
10738 struct intel_super
*super
= a
->container
->sb
;
10742 ord
= imsm_disk_slot_to_ord(a
, slot
);
10746 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10748 super
->updates_pending
++;
10752 /*******************************************************************************
10753 * Function: imsm_get_badblocks
10754 * Description: This routine get list of bad blocks for an array
10758 * slot : disk number
10760 * bb : structure containing bad blocks
10762 ******************************************************************************/
10763 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10765 int inst
= a
->info
.container_member
;
10766 struct intel_super
*super
= a
->container
->sb
;
10767 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10768 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10771 ord
= imsm_disk_slot_to_ord(a
, slot
);
10775 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10776 per_dev_array_size(map
), &super
->bb
);
10780 /*******************************************************************************
10781 * Function: examine_badblocks_imsm
10782 * Description: Prints list of bad blocks on a disk to the standard output
10785 * st : metadata handler
10786 * fd : open file descriptor for device
10787 * devname : device name
10791 ******************************************************************************/
10792 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10794 struct intel_super
*super
= st
->sb
;
10795 struct bbm_log
*log
= super
->bbm_log
;
10796 struct dl
*d
= NULL
;
10799 for (d
= super
->disks
; d
; d
= d
->next
) {
10800 if (strcmp(d
->devname
, devname
) == 0)
10804 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10805 pr_err("%s doesn't appear to be part of a raid array\n",
10812 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10814 for (i
= 0; i
< log
->entry_count
; i
++) {
10815 if (entry
[i
].disk_ordinal
== d
->index
) {
10816 unsigned long long sector
= __le48_to_cpu(
10817 &entry
[i
].defective_block_start
);
10818 int cnt
= entry
[i
].marked_count
+ 1;
10821 printf("Bad-blocks on %s:\n", devname
);
10825 printf("%20llu for %d sectors\n", sector
, cnt
);
10831 printf("No bad-blocks list configured on %s\n", devname
);
10835 /*******************************************************************************
10836 * Function: init_migr_record_imsm
10837 * Description: Function inits imsm migration record
10839 * super : imsm internal array info
10840 * dev : device under migration
10841 * info : general array info to find the smallest device
10844 ******************************************************************************/
10845 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10846 struct mdinfo
*info
)
10848 struct intel_super
*super
= st
->sb
;
10849 struct migr_record
*migr_rec
= super
->migr_rec
;
10850 int new_data_disks
;
10851 unsigned long long dsize
, dev_sectors
;
10852 long long unsigned min_dev_sectors
= -1LLU;
10853 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10854 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10855 unsigned long long num_migr_units
;
10856 unsigned long long array_blocks
;
10857 struct dl
*dl_disk
= NULL
;
10859 memset(migr_rec
, 0, sizeof(struct migr_record
));
10860 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10862 /* only ascending reshape supported now */
10863 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10865 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10866 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10867 migr_rec
->dest_depth_per_unit
*=
10868 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10869 new_data_disks
= imsm_num_data_members(map_dest
);
10870 migr_rec
->blocks_per_unit
=
10871 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10872 migr_rec
->dest_depth_per_unit
=
10873 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10874 array_blocks
= info
->component_size
* new_data_disks
;
10876 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10878 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10880 set_num_migr_units(migr_rec
, num_migr_units
);
10882 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10883 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10885 /* Find the smallest dev */
10886 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
10887 /* ignore spares in container */
10888 if (dl_disk
->index
< 0)
10890 get_dev_size(dl_disk
->fd
, NULL
, &dsize
);
10891 dev_sectors
= dsize
/ 512;
10892 if (dev_sectors
< min_dev_sectors
)
10893 min_dev_sectors
= dev_sectors
;
10895 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
10896 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10898 write_imsm_migr_rec(st
);
10903 /*******************************************************************************
10904 * Function: save_backup_imsm
10905 * Description: Function saves critical data stripes to Migration Copy Area
10906 * and updates the current migration unit status.
10907 * Use restore_stripes() to form a destination stripe,
10908 * and to write it to the Copy Area.
10910 * st : supertype information
10911 * dev : imsm device that backup is saved for
10912 * info : general array info
10913 * buf : input buffer
10914 * length : length of data to backup (blocks_per_unit)
10918 ******************************************************************************/
10919 int save_backup_imsm(struct supertype
*st
,
10920 struct imsm_dev
*dev
,
10921 struct mdinfo
*info
,
10926 struct intel_super
*super
= st
->sb
;
10927 unsigned long long *target_offsets
;
10930 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10931 int new_disks
= map_dest
->num_members
;
10932 int dest_layout
= 0;
10934 unsigned long long start
;
10935 int data_disks
= imsm_num_data_members(map_dest
);
10937 targets
= xmalloc(new_disks
* sizeof(int));
10939 for (i
= 0; i
< new_disks
; i
++) {
10940 struct dl
*dl_disk
= get_imsm_dl_disk(super
, i
);
10942 targets
[i
] = dl_disk
->fd
;
10945 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10947 start
= info
->reshape_progress
* 512;
10948 for (i
= 0; i
< new_disks
; i
++) {
10949 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
10950 /* move back copy area adderss, it will be moved forward
10951 * in restore_stripes() using start input variable
10953 target_offsets
[i
] -= start
/data_disks
;
10956 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10957 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10959 if (restore_stripes(targets
, /* list of dest devices */
10960 target_offsets
, /* migration record offsets */
10963 map_dest
->raid_level
,
10965 -1, /* source backup file descriptor */
10966 0, /* input buf offset
10967 * always 0 buf is already offseted */
10971 pr_err("Error restoring stripes\n");
10981 free(target_offsets
);
10986 /*******************************************************************************
10987 * Function: save_checkpoint_imsm
10988 * Description: Function called for current unit status update
10989 * in the migration record. It writes it to disk.
10991 * super : imsm internal array info
10992 * info : general array info
10996 * 2: failure, means no valid migration record
10997 * / no general migration in progress /
10998 ******************************************************************************/
10999 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
11001 struct intel_super
*super
= st
->sb
;
11002 unsigned long long blocks_per_unit
;
11003 unsigned long long curr_migr_unit
;
11005 if (load_imsm_migr_rec(super
) != 0) {
11006 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
11010 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
11011 if (blocks_per_unit
== 0) {
11012 dprintf("imsm: no migration in progress.\n");
11015 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
11016 /* check if array is alligned to copy area
11017 * if it is not alligned, add one to current migration unit value
11018 * this can happend on array reshape finish only
11020 if (info
->reshape_progress
% blocks_per_unit
)
11023 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
11024 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
11025 set_migr_dest_1st_member_lba(super
->migr_rec
,
11026 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
11028 if (write_imsm_migr_rec(st
) < 0) {
11029 dprintf("imsm: Cannot write migration record outside backup area\n");
11036 /*******************************************************************************
11037 * Function: recover_backup_imsm
11038 * Description: Function recovers critical data from the Migration Copy Area
11039 * while assembling an array.
