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
25 #include "platform-intel.h"
30 #include "drive_encryption.h"
32 /* MPB == Metadata Parameter Block */
33 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
34 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
36 /* Legacy IMSM versions:
37 * MPB_VERSION_RAID0 1.0.00
38 * MPB_VERSION_RAID1 1.1.00
39 * MPB_VERSION_MANY_VOLUMES_PER_ARRAY 1.2.00
40 * MPB_VERSION_3OR4_DISK_ARRAY 1.2.01
41 * MPB_VERSION_RAID5 1.2.02
42 * MPB_VERSION_5OR6_DISK_ARRAY 1.2.04
43 * MPB_VERSION_CNG 1.2.06
46 #define MPB_VERSION_ATTRIBS "1.3.00"
47 #define MPB_VERSION_ATTRIBS_JD "2.0.00"
48 #define MAX_SIGNATURE_LENGTH 32
49 #define MAX_RAID_SERIAL_LEN 16
52 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
54 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
56 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
58 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
60 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
61 /* supports RAID CNG */
62 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
63 /* supports expanded stripe sizes of 256K, 512K and 1MB */
64 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
65 /* supports RAID10 with more than 4 drives */
66 #define MPB_ATTRIB_RAID10_EXT __cpu_to_le32(0x00000080)
68 /* The OROM Support RST Caching of Volumes */
69 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
70 /* The OROM supports creating disks greater than 2TB */
71 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
72 /* The OROM supports Bad Block Management */
73 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
75 /* THe OROM Supports NVM Caching of Volumes */
76 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
77 /* The OROM supports creating volumes greater than 2TB */
78 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
79 /* originally for PMP, now it's wasted b/c. Never use this bit! */
80 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
81 /* Verify MPB contents against checksum after reading MPB */
82 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
84 /* Define all supported attributes that have to be accepted by mdadm
86 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
88 MPB_ATTRIB_2TB_DISK | \
93 MPB_ATTRIB_EXP_STRIPE_SIZE | \
94 MPB_ATTRIB_RAID10_EXT | \
97 /* Define attributes that are unused but not harmful */
98 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
100 #define MPB_SECTOR_CNT 2210
101 #define IMSM_RESERVED_SECTORS 8192
102 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2048
103 #define SECT_PER_MB_SHIFT 11
104 #define MAX_SECTOR_SIZE 4096
105 #define MULTIPLE_PPL_AREA_SIZE_IMSM (1024 * 1024) /* Size of the whole
110 * Internal Write-intent bitmap is stored in the same area where PPL.
111 * Both features are mutually exclusive, so it is not an issue.
112 * The first 8KiB of the area are reserved and shall not be used.
114 #define IMSM_BITMAP_AREA_RESERVED_SIZE 8192
116 #define IMSM_BITMAP_HEADER_OFFSET (IMSM_BITMAP_AREA_RESERVED_SIZE)
117 #define IMSM_BITMAP_HEADER_SIZE MAX_SECTOR_SIZE
119 #define IMSM_BITMAP_START_OFFSET (IMSM_BITMAP_HEADER_OFFSET + IMSM_BITMAP_HEADER_SIZE)
120 #define IMSM_BITMAP_AREA_SIZE (MULTIPLE_PPL_AREA_SIZE_IMSM - IMSM_BITMAP_START_OFFSET)
121 #define IMSM_BITMAP_AND_HEADER_SIZE (IMSM_BITMAP_AREA_SIZE + IMSM_BITMAP_HEADER_SIZE)
123 #define IMSM_DEFAULT_BITMAP_CHUNKSIZE (64 * 1024 * 1024)
124 #define IMSM_DEFAULT_BITMAP_DAEMON_SLEEP 5
127 * This macro let's us ensure that no-one accidentally
128 * changes the size of a struct
130 #define ASSERT_SIZE(_struct, size) \
131 static inline void __assert_size_##_struct(void) \
135 case (sizeof(struct _struct) == size): break; \
139 /* Disk configuration info. */
140 #define IMSM_MAX_DEVICES 255
142 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
143 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
144 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
145 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
146 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
147 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
148 #define JOURNAL_DISK __cpu_to_le32(0x2000000) /* Device marked as Journaling Drive */
149 __u32 status
; /* 0xF0 - 0xF3 */
150 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
151 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
152 #define IMSM_DISK_FILLERS 3
153 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
155 ASSERT_SIZE(imsm_disk
, 48)
157 /* map selector for map managment
163 /* RAID map configuration infos. */
165 __u32 pba_of_lba0_lo
; /* start address of partition */
166 __u32 blocks_per_member_lo
;/* blocks per member */
167 __u32 num_data_stripes_lo
; /* number of data stripes */
168 __u16 blocks_per_strip
;
169 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
170 #define IMSM_T_STATE_NORMAL 0
171 #define IMSM_T_STATE_UNINITIALIZED 1
172 #define IMSM_T_STATE_DEGRADED 2
173 #define IMSM_T_STATE_FAILED 3
175 #define IMSM_T_RAID0 0
176 #define IMSM_T_RAID1 1
177 #define IMSM_T_RAID5 5
178 #define IMSM_T_RAID10 10
179 __u8 num_members
; /* number of member disks */
180 __u8 num_domains
; /* number of parity domains */
181 __u8 failed_disk_num
; /* valid only when state is degraded */
183 __u32 pba_of_lba0_hi
;
184 __u32 blocks_per_member_hi
;
185 __u32 num_data_stripes_hi
;
186 __u32 filler
[4]; /* expansion area */
187 #define IMSM_ORD_REBUILD (1 << 24)
188 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
189 * top byte contains some flags
192 ASSERT_SIZE(imsm_map
, 52)
195 __u32 curr_migr_unit_lo
;
196 __u32 checkpoint_id
; /* id to access curr_migr_unit */
197 __u8 migr_state
; /* Normal or Migrating */
199 #define MIGR_REBUILD 1
200 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
201 #define MIGR_GEN_MIGR 3
202 #define MIGR_STATE_CHANGE 4
203 #define MIGR_REPAIR 5
204 __u8 migr_type
; /* Initializing, Rebuilding, ... */
205 #define RAIDVOL_CLEAN 0
206 #define RAIDVOL_DIRTY 1
207 #define RAIDVOL_DSRECORD_VALID 2
209 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
210 __u16 verify_errors
; /* number of mismatches */
211 __u16 bad_blocks
; /* number of bad blocks during verify */
212 __u32 curr_migr_unit_hi
;
214 struct imsm_map map
[1];
215 /* here comes another one if migr_state */
217 ASSERT_SIZE(imsm_vol
, 84)
220 __u8 volume
[MAX_RAID_SERIAL_LEN
];
223 #define DEV_BOOTABLE __cpu_to_le32(0x01)
224 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
225 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
226 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
227 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
228 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
229 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
230 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
231 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
232 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
233 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
234 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
235 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
236 __u32 status
; /* Persistent RaidDev status */
237 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
241 __u8 cng_master_disk
;
245 __u16 my_vol_raid_dev_num
; /* Used in Unique volume Id for this RaidDev */
251 /* Unique Volume Id of the NvCache Volume associated with this volume */
252 __u32 nvc_vol_orig_family_num
;
253 __u16 nvc_vol_raid_dev_num
;
256 #define RWH_DISTRIBUTED 1
257 #define RWH_JOURNALING_DRIVE 2
258 #define RWH_MULTIPLE_DISTRIBUTED 3
259 #define RWH_MULTIPLE_PPLS_JOURNALING_DRIVE 4
260 #define RWH_MULTIPLE_OFF 5
262 __u8 rwh_policy
; /* Raid Write Hole Policy */
263 __u8 jd_serial
[MAX_RAID_SERIAL_LEN
]; /* Journal Drive serial number */
266 #define IMSM_DEV_FILLERS 3
267 __u32 filler
[IMSM_DEV_FILLERS
];
270 ASSERT_SIZE(imsm_dev
, 164)
273 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
274 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
275 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
276 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
277 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
278 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
279 __u32 attributes
; /* 0x34 - 0x37 */
280 __u8 num_disks
; /* 0x38 Number of configured disks */
281 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
282 __u8 error_log_pos
; /* 0x3A */
283 __u8 fill
[1]; /* 0x3B */
284 __u32 cache_size
; /* 0x3c - 0x40 in mb */
285 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
286 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
287 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
288 __u16 num_raid_devs_created
; /* 0x4C - 0x4D Used for generating unique
289 * volume IDs for raid_dev created in this array
292 __u16 filler1
; /* 0x4E - 0x4F */
293 __u64 creation_time
; /* 0x50 - 0x57 Array creation time */
294 #define IMSM_FILLERS 32
295 __u32 filler
[IMSM_FILLERS
]; /* 0x58 - 0xD7 RAID_MPB_FILLERS */
296 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
297 /* here comes imsm_dev[num_raid_devs] */
298 /* here comes BBM logs */
300 ASSERT_SIZE(imsm_super
, 264)
302 #define BBM_LOG_MAX_ENTRIES 254
303 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
304 #define BBM_LOG_SIGNATURE 0xabadb10c
306 struct bbm_log_block_addr
{
309 } __attribute__ ((__packed__
));
311 struct bbm_log_entry
{
312 __u8 marked_count
; /* Number of blocks marked - 1 */
313 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
314 struct bbm_log_block_addr defective_block_start
;
315 } __attribute__ ((__packed__
));
318 __u32 signature
; /* 0xABADB10C */
320 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
322 ASSERT_SIZE(bbm_log
, 2040)
324 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
326 #define BLOCKS_PER_KB (1024/512)
328 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
330 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
332 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
333 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
334 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
337 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
338 * be recovered using srcMap */
339 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
340 * already been migrated and must
341 * be recovered from checkpoint area */
343 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of single PPL, without the header */
346 __u32 rec_status
; /* Status used to determine how to restart
347 * migration in case it aborts
349 __u32 curr_migr_unit_lo
; /* 0..numMigrUnits-1 */
350 __u32 family_num
; /* Family number of MPB
351 * containing the RaidDev
352 * that is migrating */
353 __u32 ascending_migr
; /* True if migrating in increasing
355 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
356 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
358 * advances per unit-of-operation */
359 __u32 ckpt_area_pba_lo
; /* Pba of first block of ckpt copy area */
360 __u32 dest_1st_member_lba_lo
; /* First member lba on first
361 * stripe of destination */
362 __u32 num_migr_units_lo
; /* Total num migration units-of-op */
363 __u32 post_migr_vol_cap
; /* Size of volume after
364 * migration completes */
365 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
366 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
367 * migration ckpt record was read from
368 * (for recovered migrations) */
369 __u32 curr_migr_unit_hi
; /* 0..numMigrUnits-1 high order 32 bits */
370 __u32 ckpt_area_pba_hi
; /* Pba of first block of ckpt copy area
371 * high order 32 bits */
372 __u32 dest_1st_member_lba_hi
; /* First member lba on first stripe of
373 * destination - high order 32 bits */
374 __u32 num_migr_units_hi
; /* Total num migration units-of-op
375 * high order 32 bits */
378 ASSERT_SIZE(migr_record
, 128)
381 * enum imsm_status - internal IMSM return values representation.
382 * @STATUS_OK: function succeeded.
383 * @STATUS_ERROR: General error ocurred (not specified).
385 * Typedefed to imsm_status_t.
387 typedef enum imsm_status
{
388 IMSM_STATUS_ERROR
= -1,
395 * 2: metadata does not match
403 struct md_list
*next
;
406 static __u8
migr_type(struct imsm_dev
*dev
)
408 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
409 dev
->status
& DEV_VERIFY_AND_FIX
)
412 return dev
->vol
.migr_type
;
415 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
417 /* for compatibility with older oroms convert MIGR_REPAIR, into
418 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
420 if (migr_type
== MIGR_REPAIR
) {
421 dev
->vol
.migr_type
= MIGR_VERIFY
;
422 dev
->status
|= DEV_VERIFY_AND_FIX
;
424 dev
->vol
.migr_type
= migr_type
;
425 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
429 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
431 return ROUND_UP(bytes
, sector_size
) / sector_size
;
434 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
435 unsigned int sector_size
)
437 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
441 struct imsm_dev
*dev
;
442 struct intel_dev
*next
;
447 enum sys_dev_type type
;
450 struct intel_hba
*next
;
457 /* internal representation of IMSM metadata */
460 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
461 struct imsm_super
*anchor
; /* immovable parameters */
464 void *migr_rec_buf
; /* buffer for I/O operations */
465 struct migr_record
*migr_rec
; /* migration record */
467 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
468 array, it indicates that mdmon is allowed to clean migration
470 size_t len
; /* size of the 'buf' allocation */
471 size_t extra_space
; /* extra space in 'buf' that is not used yet */
472 void *next_buf
; /* for realloc'ing buf from the manager */
474 int updates_pending
; /* count of pending updates for mdmon */
475 int current_vol
; /* index of raid device undergoing creation */
476 unsigned long long create_offset
; /* common start for 'current_vol' */
477 __u32 random
; /* random data for seeding new family numbers */
478 struct intel_dev
*devlist
;
479 unsigned int sector_size
; /* sector size of used member drives */
483 __u8 serial
[MAX_RAID_SERIAL_LEN
];
486 struct imsm_disk disk
;
489 struct extent
*e
; /* for determining freespace @ create */
490 int raiddisk
; /* slot to fill in autolayout */
492 } *disks
, *current_disk
;
493 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
495 struct dl
*missing
; /* disks removed while we weren't looking */
496 struct bbm_log
*bbm_log
;
497 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
498 const struct imsm_orom
*orom
; /* platform firmware support */
499 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
500 struct md_bb bb
; /* memory for get_bad_blocks call */
504 struct imsm_disk disk
;
505 #define IMSM_UNKNOWN_OWNER (-1)
507 struct intel_disk
*next
;
511 * struct extent - reserved space details.
512 * @start: start offset.
513 * @size: size of reservation, set to 0 for metadata reservation.
514 * @vol: index of the volume, meaningful if &size is set.
517 unsigned long long start
, size
;
521 /* definitions of reshape process types */
522 enum imsm_reshape_type
{
529 /* definition of messages passed to imsm_process_update */
530 enum imsm_update_type
{
531 update_activate_spare
,
535 update_add_remove_disk
,
536 update_reshape_container_disks
,
537 update_reshape_migration
,
539 update_general_migration_checkpoint
,
541 update_prealloc_badblocks_mem
,
545 struct imsm_update_activate_spare
{
546 enum imsm_update_type type
;
550 struct imsm_update_activate_spare
*next
;
556 unsigned long long size
;
563 enum takeover_direction
{
567 struct imsm_update_takeover
{
568 enum imsm_update_type type
;
570 enum takeover_direction direction
;
573 struct imsm_update_reshape
{
574 enum imsm_update_type type
;
578 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
581 struct imsm_update_reshape_migration
{
582 enum imsm_update_type type
;
585 /* fields for array migration changes
592 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
595 struct imsm_update_size_change
{
596 enum imsm_update_type type
;
601 struct imsm_update_general_migration_checkpoint
{
602 enum imsm_update_type type
;
603 __u64 curr_migr_unit
;
607 __u8 serial
[MAX_RAID_SERIAL_LEN
];
610 struct imsm_update_create_array
{
611 enum imsm_update_type type
;
616 struct imsm_update_kill_array
{
617 enum imsm_update_type type
;
621 struct imsm_update_rename_array
{
622 enum imsm_update_type type
;
623 __u8 name
[MAX_RAID_SERIAL_LEN
];
627 struct imsm_update_add_remove_disk
{
628 enum imsm_update_type type
;
631 struct imsm_update_prealloc_bb_mem
{
632 enum imsm_update_type type
;
635 struct imsm_update_rwh_policy
{
636 enum imsm_update_type type
;
641 static const char *_sys_dev_type
[] = {
642 [SYS_DEV_UNKNOWN
] = "Unknown",
643 [SYS_DEV_SAS
] = "SAS",
644 [SYS_DEV_SATA
] = "SATA",
645 [SYS_DEV_NVME
] = "NVMe",
646 [SYS_DEV_VMD
] = "VMD",
647 [SYS_DEV_SATA_VMD
] = "SATA VMD"
650 static int no_platform
= -1;
652 static int check_no_platform(void)
654 static const char search
[] = "mdadm.imsm.test=1";
657 if (no_platform
>= 0)
660 if (check_env("IMSM_NO_PLATFORM")) {
664 fp
= fopen("/proc/cmdline", "r");
666 char *l
= conf_line(fp
);
675 if (strcmp(w
, search
) == 0)
681 if (no_platform
>= 0)
688 void imsm_set_no_platform(int v
)
693 const char *get_sys_dev_type(enum sys_dev_type type
)
695 if (type
>= SYS_DEV_MAX
)
696 type
= SYS_DEV_UNKNOWN
;
698 return _sys_dev_type
[type
];
701 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
703 struct intel_hba
*result
= xmalloc(sizeof(*result
));
705 result
->type
= device
->type
;
706 result
->path
= xstrdup(device
->path
);
708 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
714 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
716 struct intel_hba
*result
;
718 for (result
= hba
; result
; result
= result
->next
) {
719 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
725 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
727 struct intel_hba
*hba
;
729 /* check if disk attached to Intel HBA */
730 hba
= find_intel_hba(super
->hba
, device
);
733 /* Check if HBA is already attached to super */
734 if (super
->hba
== NULL
) {
735 super
->hba
= alloc_intel_hba(device
);
740 /* Intel metadata allows for all disks attached to the same type HBA.
741 * Do not support HBA types mixing
743 if (device
->type
!= hba
->type
)
746 /* Multiple same type HBAs can be used if they share the same OROM */
747 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
749 if (device_orom
!= super
->orom
)
755 hba
->next
= alloc_intel_hba(device
);
759 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
761 struct sys_dev
*list
, *elem
;
764 if ((list
= find_intel_devices()) == NULL
)
767 if (!is_fd_valid(fd
))
768 disk_path
= (char *) devname
;
770 disk_path
= diskfd_to_devpath(fd
, 1, NULL
);
775 for (elem
= list
; elem
; elem
= elem
->next
)
776 if (path_attached_to_hba(disk_path
, elem
->path
))
779 if (disk_path
!= devname
)
785 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
788 static struct supertype
*match_metadata_desc_imsm(char *arg
)
790 struct supertype
*st
;
792 if (strcmp(arg
, "imsm") != 0 &&
793 strcmp(arg
, "default") != 0
797 st
= xcalloc(1, sizeof(*st
));
798 st
->ss
= &super_imsm
;
799 st
->max_devs
= IMSM_MAX_DEVICES
;
800 st
->minor_version
= 0;
805 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
807 return &mpb
->sig
[MPB_SIG_LEN
];
810 /* retrieve a disk directly from the anchor when the anchor is known to be
811 * up-to-date, currently only at load time
813 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
815 if (index
>= mpb
->num_disks
)
817 return &mpb
->disk
[index
];
820 /* retrieve the disk description based on a index of the disk
823 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
827 for (d
= super
->disks
; d
; d
= d
->next
)
828 if (d
->index
== index
)
833 /* retrieve a disk from the parsed metadata */
834 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
838 dl
= get_imsm_dl_disk(super
, index
);
845 /* generate a checksum directly from the anchor when the anchor is known to be
846 * up-to-date, currently only at load or write_super after coalescing
848 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
850 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
851 __u32
*p
= (__u32
*) mpb
;
855 sum
+= __le32_to_cpu(*p
);
859 return sum
- __le32_to_cpu(mpb
->check_sum
);
862 static size_t sizeof_imsm_map(struct imsm_map
*map
)
864 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
867 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
869 /* A device can have 2 maps if it is in the middle of a migration.