11041 * super : imsm internal array info
11042 * info : general array info
11044 * 0 : success (or there is no data to recover)
11046 ******************************************************************************/
11047 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
11049 struct intel_super
*super
= st
->sb
;
11050 struct migr_record
*migr_rec
= super
->migr_rec
;
11051 struct imsm_map
*map_dest
;
11052 struct intel_dev
*id
= NULL
;
11053 unsigned long long read_offset
;
11054 unsigned long long write_offset
;
11056 int new_disks
, err
;
11059 unsigned int sector_size
= super
->sector_size
;
11060 unsigned long long curr_migr_unit
= current_migr_unit(migr_rec
);
11061 unsigned long long num_migr_units
= get_num_migr_units(migr_rec
);
11063 int skipped_disks
= 0;
11064 struct dl
*dl_disk
;
11066 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
11070 /* recover data only during assemblation */
11071 if (strncmp(buffer
, "inactive", 8) != 0)
11073 /* no data to recover */
11074 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
11076 if (curr_migr_unit
>= num_migr_units
)
11079 /* find device during reshape */
11080 for (id
= super
->devlist
; id
; id
= id
->next
)
11081 if (is_gen_migration(id
->dev
))
11086 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
11087 new_disks
= map_dest
->num_members
;
11089 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
11091 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
11092 pba_of_lba0(map_dest
)) * 512;
11094 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11095 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
11098 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
11099 if (dl_disk
->index
< 0)
11102 if (dl_disk
->fd
< 0) {
11106 if (lseek64(dl_disk
->fd
, read_offset
, SEEK_SET
) < 0) {
11107 pr_err("Cannot seek to block: %s\n",
11112 if (read(dl_disk
->fd
, buf
, unit_len
) != (ssize_t
)unit_len
) {
11113 pr_err("Cannot read copy area block: %s\n",
11118 if (lseek64(dl_disk
->fd
, write_offset
, SEEK_SET
) < 0) {
11119 pr_err("Cannot seek to block: %s\n",
11124 if (write(dl_disk
->fd
, buf
, unit_len
) != (ssize_t
)unit_len
) {
11125 pr_err("Cannot restore block: %s\n",
11132 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11136 pr_err("Cannot restore data from backup. Too many failed disks\n");
11140 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11141 /* ignore error == 2, this can mean end of reshape here
11143 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11152 static char disk_by_path
[] = "/dev/disk/by-path/";
11154 static const char *imsm_get_disk_controller_domain(const char *path
)
11156 char disk_path
[PATH_MAX
];
11160 strcpy(disk_path
, disk_by_path
);
11161 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
11162 if (stat(disk_path
, &st
) == 0) {
11163 struct sys_dev
* hba
;
11166 path
= devt_to_devpath(st
.st_rdev
, 1, NULL
);
11169 hba
= find_disk_attached_hba(-1, path
);
11170 if (hba
&& hba
->type
== SYS_DEV_SAS
)
11172 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
11174 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
11176 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
11180 dprintf("path: %s hba: %s attached: %s\n",
11181 path
, (hba
) ? hba
->path
: "NULL", drv
);
11187 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11189 static char devnm
[32];
11190 char subdev_name
[20];
11191 struct mdstat_ent
*mdstat
;
11193 sprintf(subdev_name
, "%d", subdev
);
11194 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11198 strcpy(devnm
, mdstat
->devnm
);
11199 free_mdstat(mdstat
);
11203 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11204 struct geo_params
*geo
,
11205 int *old_raid_disks
,
11208 /* currently we only support increasing the number of devices
11209 * for a container. This increases the number of device for each
11210 * member array. They must all be RAID0 or RAID5.
11213 struct mdinfo
*info
, *member
;
11214 int devices_that_can_grow
= 0;
11216 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11218 if (geo
->size
> 0 ||
11219 geo
->level
!= UnSet
||
11220 geo
->layout
!= UnSet
||
11221 geo
->chunksize
!= 0 ||
11222 geo
->raid_disks
== UnSet
) {
11223 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11227 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11228 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11232 info
= container_content_imsm(st
, NULL
);
11233 for (member
= info
; member
; member
= member
->next
) {
11236 dprintf("imsm: checking device_num: %i\n",
11237 member
->container_member
);
11239 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11240 /* we work on container for Online Capacity Expansion
11241 * only so raid_disks has to grow
11243 dprintf("imsm: for container operation raid disks increase is required\n");
11247 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11248 /* we cannot use this container with other raid level
11250 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11251 info
->array
.level
);
11254 /* check for platform support
11255 * for this raid level configuration
11257 struct intel_super
*super
= st
->sb
;
11258 if (!is_raid_level_supported(super
->orom
,
11259 member
->array
.level
,
11260 geo
->raid_disks
)) {
11261 dprintf("platform does not support raid%d with %d disk%s\n",
11264 geo
->raid_disks
> 1 ? "s" : "");
11267 /* check if component size is aligned to chunk size
11269 if (info
->component_size
%
11270 (info
->array
.chunk_size
/512)) {
11271 dprintf("Component size is not aligned to chunk size\n");
11276 if (*old_raid_disks
&&
11277 info
->array
.raid_disks
!= *old_raid_disks
)
11279 *old_raid_disks
= info
->array
.raid_disks
;
11281 /* All raid5 and raid0 volumes in container
11282 * have to be ready for Online Capacity Expansion
11283 * so they need to be assembled. We have already
11284 * checked that no recovery etc is happening.
11286 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11287 st
->container_devnm
);
11288 if (result
== NULL
) {
11289 dprintf("imsm: cannot find array\n");
11292 devices_that_can_grow
++;
11295 if (!member
&& devices_that_can_grow
)
11299 dprintf("Container operation allowed\n");
11301 dprintf("Error: %i\n", ret_val
);
11306 /* Function: get_spares_for_grow
11307 * Description: Allocates memory and creates list of spare devices
11308 * avaliable in container. Checks if spare drive size is acceptable.
11309 * Parameters: Pointer to the supertype structure
11310 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11313 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11315 struct spare_criteria sc
;
11317 get_spare_criteria_imsm(st
, &sc
);
11318 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11321 /******************************************************************************
11322 * function: imsm_create_metadata_update_for_reshape
11323 * Function creates update for whole IMSM container.
11325 ******************************************************************************/
11326 static int imsm_create_metadata_update_for_reshape(
11327 struct supertype
*st
,
11328 struct geo_params
*geo
,
11329 int old_raid_disks
,
11330 struct imsm_update_reshape
**updatep
)
11332 struct intel_super
*super
= st
->sb
;
11333 struct imsm_super
*mpb
= super
->anchor
;
11334 int update_memory_size
;
11335 struct imsm_update_reshape
*u
;
11336 struct mdinfo
*spares
;
11339 struct mdinfo
*dev
;
11341 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11343 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11345 /* size of all update data without anchor */
11346 update_memory_size
= sizeof(struct imsm_update_reshape
);
11348 /* now add space for spare disks that we need to add. */
11349 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11351 u
= xcalloc(1, update_memory_size
);
11352 u
->type
= update_reshape_container_disks
;
11353 u
->old_raid_disks
= old_raid_disks
;
11354 u
->new_raid_disks
= geo
->raid_disks
;
11356 /* now get spare disks list
11358 spares
= get_spares_for_grow(st
);
11360 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11361 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11366 /* we have got spares
11367 * update disk list in imsm_disk list table in anchor
11369 dprintf("imsm: %i spares are available.\n\n",
11370 spares
->array
.spare_disks
);
11372 dev
= spares
->devs
;
11373 for (i
= 0; i
< delta_disks
; i
++) {
11378 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11380 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11381 dl
->index
= mpb
->num_disks
;
11389 sysfs_free(spares
);
11391 dprintf("imsm: reshape update preparation :");
11392 if (i
== delta_disks
) {
11393 dprintf_cont(" OK\n");
11395 return update_memory_size
;
11398 dprintf_cont(" Error\n");
11403 /******************************************************************************
11404 * function: imsm_create_metadata_update_for_size_change()
11405 * Creates update for IMSM array for array size change.
11407 ******************************************************************************/
11408 static int imsm_create_metadata_update_for_size_change(
11409 struct supertype
*st
,
11410 struct geo_params
*geo
,
11411 struct imsm_update_size_change
**updatep
)
11413 struct intel_super
*super
= st
->sb
;
11414 int update_memory_size
;
11415 struct imsm_update_size_change
*u
;
11417 dprintf("(enter) New size = %llu\n", geo
->size
);
11419 /* size of all update data without anchor */
11420 update_memory_size
= sizeof(struct imsm_update_size_change
);
11422 u
= xcalloc(1, update_memory_size
);
11423 u
->type
= update_size_change
;
11424 u
->subdev
= super
->current_vol
;
11425 u
->new_size
= geo
->size
;
11427 dprintf("imsm: reshape update preparation : OK\n");
11430 return update_memory_size
;
11433 /******************************************************************************
11434 * function: imsm_create_metadata_update_for_migration()
11435 * Creates update for IMSM array.
11437 ******************************************************************************/
11438 static int imsm_create_metadata_update_for_migration(
11439 struct supertype
*st
,
11440 struct geo_params
*geo
,
11441 struct imsm_update_reshape_migration
**updatep
)
11443 struct intel_super
*super
= st
->sb
;
11444 int update_memory_size
;
11445 struct imsm_update_reshape_migration
*u
;
11446 struct imsm_dev
*dev
;
11447 int previous_level
= -1;
11449 dprintf("(enter) New Level = %i\n", geo
->level
);
11451 /* size of all update data without anchor */
11452 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11454 u
= xcalloc(1, update_memory_size
);
11455 u
->type
= update_reshape_migration
;
11456 u
->subdev
= super
->current_vol
;
11457 u
->new_level
= geo
->level
;
11458 u
->new_layout
= geo
->layout
;
11459 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11460 u
->new_disks
[0] = -1;
11461 u
->new_chunksize
= -1;
11463 dev
= get_imsm_dev(super
, u
->subdev
);
11465 struct imsm_map
*map
;
11467 map
= get_imsm_map(dev
, MAP_0
);
11469 int current_chunk_size
=
11470 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11472 if (geo
->chunksize
!= current_chunk_size
) {
11473 u
->new_chunksize
= geo
->chunksize
/ 1024;
11474 dprintf("imsm: chunk size change from %i to %i\n",
11475 current_chunk_size
, u
->new_chunksize
);
11477 previous_level
= map
->raid_level
;
11480 if (geo
->level
== 5 && previous_level
== 0) {
11481 struct mdinfo
*spares
= NULL
;
11483 u
->new_raid_disks
++;
11484 spares
= get_spares_for_grow(st
);
11485 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11487 sysfs_free(spares
);
11488 update_memory_size
= 0;
11489 pr_err("cannot get spare device for requested migration\n");
11492 sysfs_free(spares
);
11494 dprintf("imsm: reshape update preparation : OK\n");
11497 return update_memory_size
;
11500 static void imsm_update_metadata_locally(struct supertype
*st
,
11501 void *buf
, int len
)
11503 struct metadata_update mu
;
11508 mu
.space_list
= NULL
;
11510 if (imsm_prepare_update(st
, &mu
))
11511 imsm_process_update(st
, &mu
);
11513 while (mu
.space_list
) {
11514 void **space
= mu
.space_list
;
11515 mu
.space_list
= *space
;
11520 /***************************************************************************
11521 * Function: imsm_analyze_change
11522 * Description: Function analyze change for single volume
11523 * and validate if transition is supported
11524 * Parameters: Geometry parameters, supertype structure,
11525 * metadata change direction (apply/rollback)
11526 * Returns: Operation type code on success, -1 if fail
11527 ****************************************************************************/
11528 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11529 struct geo_params
*geo
,
11532 struct mdinfo info
;
11534 int check_devs
= 0;
11536 /* number of added/removed disks in operation result */
11537 int devNumChange
= 0;
11538 /* imsm compatible layout value for array geometry verification */
11539 int imsm_layout
= -1;
11541 struct imsm_dev
*dev
;
11542 struct imsm_map
*map
;
11543 struct intel_super
*super
;
11544 unsigned long long current_size
;
11545 unsigned long long free_size
;
11546 unsigned long long max_size
;
11549 getinfo_super_imsm_volume(st
, &info
, NULL
);
11550 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11551 geo
->level
!= UnSet
) {
11552 switch (info
.array
.level
) {
11554 if (geo
->level
== 5) {
11555 change
= CH_MIGRATION
;
11556 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11557 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11559 goto analyse_change_exit
;
11561 imsm_layout
= geo
->layout
;
11563 devNumChange
= 1; /* parity disk added */
11564 } else if (geo
->level
== 10) {
11565 change
= CH_TAKEOVER
;
11567 devNumChange
= 2; /* two mirrors added */
11568 imsm_layout
= 0x102; /* imsm supported layout */
11573 if (geo
->level
== 0) {
11574 change
= CH_TAKEOVER
;
11576 devNumChange
= -(geo
->raid_disks
/2);
11577 imsm_layout
= 0; /* imsm raid0 layout */
11581 if (change
== -1) {
11582 pr_err("Error. Level Migration from %d to %d not supported!\n",
11583 info
.array
.level
, geo
->level
);
11584 goto analyse_change_exit
;
11587 geo
->level
= info
.array
.level
;
11589 if (geo
->layout
!= info
.array
.layout
&&
11590 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11591 change
= CH_MIGRATION
;
11592 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11593 geo
->layout
== 5) {
11594 /* reshape 5 -> 4 */
11595 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11596 geo
->layout
== 0) {
11597 /* reshape 4 -> 5 */
11601 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11602 info
.array
.layout
, geo
->layout
);
11604 goto analyse_change_exit
;
11607 geo
->layout
= info
.array
.layout
;
11608 if (imsm_layout
== -1)
11609 imsm_layout
= info
.array
.layout
;
11612 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11613 geo
->chunksize
!= info
.array
.chunk_size
) {
11614 if (info
.array
.level
== 10) {
11615 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11617 goto analyse_change_exit
;
11618 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11619 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11620 geo
->chunksize
/1024, info
.component_size
/2);
11622 goto analyse_change_exit
;
11624 change
= CH_MIGRATION
;
11626 geo
->chunksize
= info
.array
.chunk_size
;
11629 chunk
= geo
->chunksize
/ 1024;
11632 dev
= get_imsm_dev(super
, super
->current_vol
);
11633 map
= get_imsm_map(dev
, MAP_0
);
11634 data_disks
= imsm_num_data_members(map
);
11635 /* compute current size per disk member
11637 current_size
= info
.custom_array_size
/ data_disks
;
11639 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11640 /* align component size
11642 geo
->size
= imsm_component_size_alignment_check(
11643 get_imsm_raid_level(dev
->vol
.map
),
11644 chunk
* 1024, super
->sector_size
,
11646 if (geo
->size
== 0) {