871 * MAP_0 - we return the first map
872 * MAP_1 - we return the second map if it exists, else NULL
873 * MAP_X - we return the second map if it exists, else the first
875 struct imsm_map
*map
= &dev
->vol
.map
[0];
876 struct imsm_map
*map2
= NULL
;
878 if (dev
->vol
.migr_state
)
879 map2
= (void *)map
+ sizeof_imsm_map(map
);
881 switch (second_map
) {
898 /* return the size of the device.
899 * migr_state increases the returned size if map[0] were to be duplicated
901 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
903 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
904 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
906 /* migrating means an additional map */
907 if (dev
->vol
.migr_state
)
908 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
910 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
915 /* retrieve disk serial number list from a metadata update */
916 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
919 struct disk_info
*inf
;
921 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
922 sizeof_imsm_dev(&update
->dev
, 0);
928 * __get_imsm_dev() - Get device with index from imsm_super.
929 * @mpb: &imsm_super pointer, not NULL.
930 * @index: Device index.
932 * Function works as non-NULL, aborting in such a case,
933 * when NULL would be returned.
935 * Device index should be in range 0 up to num_raid_devs.
936 * Function assumes the index was already verified.
937 * Index must be valid, otherwise abort() is called.
939 * Return: Pointer to corresponding imsm_dev.
942 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
948 if (index
>= mpb
->num_raid_devs
)
951 /* devices start after all disks */
952 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
954 for (i
= 0; i
<= index
; i
++, offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0))
956 return _mpb
+ offset
;
958 pr_err("cannot find imsm_dev with index %u in imsm_super\n", index
);
963 * get_imsm_dev() - Get device with index from intel_super.
964 * @super: &intel_super pointer, not NULL.
965 * @index: Device index.
967 * Function works as non-NULL, aborting in such a case,
968 * when NULL would be returned.
970 * Device index should be in range 0 up to num_raid_devs.
971 * Function assumes the index was already verified.
972 * Index must be valid, otherwise abort() is called.
974 * Return: Pointer to corresponding imsm_dev.
977 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
979 struct intel_dev
*dv
;
981 if (index
>= super
->anchor
->num_raid_devs
)
984 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
985 if (dv
->index
== index
)
988 pr_err("cannot find imsm_dev with index %u in intel_super\n", index
);
992 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
995 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
996 __le16_to_cpu(addr
->w1
));
999 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
1001 struct bbm_log_block_addr addr
;
1003 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
1004 addr
.dw1
= __cpu_to_le32((__u32
)(sec
>> 16) & 0xffffffff);
1008 /* get size of the bbm log */
1009 static __u32
get_imsm_bbm_log_size(struct bbm_log
*log
)
1011 if (!log
|| log
->entry_count
== 0)
1014 return sizeof(log
->signature
) +
1015 sizeof(log
->entry_count
) +
1016 log
->entry_count
* sizeof(struct bbm_log_entry
);
1019 /* check if bad block is not partially stored in bbm log */
1020 static int is_stored_in_bbm(struct bbm_log
*log
, const __u8 idx
, const unsigned
1021 long long sector
, const int length
, __u32
*pos
)
1025 for (i
= *pos
; i
< log
->entry_count
; i
++) {
1026 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
1027 unsigned long long bb_start
;
1028 unsigned long long bb_end
;
1030 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1031 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1033 if ((entry
->disk_ordinal
== idx
) && (bb_start
>= sector
) &&
1034 (bb_end
<= sector
+ length
)) {
1042 /* record new bad block in bbm log */
1043 static int record_new_badblock(struct bbm_log
*log
, const __u8 idx
, unsigned
1044 long long sector
, int length
)
1048 struct bbm_log_entry
*entry
= NULL
;
1050 while (is_stored_in_bbm(log
, idx
, sector
, length
, &pos
)) {
1051 struct bbm_log_entry
*e
= &log
->marked_block_entries
[pos
];
1053 if ((e
->marked_count
+ 1 == BBM_LOG_MAX_LBA_ENTRY_VAL
) &&
1054 (__le48_to_cpu(&e
->defective_block_start
) == sector
)) {
1055 sector
+= BBM_LOG_MAX_LBA_ENTRY_VAL
;
1056 length
-= BBM_LOG_MAX_LBA_ENTRY_VAL
;
1065 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
1066 BBM_LOG_MAX_LBA_ENTRY_VAL
;
1067 entry
->defective_block_start
= __cpu_to_le48(sector
);
1068 entry
->marked_count
= cnt
- 1;
1075 new_bb
= ROUND_UP(length
, BBM_LOG_MAX_LBA_ENTRY_VAL
) /
1076 BBM_LOG_MAX_LBA_ENTRY_VAL
;
1077 if (log
->entry_count
+ new_bb
> BBM_LOG_MAX_ENTRIES
)
1080 while (length
> 0) {
1081 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
1082 BBM_LOG_MAX_LBA_ENTRY_VAL
;
1083 struct bbm_log_entry
*entry
=
1084 &log
->marked_block_entries
[log
->entry_count
];
1086 entry
->defective_block_start
= __cpu_to_le48(sector
);
1087 entry
->marked_count
= cnt
- 1;
1088 entry
->disk_ordinal
= idx
;
1099 /* clear all bad blocks for given disk */
1100 static void clear_disk_badblocks(struct bbm_log
*log
, const __u8 idx
)
1104 while (i
< log
->entry_count
) {
1105 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
1107 if (entries
[i
].disk_ordinal
== idx
) {
1108 if (i
< log
->entry_count
- 1)
1109 entries
[i
] = entries
[log
->entry_count
- 1];
1117 /* clear given bad block */
1118 static int clear_badblock(struct bbm_log
*log
, const __u8 idx
, const unsigned
1119 long long sector
, const int length
) {
1122 while (i
< log
->entry_count
) {
1123 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
1125 if ((entries
[i
].disk_ordinal
== idx
) &&
1126 (__le48_to_cpu(&entries
[i
].defective_block_start
) ==
1127 sector
) && (entries
[i
].marked_count
+ 1 == length
)) {
1128 if (i
< log
->entry_count
- 1)
1129 entries
[i
] = entries
[log
->entry_count
- 1];
1139 /* allocate and load BBM log from metadata */
1140 static int load_bbm_log(struct intel_super
*super
)
1142 struct imsm_super
*mpb
= super
->anchor
;
1143 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1145 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
1146 if (!super
->bbm_log
)
1150 struct bbm_log
*log
= (void *)mpb
+
1151 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1155 if (bbm_log_size
< sizeof(log
->signature
) +
1156 sizeof(log
->entry_count
))
1159 entry_count
= __le32_to_cpu(log
->entry_count
);
1160 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
1161 (entry_count
> BBM_LOG_MAX_ENTRIES
))
1165 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
1166 entry_count
* sizeof(struct bbm_log_entry
))
1169 memcpy(super
->bbm_log
, log
, bbm_log_size
);
1171 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
1172 super
->bbm_log
->entry_count
= 0;
1178 /* checks if bad block is within volume boundaries */
1179 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
1180 const unsigned long long start_sector
,
1181 const unsigned long long size
)
1183 unsigned long long bb_start
;
1184 unsigned long long bb_end
;
1186 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1187 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1189 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1190 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1196 /* get list of bad blocks on a drive for a volume */
1197 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1198 const unsigned long long start_sector
,
1199 const unsigned long long size
,
1205 for (i
= 0; i
< log
->entry_count
; i
++) {
1206 const struct bbm_log_entry
*ent
=
1207 &log
->marked_block_entries
[i
];
1208 struct md_bb_entry
*bb
;
1210 if ((ent
->disk_ordinal
== idx
) &&
1211 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1213 if (!bbs
->entries
) {
1214 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1220 bb
= &bbs
->entries
[count
++];
1221 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1222 bb
->length
= ent
->marked_count
+ 1;
1230 * == MAP_0 get first map
1231 * == MAP_1 get second map
1232 * == MAP_X than get map according to the current migr_state
1234 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1238 struct imsm_map
*map
;
1240 map
= get_imsm_map(dev
, second_map
);
1242 /* top byte identifies disk under rebuild */
1243 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1246 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1247 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1249 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1251 return ord_to_idx(ord
);
1254 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1256 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1259 static int get_imsm_disk_slot(struct imsm_map
*map
, const unsigned int idx
)
1264 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1265 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1266 if (ord_to_idx(ord
) == idx
)
1270 return IMSM_STATUS_ERROR
;
1273 * update_imsm_raid_level() - update raid level appropriately in &imsm_map.
1274 * @map: &imsm_map pointer.
1275 * @new_level: MD style level.
1277 * For backward compatibility reasons we need to differentiate RAID10.
1278 * In the past IMSM RAID10 was presented as RAID1.
1279 * Keep compatibility unless it is not explicitly updated by UEFI driver.
1281 * Routine needs num_members to be set and (optionally) raid_level.
1283 static void update_imsm_raid_level(struct imsm_map
*map
, int new_level
)
1285 if (new_level
!= IMSM_T_RAID10
) {
1286 map
->raid_level
= new_level
;
1290 if (map
->num_members
== 4) {
1291 if (map
->raid_level
== IMSM_T_RAID10
|| map
->raid_level
== IMSM_T_RAID1
)
1294 map
->raid_level
= IMSM_T_RAID1
;
1298 map
->raid_level
= IMSM_T_RAID10
;
1301 static int get_imsm_raid_level(struct imsm_map
*map
)
1303 if (map
->raid_level
== IMSM_T_RAID1
) {
1304 if (map
->num_members
== 2)
1305 return IMSM_T_RAID1
;
1307 return IMSM_T_RAID10
;
1310 return map
->raid_level
;
1314 * get_disk_slot_in_dev() - retrieve disk slot from &imsm_dev.
1315 * @super: &intel_super pointer, not NULL.
1316 * @dev_idx: imsm device index.
1319 * Return: Slot on success, IMSM_STATUS_ERROR otherwise.
1321 static int get_disk_slot_in_dev(struct intel_super
*super
, const __u8 dev_idx
,
1322 const unsigned int idx
)
1324 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
1325 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1327 return get_imsm_disk_slot(map
, idx
);
1330 static int cmp_extent(const void *av
, const void *bv
)
1332 const struct extent
*a
= av
;
1333 const struct extent
*b
= bv
;
1334 if (a
->start
< b
->start
)
1336 if (a
->start
> b
->start
)
1341 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1343 int memberships
= 0;
1346 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++)
1347 if (get_disk_slot_in_dev(super
, i
, dl
->index
) >= 0)
1353 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1355 static int split_ull(unsigned long long n
, void *lo
, void *hi
)
1357 if (lo
== 0 || hi
== 0)
1359 __put_unaligned32(__cpu_to_le32((__u32
)n
), lo
);
1360 __put_unaligned32(__cpu_to_le32((n
>> 32)), hi
);
1364 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1366 return (unsigned long long)__le32_to_cpu(lo
) |
1367 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1370 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1374 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1378 * imsm_num_data_members() - get data drives count for an array.
1379 * @map: Map to analyze.
1381 * num_data_members value represents minimal count of drives for level.
1382 * The name of the property could be misleading for RAID5 with asymmetric layout
1383 * because some data required to be calculated from parity.
1384 * The property is extracted from level and num_members value.
1386 * Return: num_data_members value on success, zero otherwise.
1388 static __u8
imsm_num_data_members(struct imsm_map
*map
)
1390 switch (get_imsm_raid_level(map
)) {
1392 return map
->num_members
;
1395 return map
->num_members
/ 2;
1397 return map
->num_members
- 1;
1399 dprintf("unsupported raid level\n");
1404 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1408 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1411 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1415 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1418 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1422 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1425 static unsigned long long vol_curr_migr_unit(struct imsm_dev
*dev
)
1430 return join_u32(dev
->vol
.curr_migr_unit_lo
, dev
->vol
.curr_migr_unit_hi
);
1433 static unsigned long long imsm_dev_size(struct imsm_dev
*dev
)
1437 return join_u32(dev
->size_low
, dev
->size_high
);
1440 static unsigned long long migr_chkp_area_pba(struct migr_record
*migr_rec
)
1442 if (migr_rec
== NULL
)
1444 return join_u32(migr_rec
->ckpt_area_pba_lo
,
1445 migr_rec
->ckpt_area_pba_hi
);
1448 static unsigned long long current_migr_unit(struct migr_record
*migr_rec
)
1450 if (migr_rec
== NULL
)
1452 return join_u32(migr_rec
->curr_migr_unit_lo
,
1453 migr_rec
->curr_migr_unit_hi
);
1456 static unsigned long long migr_dest_1st_member_lba(struct migr_record
*migr_rec
)
1458 if (migr_rec
== NULL
)
1460 return join_u32(migr_rec
->dest_1st_member_lba_lo
,
1461 migr_rec
->dest_1st_member_lba_hi
);
1464 static unsigned long long get_num_migr_units(struct migr_record
*migr_rec
)
1466 if (migr_rec
== NULL
)
1468 return join_u32(migr_rec
->num_migr_units_lo
,
1469 migr_rec
->num_migr_units_hi
);
1472 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1474 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1478 * set_num_domains() - Set number of domains for an array.
1479 * @map: Map to be updated.
1481 * num_domains property represents copies count of each data drive, thus make
1482 * it meaningful only for RAID1 and RAID10. IMSM supports two domains for
1485 static void set_num_domains(struct imsm_map
*map
)
1487 int level
= get_imsm_raid_level(map
);
1489 if (level
== 1 || level
== 10)
1490 map
->num_domains
= 2;
1492 map
->num_domains
= 1;
1495 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1497 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1500 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1502 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1505 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1507 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1511 * update_num_data_stripes() - Calculate and update num_data_stripes value.
1512 * @map: map to be updated.
1513 * @dev_size: size of volume.
1515 * num_data_stripes value is addictionally divided by num_domains, therefore for
1516 * levels where num_domains is not 1, nds is a part of real value.
1518 static void update_num_data_stripes(struct imsm_map
*map
,
1519 unsigned long long dev_size
)
1521 unsigned long long nds
= dev_size
/ imsm_num_data_members(map
);
1523 nds
/= map
->num_domains
;
1524 nds
/= map
->blocks_per_strip
;
1525 set_num_data_stripes(map
, nds
);
1528 static void set_vol_curr_migr_unit(struct imsm_dev
*dev
, unsigned long long n
)
1533 split_ull(n
, &dev
->vol
.curr_migr_unit_lo
, &dev
->vol
.curr_migr_unit_hi
);
1536 static void set_imsm_dev_size(struct imsm_dev
*dev
, unsigned long long n
)
1538 split_ull(n
, &dev
->size_low
, &dev
->size_high
);
1541 static void set_migr_chkp_area_pba(struct migr_record
*migr_rec
,
1542 unsigned long long n
)
1544 split_ull(n
, &migr_rec
->ckpt_area_pba_lo
, &migr_rec
->ckpt_area_pba_hi
);
1547 static void set_current_migr_unit(struct migr_record
*migr_rec
,
1548 unsigned long long n
)
1550 split_ull(n
, &migr_rec
->curr_migr_unit_lo
,
1551 &migr_rec
->curr_migr_unit_hi
);
1554 static void set_migr_dest_1st_member_lba(struct migr_record
*migr_rec
,
1555 unsigned long long n
)
1557 split_ull(n
, &migr_rec
->dest_1st_member_lba_lo
,
1558 &migr_rec
->dest_1st_member_lba_hi
);
1561 static void set_num_migr_units(struct migr_record
*migr_rec
,
1562 unsigned long long n
)
1564 split_ull(n
, &migr_rec
->num_migr_units_lo
,
1565 &migr_rec
->num_migr_units_hi
);
1568 static unsigned long long per_dev_array_size(struct imsm_map
*map
)
1570 unsigned long long array_size
= 0;
1575 array_size
= num_data_stripes(map
) * map
->blocks_per_strip
;
1576 if (get_imsm_raid_level(map
) == 1 || get_imsm_raid_level(map
) == 10)
1582 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
,
1583 int get_minimal_reservation
)
1585 /* find a list of used extents on the given physical device */
1586 int memberships
= count_memberships(dl
, super
);
1587 struct extent
*rv
= xcalloc(memberships
+ 1, sizeof(struct extent
));
1588 struct extent
*e
= rv
;
1592 /* trim the reserved area for spares, so they can join any array
1593 * regardless of whether the OROM has assigned sectors from the
1594 * IMSM_RESERVED_SECTORS region
1596 if (dl
->index
== -1 || get_minimal_reservation
)
1597 reservation
= imsm_min_reserved_sectors(super
);
1599 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1601 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1602 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1603 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1605 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1606 e
->start
= pba_of_lba0(map
);
1607 e
->size
= per_dev_array_size(map
);
1612 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1614 /* determine the start of the metadata
1615 * when no raid devices are defined use the default
1616 * ...otherwise allow the metadata to truncate the value
1617 * as is the case with older versions of imsm
1620 struct extent
*last
= &rv
[memberships
- 1];
1621 unsigned long long remainder
;
1623 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1624 /* round down to 1k block to satisfy precision of the kernel
1628 /* make sure remainder is still sane */
1629 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1630 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1631 if (reservation
> remainder
)
1632 reservation
= remainder
;
1634 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1639 /* try to determine how much space is reserved for metadata from
1640 * the last get_extents() entry, otherwise fallback to the
1643 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1649 /* for spares just return a minimal reservation which will grow
1650 * once the spare is picked up by an array
1652 if (dl
->index
== -1)
1653 return MPB_SECTOR_CNT
;
1655 e
= get_extents(super
, dl
, 0);
1657 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1659 /* scroll to last entry */
1660 for (i
= 0; e
[i
].size
; i
++)
1663 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1670 static int is_spare(struct imsm_disk
*disk
)
1672 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1675 static int is_configured(struct imsm_disk
*disk
)
1677 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1680 static int is_failed(struct imsm_disk
*disk
)
1682 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1685 static int is_journal(struct imsm_disk
*disk
)
1687 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1691 * round_member_size_to_mb()- Round given size to closest MiB.