11647 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11649 goto analyse_change_exit
;
11653 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11654 if (change
!= -1) {
11655 pr_err("Error. Size change should be the only one at a time.\n");
11657 goto analyse_change_exit
;
11659 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11660 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11661 super
->current_vol
, st
->devnm
);
11662 goto analyse_change_exit
;
11664 /* check the maximum available size
11666 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11667 0, chunk
, &free_size
);
11669 /* Cannot find maximum available space
11673 max_size
= free_size
+ current_size
;
11674 /* align component size
11676 max_size
= imsm_component_size_alignment_check(
11677 get_imsm_raid_level(dev
->vol
.map
),
11678 chunk
* 1024, super
->sector_size
,
11681 if (geo
->size
== MAX_SIZE
) {
11682 /* requested size change to the maximum available size
11684 if (max_size
== 0) {
11685 pr_err("Error. Cannot find maximum available space.\n");
11687 goto analyse_change_exit
;
11689 geo
->size
= max_size
;
11692 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11693 /* accept size for rollback only
11696 /* round size due to metadata compatibility
11698 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11699 << SECT_PER_MB_SHIFT
;
11700 dprintf("Prepare update for size change to %llu\n",
11702 if (current_size
>= geo
->size
) {
11703 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11704 current_size
, geo
->size
);
11705 goto analyse_change_exit
;
11707 if (max_size
&& geo
->size
> max_size
) {
11708 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11709 max_size
, geo
->size
);
11710 goto analyse_change_exit
;
11713 geo
->size
*= data_disks
;
11714 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11715 change
= CH_ARRAY_SIZE
;
11717 if (!validate_geometry_imsm(st
,
11720 geo
->raid_disks
+ devNumChange
,
11722 geo
->size
, INVALID_SECTORS
,
11723 0, 0, info
.consistency_policy
, 1))
11727 struct intel_super
*super
= st
->sb
;
11728 struct imsm_super
*mpb
= super
->anchor
;
11730 if (mpb
->num_raid_devs
> 1) {
11731 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11737 analyse_change_exit
:
11738 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11739 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11740 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11746 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11748 struct intel_super
*super
= st
->sb
;
11749 struct imsm_update_takeover
*u
;
11751 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11753 u
->type
= update_takeover
;
11754 u
->subarray
= super
->current_vol
;
11756 /* 10->0 transition */
11757 if (geo
->level
== 0)
11758 u
->direction
= R10_TO_R0
;
11760 /* 0->10 transition */
11761 if (geo
->level
== 10)
11762 u
->direction
= R0_TO_R10
;
11764 /* update metadata locally */
11765 imsm_update_metadata_locally(st
, u
,
11766 sizeof(struct imsm_update_takeover
));
11767 /* and possibly remotely */
11768 if (st
->update_tail
)
11769 append_metadata_update(st
, u
,
11770 sizeof(struct imsm_update_takeover
));
11777 /* Flush size update if size calculated by num_data_stripes is higher than
11778 * imsm_dev_size to eliminate differences during reshape.
11779 * Mdmon will recalculate them correctly.
11780 * If subarray index is not set then check whole container.
11782 * 0 - no error occurred
11783 * 1 - error detected
11785 static int imsm_fix_size_mismatch(struct supertype
*st
, int subarray_index
)
11787 struct intel_super
*super
= st
->sb
;
11788 int tmp
= super
->current_vol
;
11792 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
11793 if (subarray_index
>= 0 && i
!= subarray_index
)
11795 super
->current_vol
= i
;
11796 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11797 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11798 unsigned int disc_count
= imsm_num_data_members(map
);
11799 struct geo_params geo
;
11800 struct imsm_update_size_change
*update
;
11801 unsigned long long calc_size
= per_dev_array_size(map
) * disc_count
;
11802 unsigned long long d_size
= imsm_dev_size(dev
);
11805 if (calc_size
== d_size
|| dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
11808 /* There is a difference, confirm that imsm_dev_size is
11809 * smaller and push update.
11811 if (d_size
> calc_size
) {
11812 pr_err("imsm: dev size of subarray %d is incorrect\n",
11816 memset(&geo
, 0, sizeof(struct geo_params
));
11818 u_size
= imsm_create_metadata_update_for_size_change(st
, &geo
,
11821 dprintf("imsm: Cannot prepare size change update\n");
11824 imsm_update_metadata_locally(st
, update
, u_size
);
11825 if (st
->update_tail
) {
11826 append_metadata_update(st
, update
, u_size
);
11827 flush_metadata_updates(st
);
11828 st
->update_tail
= &st
->updates
;
11830 imsm_sync_metadata(st
);
11835 super
->current_vol
= tmp
;
11839 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11841 int layout
, int chunksize
, int raid_disks
,
11842 int delta_disks
, char *backup
, char *dev
,
11843 int direction
, int verbose
)
11846 struct geo_params geo
;
11848 dprintf("(enter)\n");
11850 memset(&geo
, 0, sizeof(struct geo_params
));
11852 geo
.dev_name
= dev
;
11853 strcpy(geo
.devnm
, st
->devnm
);
11856 geo
.layout
= layout
;
11857 geo
.chunksize
= chunksize
;
11858 geo
.raid_disks
= raid_disks
;
11859 if (delta_disks
!= UnSet
)
11860 geo
.raid_disks
+= delta_disks
;
11862 dprintf("for level : %i\n", geo
.level
);
11863 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11865 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11866 /* On container level we can only increase number of devices. */
11867 dprintf("imsm: info: Container operation\n");
11868 int old_raid_disks
= 0;
11870 if (imsm_reshape_is_allowed_on_container(
11871 st
, &geo
, &old_raid_disks
, direction
)) {
11872 struct imsm_update_reshape
*u
= NULL
;
11875 if (imsm_fix_size_mismatch(st
, -1)) {
11876 dprintf("imsm: Cannot fix size mismatch\n");
11877 goto exit_imsm_reshape_super
;
11880 len
= imsm_create_metadata_update_for_reshape(
11881 st
, &geo
, old_raid_disks
, &u
);
11884 dprintf("imsm: Cannot prepare update\n");
11885 goto exit_imsm_reshape_super
;
11889 /* update metadata locally */
11890 imsm_update_metadata_locally(st
, u
, len
);
11891 /* and possibly remotely */
11892 if (st
->update_tail
)
11893 append_metadata_update(st
, u
, len
);
11898 pr_err("(imsm) Operation is not allowed on this container\n");
11901 /* On volume level we support following operations
11902 * - takeover: raid10 -> raid0; raid0 -> raid10
11903 * - chunk size migration
11904 * - migration: raid5 -> raid0; raid0 -> raid5
11906 struct intel_super
*super
= st
->sb
;
11907 struct intel_dev
*dev
= super
->devlist
;
11909 dprintf("imsm: info: Volume operation\n");
11910 /* find requested device */
11913 imsm_find_array_devnm_by_subdev(
11914 dev
->index
, st
->container_devnm
);