1692 * @size: size to round in sectors.
1694 static inline unsigned long long round_member_size_to_mb(unsigned long long size
)
1696 return (size
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
1700 * round_size_to_mb()- Round given size.
1701 * @array_size: size to round in sectors.
1702 * @disk_count: count of data members.
1704 * Get size per each data member and round it to closest MiB to ensure that data
1705 * splits evenly between members.
1707 * Return: Array size, rounded down.
1709 static inline unsigned long long round_size_to_mb(unsigned long long array_size
,
1710 unsigned int disk_count
)
1712 return round_member_size_to_mb(array_size
/ disk_count
) * disk_count
;
1715 static int able_to_resync(int raid_level
, int missing_disks
)
1717 int max_missing_disks
= 0;
1719 switch (raid_level
) {
1721 max_missing_disks
= 1;
1724 max_missing_disks
= 0;
1726 return missing_disks
<= max_missing_disks
;
1729 /* try to determine how much space is reserved for metadata from
1730 * the last get_extents() entry on the smallest active disk,
1731 * otherwise fallback to the default
1733 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1737 unsigned long long min_active
;
1739 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1740 struct dl
*dl
, *dl_min
= NULL
;
1746 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1749 unsigned long long blocks
= total_blocks(&dl
->disk
);
1750 if (blocks
< min_active
|| min_active
== 0) {
1752 min_active
= blocks
;
1758 /* find last lba used by subarrays on the smallest active disk */
1759 e
= get_extents(super
, dl_min
, 0);
1762 for (i
= 0; e
[i
].size
; i
++)
1765 remainder
= min_active
- e
[i
].start
;
1768 /* to give priority to recovery we should not require full
1769 IMSM_RESERVED_SECTORS from the spare */
1770 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1772 /* if real reservation is smaller use that value */
1773 return (remainder
< rv
) ? remainder
: rv
;
1776 static bool is_gen_migration(struct imsm_dev
*dev
);
1778 #define IMSM_4K_DIV 8
1780 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1781 struct imsm_dev
*dev
);
1783 static void print_imsm_dev(struct intel_super
*super
,
1784 struct imsm_dev
*dev
,
1790 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1791 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1795 printf("[%.16s]:\n", dev
->volume
);
1796 printf(" Subarray : %d\n", super
->current_vol
);
1797 printf(" UUID : %s\n", uuid
);
1798 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1800 printf(" <-- %d", get_imsm_raid_level(map2
));
1802 printf(" Members : %d", map
->num_members
);
1804 printf(" <-- %d", map2
->num_members
);
1806 printf(" Slots : [");
1807 for (i
= 0; i
< map
->num_members
; i
++) {
1808 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1809 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1814 for (i
= 0; i
< map2
->num_members
; i
++) {
1815 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1816 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1821 printf(" Failed disk : ");
1822 if (map
->failed_disk_num
== 0xff)
1823 printf(STR_COMMON_NONE
);
1825 printf("%i", map
->failed_disk_num
);
1827 slot
= get_imsm_disk_slot(map
, disk_idx
);
1829 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1830 printf(" This Slot : %d%s\n", slot
,
1831 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1833 printf(" This Slot : ?\n");
1834 printf(" Sector Size : %u\n", super
->sector_size
);
1835 sz
= imsm_dev_size(dev
);
1836 printf(" Array Size : %llu%s\n",
1837 (unsigned long long)sz
* 512 / super
->sector_size
,
1838 human_size(sz
* 512));
1839 sz
= blocks_per_member(map
);
1840 printf(" Per Dev Size : %llu%s\n",
1841 (unsigned long long)sz
* 512 / super
->sector_size
,
1842 human_size(sz
* 512));
1843 printf(" Sector Offset : %llu\n",
1844 pba_of_lba0(map
) * 512 / super
->sector_size
);
1845 printf(" Num Stripes : %llu\n",
1846 num_data_stripes(map
));
1847 printf(" Chunk Size : %u KiB",
1848 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1850 printf(" <-- %u KiB",
1851 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1853 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1854 printf(" Migrate State : ");
1855 if (dev
->vol
.migr_state
) {
1856 if (migr_type(dev
) == MIGR_INIT
)
1857 printf("initialize\n");
1858 else if (migr_type(dev
) == MIGR_REBUILD
)
1859 printf("rebuild\n");
1860 else if (migr_type(dev
) == MIGR_VERIFY
)
1862 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1863 printf("general migration\n");
1864 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1865 printf("state change\n");
1866 else if (migr_type(dev
) == MIGR_REPAIR
)
1869 printf("<unknown:%d>\n", migr_type(dev
));
1872 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1873 if (dev
->vol
.migr_state
) {
1874 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1876 printf(" <-- %s", map_state_str
[map
->map_state
]);
1877 printf("\n Checkpoint : %llu ", vol_curr_migr_unit(dev
));
1878 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1881 printf("(%llu)", (unsigned long long)
1882 blocks_per_migr_unit(super
, dev
));
1885 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1887 printf(" RWH Policy : ");
1888 if (dev
->rwh_policy
== RWH_OFF
|| dev
->rwh_policy
== RWH_MULTIPLE_OFF
)
1890 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1891 printf("PPL distributed\n");
1892 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1893 printf("PPL journaling drive\n");
1894 else if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
1895 printf("Multiple distributed PPLs\n");
1896 else if (dev
->rwh_policy
== RWH_MULTIPLE_PPLS_JOURNALING_DRIVE
)
1897 printf("Multiple PPLs on journaling drive\n");
1898 else if (dev
->rwh_policy
== RWH_BITMAP
)
1899 printf("Write-intent bitmap\n");
1901 printf("<unknown:%d>\n", dev
->rwh_policy
);
1903 printf(" Volume ID : %u\n", dev
->my_vol_raid_dev_num
);
1906 static void print_imsm_disk(struct imsm_disk
*disk
,
1909 unsigned int sector_size
) {
1910 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1913 if (index
< -1 || !disk
)
1917 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1919 printf(" Disk%02d Serial : %s\n", index
, str
);
1921 printf(" Disk Serial : %s\n", str
);
1922 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1923 is_configured(disk
) ? " active" : "",
1924 is_failed(disk
) ? " failed" : "",
1925 is_journal(disk
) ? " journal" : "");
1926 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1927 sz
= total_blocks(disk
) - reserved
;
1928 printf(" Usable Size : %llu%s\n",
1929 (unsigned long long)sz
* 512 / sector_size
,
1930 human_size(sz
* 512));
1933 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1935 struct migr_record
*migr_rec
= super
->migr_rec
;
1937 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1938 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1939 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1940 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1941 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1942 set_migr_chkp_area_pba(migr_rec
,
1943 migr_chkp_area_pba(migr_rec
) / IMSM_4K_DIV
);
1944 set_migr_dest_1st_member_lba(migr_rec
,
1945 migr_dest_1st_member_lba(migr_rec
) / IMSM_4K_DIV
);
1948 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1950 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1953 void convert_to_4k(struct intel_super
*super
)
1955 struct imsm_super
*mpb
= super
->anchor
;
1956 struct imsm_disk
*disk
;
1958 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1960 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1961 disk
= __get_imsm_disk(mpb
, i
);
1963 convert_to_4k_imsm_disk(disk
);
1965 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1966 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1967 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1969 set_imsm_dev_size(dev
, imsm_dev_size(dev
)/IMSM_4K_DIV
);
1970 set_vol_curr_migr_unit(dev
,
1971 vol_curr_migr_unit(dev
) / IMSM_4K_DIV
);
1974 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1975 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1976 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1978 if (dev
->vol
.migr_state
) {
1980 map
= get_imsm_map(dev
, MAP_1
);
1981 set_blocks_per_member(map
,
1982 blocks_per_member(map
)/IMSM_4K_DIV
);
1983 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1984 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1988 struct bbm_log
*log
= (void *)mpb
+
1989 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1992 for (i
= 0; i
< log
->entry_count
; i
++) {
1993 struct bbm_log_entry
*entry
=
1994 &log
->marked_block_entries
[i
];
1996 __u8 count
= entry
->marked_count
+ 1;
1997 unsigned long long sector
=
1998 __le48_to_cpu(&entry
->defective_block_start
);
2000 entry
->defective_block_start
=
2001 __cpu_to_le48(sector
/IMSM_4K_DIV
);
2002 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
2006 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
2009 void examine_migr_rec_imsm(struct intel_super
*super
)
2011 struct migr_record
*migr_rec
= super
->migr_rec
;
2012 struct imsm_super
*mpb
= super
->anchor
;
2015 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2016 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2017 struct imsm_map
*map
;
2020 if (is_gen_migration(dev
) == false)
2023 printf("\nMigration Record Information:");
2025 /* first map under migration */
2026 map
= get_imsm_map(dev
, MAP_0
);
2029 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
2030 if (map
== NULL
|| slot
> 1 || slot
< 0) {
2031 printf(" Empty\n ");
2032 printf("Examine one of first two disks in array\n");
2035 printf("\n Status : ");
2036 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
2039 printf("Contains Data\n");
2040 printf(" Current Unit : %llu\n",
2041 current_migr_unit(migr_rec
));
2042 printf(" Family : %u\n",
2043 __le32_to_cpu(migr_rec
->family_num
));
2044 printf(" Ascending : %u\n",
2045 __le32_to_cpu(migr_rec
->ascending_migr
));
2046 printf(" Blocks Per Unit : %u\n",
2047 __le32_to_cpu(migr_rec
->blocks_per_unit
));
2048 printf(" Dest. Depth Per Unit : %u\n",
2049 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
2050 printf(" Checkpoint Area pba : %llu\n",
2051 migr_chkp_area_pba(migr_rec
));
2052 printf(" First member lba : %llu\n",
2053 migr_dest_1st_member_lba(migr_rec
));
2054 printf(" Total Number of Units : %llu\n",
2055 get_num_migr_units(migr_rec
));
2056 printf(" Size of volume : %llu\n",
2057 join_u32(migr_rec
->post_migr_vol_cap
,
2058 migr_rec
->post_migr_vol_cap_hi
));
2059 printf(" Record was read from : %u\n",
2060 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
2066 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
2068 struct migr_record
*migr_rec
= super
->migr_rec
;
2070 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
2071 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
2072 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
2073 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
2074 &migr_rec
->post_migr_vol_cap
,
2075 &migr_rec
->post_migr_vol_cap_hi
);
2076 set_migr_chkp_area_pba(migr_rec
,
2077 migr_chkp_area_pba(migr_rec
) * IMSM_4K_DIV
);
2078 set_migr_dest_1st_member_lba(migr_rec
,
2079 migr_dest_1st_member_lba(migr_rec
) * IMSM_4K_DIV
);
2082 void convert_from_4k(struct intel_super
*super
)
2084 struct imsm_super
*mpb
= super
->anchor
;
2085 struct imsm_disk
*disk
;
2087 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
2089 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2090 disk
= __get_imsm_disk(mpb
, i
);
2092 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
2095 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2096 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2097 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2099 set_imsm_dev_size(dev
, imsm_dev_size(dev
)*IMSM_4K_DIV
);
2100 set_vol_curr_migr_unit(dev
,
2101 vol_curr_migr_unit(dev
) * IMSM_4K_DIV
);
2104 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
2105 map
->blocks_per_strip
*= IMSM_4K_DIV
;
2106 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
2108 if (dev
->vol
.migr_state
) {
2110 map
= get_imsm_map(dev
, MAP_1
);
2111 set_blocks_per_member(map
,
2112 blocks_per_member(map
)*IMSM_4K_DIV
);
2113 map
->blocks_per_strip
*= IMSM_4K_DIV
;
2114 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
2118 struct bbm_log
*log
= (void *)mpb
+
2119 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
2122 for (i
= 0; i
< log
->entry_count
; i
++) {
2123 struct bbm_log_entry
*entry
=
2124 &log
->marked_block_entries
[i
];
2126 __u8 count
= entry
->marked_count
+ 1;
2127 unsigned long long sector
=
2128 __le48_to_cpu(&entry
->defective_block_start
);
2130 entry
->defective_block_start
=
2131 __cpu_to_le48(sector
*IMSM_4K_DIV
);
2132 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
2136 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
2140 * imsm_check_attributes() - Check if features represented by attributes flags are supported.
2142 * @attributes: attributes read from metadata.
2143 * Returns: true if all features are supported, false otherwise.
2145 static bool imsm_check_attributes(__u32 attributes
)
2147 if ((attributes
& (MPB_ATTRIB_SUPPORTED
| MPB_ATTRIB_IGNORED
)) == attributes
)
2153 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
2155 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
2157 struct intel_super
*super
= st
->sb
;
2158 struct imsm_super
*mpb
= super
->anchor
;
2159 char str
[MAX_SIGNATURE_LENGTH
];
2164 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2166 time_t creation_time
;
2168 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
2169 str
[MPB_SIG_LEN
-1] = '\0';
2170 printf(" Magic : %s\n", str
);
2171 printf(" Version : %s\n", get_imsm_version(mpb
));
2172 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
2173 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
2174 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
2175 creation_time
= __le64_to_cpu(mpb
->creation_time
);
2176 printf(" Creation Time : %.24s\n",
2177 creation_time
? ctime(&creation_time
) : "Unknown");
2179 printf(" Attributes : %08x (%s)\n", mpb
->attributes
,
2180 imsm_check_attributes(mpb
->attributes
) ? "supported" : "not supported");
2182 getinfo_super_imsm(st
, &info
, NULL
);
2183 fname_from_uuid(&info
, nbuf
);
2184 printf(" UUID : %s\n", nbuf
+ 5);
2185 sum
= __le32_to_cpu(mpb
->check_sum
);
2186 printf(" Checksum : %08x %s\n", sum
,
2187 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
2188 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
2189 printf(" Disks : %d\n", mpb
->num_disks
);
2190 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
2191 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
2192 super
->disks
->index
, reserved
, super
->sector_size
);
2193 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
2194 struct bbm_log
*log
= super
->bbm_log
;
2197 printf("Bad Block Management Log:\n");
2198 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
2199 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
2200 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
2202 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2204 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2206 super
->current_vol
= i
;
2207 getinfo_super_imsm(st
, &info
, NULL
);
2208 fname_from_uuid(&info
, nbuf
);
2209 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
2211 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2212 if (i
== super
->disks
->index
)
2214 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
2215 super
->sector_size
);
2218 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2219 if (dl
->index
== -1)
2220 print_imsm_disk(&dl
->disk
, -1, reserved
,
2221 super
->sector_size
);
2223 examine_migr_rec_imsm(super
);
2226 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
2228 /* We just write a generic IMSM ARRAY entry */
2232 getinfo_super_imsm(st
, &info
, NULL
);
2233 fname_from_uuid(&info
, nbuf
);
2234 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
2237 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
2239 /* We just write a generic IMSM ARRAY entry */
2243 struct intel_super
*super
= st
->sb
;
2246 if (!super
->anchor
->num_raid_devs
)
2249 getinfo_super_imsm(st
, &info
, NULL
);
2250 fname_from_uuid(&info
, nbuf
);
2251 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2252 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2254 super
->current_vol
= i
;
2255 getinfo_super_imsm(st
, &info
, NULL
);
2256 fname_from_uuid(&info
, nbuf1
);
2257 printf("ARRAY " DEV_MD_DIR
"%.16s container=%s member=%d UUID=%s\n",
2258 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
2262 static void export_examine_super_imsm(struct supertype
*st
)
2264 struct intel_super
*super
= st
->sb
;
2265 struct imsm_super
*mpb
= super
->anchor
;
2269 getinfo_super_imsm(st
, &info
, NULL
);
2270 fname_from_uuid(&info
, nbuf
);
2271 printf("MD_METADATA=imsm\n");
2272 printf("MD_LEVEL=container\n");
2273 printf("MD_UUID=%s\n", nbuf
+5);
2274 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
2275 printf("MD_CREATION_TIME=%llu\n", __le64_to_cpu(mpb
->creation_time
));
2278 static void detail_super_imsm(struct supertype
*st
, char *homehost
,
2283 struct intel_super
*super
= st
->sb
;
2284 int temp_vol
= super
->current_vol
;
2287 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2289 getinfo_super_imsm(st
, &info
, NULL
);
2290 fname_from_uuid(&info
, nbuf
);
2291 printf("\n UUID : %s\n", nbuf
+ 5);
2293 super
->current_vol
= temp_vol
;
2296 static void brief_detail_super_imsm(struct supertype
*st
, char *subarray
)
2300 struct intel_super
*super
= st
->sb
;
2301 int temp_vol
= super
->current_vol
;
2304 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2306 getinfo_super_imsm(st
, &info
, NULL
);
2307 fname_from_uuid(&info
, nbuf
);
2308 printf(" UUID=%s", nbuf
+ 5);
2310 super
->current_vol
= temp_vol
;
2313 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
,
2314 size_t serial_buf_len
);
2315 static void fd2devname(int fd
, char *name
);
2317 void print_encryption_information(int disk_fd
, enum sys_dev_type hba_type
)
2319 struct encryption_information information
= {0};
2320 mdadm_status_t status
= MDADM_STATUS_SUCCESS
;
2321 const char *indent
= " ";
2326 status
= get_nvme_opal_encryption_information(disk_fd
, &information
, 1);
2329 case SYS_DEV_SATA_VMD
:
2330 status
= get_ata_encryption_information(disk_fd
, &information
, 1);
2337 pr_err("Failed to get drive encryption information.\n");
2341 printf("%sEncryption(Ability|Status): %s|%s\n", indent
,
2342 get_encryption_ability_string(information
.ability
),
2343 get_encryption_status_string(information
.status
));
2346 static int ahci_enumerate_ports(struct sys_dev
*hba
, int port_count
, int host_base
, int verbose
)
2348 /* dump an unsorted list of devices attached to AHCI Intel storage
2349 * controller, as well as non-connected ports
2351 int hba_len
= strlen(hba
->path
) + 1;
2356 unsigned long port_mask
= (1 << port_count
) - 1;
2358 if (port_count
> (int)sizeof(port_mask
) * 8) {
2360 pr_err("port_count %d out of range\n", port_count
);
2364 /* scroll through /sys/dev/block looking for devices attached to
2367 dir
= opendir("/sys/dev/block");
2371 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2377 char device
[PATH_MAX
];
2382 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2384 path
= devt_to_devpath(makedev(major
, minor
), 1, NULL
);
2387 if (!path_attached_to_hba(path
, hba
->path
)) {
2393 /* retrieve the scsi device */
2394 if (!devt_to_devpath(makedev(major
, minor
), 1, device
)) {
2396 pr_err("failed to get device\n");
2400 if (devpath_to_char(device
, "type", buf
, sizeof(buf
), 0)) {
2404 type
= strtoul(buf
, NULL
, 10);
2406 /* if it's not a disk print the vendor and model */
2407 if (!(type
== 0 || type
== 7 || type
== 14)) {
2411 if (devpath_to_char(device
, "vendor", buf
,
2412 sizeof(buf
), 0) == 0) {
2413 strncpy(vendor
, buf
, sizeof(vendor
));
2414 vendor
[sizeof(vendor
) - 1] = '\0';
2415 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2416 while (isspace(*c
) || *c
== '\0')
2421 if (devpath_to_char(device
, "model", buf
,
2422 sizeof(buf
), 0) == 0) {
2423 strncpy(model
, buf
, sizeof(model
));
2424 model
[sizeof(model
) - 1] = '\0';
2425 c
= (char *) &model
[sizeof(model
) - 1];
2426 while (isspace(*c
) || *c
== '\0')
2430 if (vendor
[0] && model
[0])
2431 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2433 switch (type
) { /* numbers from hald/linux/device.c */
2434 case 1: sprintf(buf
, "tape"); break;
2435 case 2: sprintf(buf
, "printer"); break;
2436 case 3: sprintf(buf
, "processor"); break;
2438 case 5: sprintf(buf
, "cdrom"); break;
2439 case 6: sprintf(buf
, "scanner"); break;
2440 case 8: sprintf(buf
, "media_changer"); break;
2441 case 9: sprintf(buf
, "comm"); break;
2442 case 12: sprintf(buf
, "raid"); break;
2443 default: sprintf(buf
, "unknown");
2448 /* chop device path to 'host%d' and calculate the port number */
2449 c
= strchr(&path
[hba_len
], '/');
2452 pr_err("%s - invalid path name\n", path
+ hba_len
);
2457 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2458 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2462 *c
= '/'; /* repair the full string */
2463 pr_err("failed to determine port number for %s\n",
2470 /* mark this port as used */
2471 port_mask
&= ~(1 << port
);
2473 /* print out the device information */
2475 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2479 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2480 if (!is_fd_valid(fd
))
2481 printf(" Port%d : - disk info unavailable -\n", port
);
2483 fd2devname(fd
, buf
);
2484 printf(" Port%d : %s", port
, buf
);
2485 if (imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2487 printf(" (%s)\n", buf
);
2491 print_encryption_information(fd
, hba
->type
);
2504 for (i
= 0; i
< port_count
; i
++)
2505 if (port_mask
& (1 << i
))
2506 printf(" Port%d : - no device attached -\n", i
);
2512 static int print_nvme_info(struct sys_dev
*hba
)
2517 dir
= opendir("/sys/block/");
2521 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2522 char ns_path
[PATH_MAX
];
2523 char cntrl_path
[PATH_MAX
];
2527 if (!strstr(ent
->d_name
, "nvme"))
2530 fd
= open_dev(ent
->d_name
);
2531 if (!is_fd_valid(fd
))
2534 if (!diskfd_to_devpath(fd
, 0, ns_path
) ||
2535 !diskfd_to_devpath(fd
, 1, cntrl_path
))
2538 if (!path_attached_to_hba(cntrl_path
, hba
->path
))
2541 if (!imsm_is_nvme_namespace_supported(fd
, 0))
2544 fd2devname(fd
, buf
);
2545 if (hba
->type
== SYS_DEV_VMD
)
2546 printf(" NVMe under VMD : %s", buf
);
2547 else if (hba
->type
== SYS_DEV_NVME
)
2548 printf(" NVMe Device : %s", buf
);
2550 if (!imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2552 printf(" (%s)\n", buf
);
2556 print_encryption_information(fd
, hba
->type
);
2566 static void print_found_intel_controllers(struct sys_dev
*elem
)
2568 for (; elem
; elem
= elem
->next
) {
2569 pr_err("found Intel(R) ");
2570 if (elem
->type
== SYS_DEV_SATA
)
2571 fprintf(stderr
, "SATA ");
2572 else if (elem
->type
== SYS_DEV_SAS
)
2573 fprintf(stderr
, "SAS ");
2574 else if (elem
->type
== SYS_DEV_NVME
)
2575 fprintf(stderr
, "NVMe ");
2577 if (elem
->type
== SYS_DEV_VMD
)
2578 fprintf(stderr
, "VMD domain");
2579 else if (elem
->type
== SYS_DEV_SATA_VMD
)
2580 fprintf(stderr
, "SATA VMD domain");
2582 fprintf(stderr
, "RAID controller");
2585 fprintf(stderr
, " at %s", elem
->pci_id
);
2586 fprintf(stderr
, ".\n");
2591 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2598 if ((dir
= opendir(hba_path
)) == NULL
)
2601 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2604 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2605 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2607 if (*port_count
== 0)
2609 else if (host
< host_base
)
2612 if (host
+ 1 > *port_count
+ host_base
)
2613 *port_count
= host
+ 1 - host_base
;
2619 static void print_imsm_level_capability(const struct imsm_orom
*orom
)
2623 for (idx
= 0; imsm_level_ops
[idx
].name
; idx
++)
2624 if (imsm_level_ops
[idx
].is_level_supported(orom
))
2625 printf("%s ", imsm_level_ops
[idx
].name
);
2628 static void print_imsm_capability(const struct imsm_orom
*orom
)
2630 printf(" Platform : Intel(R) ");
2631 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2632 printf("Matrix Storage Manager\n");
2633 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2634 printf("Virtual RAID on CPU\n");
2636 printf("Rapid Storage Technology%s\n",
2637 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2638 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
) {
2639 if (imsm_orom_is_vmd_without_efi(orom
))
2640 printf(" Version : %d.%d\n", orom
->major_ver
,
2643 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2644 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2647 printf(" RAID Levels : ");
2648 print_imsm_level_capability(orom
);
2651 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2652 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2653 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2654 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2655 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2656 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2657 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2658 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2659 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2660 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2661 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2662 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2663 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2664 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2665 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2666 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2667 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2668 printf(" 2TB volumes :%s supported\n",
2669 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2670 printf(" 2TB disks :%s supported\n",
2671 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2672 printf(" Max Disks : %d\n", orom
->tds
);
2673 printf(" Max Volumes : %d per array, %d per %s\n",
2674 orom
->vpa
, orom
->vphba
,
2675 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2679 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2681 printf("MD_FIRMWARE_TYPE=imsm\n");
2682 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2683 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2684 orom
->hotfix_ver
, orom
->build
);
2686 printf("IMSM_SUPPORTED_RAID_LEVELS=");
2687 print_imsm_level_capability(orom
);
2690 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2691 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2692 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2693 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2694 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2695 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2696 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2697 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2698 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2699 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2700 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2701 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2702 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2703 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2704 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2705 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2706 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2707 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2708 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2709 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2710 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2711 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2714 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2716 /* There are two components to imsm platform support, the ahci SATA
2717 * controller and the option-rom. To find the SATA controller we
2718 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2719 * controller with the Intel vendor id is present. This approach
2720 * allows mdadm to leverage the kernel's ahci detection logic, with the
2721 * caveat that if ahci.ko is not loaded mdadm will not be able to
2722 * detect platform raid capabilities. The option-rom resides in a
2723 * platform "Adapter ROM". We scan for its signature to retrieve the
2724 * platform capabilities. If raid support is disabled in the BIOS the
2725 * option-rom capability structure will not be available.
2727 struct sys_dev
*list
, *hba
;
2732 if (enumerate_only
) {
2733 if (check_no_platform())
2735 list
= find_intel_devices();
2738 for (hba
= list
; hba
; hba
= hba
->next
) {
2739 if (find_imsm_capability(hba
)) {
2749 list
= find_intel_devices();
2752 pr_err("no active Intel(R) RAID controller found.\n");
2754 } else if (verbose
> 0)
2755 print_found_intel_controllers(list
);
2757 for (hba
= list
; hba
; hba
= hba
->next
) {
2758 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2760 if (!find_imsm_capability(hba
)) {
2762 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2763 hba
->type
== SYS_DEV_VMD
|| hba
->type
== SYS_DEV_SATA_VMD
?
2764 vmd_domain_to_controller(hba
, buf
) :
2765 hba
->path
, get_sys_dev_type(hba
->type
));
2771 if (controller_path
&& result
== 1) {
2772 pr_err("no active Intel(R) RAID controller found under %s\n",
2777 const struct orom_entry
*entry
;
2779 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2780 if (entry
->type
== SYS_DEV_VMD
) {
2781 print_imsm_capability(&entry
->orom
);
2782 printf(" 3rd party NVMe :%s supported\n",
2783 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2784 for (hba
= list
; hba
; hba
= hba
->next
) {
2785 if (hba
->type
== SYS_DEV_VMD
) {
2787 printf(" I/O Controller : %s (%s)\n",
2788 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2789 if (print_nvme_info(hba
)) {
2791 pr_err("failed to get devices attached to VMD domain.\n");
2800 print_imsm_capability(&entry
->orom
);
2801 if (entry
->type
== SYS_DEV_NVME
) {
2802 for (hba
= list
; hba
; hba
= hba
->next
) {
2803 if (hba
->type
== SYS_DEV_NVME
)
2804 print_nvme_info(hba
);
2810 struct devid_list
*devid
;
2811 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2812 hba
= device_by_id(devid
->devid
);
2816 printf(" I/O Controller : %s (%s)\n",
2817 hba
->path
, get_sys_dev_type(hba
->type
));
2818 if (hba
->type
== SYS_DEV_SATA
|| hba
->type
== SYS_DEV_SATA_VMD
) {
2819 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2820 if (ahci_enumerate_ports(hba
, port_count
, host_base
, verbose
)) {
2822 pr_err("failed to enumerate ports on %s controller at %s.\n",
2823 get_sys_dev_type(hba
->type
), hba
->pci_id
);
2834 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2836 struct sys_dev
*list
, *hba
;
2839 list
= find_intel_devices();
2842 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2847 for (hba
= list
; hba
; hba
= hba
->next
) {
2848 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2850 if (!find_imsm_capability(hba
) && verbose
> 0) {
2852 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2853 hba
->type
== SYS_DEV_VMD
|| hba
->type
== SYS_DEV_SATA_VMD
?
2854 vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2860 const struct orom_entry
*entry
;
2862 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2863 if (entry
->type
== SYS_DEV_VMD
|| entry
->type
== SYS_DEV_SATA_VMD
) {
2864 for (hba
= list
; hba
; hba
= hba
->next
)
2865 print_imsm_capability_export(&entry
->orom
);
2868 print_imsm_capability_export(&entry
->orom
);
2874 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2876 /* the imsm metadata format does not specify any host
2877 * identification information. We return -1 since we can never
2878 * confirm nor deny whether a given array is "meant" for this
2879 * host. We rely on compare_super and the 'family_num' fields to
2880 * exclude member disks that do not belong, and we rely on
2881 * mdadm.conf to specify the arrays that should be assembled.
2882 * Auto-assembly may still pick up "foreign" arrays.
2888 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2890 /* The uuid returned here is used for:
2891 * uuid to put into bitmap file (Create, Grow)
2892 * uuid for backup header when saving critical section (Grow)
2893 * comparing uuids when re-adding a device into an array
2894 * In these cases the uuid required is that of the data-array,
2895 * not the device-set.
2896 * uuid to recognise same set when adding a missing device back
2897 * to an array. This is a uuid for the device-set.
2899 * For each of these we can make do with a truncated
2900 * or hashed uuid rather than the original, as long as
2902 * In each case the uuid required is that of the data-array,
2903 * not the device-set.
2905 /* imsm does not track uuid's so we synthesis one using sha1 on
2906 * - The signature (Which is constant for all imsm array, but no matter)
2907 * - the orig_family_num of the container
2908 * - the index number of the volume
2909 * - the 'serial' number of the volume.
2910 * Hopefully these are all constant.
2912 struct intel_super
*super
= st
->sb
;
2915 struct sha1_ctx ctx
;
2916 struct imsm_dev
*dev
= NULL
;
2919 /* some mdadm versions failed to set ->orig_family_num, in which
2920 * case fall back to ->family_num. orig_family_num will be
2921 * fixed up with the first metadata update.
2923 family_num
= super
->anchor
->orig_family_num
;
2924 if (family_num
== 0)
2925 family_num
= super
->anchor
->family_num
;
2926 sha1_init_ctx(&ctx
);
2927 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2928 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2929 if (super
->current_vol
>= 0)
2930 dev
= get_imsm_dev(super
, super
->current_vol
);
2932 __u32 vol
= super
->current_vol
;
2933 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2934 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2936 sha1_finish_ctx(&ctx
, buf
);
2937 memcpy(uuid
, buf
, 4*4);
2940 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2942 /* migr_strip_size when repairing or initializing parity */
2943 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2944 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2946 switch (get_imsm_raid_level(map
)) {
2951 return 128*1024 >> 9;
2955 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2957 /* migr_strip_size when rebuilding a degraded disk, no idea why
2958 * this is different than migr_strip_size_resync(), but it's good
2961 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2962 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2964 switch (get_imsm_raid_level(map
)) {
2967 if (map
->num_members
% map
->num_domains
== 0)
2968 return 128*1024 >> 9;
2972 return max((__u32
) 64*1024 >> 9, chunk
);
2974 return 128*1024 >> 9;
2978 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2980 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2981 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2982 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2983 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2985 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2988 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2990 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2991 int level
= get_imsm_raid_level(lo
);
2993 if (level
== 1 || level
== 10) {
2994 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2996 return hi
->num_domains
;
2998 return num_stripes_per_unit_resync(dev
);
3001 static unsigned long long calc_component_size(struct imsm_map
*map
,
3002 struct imsm_dev
*dev
)
3004 unsigned long long component_size
;
3005 unsigned long long dev_size
= imsm_dev_size(dev
);
3006 long long calc_dev_size
= 0;
3007 unsigned int member_disks
= imsm_num_data_members(map
);
3009 if (member_disks
== 0)
3012 component_size
= per_dev_array_size(map
);
3013 calc_dev_size
= component_size
* member_disks
;
3015 /* Component size is rounded to 1MB so difference between size from
3016 * metadata and size calculated from num_data_stripes equals up to
3017 * 2048 blocks per each device. If the difference is higher it means
3018 * that array size was expanded and num_data_stripes was not updated.
3020 if (llabs(calc_dev_size
- (long long)dev_size
) >
3021 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
3022 component_size
= dev_size
/ member_disks
;
3023 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
3024 component_size
/ map
->blocks_per_strip
,
3025 num_data_stripes(map
));
3028 return component_size
;
3031 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
3033 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3034 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
3036 switch(get_imsm_raid_level(map
)) {
3039 return chunk
* map
->num_domains
;
3041 return chunk
* map
->num_members
;
3047 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
3049 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
3050 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
3051 __u32 strip
= block
/ chunk
;
3053 switch (get_imsm_raid_level(map
)) {
3056 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
3057 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
3059 return vol_stripe
* chunk
+ block
% chunk
;
3061 __u32 stripe
= strip
/ (map
->num_members
- 1);
3063 return stripe
* chunk
+ block
% chunk
;
3070 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
3071 struct imsm_dev
*dev
)
3073 /* calculate the conversion factor between per member 'blocks'
3074 * (md/{resync,rebuild}_start) and imsm migration units, return
3075 * 0 for the 'not migrating' and 'unsupported migration' cases
3077 if (!dev
->vol
.migr_state
)
3080 switch (migr_type(dev
)) {
3081 case MIGR_GEN_MIGR
: {
3082 struct migr_record
*migr_rec
= super
->migr_rec
;
3083 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
3088 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3089 __u32 stripes_per_unit
;
3090 __u32 blocks_per_unit
;
3099 /* yes, this is really the translation of migr_units to
3100 * per-member blocks in the 'resync' case
3102 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
3103 migr_chunk
= migr_strip_blocks_resync(dev
);
3104 disks
= imsm_num_data_members(map
);
3105 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
3106 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
3107 segment
= blocks_per_unit
/ stripe
;
3108 block_rel
= blocks_per_unit
- segment
* stripe
;
3109 parity_depth
= parity_segment_depth(dev
);
3110 block_map
= map_migr_block(dev
, block_rel
);
3111 return block_map
+ parity_depth
* segment
;
3113 case MIGR_REBUILD
: {
3114 __u32 stripes_per_unit
;
3117 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
3118 migr_chunk
= migr_strip_blocks_rebuild(dev
);
3119 return migr_chunk
* stripes_per_unit
;
3121 case MIGR_STATE_CHANGE
:
3127 static int imsm_level_to_layout(int level
)
3135 return ALGORITHM_LEFT_ASYMMETRIC
;
3142 /*******************************************************************************
3143 * Function: read_imsm_migr_rec
3144 * Description: Function reads imsm migration record from last sector of disk
3146 * fd : disk descriptor
3147 * super : metadata info
3151 ******************************************************************************/
3152 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
3155 unsigned int sector_size
= super
->sector_size
;
3156 unsigned long long dsize
;
3158 get_dev_size(fd
, NULL
, &dsize
);
3159 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
3161 pr_err("Cannot seek to anchor block: %s\n",
3165 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
3166 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3167 MIGR_REC_BUF_SECTORS
*sector_size
) {
3168 pr_err("Cannot read migr record block: %s\n",
3173 if (sector_size
== 4096)
3174 convert_from_4k_imsm_migr_rec(super
);
3180 static struct imsm_dev
*imsm_get_device_during_migration(
3181 struct intel_super
*super
)
3184 struct intel_dev
*dv
;
3186 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
3187 if (is_gen_migration(dv
->dev
))
3193 /*******************************************************************************
3194 * Function: load_imsm_migr_rec
3195 * Description: Function reads imsm migration record (it is stored at the last
3198 * super : imsm internal array info
3202 * -2 : no migration in progress
3203 ******************************************************************************/
3204 static int load_imsm_migr_rec(struct intel_super
*super
)
3210 struct imsm_dev
*dev
;
3211 struct imsm_map
*map
;
3215 /* find map under migration */
3216 dev
= imsm_get_device_during_migration(super
);
3217 /* nothing to load,no migration in progress?
3222 map
= get_imsm_map(dev
, MAP_0
);
3226 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3227 /* skip spare and failed disks
3231 /* read only from one of the first two slots
3233 slot
= get_imsm_disk_slot(map
, dl
->index
);
3234 if (slot
> 1 || slot
< 0)
3237 if (!is_fd_valid(dl
->fd
)) {
3238 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3239 fd
= dev_open(nm
, O_RDONLY
);
3241 if (is_fd_valid(fd
)) {
3251 if (!is_fd_valid(fd
))
3253 retval
= read_imsm_migr_rec(fd
, super
);
3260 /*******************************************************************************
3261 * function: imsm_create_metadata_checkpoint_update
3262 * Description: It creates update for checkpoint change.