11915 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11920 pr_err("Cannot find %s (%s) subarray\n",
11921 geo
.dev_name
, geo
.devnm
);
11922 goto exit_imsm_reshape_super
;
11924 super
->current_vol
= dev
->index
;
11925 change
= imsm_analyze_change(st
, &geo
, direction
);
11928 ret_val
= imsm_takeover(st
, &geo
);
11930 case CH_MIGRATION
: {
11931 struct imsm_update_reshape_migration
*u
= NULL
;
11933 imsm_create_metadata_update_for_migration(
11936 dprintf("imsm: Cannot prepare update\n");
11940 /* update metadata locally */
11941 imsm_update_metadata_locally(st
, u
, len
);
11942 /* and possibly remotely */
11943 if (st
->update_tail
)
11944 append_metadata_update(st
, u
, len
);
11949 case CH_ARRAY_SIZE
: {
11950 struct imsm_update_size_change
*u
= NULL
;
11952 imsm_create_metadata_update_for_size_change(
11955 dprintf("imsm: Cannot prepare update\n");
11959 /* update metadata locally */
11960 imsm_update_metadata_locally(st
, u
, len
);
11961 /* and possibly remotely */
11962 if (st
->update_tail
)
11963 append_metadata_update(st
, u
, len
);
11973 exit_imsm_reshape_super
:
11974 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11978 #define COMPLETED_OK 0
11979 #define COMPLETED_NONE 1
11980 #define COMPLETED_DELAYED 2
11982 static int read_completed(int fd
, unsigned long long *val
)
11987 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11991 ret
= COMPLETED_OK
;
11992 if (strncmp(buf
, "none", 4) == 0) {
11993 ret
= COMPLETED_NONE
;
11994 } else if (strncmp(buf
, "delayed", 7) == 0) {
11995 ret
= COMPLETED_DELAYED
;
11998 *val
= strtoull(buf
, &ep
, 0);
11999 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
12005 /*******************************************************************************
12006 * Function: wait_for_reshape_imsm
12007 * Description: Function writes new sync_max value and waits until
12008 * reshape process reach new position
12010 * sra : general array info
12011 * ndata : number of disks in new array's layout
12014 * 1 : there is no reshape in progress,
12016 ******************************************************************************/
12017 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
12019 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
12021 unsigned long long completed
;
12022 /* to_complete : new sync_max position */
12023 unsigned long long to_complete
= sra
->reshape_progress
;
12024 unsigned long long position_to_set
= to_complete
/ ndata
;
12027 dprintf("cannot open reshape_position\n");
12032 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
12034 dprintf("cannot read reshape_position (no reshape in progres)\n");
12043 if (completed
> position_to_set
) {
12044 dprintf("wrong next position to set %llu (%llu)\n",
12045 to_complete
, position_to_set
);
12049 dprintf("Position set: %llu\n", position_to_set
);
12050 if (sysfs_set_num(sra
, NULL
, "sync_max",
12051 position_to_set
) != 0) {
12052 dprintf("cannot set reshape position to %llu\n",
12061 int timeout
= 3000;
12063 sysfs_wait(fd
, &timeout
);
12064 if (sysfs_get_str(sra
, NULL
, "sync_action",
12066 strncmp(action
, "reshape", 7) != 0) {
12067 if (strncmp(action
, "idle", 4) == 0)
12073 rc
= read_completed(fd
, &completed
);
12075 dprintf("cannot read reshape_position (in loop)\n");
12078 } else if (rc
== COMPLETED_NONE
)
12080 } while (completed
< position_to_set
);
12086 /*******************************************************************************
12087 * Function: check_degradation_change
12088 * Description: Check that array hasn't become failed.
12090 * info : for sysfs access
12091 * sources : source disks descriptors
12092 * degraded: previous degradation level
12094 * degradation level
12095 ******************************************************************************/
12096 int check_degradation_change(struct mdinfo
*info
,
12100 unsigned long long new_degraded
;
12103 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
12104 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
12105 /* check each device to ensure it is still working */
12108 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
12109 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
12111 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
12114 if (sysfs_get_str(info
,
12115 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
12116 strstr(sbuf
, "faulty") ||
12117 strstr(sbuf
, "in_sync") == NULL
) {
12118 /* this device is dead */
12119 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
12120 if (sd
->disk
.raid_disk
>= 0 &&
12121 sources
[sd
->disk
.raid_disk
] >= 0) {
12123 sd
->disk
.raid_disk
]);
12124 sources
[sd
->disk
.raid_disk
] =
12133 return new_degraded
;
12136 /*******************************************************************************
12137 * Function: imsm_manage_reshape
12138 * Description: Function finds array under reshape and it manages reshape
12139 * process. It creates stripes backups (if required) and sets
12142 * afd : Backup handle (nattive) - not used
12143 * sra : general array info
12144 * reshape : reshape parameters - not used
12145 * st : supertype structure
12146 * blocks : size of critical section [blocks]
12147 * fds : table of source device descriptor
12148 * offsets : start of array (offest per devices)
12150 * destfd : table of destination device descriptor
12151 * destoffsets : table of destination offsets (per device)
12153 * 1 : success, reshape is done
12155 ******************************************************************************/
12156 static int imsm_manage_reshape(
12157 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
12158 struct supertype
*st
, unsigned long backup_blocks
,
12159 int *fds
, unsigned long long *offsets
,
12160 int dests
, int *destfd
, unsigned long long *destoffsets
)
12163 struct intel_super
*super
= st
->sb
;
12164 struct intel_dev
*dv
;
12165 unsigned int sector_size
= super
->sector_size
;
12166 struct imsm_dev
*dev
= NULL
;
12167 struct imsm_map
*map_src
, *map_dest
;
12168 int migr_vol_qan
= 0;
12169 int ndata
, odata
; /* [bytes] */
12170 int chunk
; /* [bytes] */
12171 struct migr_record
*migr_rec
;
12173 unsigned int buf_size
; /* [bytes] */
12174 unsigned long long max_position
; /* array size [bytes] */
12175 unsigned long long next_step
; /* [blocks]/[bytes] */
12176 unsigned long long old_data_stripe_length
;
12177 unsigned long long start_src
; /* [bytes] */
12178 unsigned long long start
; /* [bytes] */
12179 unsigned long long start_buf_shift
; /* [bytes] */
12181 int source_layout
= 0;
12182 int subarray_index
= -1;
12187 if (!fds
|| !offsets
)
12190 /* Find volume during the reshape */
12191 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
12192 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12193 dv
->dev
->vol
.migr_state
== 1) {
12196 subarray_index
= dv
->index
;
12199 /* Only one volume can migrate at the same time */
12200 if (migr_vol_qan
!= 1) {
12201 pr_err("%s", migr_vol_qan
?