3264 * super : imsm internal array info
3265 * u : pointer to prepared update
3268 * If length is equal to 0, input pointer u contains no update
3269 ******************************************************************************/
3270 static int imsm_create_metadata_checkpoint_update(
3271 struct intel_super
*super
,
3272 struct imsm_update_general_migration_checkpoint
**u
)
3275 int update_memory_size
= 0;
3277 dprintf("(enter)\n");
3283 /* size of all update data without anchor */
3284 update_memory_size
=
3285 sizeof(struct imsm_update_general_migration_checkpoint
);
3287 *u
= xcalloc(1, update_memory_size
);
3289 dprintf("error: cannot get memory\n");
3292 (*u
)->type
= update_general_migration_checkpoint
;
3293 (*u
)->curr_migr_unit
= current_migr_unit(super
->migr_rec
);
3294 dprintf("prepared for %llu\n", (unsigned long long)(*u
)->curr_migr_unit
);
3296 return update_memory_size
;
3299 static void imsm_update_metadata_locally(struct supertype
*st
,
3300 void *buf
, int len
);
3302 /*******************************************************************************
3303 * Function: write_imsm_migr_rec
3304 * Description: Function writes imsm migration record
3305 * (at the last sector of disk)
3307 * super : imsm internal array info
3311 ******************************************************************************/
3312 static int write_imsm_migr_rec(struct supertype
*st
)
3314 struct intel_super
*super
= st
->sb
;
3315 unsigned int sector_size
= super
->sector_size
;
3316 unsigned long long dsize
;
3320 struct imsm_update_general_migration_checkpoint
*u
;
3321 struct imsm_dev
*dev
;
3322 struct imsm_map
*map
;
3324 /* find map under migration */
3325 dev
= imsm_get_device_during_migration(super
);
3326 /* if no migration, write buffer anyway to clear migr_record
3327 * on disk based on first available device
3330 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3331 super
->current_vol
);
3333 map
= get_imsm_map(dev
, MAP_0
);
3335 if (sector_size
== 4096)
3336 convert_to_4k_imsm_migr_rec(super
);
3337 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3340 /* skip failed and spare devices */
3343 /* write to 2 first slots only */
3345 slot
= get_imsm_disk_slot(map
, sd
->index
);
3346 if (map
== NULL
|| slot
> 1 || slot
< 0)
3349 get_dev_size(sd
->fd
, NULL
, &dsize
);
3350 if (lseek64(sd
->fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*
3353 pr_err("Cannot seek to anchor block: %s\n",
3357 if ((unsigned int)write(sd
->fd
, super
->migr_rec_buf
,
3358 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3359 MIGR_REC_BUF_SECTORS
*sector_size
) {
3360 pr_err("Cannot write migr record block: %s\n",
3365 if (sector_size
== 4096)
3366 convert_from_4k_imsm_migr_rec(super
);
3367 /* update checkpoint information in metadata */
3368 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3370 dprintf("imsm: Cannot prepare update\n");
3373 /* update metadata locally */
3374 imsm_update_metadata_locally(st
, u
, len
);
3375 /* and possibly remotely */
3376 if (st
->update_tail
) {
3377 append_metadata_update(st
, u
, len
);
3378 /* during reshape we do all work inside metadata handler
3379 * manage_reshape(), so metadata update has to be triggered
3382 flush_metadata_updates(st
);
3383 st
->update_tail
= &st
->updates
;
3392 /* spare/missing disks activations are not allowe when
3393 * array/container performs reshape operation, because
3394 * all arrays in container works on the same disks set
3396 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3399 struct intel_dev
*i_dev
;
3400 struct imsm_dev
*dev
;
3402 /* check whole container
3404 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3406 if (is_gen_migration(dev
)) {
3407 /* No repair during any migration in container
3415 static unsigned long long imsm_component_size_alignment_check(int level
,
3417 unsigned int sector_size
,
3418 unsigned long long component_size
)
3420 unsigned int component_size_alignment
;
3422 /* check component size alignment
3424 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3426 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3427 level
, chunk_size
, component_size
,
3428 component_size_alignment
);
3430 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3431 dprintf("imsm: reported component size aligned from %llu ",
3433 component_size
-= component_size_alignment
;
3434 dprintf_cont("to %llu (%i).\n",
3435 component_size
, component_size_alignment
);
3438 return component_size
;
3441 /*******************************************************************************
3442 * Function: get_bitmap_header_sector
3443 * Description: Returns the sector where the bitmap header is placed.
3445 * st : supertype information
3446 * dev_idx : index of the device with bitmap
3449 * The sector where the bitmap header is placed
3450 ******************************************************************************/
3451 static unsigned long long get_bitmap_header_sector(struct intel_super
*super
,
3454 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3455 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3457 if (!super
->sector_size
) {
3458 dprintf("sector size is not set\n");
3462 return pba_of_lba0(map
) + calc_component_size(map
, dev
) +
3463 (IMSM_BITMAP_HEADER_OFFSET
/ super
->sector_size
);
3466 /*******************************************************************************
3467 * Function: get_bitmap_sector
3468 * Description: Returns the sector where the bitmap is placed.
3470 * st : supertype information
3471 * dev_idx : index of the device with bitmap
3474 * The sector where the bitmap is placed
3475 ******************************************************************************/
3476 static unsigned long long get_bitmap_sector(struct intel_super
*super
,
3479 if (!super
->sector_size
) {
3480 dprintf("sector size is not set\n");
3484 return get_bitmap_header_sector(super
, dev_idx
) +
3485 (IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
);
3488 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3490 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3491 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3493 return pba_of_lba0(map
) +
3494 (num_data_stripes(map
) * map
->blocks_per_strip
);
3497 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3499 struct intel_super
*super
= st
->sb
;
3500 struct migr_record
*migr_rec
= super
->migr_rec
;
3501 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3502 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3503 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3504 struct imsm_map
*map_to_analyse
= map
;
3506 int map_disks
= info
->array
.raid_disks
;
3508 memset(info
, 0, sizeof(*info
));
3510 map_to_analyse
= prev_map
;
3512 dl
= super
->current_disk
;
3514 info
->container_member
= super
->current_vol
;
3515 info
->array
.raid_disks
= map
->num_members
;
3516 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3517 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3518 info
->array
.md_minor
= -1;
3519 info
->array
.ctime
= 0;
3520 info
->array
.utime
= 0;
3521 info
->array
.chunk_size
=
3522 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3523 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3524 info
->custom_array_size
= imsm_dev_size(dev
);
3525 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3527 if (is_gen_migration(dev
)) {
3529 * device prev_map should be added if it is in the middle
3534 info
->reshape_active
= 1;
3535 info
->new_level
= get_imsm_raid_level(map
);
3536 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3537 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3538 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3539 if (info
->delta_disks
) {
3540 /* this needs to be applied to every array
3543 info
->reshape_active
= CONTAINER_RESHAPE
;
3545 /* We shape information that we give to md might have to be
3546 * modify to cope with md's requirement for reshaping arrays.
3547 * For example, when reshaping a RAID0, md requires it to be
3548 * presented as a degraded RAID4.
3549 * Also if a RAID0 is migrating to a RAID5 we need to specify
3550 * the array as already being RAID5, but the 'before' layout
3551 * is a RAID4-like layout.
3553 switch (info
->array
.level
) {
3555 switch(info
->new_level
) {
3557 /* conversion is happening as RAID4 */
3558 info
->array
.level
= 4;
3559 info
->array
.raid_disks
+= 1;
3562 /* conversion is happening as RAID5 */
3563 info
->array
.level
= 5;
3564 info
->array
.layout
= ALGORITHM_PARITY_N
;
3565 info
->delta_disks
-= 1;
3568 /* FIXME error message */
3569 info
->array
.level
= UnSet
;
3575 info
->new_level
= UnSet
;
3576 info
->new_layout
= UnSet
;
3577 info
->new_chunk
= info
->array
.chunk_size
;
3578 info
->delta_disks
= 0;
3582 info
->disk
.major
= dl
->major
;
3583 info
->disk
.minor
= dl
->minor
;
3584 info
->disk
.number
= dl
->index
;
3585 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3589 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3590 info
->component_size
= calc_component_size(map
, dev
);
3591 info
->component_size
= imsm_component_size_alignment_check(
3593 info
->array
.chunk_size
,
3595 info
->component_size
);
3596 info
->bb
.supported
= 1;
3598 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3599 info
->recovery_start
= MaxSector
;
3601 if (info
->array
.level
== 5 &&
3602 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3603 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3604 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3605 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3606 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3607 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3609 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3611 } else if (info
->array
.level
<= 0) {
3612 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3614 if (dev
->rwh_policy
== RWH_BITMAP
) {
3615 info
->bitmap_offset
= get_bitmap_sector(super
, super
->current_vol
);
3616 info
->consistency_policy
= CONSISTENCY_POLICY_BITMAP
;
3618 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3622 info
->reshape_progress
= 0;
3623 info
->resync_start
= MaxSector
;
3624 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3625 !(info
->array
.state
& 1)) &&
3626 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3627 info
->resync_start
= 0;
3629 if (dev
->vol
.migr_state
) {
3630 switch (migr_type(dev
)) {
3633 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3635 __u64 units
= vol_curr_migr_unit(dev
);
3637 info
->resync_start
= blocks_per_unit
* units
;
3640 case MIGR_GEN_MIGR
: {
3641 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3643 __u64 units
= current_migr_unit(migr_rec
);
3646 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3648 (get_num_migr_units(migr_rec
)-1)) &&
3649 (super
->migr_rec
->rec_status
==
3650 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3653 info
->reshape_progress
= blocks_per_unit
* units
;
3655 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3656 (unsigned long long)units
,
3657 (unsigned long long)blocks_per_unit
,
3658 info
->reshape_progress
);
3660 used_disks
= imsm_num_data_members(prev_map
);
3661 if (used_disks
> 0) {
3662 info
->custom_array_size
= per_dev_array_size(map
) *
3667 /* we could emulate the checkpointing of
3668 * 'sync_action=check' migrations, but for now
3669 * we just immediately complete them
3672 /* this is handled by container_content_imsm() */
3673 case MIGR_STATE_CHANGE
:
3674 /* FIXME handle other migrations */
3676 /* we are not dirty, so... */
3677 info
->resync_start
= MaxSector
;
3681 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3682 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3684 info
->array
.major_version
= -1;
3685 info
->array
.minor_version
= -2;
3686 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3687 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3688 uuid_from_super_imsm(st
, info
->uuid
);
3692 for (i
=0; i
<map_disks
; i
++) {
3694 if (i
< info
->array
.raid_disks
) {
3695 struct imsm_disk
*dsk
;
3696 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3697 dsk
= get_imsm_disk(super
, j
);
3698 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3705 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3706 int failed
, int look_in_map
);
3708 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3711 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3713 if (is_gen_migration(dev
)) {
3716 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3718 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3719 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3720 if (map2
->map_state
!= map_state
) {
3721 map2
->map_state
= map_state
;
3722 super
->updates_pending
++;
3727 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3731 for (d
= super
->missing
; d
; d
= d
->next
)
3732 if (d
->index
== index
)
3737 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3739 struct intel_super
*super
= st
->sb
;
3740 struct imsm_disk
*disk
;
3741 int map_disks
= info
->array
.raid_disks
;
3742 int max_enough
= -1;
3744 struct imsm_super
*mpb
;
3746 if (super
->current_vol
>= 0) {
3747 getinfo_super_imsm_volume(st
, info
, map
);
3750 memset(info
, 0, sizeof(*info
));
3752 /* Set raid_disks to zero so that Assemble will always pull in valid
3755 info
->array
.raid_disks
= 0;
3756 info
->array
.level
= LEVEL_CONTAINER
;
3757 info
->array
.layout
= 0;
3758 info
->array
.md_minor
= -1;
3759 info
->array
.ctime
= 0; /* N/A for imsm */
3760 info
->array
.utime
= 0;
3761 info
->array
.chunk_size
= 0;
3763 info
->disk
.major
= 0;
3764 info
->disk
.minor
= 0;
3765 info
->disk
.raid_disk
= -1;
3766 info
->reshape_active
= 0;
3767 info
->array
.major_version
= -1;
3768 info
->array
.minor_version
= -2;
3769 strcpy(info
->text_version
, "imsm");
3770 info
->safe_mode_delay
= 0;
3771 info
->disk
.number
= -1;
3772 info
->disk
.state
= 0;
3774 info
->recovery_start
= MaxSector
;
3775 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3776 info
->bb
.supported
= 1;
3778 /* do we have the all the insync disks that we expect? */
3779 mpb
= super
->anchor
;
3780 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3782 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3783 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3784 int failed
, enough
, j
, missing
= 0;
3785 struct imsm_map
*map
;
3788 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3789 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3790 map
= get_imsm_map(dev
, MAP_0
);
3792 /* any newly missing disks?
3793 * (catches single-degraded vs double-degraded)
3795 for (j
= 0; j
< map
->num_members
; j
++) {
3796 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3797 __u32 idx
= ord_to_idx(ord
);
3799 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3800 info
->disk
.raid_disk
= j
;
3802 if (!(ord
& IMSM_ORD_REBUILD
) &&
3803 get_imsm_missing(super
, idx
)) {
3809 if (state
== IMSM_T_STATE_FAILED
)
3811 else if (state
== IMSM_T_STATE_DEGRADED
&&
3812 (state
!= map
->map_state
|| missing
))
3814 else /* we're normal, or already degraded */
3816 if (is_gen_migration(dev
) && missing
) {
3817 /* during general migration we need all disks
3818 * that process is running on.
3819 * No new missing disk is allowed.
3823 /* no more checks necessary
3827 /* in the missing/failed disk case check to see
3828 * if at least one array is runnable
3830 max_enough
= max(max_enough
, enough
);
3833 info
->container_enough
= max_enough
;
3836 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3838 disk
= &super
->disks
->disk
;
3839 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3840 info
->component_size
= reserved
;
3841 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3842 /* we don't change info->disk.raid_disk here because
3843 * this state will be finalized in mdmon after we have
3844 * found the 'most fresh' version of the metadata
3846 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3847 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3848 0 : (1 << MD_DISK_SYNC
);
3851 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3852 * ->compare_super may have updated the 'num_raid_devs' field for spares
3854 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3855 uuid_from_super_imsm(st
, info
->uuid
);
3857 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3859 /* I don't know how to compute 'map' on imsm, so use safe default */
3862 for (i
= 0; i
< map_disks
; i
++)
3868 /* allocates memory and fills disk in mdinfo structure
3869 * for each disk in array */
3870 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3872 struct mdinfo
*mddev
;
3873 struct intel_super
*super
= st
->sb
;
3874 struct imsm_disk
*disk
;
3877 if (!super
|| !super
->disks
)
3880 mddev
= xcalloc(1, sizeof(*mddev
));
3884 tmp
= xcalloc(1, sizeof(*tmp
));
3886 tmp
->next
= mddev
->devs
;
3888 tmp
->disk
.number
= count
++;
3889 tmp
->disk
.major
= dl
->major
;
3890 tmp
->disk
.minor
= dl
->minor
;
3891 tmp
->disk
.state
= is_configured(disk
) ?
3892 (1 << MD_DISK_ACTIVE
) : 0;
3893 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3894 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3895 tmp
->disk
.raid_disk
= -1;
3901 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3902 enum update_opt update
, char *devname
,
3903 int verbose
, int uuid_set
, char *homehost
)
3905 /* For 'assemble' and 'force' we need to return non-zero if any
3906 * change was made. For others, the return value is ignored.
3907 * Update options are:
3908 * force-one : This device looks a bit old but needs to be included,
3909 * update age info appropriately.
3910 * assemble: clear any 'faulty' flag to allow this device to
3912 * force-array: Array is degraded but being forced, mark it clean
3913 * if that will be needed to assemble it.
3915 * newdev: not used ????
3916 * grow: Array has gained a new device - this is currently for
3918 * resync: mark as dirty so a resync will happen.
3919 * name: update the name - preserving the homehost
3920 * uuid: Change the uuid of the array to match watch is given
3922 * Following are not relevant for this imsm:
3923 * sparc2.2 : update from old dodgey metadata
3924 * super-minor: change the preferred_minor number
3925 * summaries: update redundant counters.
3926 * homehost: update the recorded homehost
3927 * _reshape_progress: record new reshape_progress position.
3930 struct intel_super
*super
= st
->sb
;
3931 struct imsm_super
*mpb
;
3933 /* we can only update container info */
3934 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3937 mpb
= super
->anchor
;
3941 /* We take this to mean that the family_num should be updated.
3942 * However that is much smaller than the uuid so we cannot really
3943 * allow an explicit uuid to be given. And it is hard to reliably
3945 * So if !uuid_set we know the current uuid is random and just used
3946 * the first 'int' and copy it to the other 3 positions.
3947 * Otherwise we require the 4 'int's to be the same as would be the
3948 * case if we are using a random uuid. So an explicit uuid will be
3949 * accepted as long as all for ints are the same... which shouldn't hurt
3952 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3955 if (info
->uuid
[0] != info
->uuid
[1] ||
3956 info
->uuid
[1] != info
->uuid
[2] ||
3957 info
->uuid
[2] != info
->uuid
[3])
3963 mpb
->orig_family_num
= info
->uuid
[0];
3965 case UOPT_SPEC_ASSEMBLE
:
3973 /* successful update? recompute checksum */
3975 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3980 static size_t disks_to_mpb_size(int disks
)
3984 size
= sizeof(struct imsm_super
);
3985 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3986 size
+= 2 * sizeof(struct imsm_dev
);
3987 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3988 size
+= (4 - 2) * sizeof(struct imsm_map
);
3989 /* 4 possible disk_ord_tbl's */
3990 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3991 /* maximum bbm log */
3992 size
+= sizeof(struct bbm_log
);
3997 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3998 unsigned long long data_offset
)
4000 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
4003 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
4006 static void free_devlist(struct intel_super
*super
)
4008 struct intel_dev
*dv
;
4010 while (super
->devlist
) {
4011 dv
= super
->devlist
->next
;
4012 free(super
->devlist
->dev
);
4013 free(super
->devlist
);
4014 super
->devlist
= dv
;
4018 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
4020 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
4023 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
,
4027 * 0 same, or first was empty, and second was copied
4028 * 1 sb are different
4030 struct intel_super
*first
= st
->sb
;
4031 struct intel_super
*sec
= tst
->sb
;
4039 /* in platform dependent environment test if the disks
4040 * use the same Intel hba
4041 * if not on Intel hba at all, allow anything.
4042 * doesn't check HBAs if num_raid_devs is not set, as it means
4043 * it is a free floating spare, and all spares regardless of HBA type
4044 * will fall into separate container during the assembly
4046 if (first
->hba
&& sec
->hba
&& first
->anchor
->num_raid_devs
!= 0) {
4047 if (first
->hba
->type
!= sec
->hba
->type
) {
4049 pr_err("HBAs of devices do not match %s != %s\n",
4050 get_sys_dev_type(first
->hba
->type
),
4051 get_sys_dev_type(sec
->hba
->type
));
4054 if (first
->orom
!= sec
->orom
) {
4056 pr_err("HBAs of devices do not match %s != %s\n",
4057 first
->hba
->pci_id
, sec
->hba
->pci_id
);
4062 if (first
->anchor
->num_raid_devs
> 0 &&
4063 sec
->anchor
->num_raid_devs
> 0) {
4064 /* Determine if these disks might ever have been
4065 * related. Further disambiguation can only take place
4066 * in load_super_imsm_all
4068 __u32 first_family
= first
->anchor
->orig_family_num
;
4069 __u32 sec_family
= sec
->anchor
->orig_family_num
;
4071 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
4072 MAX_SIGNATURE_LENGTH
) != 0)
4075 if (first_family
== 0)
4076 first_family
= first
->anchor
->family_num
;
4077 if (sec_family
== 0)
4078 sec_family
= sec
->anchor
->family_num
;
4080 if (first_family
!= sec_family
)
4085 /* if an anchor does not have num_raid_devs set then it is a free
4086 * floating spare. don't assosiate spare with any array, as during assembly
4087 * spares shall fall into separate container, from which they can be moved
4090 if (first
->anchor
->num_raid_devs
^ sec
->anchor
->num_raid_devs
)
4096 static void fd2devname(int fd
, char *name
)
4104 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
4107 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
4109 char path
[PATH_MAX
];
4110 char *name
= fd2kname(fd
);
4115 if (strncmp(name
, "nvme", 4) != 0)
4118 if (!diskfd_to_devpath(fd
, 1, path
))
4121 return devpath_to_char(path
, "serial", buf
, buf_len
, 0);
4124 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
4126 static int imsm_read_serial(int fd
, char *devname
,
4127 __u8
*serial
, size_t serial_buf_len
)
4136 memset(buf
, 0, sizeof(buf
));
4138 if (check_env("IMSM_DEVNAME_AS_SERIAL")) {
4139 memset(serial
, 0, serial_buf_len
);
4140 fd2devname(fd
, (char *) serial
);
4144 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
4147 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
4151 pr_err("Failed to retrieve serial for %s\n",
4156 /* trim all whitespace and non-printable characters and convert
4159 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4162 /* ':' is reserved for use in placeholder serial
4163 * numbers for missing disks
4174 if (len
> serial_buf_len
) {
4175 /* truncate leading characters */
4176 dest
+= len
- serial_buf_len
;
4177 len
= serial_buf_len
;
4180 memset(serial
, 0, serial_buf_len
);
4181 memcpy(serial
, dest
, len
);
4186 static int serialcmp(__u8
*s1
, __u8
*s2
)
4188 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4191 static void serialcpy(__u8
*dest
, __u8
*src
)
4193 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4196 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4200 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4201 if (serialcmp(dl
->serial
, serial
) == 0)
4207 static struct imsm_disk
*
4208 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4212 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4213 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4215 if (serialcmp(disk
->serial
, serial
) == 0) {
4226 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4228 struct imsm_disk
*disk
;
4233 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4235 rv
= imsm_read_serial(fd
, devname
, serial
, MAX_RAID_SERIAL_LEN
);
4240 dl
= xcalloc(1, sizeof(*dl
));
4243 dl
->major
= major(stb
.st_rdev
);
4244 dl
->minor
= minor(stb
.st_rdev
);
4245 dl
->next
= super
->disks
;
4246 dl
->fd
= keep_fd
? fd
: -1;
4247 assert(super
->disks
== NULL
);
4249 serialcpy(dl
->serial
, serial
);
4252 fd2devname(fd
, name
);
4254 dl
->devname
= xstrdup(devname
);
4256 dl
->devname
= xstrdup(name
);
4258 /* look up this disk's index in the current anchor */
4259 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4262 /* only set index on disks that are a member of a
4263 * populated contianer, i.e. one with raid_devs
4265 if (is_failed(&dl
->disk
))
4267 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4274 /* When migrating map0 contains the 'destination' state while map1
4275 * contains the current state. When not migrating map0 contains the
4276 * current state. This routine assumes that map[0].map_state is set to
4277 * the current array state before being called.
4279 * Migration is indicated by one of the following states
4280 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4281 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4282 * map1state=unitialized)
4283 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4285 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4286 * map1state=degraded)
4287 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4290 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4291 __u8 to_state
, int migr_type
)
4293 struct imsm_map
*dest
;
4294 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4296 dev
->vol
.migr_state
= 1;
4297 set_migr_type(dev
, migr_type
);
4298 set_vol_curr_migr_unit(dev
, 0);
4299 dest
= get_imsm_map(dev
, MAP_1
);
4301 /* duplicate and then set the target end state in map[0] */
4302 memcpy(dest
, src
, sizeof_imsm_map(src
));
4303 if (migr_type
== MIGR_GEN_MIGR
) {
4307 for (i
= 0; i
< src
->num_members
; i
++) {
4308 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4309 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4313 if (migr_type
== MIGR_GEN_MIGR
)
4314 /* Clear migration record */
4315 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4317 src
->map_state
= to_state
;
4320 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4323 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4324 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4328 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4329 * completed in the last migration.
4331 * FIXME add support for raid-level-migration
4333 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == false) &&
4334 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4335 /* when final map state is other than expected
4336 * merge maps (not for migration)
4340 for (i
= 0; i
< prev
->num_members
; i
++)
4341 for (j
= 0; j
< map
->num_members
; j
++)
4342 /* during online capacity expansion
4343 * disks position can be changed
4344 * if takeover is used
4346 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4347 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4348 map
->disk_ord_tbl
[j
] |=
4349 prev
->disk_ord_tbl
[i
];
4352 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4353 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4356 dev
->vol
.migr_state
= 0;
4357 set_migr_type(dev
, 0);
4358 set_vol_curr_migr_unit(dev
, 0);
4359 map
->map_state
= map_state
;
4362 static int parse_raid_devices(struct intel_super
*super
)
4365 struct imsm_dev
*dev_new
;
4366 size_t len
, len_migr
;
4368 size_t space_needed
= 0;
4369 struct imsm_super
*mpb
= super
->anchor
;
4371 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4372 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4373 struct intel_dev
*dv
;
4375 len
= sizeof_imsm_dev(dev_iter
, 0);
4376 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4378 space_needed
+= len_migr
- len
;
4380 dv
= xmalloc(sizeof(*dv
));
4381 if (max_len
< len_migr
)
4383 if (max_len
> len_migr
)
4384 space_needed
+= max_len
- len_migr
;
4385 dev_new
= xmalloc(max_len
);
4386 imsm_copy_dev(dev_new
, dev_iter
);
4389 dv
->next
= super
->devlist
;
4390 super
->devlist
= dv
;
4393 /* ensure that super->buf is large enough when all raid devices
4396 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4399 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4400 super
->sector_size
);
4401 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4404 memcpy(buf
, super
->buf
, super
->len
);
4405 memset(buf
+ super
->len
, 0, len
- super
->len
);
4411 super
->extra_space
+= space_needed
;
4416 /*******************************************************************************
4417 * Function: check_mpb_migr_compatibility
4418 * Description: Function checks for unsupported migration features:
4419 * - migration optimization area (pba_of_lba0)
4420 * - descending reshape (ascending_migr)
4422 * super : imsm metadata information
4424 * 0 : migration is compatible
4425 * -1 : migration is not compatible
4426 ******************************************************************************/
4427 int check_mpb_migr_compatibility(struct intel_super
*super
)
4429 struct imsm_map
*map0
, *map1
;
4430 struct migr_record
*migr_rec
= super
->migr_rec
;
4433 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4434 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4436 if (dev_iter
->vol
.migr_state
== 1 &&
4437 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4438 /* This device is migrating */
4439 map0
= get_imsm_map(dev_iter
, MAP_0
);
4440 map1
= get_imsm_map(dev_iter
, MAP_1
);
4441 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4442 /* migration optimization area was used */
4444 if (migr_rec
->ascending_migr
== 0 &&
4445 migr_rec
->dest_depth_per_unit
> 0)
4446 /* descending reshape not supported yet */
4453 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4455 /* load_imsm_mpb - read matrix metadata
4456 * allocates super->mpb to be freed by free_imsm
4458 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4460 unsigned long long dsize
;
4461 unsigned long long sectors
;
4462 unsigned int sector_size
= super
->sector_size
;
4464 struct imsm_super
*anchor
;
4467 get_dev_size(fd
, NULL
, &dsize
);
4468 if (dsize
< 2*sector_size
) {
4470 pr_err("%s: device to small for imsm\n",
4475 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4477 pr_err("Cannot seek to anchor block on %s: %s\n",
4478 devname
, strerror(errno
));
4482 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4484 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4487 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4489 pr_err("Cannot read anchor block on %s: %s\n",
4490 devname
, strerror(errno
));
4495 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4497 pr_err("no IMSM anchor on %s\n", devname
);
4502 __free_imsm(super
, 0);
4503 /* reload capability and hba */
4505 /* capability and hba must be updated with new super allocation */
4506 find_intel_hba_capability(fd
, super
, devname
);
4507 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4508 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4510 pr_err("unable to allocate %zu byte mpb buffer\n",
4515 memcpy(super
->buf
, anchor
, sector_size
);
4517 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4520 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4521 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4522 pr_err("could not allocate migr_rec buffer\n");
4527 super
->clean_migration_record_by_mdmon
= 0;
4530 check_sum
= __gen_imsm_checksum(super
->anchor
);
4531 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4533 pr_err("IMSM checksum %x != %x on %s\n",
4535 __le32_to_cpu(super
->anchor
->check_sum
),
4543 /* read the extended mpb */
4544 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4546 pr_err("Cannot seek to extended mpb on %s: %s\n",
4547 devname
, strerror(errno
));
4551 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4552 super
->len
- sector_size
) != super
->len
- sector_size
) {
4554 pr_err("Cannot read extended mpb on %s: %s\n",
4555 devname
, strerror(errno
));
4559 check_sum
= __gen_imsm_checksum(super
->anchor
);
4560 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4562 pr_err("IMSM checksum %x != %x on %s\n",
4563 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4571 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4573 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4574 static void clear_hi(struct intel_super
*super
)
4576 struct imsm_super
*mpb
= super
->anchor
;
4578 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4580 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4581 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4582 disk
->total_blocks_hi
= 0;
4584 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4585 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4586 for (n
= 0; n
< 2; ++n
) {
4587 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4590 map
->pba_of_lba0_hi
= 0;
4591 map
->blocks_per_member_hi
= 0;
4592 map
->num_data_stripes_hi
= 0;
4598 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4602 err
= load_imsm_mpb(fd
, super
, devname
);
4605 if (super
->sector_size
== 4096)
4606 convert_from_4k(super
);
4607 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4610 err
= parse_raid_devices(super
);
4613 err
= load_bbm_log(super
);
4618 static void __free_imsm_disk(struct dl
*d
, int do_close
)
4630 static void free_imsm_disks(struct intel_super
*super
)
4634 while (super
->disks
) {
4636 super
->disks
= d
->next
;
4637 __free_imsm_disk(d
, 1);
4639 while (super
->disk_mgmt_list
) {
4640 d
= super
->disk_mgmt_list
;
4641 super
->disk_mgmt_list
= d
->next
;
4642 __free_imsm_disk(d
, 1);
4644 while (super
->missing
) {
4646 super
->missing
= d
->next
;
4647 __free_imsm_disk(d
, 1);
4652 /* free all the pieces hanging off of a super pointer */
4653 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4655 struct intel_hba
*elem
, *next
;
4661 /* unlink capability description */
4663 if (super
->migr_rec_buf
) {
4664 free(super
->migr_rec_buf
);
4665 super
->migr_rec_buf
= NULL
;
4668 free_imsm_disks(super
);
4669 free_devlist(super
);
4673 free((void *)elem
->path
);
4679 free(super
->bbm_log
);
4683 static void free_imsm(struct intel_super
*super
)
4685 __free_imsm(super
, 1);
4686 free(super
->bb
.entries
);
4690 static void free_super_imsm(struct supertype
*st
)
4692 struct intel_super
*super
= st
->sb
;
4701 static struct intel_super
*alloc_super(void)
4703 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4705 super
->current_vol
= -1;
4706 super
->create_offset
= ~((unsigned long long) 0);
4708 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4709 sizeof(struct md_bb_entry
));
4710 if (!super
->bb
.entries
) {
4719 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4721 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4723 struct sys_dev
*hba_name
;
4726 if (is_fd_valid(fd
) && test_partition(fd
)) {
4727 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4731 if (!is_fd_valid(fd
) || check_no_platform()) {
4736 hba_name
= find_disk_attached_hba(fd
, NULL
);
4739 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4743 rv
= attach_hba_to_super(super
, hba_name
);
4746 struct intel_hba
*hba
= super
->hba
;
4748 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4749 " but the container is assigned to Intel(R) %s %s (",
4751 get_sys_dev_type(hba_name
->type
),
4752 hba_name
->type
== SYS_DEV_VMD
|| hba_name
->type
== SYS_DEV_SATA_VMD
?
4753 "domain" : "RAID controller",
4754 hba_name
->pci_id
? : "Err!",
4755 get_sys_dev_type(super
->hba
->type
),
4756 hba
->type
== SYS_DEV_VMD
|| hba_name
->type
== SYS_DEV_SATA_VMD
?
4757 "domain" : "RAID controller");
4760 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4762 fprintf(stderr
, ", ");
4765 fprintf(stderr
, ").\n"
4766 " Mixing devices attached to different controllers is not allowed.\n");
4770 super
->orom
= find_imsm_capability(hba_name
);
4777 /* find_missing - helper routine for load_super_imsm_all that identifies
4778 * disks that have disappeared from the system. This routine relies on
4779 * the mpb being uptodate, which it is at load time.
4781 static int find_missing(struct intel_super
*super
)
4784 struct imsm_super
*mpb
= super
->anchor
;
4786 struct imsm_disk
*disk
;
4788 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4789 disk
= __get_imsm_disk(mpb
, i
);
4790 dl
= serial_to_dl(disk
->serial
, super
);
4794 dl
= xmalloc(sizeof(*dl
));
4798 dl
->devname
= xstrdup("missing");
4800 serialcpy(dl
->serial
, disk
->serial
);
4803 dl
->next
= super
->missing
;
4804 super
->missing
= dl
;
4810 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4812 struct intel_disk
*idisk
= disk_list
;
4815 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4817 idisk
= idisk
->next
;
4823 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4824 struct intel_super
*super
,
4825 struct intel_disk
**disk_list
)
4827 struct imsm_disk
*d
= &super
->disks
->disk
;
4828 struct imsm_super
*mpb
= super
->anchor
;
4831 for (i
= 0; i
< tbl_size
; i
++) {
4832 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4833 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4835 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4836 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4837 dprintf("mpb from %d:%d matches %d:%d\n",
4838 super
->disks
->major
,
4839 super
->disks
->minor
,
4840 table
[i
]->disks
->major
,
4841 table
[i
]->disks
->minor
);
4845 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4846 is_configured(d
) == is_configured(tbl_d
)) &&
4847 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4848 /* current version of the mpb is a
4849 * better candidate than the one in
4850 * super_table, but copy over "cross
4851 * generational" status
4853 struct intel_disk
*idisk
;
4855 dprintf("mpb from %d:%d replaces %d:%d\n",
4856 super
->disks
->major
,
4857 super
->disks
->minor
,
4858 table
[i
]->disks
->major
,
4859 table
[i
]->disks
->minor
);
4861 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4862 if (idisk
&& is_failed(&idisk
->disk
))
4863 tbl_d
->status
|= FAILED_DISK
;
4866 struct intel_disk
*idisk
;
4867 struct imsm_disk
*disk
;
4869 /* tbl_mpb is more up to date, but copy
4870 * over cross generational status before
4873 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4874 if (disk
&& is_failed(disk
))
4875 d
->status
|= FAILED_DISK
;
4877 idisk
= disk_list_get(d
->serial
, *disk_list
);
4880 if (disk
&& is_configured(disk
))
4881 idisk
->disk
.status
|= CONFIGURED_DISK
;
4884 dprintf("mpb from %d:%d prefer %d:%d\n",
4885 super
->disks
->major
,
4886 super
->disks
->minor
,
4887 table
[i
]->disks
->major
,
4888 table
[i
]->disks
->minor
);
4896 table
[tbl_size
++] = super
;
4900 /* update/extend the merged list of imsm_disk records */
4901 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4902 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4903 struct intel_disk
*idisk
;
4905 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4907 idisk
->disk
.status
|= disk
->status
;
4908 if (is_configured(&idisk
->disk
) ||
4909 is_failed(&idisk
->disk
))
4910 idisk
->disk
.status
&= ~(SPARE_DISK
);
4912 idisk
= xcalloc(1, sizeof(*idisk
));
4913 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4914 idisk
->disk
= *disk
;
4915 idisk
->next
= *disk_list
;
4919 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4926 static struct intel_super
*
4927 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4930 struct imsm_super
*mpb
= super
->anchor
;
4934 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4935 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4936 struct intel_disk
*idisk
;
4938 idisk
= disk_list_get(disk
->serial
, disk_list
);
4940 if (idisk
->owner
== owner
||
4941 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4944 dprintf("'%.16s' owner %d != %d\n",
4945 disk
->serial
, idisk
->owner
,
4948 dprintf("unknown disk %x [%d]: %.16s\n",
4949 __le32_to_cpu(mpb
->family_num
), i
,
4955 if (ok_count
== mpb
->num_disks
)
4960 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4962 struct intel_super
*s
;
4964 for (s
= super_list
; s
; s
= s
->next
) {
4965 if (family_num
!= s
->anchor
->family_num
)
4967 pr_err("Conflict, offlining family %#x on '%s'\n",
4968 __le32_to_cpu(family_num
), s
->disks
->devname
);
4972 static struct intel_super
*
4973 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4975 struct intel_super
*super_table
[len
];
4976 struct intel_disk
*disk_list
= NULL
;
4977 struct intel_super
*champion
, *spare
;
4978 struct intel_super
*s
, **del
;
4983 memset(super_table
, 0, sizeof(super_table
));
4984 for (s
= *super_list
; s
; s
= s
->next
)
4985 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4987 for (i
= 0; i
< tbl_size
; i
++) {
4988 struct imsm_disk
*d
;
4989 struct intel_disk
*idisk
;
4990 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4993 d
= &s
->disks
->disk
;
4995 /* 'd' must appear in merged disk list for its
4996 * configuration to be valid
4998 idisk
= disk_list_get(d
->serial
, disk_list
);
4999 if (idisk
&& idisk
->owner
== i
)
5000 s
= validate_members(s
, disk_list
, i
);
5005 dprintf("marking family: %#x from %d:%d offline\n",
5007 super_table
[i
]->disks
->major
,
5008 super_table
[i
]->disks
->minor
);
5012 /* This is where the mdadm implementation differs from the Windows
5013 * driver which has no strict concept of a container. We can only
5014 * assemble one family from a container, so when returning a prodigal
5015 * array member to this system the code will not be able to disambiguate
5016 * the container contents that should be assembled ("foreign" versus
5017 * "local"). It requires user intervention to set the orig_family_num
5018 * to a new value to establish a new container. The Windows driver in
5019 * this situation fixes up the volume name in place and manages the
5020 * foreign array as an independent entity.