12202 "Number of migrating volumes greater than 1\n" :
12203 "There is no volume during migrationg\n");
12207 map_dest
= get_imsm_map(dev
, MAP_0
);
12208 map_src
= get_imsm_map(dev
, MAP_1
);
12209 if (map_src
== NULL
)
12212 ndata
= imsm_num_data_members(map_dest
);
12213 odata
= imsm_num_data_members(map_src
);
12215 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12216 old_data_stripe_length
= odata
* chunk
;
12218 migr_rec
= super
->migr_rec
;
12220 /* initialize migration record for start condition */
12221 if (sra
->reshape_progress
== 0)
12222 init_migr_record_imsm(st
, dev
, sra
);
12224 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12225 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12228 /* Save checkpoint to update migration record for current
12229 * reshape position (in md). It can be farther than current
12230 * reshape position in metadata.
12232 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12233 /* ignore error == 2, this can mean end of reshape here
12235 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12240 /* size for data */
12241 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12242 /* extend buffer size for parity disk */
12243 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12244 /* add space for stripe alignment */
12245 buf_size
+= old_data_stripe_length
;
12246 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12247 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12251 max_position
= sra
->component_size
* ndata
;
12252 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12254 while (current_migr_unit(migr_rec
) <
12255 get_num_migr_units(migr_rec
)) {
12256 /* current reshape position [blocks] */
12257 unsigned long long current_position
=
12258 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12259 * current_migr_unit(migr_rec
);
12260 unsigned long long border
;
12262 /* Check that array hasn't become failed.
12264 degraded
= check_degradation_change(sra
, fds
, degraded
);
12265 if (degraded
> 1) {
12266 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12270 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12272 if ((current_position
+ next_step
) > max_position
)
12273 next_step
= max_position
- current_position
;
12275 start
= current_position
* 512;
12277 /* align reading start to old geometry */
12278 start_buf_shift
= start
% old_data_stripe_length
;
12279 start_src
= start
- start_buf_shift
;
12281 border
= (start_src
/ odata
) - (start
/ ndata
);
12283 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12284 /* save critical stripes to buf
12285 * start - start address of current unit
12286 * to backup [bytes]
12287 * start_src - start address of current unit
12288 * to backup alligned to source array
12291 unsigned long long next_step_filler
;
12292 unsigned long long copy_length
= next_step
* 512;
12294 /* allign copy area length to stripe in old geometry */
12295 next_step_filler
= ((copy_length
+ start_buf_shift
)
12296 % old_data_stripe_length
);
12297 if (next_step_filler
)
12298 next_step_filler
= (old_data_stripe_length
12299 - next_step_filler
);
12300 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12301 start
, start_src
, copy_length
,
12302 start_buf_shift
, next_step_filler
);
12304 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12305 chunk
, map_src
->raid_level
,
12306 source_layout
, 0, NULL
, start_src
,
12308 next_step_filler
+ start_buf_shift
,
12310 dprintf("imsm: Cannot save stripes to buffer\n");
12313 /* Convert data to destination format and store it
12314 * in backup general migration area
12316 if (save_backup_imsm(st
, dev
, sra
,
12317 buf
+ start_buf_shift
, copy_length
)) {
12318 dprintf("imsm: Cannot save stripes to target devices\n");
12321 if (save_checkpoint_imsm(st
, sra
,
12322 UNIT_SRC_IN_CP_AREA
)) {
12323 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12327 /* set next step to use whole border area */
12328 border
/= next_step
;
12330 next_step
*= border
;
12332 /* When data backed up, checkpoint stored,
12333 * kick the kernel to reshape unit of data
12335 next_step
= next_step
+ sra
->reshape_progress
;
12336 /* limit next step to array max position */
12337 if (next_step
> max_position
)
12338 next_step
= max_position
;
12339 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12340 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12341 sra
->reshape_progress
= next_step
;
12343 /* wait until reshape finish */
12344 if (wait_for_reshape_imsm(sra
, ndata
)) {
12345 dprintf("wait_for_reshape_imsm returned error!\n");
12351 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12352 /* ignore error == 2, this can mean end of reshape here
12354 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12360 /* clear migr_rec on disks after successful migration */
12363 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12364 for (d
= super
->disks
; d
; d
= d
->next
) {
12365 if (d
->index
< 0 || is_failed(&d
->disk
))
12367 unsigned long long dsize
;
12369 get_dev_size(d
->fd
, NULL
, &dsize
);
12370 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12372 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12373 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12374 MIGR_REC_BUF_SECTORS
*sector_size
)
12375 perror("Write migr_rec failed");
12379 /* return '1' if done */
12382 /* After the reshape eliminate size mismatch in metadata.
12383 * Don't update md/component_size here, volume hasn't
12384 * to take whole space. It is allowed by kernel.
12385 * md/component_size will be set propoperly after next assembly.
12387 imsm_fix_size_mismatch(st
, subarray_index
);
12391 /* See Grow.c: abort_reshape() for further explanation */
12392 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12393 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12394 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12399 /*******************************************************************************
12400 * Function: calculate_bitmap_min_chunksize
12401 * Description: Calculates the minimal valid bitmap chunk size
12403 * max_bits : indicate how many bits can be used for the bitmap
12404 * data_area_size : the size of the data area covered by the bitmap
12407 * The bitmap chunk size
12408 ******************************************************************************/
12409 static unsigned long long
12410 calculate_bitmap_min_chunksize(unsigned long long max_bits
,
12411 unsigned long long data_area_size
)
12413 unsigned long long min_chunk
=
12414 4096; /* sub-page chunks don't work yet.. */
12415 unsigned long long bits
= data_area_size
/ min_chunk
+ 1;
12417 while (bits
> max_bits
) {
12419 bits
= (bits
+ 1) / 2;
12424 /*******************************************************************************
12425 * Function: calculate_bitmap_chunksize
12426 * Description: Calculates the bitmap chunk size for the given device
12428 * st : supertype information
12429 * dev : device for the bitmap
12432 * The bitmap chunk size
12433 ******************************************************************************/
12434 static unsigned long long calculate_bitmap_chunksize(struct supertype
*st
,
12435 struct imsm_dev
*dev
)
12437 struct intel_super
*super
= st
->sb
;
12438 unsigned long long min_chunksize
;
12439 unsigned long long result
= IMSM_DEFAULT_BITMAP_CHUNKSIZE
;
12440 size_t dev_size
= imsm_dev_size(dev
);
12442 min_chunksize
= calculate_bitmap_min_chunksize(
12443 IMSM_BITMAP_AREA_SIZE
* super
->sector_size
, dev_size
);
12445 if (result
< min_chunksize
)
12446 result
= min_chunksize
;
12451 /*******************************************************************************
12452 * Function: init_bitmap_header
12453 * Description: Initialize the bitmap header structure
12455 * st : supertype information
12456 * bms : bitmap header struct to initialize
12457 * dev : device for the bitmap
12462 ******************************************************************************/
12463 static int init_bitmap_header(struct supertype
*st
, struct bitmap_super_s
*bms
,
12464 struct imsm_dev
*dev
)
12471 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
12472 bms
->version
= __cpu_to_le32(BITMAP_MAJOR_HI
);
12473 bms
->daemon_sleep
= __cpu_to_le32(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
);
12474 bms
->sync_size
= __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
);
12475 bms
->write_behind
= __cpu_to_le32(0);
12477 uuid_from_super_imsm(st
, vol_uuid
);
12478 memcpy(bms
->uuid
, vol_uuid
, 16);
12480 bms
->chunksize
= calculate_bitmap_chunksize(st
, dev
);
12485 /*******************************************************************************
12486 * Function: validate_internal_bitmap_for_drive
12487 * Description: Verify if the bitmap header for a given drive.