5025 for (i
= 0; i
< tbl_size
; i
++) {
5026 struct intel_super
*tbl_ent
= super_table
[i
];
5032 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
5037 if (s
&& !is_spare
) {
5038 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
5040 } else if (!s
&& !is_spare
)
5053 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
5054 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
5056 /* collect all dl's onto 'champion', and update them to
5057 * champion's version of the status
5059 for (s
= *super_list
; s
; s
= s
->next
) {
5060 struct imsm_super
*mpb
= champion
->anchor
;
5061 struct dl
*dl
= s
->disks
;
5066 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
5068 for (i
= 0; i
< mpb
->num_disks
; i
++) {
5069 struct imsm_disk
*disk
;
5071 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
5074 /* only set index on disks that are a member of
5075 * a populated contianer, i.e. one with
5078 if (is_failed(&dl
->disk
))
5080 else if (is_spare(&dl
->disk
))
5086 if (i
>= mpb
->num_disks
) {
5087 struct intel_disk
*idisk
;
5089 idisk
= disk_list_get(dl
->serial
, disk_list
);
5090 if (idisk
&& is_spare(&idisk
->disk
) &&
5091 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
5099 dl
->next
= champion
->disks
;
5100 champion
->disks
= dl
;
5104 /* delete 'champion' from super_list */
5105 for (del
= super_list
; *del
; ) {
5106 if (*del
== champion
) {
5107 *del
= (*del
)->next
;
5110 del
= &(*del
)->next
;
5112 champion
->next
= NULL
;
5116 struct intel_disk
*idisk
= disk_list
;
5118 disk_list
= disk_list
->next
;
5126 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
5127 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5128 int major
, int minor
, int keep_fd
);
5130 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5131 int *max
, int keep_fd
);
5133 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
5134 char *devname
, struct md_list
*devlist
,
5137 struct intel_super
*super_list
= NULL
;
5138 struct intel_super
*super
= NULL
;
5142 if (is_fd_valid(fd
))
5143 /* 'fd' is an opened container */
5144 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
5146 /* get super block from devlist devices */
5147 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
5150 /* all mpbs enter, maybe one leaves */
5151 super
= imsm_thunderdome(&super_list
, i
);
5157 if (find_missing(super
) != 0) {
5163 /* load migration record */
5164 err
= load_imsm_migr_rec(super
);
5166 /* migration is in progress,
5167 * but migr_rec cannot be loaded,
5173 /* Check migration compatibility */
5174 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5175 pr_err("Unsupported migration detected");
5177 fprintf(stderr
, " on %s\n", devname
);
5179 fprintf(stderr
, " (IMSM).\n");
5188 while (super_list
) {
5189 struct intel_super
*s
= super_list
;
5191 super_list
= super_list
->next
;
5199 if (is_fd_valid(fd
))
5200 strcpy(st
->container_devnm
, fd2devnm(fd
));
5202 st
->container_devnm
[0] = 0;
5203 if (err
== 0 && st
->ss
== NULL
) {
5204 st
->ss
= &super_imsm
;
5205 st
->minor_version
= 0;
5206 st
->max_devs
= IMSM_MAX_DEVICES
;
5212 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5213 int *max
, int keep_fd
)
5215 struct md_list
*tmpdev
;
5219 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5220 if (tmpdev
->used
!= 1)
5222 if (tmpdev
->container
== 1) {
5224 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5225 if (!is_fd_valid(fd
)) {
5226 pr_err("cannot open device %s: %s\n",
5227 tmpdev
->devname
, strerror(errno
));
5231 err
= get_sra_super_block(fd
, super_list
,
5232 tmpdev
->devname
, &lmax
,
5241 int major
= major(tmpdev
->st_rdev
);
5242 int minor
= minor(tmpdev
->st_rdev
);
5243 err
= get_super_block(super_list
,
5260 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5261 int major
, int minor
, int keep_fd
)
5263 struct intel_super
*s
;
5275 sprintf(nm
, "%d:%d", major
, minor
);
5276 dfd
= dev_open(nm
, O_RDWR
);
5277 if (!is_fd_valid(dfd
)) {
5282 if (!get_dev_sector_size(dfd
, NULL
, &s
->sector_size
)) {
5286 find_intel_hba_capability(dfd
, s
, devname
);
5287 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5289 /* retry the load if we might have raced against mdmon */
5290 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5291 for (retry
= 0; retry
< 3; retry
++) {
5292 sleep_for(0, MSEC_TO_NSEC(3), true);
5293 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5299 s
->next
= *super_list
;
5313 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5320 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5324 if (sra
->array
.major_version
!= -1 ||
5325 sra
->array
.minor_version
!= -2 ||
5326 strcmp(sra
->text_version
, "imsm") != 0) {
5331 devnm
= fd2devnm(fd
);
5332 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5333 if (get_super_block(super_list
, devnm
, devname
,
5334 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5345 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5347 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5350 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5352 struct intel_super
*super
;
5356 if (test_partition(fd
))
5357 /* IMSM not allowed on partitions */
5360 free_super_imsm(st
);
5362 super
= alloc_super();
5366 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
5370 /* Load hba and capabilities if they exist.
5371 * But do not preclude loading metadata in case capabilities or hba are
5372 * non-compliant and ignore_hw_compat is set.
5374 rv
= find_intel_hba_capability(fd
, super
, devname
);
5375 /* no orom/efi or non-intel hba of the disk */
5376 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5378 pr_err("No OROM/EFI properties for %s\n", devname
);
5382 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5384 /* retry the load if we might have raced against mdmon */
5386 struct mdstat_ent
*mdstat
= NULL
;
5387 char *name
= fd2kname(fd
);
5390 mdstat
= mdstat_by_component(name
);
5392 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5393 for (retry
= 0; retry
< 3; retry
++) {
5394 sleep_for(0, MSEC_TO_NSEC(3), true);
5395 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5401 free_mdstat(mdstat
);
5406 pr_err("Failed to load all information sections on %s\n", devname
);
5412 if (st
->ss
== NULL
) {
5413 st
->ss
= &super_imsm
;
5414 st
->minor_version
= 0;
5415 st
->max_devs
= IMSM_MAX_DEVICES
;
5418 /* load migration record */
5419 if (load_imsm_migr_rec(super
) == 0) {
5420 /* Check for unsupported migration features */
5421 if (check_mpb_migr_compatibility(super
) != 0) {
5422 pr_err("Unsupported migration detected");
5424 fprintf(stderr
, " on %s\n", devname
);
5426 fprintf(stderr
, " (IMSM).\n");
5434 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5436 if (info
->level
== 1)
5438 return info
->chunk_size
>> 9;
5441 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5442 unsigned long long size
)
5444 if (info
->level
== 1)
5447 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5450 static void imsm_write_signature(struct imsm_super
*mpb
)
5452 /* It is safer to eventually truncate version rather than left it not NULL ended */
5453 snprintf((char *) mpb
->sig
, MAX_SIGNATURE_LENGTH
, MPB_SIGNATURE MPB_VERSION_ATTRIBS
);
5456 static void imsm_update_version_info(struct intel_super
*super
)
5458 /* update the version and attributes */
5459 struct imsm_super
*mpb
= super
->anchor
;
5460 struct imsm_dev
*dev
;
5461 struct imsm_map
*map
;
5464 mpb
->attributes
|= MPB_ATTRIB_CHECKSUM_VERIFY
;
5466 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5467 dev
= get_imsm_dev(super
, i
);
5468 map
= get_imsm_map(dev
, MAP_0
);
5470 if (__le32_to_cpu(dev
->size_high
) > 0)
5471 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5473 switch (get_imsm_raid_level(map
)) {
5475 mpb
->attributes
|= MPB_ATTRIB_RAID0
;
5478 mpb
->attributes
|= MPB_ATTRIB_RAID1
;
5481 mpb
->attributes
|= MPB_ATTRIB_RAID5
;
5484 mpb
->attributes
|= MPB_ATTRIB_RAID10
;
5485 if (map
->num_members
> 4)
5486 mpb
->attributes
|= MPB_ATTRIB_RAID10_EXT
;
5491 imsm_write_signature(mpb
);
5495 * imsm_check_name() - check imsm naming criteria.
5496 * @super: &intel_super pointer, not NULL.
5497 * @name: name to check.
5498 * @verbose: verbose level.
5500 * Name must be no longer than &MAX_RAID_SERIAL_LEN and must be unique across volumes.
5502 * Returns: &true if @name matches, &false otherwise.
5504 static bool imsm_is_name_allowed(struct intel_super
*super
, const char * const name
,
5507 struct imsm_super
*mpb
= super
->anchor
;
5510 if (is_string_lq(name
, MAX_RAID_SERIAL_LEN
+ 1) == false) {
5511 pr_vrb("imsm: Name \"%s\" is too long\n", name
);
5515 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5516 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5518 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5519 pr_vrb("imsm: Name \"%s\" already exists\n", name
);
5527 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5528 struct shape
*s
, char *name
,
5529 char *homehost
, int *uuid
,
5530 long long data_offset
)
5532 /* We are creating a volume inside a pre-existing container.
5533 * so st->sb is already set.
5535 struct intel_super
*super
= st
->sb
;
5536 unsigned int sector_size
= super
->sector_size
;
5537 struct imsm_super
*mpb
= super
->anchor
;
5538 struct intel_dev
*dv
;
5539 struct imsm_dev
*dev
;
5540 struct imsm_vol
*vol
;
5541 struct imsm_map
*map
;
5542 int idx
= mpb
->num_raid_devs
;
5545 unsigned long long array_blocks
;
5546 size_t size_old
, size_new
;
5547 unsigned int data_disks
;
5548 unsigned long long size_per_member
;
5550 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5551 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5555 /* ensure the mpb is large enough for the new data */
5556 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5557 size_new
= disks_to_mpb_size(info
->nr_disks
);
5558 if (size_new
> size_old
) {
5560 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5562 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5563 pr_err("could not allocate new mpb\n");
5566 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5567 MIGR_REC_BUF_SECTORS
*
5568 MAX_SECTOR_SIZE
) != 0) {
5569 pr_err("could not allocate migr_rec buffer\n");
5575 memcpy(mpb_new
, mpb
, size_old
);
5578 super
->anchor
= mpb_new
;
5579 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5580 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5581 super
->len
= size_round
;
5583 super
->current_vol
= idx
;
5585 /* handle 'failed_disks' by either:
5586 * a) create dummy disk entries in the table if this the first
5587 * volume in the array. We add them here as this is the only
5588 * opportunity to add them. add_to_super_imsm_volume()
5589 * handles the non-failed disks and continues incrementing
5591 * b) validate that 'failed_disks' matches the current number
5592 * of missing disks if the container is populated
5594 if (super
->current_vol
== 0) {
5596 for (i
= 0; i
< info
->failed_disks
; i
++) {
5597 struct imsm_disk
*disk
;
5600 disk
= __get_imsm_disk(mpb
, i
);
5601 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5602 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5603 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5604 "missing:%d", (__u8
)i
);
5606 find_missing(super
);
5611 for (d
= super
->missing
; d
; d
= d
->next
)
5613 if (info
->failed_disks
> missing
) {
5614 pr_err("unable to add 'missing' disk to container\n");
5619 if (imsm_is_name_allowed(super
, name
, 1) == false)
5622 dv
= xmalloc(sizeof(*dv
));
5623 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5625 * Explicitly allow truncating to not confuse gcc's
5626 * -Werror=stringop-truncation
5628 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5629 memcpy(dev
->volume
, name
, namelen
);
5630 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5631 info
->layout
, info
->chunk_size
,
5632 s
->size
* BLOCKS_PER_KB
);
5633 data_disks
= get_data_disks(info
->level
, info
->layout
,
5635 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5636 size_per_member
= array_blocks
/ data_disks
;
5638 set_imsm_dev_size(dev
, array_blocks
);
5639 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5641 vol
->migr_state
= 0;
5642 set_migr_type(dev
, MIGR_INIT
);
5643 vol
->dirty
= !info
->state
;
5644 set_vol_curr_migr_unit(dev
, 0);
5645 map
= get_imsm_map(dev
, MAP_0
);
5646 set_pba_of_lba0(map
, super
->create_offset
);
5647 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5648 map
->failed_disk_num
= ~0;
5649 if (info
->level
> IMSM_T_RAID0
)
5650 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5651 : IMSM_T_STATE_UNINITIALIZED
);
5653 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5654 IMSM_T_STATE_NORMAL
;
5657 if (info
->level
== IMSM_T_RAID1
&& info
->raid_disks
> 2) {
5660 pr_err("imsm does not support more than 2 disks in a raid1 volume\n");
5663 map
->num_members
= info
->raid_disks
;
5665 update_imsm_raid_level(map
, info
->level
);
5666 set_num_domains(map
);
5668 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5669 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5673 update_num_data_stripes(map
, array_blocks
);
5674 for (i
= 0; i
< map
->num_members
; i
++) {
5675 /* initialized in add_to_super */
5676 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5678 mpb
->num_raid_devs
++;
5679 mpb
->num_raid_devs_created
++;
5680 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5682 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5683 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5684 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5685 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5689 pr_err("imsm does not support consistency policy %s\n",
5690 map_num_s(consistency_policies
, s
->consistency_policy
));
5695 dv
->index
= super
->current_vol
;
5696 dv
->next
= super
->devlist
;
5697 super
->devlist
= dv
;
5699 imsm_update_version_info(super
);
5704 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5705 struct shape
*s
, char *name
,
5706 char *homehost
, int *uuid
,
5707 unsigned long long data_offset
)
5709 /* This is primarily called by Create when creating a new array.
5710 * We will then get add_to_super called for each component, and then
5711 * write_init_super called to write it out to each device.
5712 * For IMSM, Create can create on fresh devices or on a pre-existing
5714 * To create on a pre-existing array a different method will be called.
5715 * This one is just for fresh drives.
5717 struct intel_super
*super
;
5718 struct imsm_super
*mpb
;
5721 if (data_offset
!= INVALID_SECTORS
) {
5722 pr_err("data-offset not supported by imsm\n");
5727 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5731 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5733 mpb_size
= MAX_SECTOR_SIZE
;
5735 super
= alloc_super();
5737 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5742 pr_err("could not allocate superblock\n");
5745 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5746 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5747 pr_err("could not allocate migr_rec buffer\n");
5752 memset(super
->buf
, 0, mpb_size
);
5754 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5758 /* zeroing superblock */
5762 imsm_update_version_info(super
);
5766 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5768 unsigned int member_sector_size
;
5770 if (!is_fd_valid(dl
->fd
)) {
5771 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5775 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5777 if (member_sector_size
!= super
->sector_size
)
5782 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5783 int fd
, char *devname
)
5785 struct intel_super
*super
= st
->sb
;
5786 struct imsm_super
*mpb
= super
->anchor
;
5787 struct imsm_disk
*_disk
;
5788 struct imsm_dev
*dev
;
5789 struct imsm_map
*map
;
5794 if (!is_fd_valid(fd
))
5797 dev
= get_imsm_dev(super
, super
->current_vol
);
5798 map
= get_imsm_map(dev
, MAP_0
);
5800 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5801 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5806 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5808 if (dl
->raiddisk
== dk
->raid_disk
)
5810 } else if (dl
->major
== dk
->major
&& dl
->minor
== dk
->minor
)
5816 pr_err("%s is not a member of the same container.\n",
5821 if (!autolayout
&& super
->current_vol
> 0) {
5822 int _slot
= get_disk_slot_in_dev(super
, 0, dl
->index
);
5824 if (_slot
!= dk
->raid_disk
) {
5825 pr_err("Member %s is in %d slot for the first volume, but is in %d slot for a new volume.\n",
5826 dl
->devname
, _slot
, dk
->raid_disk
);
5827 pr_err("Raid members are in different order than for the first volume, aborting.\n");
5832 if (mpb
->num_disks
== 0)
5833 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5834 &super
->sector_size
))
5837 if (!drive_validate_sector_size(super
, dl
)) {
5838 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5842 /* add a pristine spare to the metadata */
5843 if (dl
->index
< 0) {
5844 dl
->index
= super
->anchor
->num_disks
;
5845 super
->anchor
->num_disks
++;
5847 /* Check the device has not already been added */
5848 slot
= get_imsm_disk_slot(map
, dl
->index
);
5850 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5851 pr_err("%s has been included in this array twice\n",
5855 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5856 dl
->disk
.status
= CONFIGURED_DISK
;
5858 /* update size of 'missing' disks to be at least as large as the
5859 * largest acitve member (we only have dummy missing disks when
5860 * creating the first volume)
5862 if (super
->current_vol
== 0) {
5863 for (df
= super
->missing
; df
; df
= df
->next
) {
5864 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5865 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5866 _disk
= __get_imsm_disk(mpb
, df
->index
);
5871 /* refresh unset/failed slots to point to valid 'missing' entries */
5872 for (df
= super
->missing
; df
; df
= df
->next
)
5873 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5874 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5876 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5878 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5879 if (is_gen_migration(dev
)) {
5880 struct imsm_map
*map2
= get_imsm_map(dev
,
5882 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5883 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5884 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5887 if ((unsigned)df
->index
==
5889 set_imsm_ord_tbl_ent(map2
,
5895 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5899 /* if we are creating the first raid device update the family number */
5900 if (super
->current_vol
== 0) {
5902 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5904 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5906 pr_err("BUG mpb setup error\n");
5912 sum
+= __gen_imsm_checksum(mpb
);
5913 mpb
->family_num
= __cpu_to_le32(sum
);
5914 mpb
->orig_family_num
= mpb
->family_num
;
5915 mpb
->creation_time
= __cpu_to_le64((__u64
)time(NULL
));
5917 super
->current_disk
= dl
;
5922 * Function marks disk as spare and restores disk serial
5923 * in case it was previously marked as failed by takeover operation
5925 * -1 : critical error
5926 * 0 : disk is marked as spare but serial is not set
5929 int mark_spare(struct dl
*disk
)
5931 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5938 if (!imsm_read_serial(disk
->fd
, NULL
, serial
, MAX_RAID_SERIAL_LEN
)) {
5939 /* Restore disk serial number, because takeover marks disk
5940 * as failed and adds to serial ':0' before it becomes
5943 serialcpy(disk
->serial
, serial
);
5944 serialcpy(disk
->disk
.serial
, serial
);
5947 disk
->disk
.status
= SPARE_DISK
;
5954 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
);
5956 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5957 int fd
, char *devname
,
5958 unsigned long long data_offset
)
5960 struct intel_super
*super
= st
->sb
;
5962 unsigned long long size
;
5963 unsigned int member_sector_size
;
5968 /* If we are on an RAID enabled platform check that the disk is
5969 * attached to the raid controller.