12489 * st : supertype information
12490 * offset : The offset from the beginning of the drive where to look for
12491 * the bitmap header.
12492 * d : the drive info
12497 ******************************************************************************/
12498 static int validate_internal_bitmap_for_drive(struct supertype
*st
,
12499 unsigned long long offset
,
12502 struct intel_super
*super
= st
->sb
;
12505 bitmap_super_t
*bms
;
12513 if (posix_memalign(&read_buf
, MAX_SECTOR_SIZE
, IMSM_BITMAP_HEADER_SIZE
))
12518 fd
= open(d
->devname
, O_RDONLY
, 0);
12520 dprintf("cannot open the device %s\n", d
->devname
);
12525 if (lseek64(fd
, offset
* super
->sector_size
, SEEK_SET
) < 0)
12527 if (read(fd
, read_buf
, IMSM_BITMAP_HEADER_SIZE
) !=
12528 IMSM_BITMAP_HEADER_SIZE
)
12531 uuid_from_super_imsm(st
, vol_uuid
);
12534 if ((bms
->magic
!= __cpu_to_le32(BITMAP_MAGIC
)) ||
12535 (bms
->version
!= __cpu_to_le32(BITMAP_MAJOR_HI
)) ||
12536 (!same_uuid((int *)bms
->uuid
, vol_uuid
, st
->ss
->swapuuid
))) {
12537 dprintf("wrong bitmap header detected\n");
12543 if ((d
->fd
< 0) && (fd
>= 0))
12551 /*******************************************************************************
12552 * Function: validate_internal_bitmap_imsm
12553 * Description: Verify if the bitmap header is in place and with proper data.
12555 * st : supertype information
12558 * 0 : success or device w/o RWH_BITMAP
12560 ******************************************************************************/
12561 static int validate_internal_bitmap_imsm(struct supertype
*st
)
12563 struct intel_super
*super
= st
->sb
;
12564 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
12565 unsigned long long offset
;
12571 if (dev
->rwh_policy
!= RWH_BITMAP
)
12574 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12575 for (d
= super
->disks
; d
; d
= d
->next
) {
12576 if (d
->index
< 0 || is_failed(&d
->disk
))
12579 if (validate_internal_bitmap_for_drive(st
, offset
, d
)) {
12580 pr_err("imsm: bitmap validation failed\n");
12587 /*******************************************************************************
12588 * Function: add_internal_bitmap_imsm
12589 * Description: Mark the volume to use the bitmap and updates the chunk size value.
12591 * st : supertype information
12592 * chunkp : bitmap chunk size
12593 * delay : not used for imsm
12594 * write_behind : not used for imsm
12595 * size : not used for imsm
12596 * may_change : not used for imsm
12597 * amajor : not used for imsm
12602 ******************************************************************************/
12603 static int add_internal_bitmap_imsm(struct supertype
*st
, int *chunkp
,
12604 int delay
, int write_behind
,
12605 unsigned long long size
, int may_change
,
12608 struct intel_super
*super
= st
->sb
;
12609 int vol_idx
= super
->current_vol
;
12610 struct imsm_dev
*dev
;
12612 if (!super
->devlist
|| vol_idx
== -1 || !chunkp
)
12615 dev
= get_imsm_dev(super
, vol_idx
);
12618 dprintf("cannot find the device for volume index %d\n",
12622 dev
->rwh_policy
= RWH_BITMAP
;
12624 *chunkp
= calculate_bitmap_chunksize(st
, dev
);
12629 /*******************************************************************************
12630 * Function: locate_bitmap_imsm
12631 * Description: Seek 'fd' to start of write-intent-bitmap.
12633 * st : supertype information
12634 * fd : file descriptor for the device
12635 * node_num : not used for imsm
12640 ******************************************************************************/
12641 static int locate_bitmap_imsm(struct supertype
*st
, int fd
, int node_num
)
12643 struct intel_super
*super
= st
->sb
;
12644 unsigned long long offset
;
12645 int vol_idx
= super
->current_vol
;
12647 if (!super
->devlist
|| vol_idx
== -1)
12650 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12651 dprintf("bitmap header offset is %llu\n", offset
);
12653 lseek64(fd
, offset
<< 9, 0);
12658 /*******************************************************************************
12659 * Function: write_init_bitmap_imsm
12660 * Description: Write a bitmap header and prepares the area for the bitmap.
12662 * st : supertype information
12663 * fd : file descriptor for the device
12664 * update : not used for imsm
12669 ******************************************************************************/
12670 static int write_init_bitmap_imsm(struct supertype
*st
, int fd
,
12671 enum bitmap_update update
)
12673 struct intel_super
*super
= st
->sb
;
12674 int vol_idx
= super
->current_vol
;
12676 unsigned long long offset
;
12677 bitmap_super_t bms
= { 0 };
12678 size_t written
= 0;
12683 if (!super
->devlist
|| !super
->sector_size
|| vol_idx
== -1)
12686 struct imsm_dev
*dev
= get_imsm_dev(super
, vol_idx
);
12688 /* first clear the space for bitmap header */
12689 unsigned long long bitmap_area_start
=
12690 get_bitmap_header_sector(super
, vol_idx
);
12692 dprintf("zeroing area start (%llu) and size (%u)\n", bitmap_area_start
,
12693 IMSM_BITMAP_AND_HEADER_SIZE
/ super
->sector_size
);
12694 if (zero_disk_range(fd
, bitmap_area_start
,
12695 IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
)) {
12696 pr_err("imsm: cannot zeroing the space for the bitmap\n");
12700 /* The bitmap area should be filled with "1"s to perform initial
12703 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
))
12705 memset(buf
, 0xFF, MAX_SECTOR_SIZE
);
12706 offset
= get_bitmap_sector(super
, vol_idx
);
12707 lseek64(fd
, offset
<< 9, 0);
12708 while (written
< IMSM_BITMAP_AREA_SIZE
) {
12709 to_write
= IMSM_BITMAP_AREA_SIZE
- written
;
12710 if (to_write
> MAX_SECTOR_SIZE
)
12711 to_write
= MAX_SECTOR_SIZE
;
12712 rv_num
= write(fd
, buf
, MAX_SECTOR_SIZE
);
12713 if (rv_num
!= MAX_SECTOR_SIZE
) {
12715 dprintf("cannot initialize bitmap area\n");
12721 /* write a bitmap header */
12722 init_bitmap_header(st
, &bms
, dev
);
12723 memset(buf
, 0, MAX_SECTOR_SIZE
);
12724 memcpy(buf
, &bms
, sizeof(bitmap_super_t
));
12725 if (locate_bitmap_imsm(st
, fd
, 0)) {
12727 dprintf("cannot locate the bitmap\n");
12730 if (write(fd
, buf
, MAX_SECTOR_SIZE
) != MAX_SECTOR_SIZE
) {
12732 dprintf("cannot write the bitmap header\n");
12743 /*******************************************************************************
12744 * Function: is_vol_to_setup_bitmap
12745 * Description: Checks if a bitmap should be activated on the dev.
12747 * info : info about the volume to setup the bitmap
12748 * dev : the device to check against bitmap creation
12751 * 0 : bitmap should be set up on the device
12753 ******************************************************************************/
12754 static int is_vol_to_setup_bitmap(struct mdinfo
*info
, struct imsm_dev
*dev
)
12759 if ((strcmp((char *)dev
->volume
, info
->name
) == 0) &&
12760 (dev
->rwh_policy
== RWH_BITMAP
))
12766 /*******************************************************************************
12767 * Function: set_bitmap_sysfs
12768 * Description: Set the sysfs atributes of a given volume to activate the bitmap.
12770 * info : info about the volume where the bitmap should be setup
12771 * chunksize : bitmap chunk size
12772 * location : location of the bitmap
12777 ******************************************************************************/
12778 static int set_bitmap_sysfs(struct mdinfo
*info
, unsigned long long chunksize
,
12781 /* The bitmap/metadata is set to external to allow changing of value for
12782 * bitmap/location. When external is used, the kernel will treat an offset
12783 * related to the device's first lba (in opposition to the "internal" case
12784 * when this value is related to the beginning of the superblock).
12786 if (sysfs_set_str(info
, NULL
, "bitmap/metadata", "external")) {
12787 dprintf("failed to set bitmap/metadata\n");
12791 /* It can only be changed when no bitmap is active.
12792 * Should be bigger than 512 and must be power of 2.
12793 * It is expecting the value in bytes.
12795 if (sysfs_set_num(info
, NULL
, "bitmap/chunksize",
12796 __cpu_to_le32(chunksize
))) {
12797 dprintf("failed to set bitmap/chunksize\n");
12801 /* It is expecting the value in sectors. */
12802 if (sysfs_set_num(info
, NULL
, "bitmap/space",
12803 __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
))) {
12804 dprintf("failed to set bitmap/space\n");
12808 /* Determines the delay between the bitmap updates.
12809 * It is expecting the value in seconds.
12811 if (sysfs_set_num(info
, NULL
, "bitmap/time_base",
12812 __cpu_to_le64(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
))) {
12813 dprintf("failed to set bitmap/time_base\n");
12817 /* It is expecting the value in sectors with a sign at the beginning. */
12818 if (sysfs_set_str(info
, NULL
, "bitmap/location", location
)) {
12819 dprintf("failed to set bitmap/location\n");
12826 /*******************************************************************************
12827 * Function: set_bitmap_imsm
12828 * Description: Setup the bitmap for the given volume
12830 * st : supertype information
12831 * info : info about the volume where the bitmap should be setup
12836 ******************************************************************************/
12837 static int set_bitmap_imsm(struct supertype
*st
, struct mdinfo
*info
)
12839 struct intel_super
*super
= st
->sb
;
12840 int prev_current_vol
= super
->current_vol
;
12841 struct imsm_dev
*dev
;
12843 char location
[16] = "";
12844 unsigned long long chunksize
;
12845 struct intel_dev
*dev_it
;
12847 for (dev_it
= super
->devlist
; dev_it
; dev_it
= dev_it
->next
) {
12848 super
->current_vol
= dev_it
->index
;
12849 dev
= get_imsm_dev(super
, super
->current_vol
);
12851 if (is_vol_to_setup_bitmap(info
, dev
)) {
12852 if (validate_internal_bitmap_imsm(st
)) {
12853 dprintf("bitmap header validation failed\n");
12857 chunksize
= calculate_bitmap_chunksize(st
, dev
);
12858 dprintf("chunk size is %llu\n", chunksize
);
12860 snprintf(location
, sizeof(location
), "+%llu",
12861 get_bitmap_sector(super
, super
->current_vol
));
12862 dprintf("bitmap offset is %s\n", location
);
12864 if (set_bitmap_sysfs(info
, chunksize
, location
)) {
12865 dprintf("cannot setup the bitmap\n");
12872 super
->current_vol
= prev_current_vol
;
12876 struct superswitch super_imsm
= {
12877 .examine_super
= examine_super_imsm
,
12878 .brief_examine_super
= brief_examine_super_imsm
,
12879 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12880 .export_examine_super
= export_examine_super_imsm
,
12881 .detail_super
= detail_super_imsm
,
12882 .brief_detail_super
= brief_detail_super_imsm
,
12883 .write_init_super
= write_init_super_imsm
,
12884 .validate_geometry
= validate_geometry_imsm
,
12885 .add_to_super
= add_to_super_imsm
,
12886 .remove_from_super
= remove_from_super_imsm
,
12887 .detail_platform
= detail_platform_imsm
,
12888 .export_detail_platform
= export_detail_platform_imsm
,
12889 .kill_subarray
= kill_subarray_imsm
,
12890 .update_subarray
= update_subarray_imsm
,
12891 .load_container
= load_container_imsm
,
12892 .default_geometry
= default_geometry_imsm
,
12893 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12894 .reshape_super
= imsm_reshape_super
,
12895 .manage_reshape
= imsm_manage_reshape
,
12896 .recover_backup
= recover_backup_imsm
,
12897 .examine_badblocks
= examine_badblocks_imsm
,
12898 .match_home
= match_home_imsm
,
12899 .uuid_from_super
= uuid_from_super_imsm
,
12900 .getinfo_super
= getinfo_super_imsm
,
12901 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12902 .update_super
= update_super_imsm
,
12904 .avail_size
= avail_size_imsm
,
12905 .get_spare_criteria
= get_spare_criteria_imsm
,
12907 .compare_super
= compare_super_imsm
,
12909 .load_super
= load_super_imsm
,
12910 .init_super
= init_super_imsm
,
12911 .store_super
= store_super_imsm
,
12912 .free_super
= free_super_imsm
,
12913 .match_metadata_desc
= match_metadata_desc_imsm
,
12914 .container_content
= container_content_imsm
,
12915 .validate_container
= validate_container_imsm
,
12917 .add_internal_bitmap
= add_internal_bitmap_imsm
,
12918 .locate_bitmap
= locate_bitmap_imsm
,
12919 .write_bitmap
= write_init_bitmap_imsm
,
12920 .set_bitmap
= set_bitmap_imsm
,
12922 .write_init_ppl
= write_init_ppl_imsm
,
12923 .validate_ppl
= validate_ppl_imsm
,
12929 .open_new
= imsm_open_new
,
12930 .set_array_state
= imsm_set_array_state
,
12931 .set_disk
= imsm_set_disk
,
12932 .sync_metadata
= imsm_sync_metadata
,
12933 .activate_spare
= imsm_activate_spare
,
12934 .process_update
= imsm_process_update
,
12935 .prepare_update
= imsm_prepare_update
,
12936 .record_bad_block
= imsm_record_badblock
,
12937 .clear_bad_block
= imsm_clear_badblock
,
12938 .get_bad_blocks
= imsm_get_badblocks
,