5970 * We do not need to test disks attachment for container based additions,
5971 * they shall be already tested when container was created/assembled.
5973 rv
= find_intel_hba_capability(fd
, super
, devname
);
5974 /* no orom/efi or non-intel hba of the disk */
5976 dprintf("capability: %p fd: %d ret: %d\n",
5977 super
->orom
, fd
, rv
);
5981 if (super
->current_vol
>= 0)
5982 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5985 dd
= xcalloc(sizeof(*dd
), 1);
5986 dd
->major
= major(stb
.st_rdev
);
5987 dd
->minor
= minor(stb
.st_rdev
);
5988 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5991 dd
->action
= DISK_ADD
;
5992 rv
= imsm_read_serial(fd
, devname
, dd
->serial
, MAX_RAID_SERIAL_LEN
);
5994 pr_err("failed to retrieve scsi serial, aborting\n");
5995 __free_imsm_disk(dd
, 0);
5999 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
6000 (super
->hba
->type
== SYS_DEV_VMD
))) {
6002 char cntrl_path
[PATH_MAX
];
6004 char pci_dev_path
[PATH_MAX
];
6006 if (!diskfd_to_devpath(fd
, 2, pci_dev_path
) ||
6007 !diskfd_to_devpath(fd
, 1, cntrl_path
)) {
6008 pr_err("failed to get dev paths, aborting\n");
6009 __free_imsm_disk(dd
, 0);
6013 cntrl_name
= basename(cntrl_path
);
6014 if (is_multipath_nvme(fd
))
6015 pr_err("%s controller supports Multi-Path I/O, Intel (R) VROC does not support multipathing\n",
6018 if (devpath_to_vendor(pci_dev_path
) == 0x8086) {
6020 * If Intel's NVMe drive has serial ended with
6021 * "-A","-B","-1" or "-2" it means that this is "x8"
6022 * device (double drive on single PCIe card).
6023 * User should be warned about potential data loss.
6025 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
6026 /* Skip empty character at the end */
6027 if (dd
->serial
[i
] == 0)
6030 if (((dd
->serial
[i
] == 'A') ||
6031 (dd
->serial
[i
] == 'B') ||
6032 (dd
->serial
[i
] == '1') ||
6033 (dd
->serial
[i
] == '2')) &&
6034 (dd
->serial
[i
-1] == '-'))
6035 pr_err("\tThe action you are about to take may put your data at risk.\n"
6036 "\tPlease note that x8 devices may consist of two separate x4 devices "
6037 "located on a single PCIe port.\n"
6038 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
6041 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
6042 !imsm_orom_has_tpv_support(super
->orom
)) {
6043 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
6044 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
6045 __free_imsm_disk(dd
, 0);
6050 get_dev_size(fd
, NULL
, &size
);
6051 if (!get_dev_sector_size(fd
, NULL
, &member_sector_size
)) {
6052 __free_imsm_disk(dd
, 0);
6056 if (super
->sector_size
== 0) {
6057 /* this a first device, so sector_size is not set yet */
6058 super
->sector_size
= member_sector_size
;
6061 /* clear migr_rec when adding disk to container */
6062 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6063 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
6065 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
6066 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
6067 MIGR_REC_BUF_SECTORS
*member_sector_size
)
6068 perror("Write migr_rec failed");
6072 serialcpy(dd
->disk
.serial
, dd
->serial
);
6073 set_total_blocks(&dd
->disk
, size
);
6074 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
6075 struct imsm_super
*mpb
= super
->anchor
;
6076 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6079 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
6080 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
6082 dd
->disk
.scsi_id
= __cpu_to_le32(0);
6084 if (st
->update_tail
) {
6085 dd
->next
= super
->disk_mgmt_list
;
6086 super
->disk_mgmt_list
= dd
;
6088 /* this is called outside of mdmon
6089 * write initial spare metadata
6090 * mdmon will overwrite it.
6092 dd
->next
= super
->disks
;
6094 write_super_imsm_spare(super
, dd
);
6100 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
6102 struct intel_super
*super
= st
->sb
;
6105 /* remove from super works only in mdmon - for communication
6106 * manager - monitor. Check if communication memory buffer
6109 if (!st
->update_tail
) {
6110 pr_err("shall be used in mdmon context only\n");
6113 dd
= xcalloc(1, sizeof(*dd
));
6114 dd
->major
= dk
->major
;
6115 dd
->minor
= dk
->minor
;
6118 dd
->action
= DISK_REMOVE
;
6120 dd
->next
= super
->disk_mgmt_list
;
6121 super
->disk_mgmt_list
= dd
;
6126 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
6129 char buf
[MAX_SECTOR_SIZE
];
6130 struct imsm_super anchor
;
6131 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
6134 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
)
6136 struct imsm_super
*spare
= &spare_record
.anchor
;
6142 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
6143 spare
->generation_num
= __cpu_to_le32(1UL);
6144 spare
->num_disks
= 1;
6145 spare
->num_raid_devs
= 0;
6146 spare
->pwr_cycle_count
= __cpu_to_le32(1);
6148 imsm_write_signature(spare
);
6150 spare
->disk
[0] = d
->disk
;
6151 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
6152 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6154 if (super
->sector_size
== 4096)
6155 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6157 sum
= __gen_imsm_checksum(spare
);
6158 spare
->family_num
= __cpu_to_le32(sum
);
6159 spare
->orig_family_num
= 0;
6160 sum
= __gen_imsm_checksum(spare
);
6161 spare
->check_sum
= __cpu_to_le32(sum
);
6163 if (store_imsm_mpb(d
->fd
, spare
)) {
6164 pr_err("failed for device %d:%d %s\n",
6165 d
->major
, d
->minor
, strerror(errno
));
6171 /* spare records have their own family number and do not have any defined raid
6174 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
6178 for (d
= super
->disks
; d
; d
= d
->next
) {
6182 if (write_super_imsm_spare(super
, d
))
6192 static int write_super_imsm(struct supertype
*st
, int doclose
)
6194 struct intel_super
*super
= st
->sb
;
6195 unsigned int sector_size
= super
->sector_size
;
6196 struct imsm_super
*mpb
= super
->anchor
;
6202 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6204 int clear_migration_record
= 1;
6207 /* 'generation' is incremented everytime the metadata is written */
6208 generation
= __le32_to_cpu(mpb
->generation_num
);
6210 mpb
->generation_num
= __cpu_to_le32(generation
);
6212 /* fix up cases where previous mdadm releases failed to set
6215 if (mpb
->orig_family_num
== 0)
6216 mpb
->orig_family_num
= mpb
->family_num
;
6218 for (d
= super
->disks
; d
; d
= d
->next
) {
6222 mpb
->disk
[d
->index
] = d
->disk
;
6226 for (d
= super
->missing
; d
; d
= d
->next
) {
6227 mpb
->disk
[d
->index
] = d
->disk
;
6230 mpb
->num_disks
= num_disks
;
6231 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6233 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6234 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6235 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6237 imsm_copy_dev(dev
, dev2
);
6238 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6240 if (is_gen_migration(dev2
))
6241 clear_migration_record
= 0;
6244 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6247 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6248 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6250 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6252 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6253 mpb_size
+= bbm_log_size
;
6254 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6257 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6260 /* recalculate checksum */
6261 sum
= __gen_imsm_checksum(mpb
);
6262 mpb
->check_sum
= __cpu_to_le32(sum
);
6264 if (super
->clean_migration_record_by_mdmon
) {
6265 clear_migration_record
= 1;
6266 super
->clean_migration_record_by_mdmon
= 0;
6268 if (clear_migration_record
)
6269 memset(super
->migr_rec_buf
, 0,
6270 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6272 if (sector_size
== 4096)
6273 convert_to_4k(super
);
6275 /* write the mpb for disks that compose raid devices */
6276 for (d
= super
->disks
; d
; d
= d
->next
) {
6277 if (d
->index
< 0 || is_failed(&d
->disk
))
6280 if (clear_migration_record
) {
6281 unsigned long long dsize
;
6283 get_dev_size(d
->fd
, NULL
, &dsize
);
6284 if (lseek64(d
->fd
, dsize
- sector_size
,
6286 if ((unsigned int)write(d
->fd
,
6287 super
->migr_rec_buf
,
6288 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6289 MIGR_REC_BUF_SECTORS
*sector_size
)
6290 perror("Write migr_rec failed");
6294 if (store_imsm_mpb(d
->fd
, mpb
))
6296 "failed for device %d:%d (fd: %d)%s\n",
6298 d
->fd
, strerror(errno
));
6305 return write_super_imsm_spares(super
, doclose
);
6310 static int create_array(struct supertype
*st
, int dev_idx
)
6313 struct imsm_update_create_array
*u
;
6314 struct intel_super
*super
= st
->sb
;
6315 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6316 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6317 struct disk_info
*inf
;
6318 struct imsm_disk
*disk
;
6321 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6322 sizeof(*inf
) * map
->num_members
;
6324 u
->type
= update_create_array
;
6325 u
->dev_idx
= dev_idx
;
6326 imsm_copy_dev(&u
->dev
, dev
);
6327 inf
= get_disk_info(u
);
6328 for (i
= 0; i
< map
->num_members
; i
++) {
6329 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6331 disk
= get_imsm_disk(super
, idx
);
6333 disk
= get_imsm_missing(super
, idx
);
6334 serialcpy(inf
[i
].serial
, disk
->serial
);
6336 append_metadata_update(st
, u
, len
);
6341 static int mgmt_disk(struct supertype
*st
)
6343 struct intel_super
*super
= st
->sb
;
6345 struct imsm_update_add_remove_disk
*u
;
6347 if (!super
->disk_mgmt_list
)
6352 u
->type
= update_add_remove_disk
;
6353 append_metadata_update(st
, u
, len
);
6358 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6360 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6362 struct ppl_header
*ppl_hdr
= buf
;
6365 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6367 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6369 perror("Failed to seek to PPL header location");
6373 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6375 perror("Write PPL header failed");
6384 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6386 struct intel_super
*super
= st
->sb
;
6388 struct ppl_header
*ppl_hdr
;
6391 /* first clear entire ppl space */
6392 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6396 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6398 pr_err("Failed to allocate PPL header buffer\n");
6402 memset(buf
, 0, PPL_HEADER_SIZE
);
6404 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6405 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6407 if (info
->mismatch_cnt
) {
6409 * We are overwriting an invalid ppl. Make one entry with wrong
6410 * checksum to prevent the kernel from skipping resync.
6412 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6413 ppl_hdr
->entries
[0].checksum
= ~0;
6416 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6422 static int is_rebuilding(struct imsm_dev
*dev
);
6424 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6425 struct mdinfo
*disk
)
6427 struct intel_super
*super
= st
->sb
;
6429 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6431 struct ppl_header
*ppl_hdr
= NULL
;
6433 struct imsm_dev
*dev
;
6436 unsigned long long ppl_offset
= 0;
6437 unsigned long long prev_gen_num
= 0;
6439 if (disk
->disk
.raid_disk
< 0)
6442 dev
= get_imsm_dev(super
, info
->container_member
);
6443 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6444 d
= get_imsm_dl_disk(super
, idx
);
6446 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6449 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6450 pr_err("Failed to allocate PPL header buffer\n");
6456 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6459 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6461 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6463 perror("Failed to seek to PPL header location");
6468 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6469 perror("Read PPL header failed");
6476 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6477 ppl_hdr
->checksum
= 0;
6479 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6480 dprintf("Wrong PPL header checksum on %s\n",
6485 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6486 /* previous was newest, it was already checked */
6490 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6491 super
->anchor
->orig_family_num
)) {
6492 dprintf("Wrong PPL header signature on %s\n",
6499 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6501 ppl_offset
+= PPL_HEADER_SIZE
;
6502 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6504 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6507 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6519 * Update metadata to use mutliple PPLs area (1MB).
6520 * This is done once for all RAID members
6522 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6523 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6525 struct mdinfo
*member_dev
;
6527 sprintf(subarray
, "%d", info
->container_member
);
6529 if (mdmon_running(st
->container_devnm
))
6530 st
->update_tail
= &st
->updates
;
6532 if (st
->ss
->update_subarray(st
, subarray
, UOPT_PPL
, NULL
)) {
6533 pr_err("Failed to update subarray %s\n",
6536 if (st
->update_tail
)
6537 flush_metadata_updates(st
);
6539 st
->ss
->sync_metadata(st
);
6540 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6541 for (member_dev
= info
->devs
; member_dev
;
6542 member_dev
= member_dev
->next
)
6543 member_dev
->ppl_size
=
6544 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6549 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6551 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6552 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6553 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6554 (is_rebuilding(dev
) &&
6555 vol_curr_migr_unit(dev
) == 0 &&
6556 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6557 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6559 info
->mismatch_cnt
++;
6560 } else if (ret
== 0 &&
6561 ppl_hdr
->entries_count
== 0 &&
6562 is_rebuilding(dev
) &&
6563 info
->resync_start
== 0) {
6565 * The header has no entries - add a single empty entry and
6566 * rewrite the header to prevent the kernel from going into
6567 * resync after an interrupted rebuild.
6569 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6570 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6578 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6580 struct intel_super
*super
= st
->sb
;
6584 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6585 info
->array
.level
!= 5)
6588 for (d
= super
->disks
; d
; d
= d
->next
) {
6589 if (d
->index
< 0 || is_failed(&d
->disk
))
6592 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6600 /*******************************************************************************
6601 * Function: write_init_bitmap_imsm_vol
6602 * Description: Write a bitmap header and prepares the area for the bitmap.
6604 * st : supertype information
6605 * vol_idx : the volume index to use
6610 ******************************************************************************/
6611 static int write_init_bitmap_imsm_vol(struct supertype
*st
, int vol_idx
)
6613 struct intel_super
*super
= st
->sb
;
6614 int prev_current_vol
= super
->current_vol
;
6618 super
->current_vol
= vol_idx
;
6619 for (d
= super
->disks
; d
; d
= d
->next
) {
6620 if (d
->index
< 0 || is_failed(&d
->disk
))
6622 ret
= st
->ss
->write_bitmap(st
, d
->fd
, NoUpdate
);
6626 super
->current_vol
= prev_current_vol
;
6630 /*******************************************************************************
6631 * Function: write_init_bitmap_imsm_all
6632 * Description: Write a bitmap header and prepares the area for the bitmap.
6633 * Operation is executed for volumes with CONSISTENCY_POLICY_BITMAP.
6635 * st : supertype information
6636 * info : info about the volume where the bitmap should be written
6637 * vol_idx : the volume index to use
6642 ******************************************************************************/
6643 static int write_init_bitmap_imsm_all(struct supertype
*st
, struct mdinfo
*info
,
6648 if (info
&& (info
->consistency_policy
== CONSISTENCY_POLICY_BITMAP
))
6649 ret
= write_init_bitmap_imsm_vol(st
, vol_idx
);
6654 static int write_init_super_imsm(struct supertype
*st
)
6656 struct intel_super
*super
= st
->sb
;
6657 int current_vol
= super
->current_vol
;
6661 getinfo_super_imsm(st
, &info
, NULL
);
6663 /* we are done with current_vol reset it to point st at the container */
6664 super
->current_vol
= -1;
6666 if (st
->update_tail
) {
6667 /* queue the recently created array / added disk
6668 * as a metadata update */
6670 /* determine if we are creating a volume or adding a disk */
6671 if (current_vol
< 0) {
6672 /* in the mgmt (add/remove) disk case we are running
6673 * in mdmon context, so don't close fd's
6677 /* adding the second volume to the array */
6678 rv
= write_init_ppl_imsm_all(st
, &info
);
6680 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6682 rv
= create_array(st
, current_vol
);
6686 for (d
= super
->disks
; d
; d
= d
->next
)
6687 Kill(d
->devname
, NULL
, 0, -1, 1);
6688 if (current_vol
>= 0) {
6689 rv
= write_init_ppl_imsm_all(st
, &info
);
6691 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6695 rv
= write_super_imsm(st
, 1);
6701 static int store_super_imsm(struct supertype
*st
, int fd
)
6703 struct intel_super
*super
= st
->sb
;
6704 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6709 if (super
->sector_size
== 4096)
6710 convert_to_4k(super
);
6711 return store_imsm_mpb(fd
, mpb
);
6714 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6716 unsigned long long data_offset
,
6718 unsigned long long *freesize
,
6722 unsigned long long ldsize
;
6723 struct intel_super
*super
= NULL
;
6726 if (!is_container(level
))
6731 fd
= dev_open(dev
, O_RDONLY
|O_EXCL
);
6732 if (!is_fd_valid(fd
)) {
6733 pr_vrb("imsm: Cannot open %s: %s\n", dev
, strerror(errno
));
6736 if (!get_dev_size(fd
, dev
, &ldsize
))
6739 /* capabilities retrieve could be possible
6740 * note that there is no fd for the disks in array.
6742 super
= alloc_super();
6746 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
))
6749 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6753 fd2devname(fd
, str
);
6754 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6755 fd
, str
, super
->orom
, rv
, raiddisks
);
6757 /* no orom/efi or non-intel hba of the disk */
6762 if (raiddisks
> super
->orom
->tds
) {
6764 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6765 raiddisks
, super
->orom
->tds
);
6768 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6769 (ldsize
>> 9) >> 32 > 0) {
6771 pr_err("%s exceeds maximum platform supported size\n", dev
);
6775 if (super
->hba
->type
== SYS_DEV_VMD
||
6776 super
->hba
->type
== SYS_DEV_NVME
) {
6777 if (!imsm_is_nvme_namespace_supported(fd
, 1)) {
6779 pr_err("NVMe namespace %s is not supported by IMSM\n",
6786 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6796 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6798 const unsigned long long base_start
= e
[*idx
].start
;
6799 unsigned long long end
= base_start
+ e
[*idx
].size
;
6802 if (base_start
== end
)
6806 for (i
= *idx
; i
< num_extents
; i
++) {
6807 /* extend overlapping extents */
6808 if (e
[i
].start
>= base_start
&&
6809 e
[i
].start
<= end
) {
6812 if (e
[i
].start
+ e
[i
].size
> end
)
6813 end
= e
[i
].start
+ e
[i
].size
;
6814 } else if (e
[i
].start
> end
) {
6820 return end
- base_start
;
6823 /** merge_extents() - analyze extents and get free size.
6824 * @super: Intel metadata, not NULL.
6825 * @expanding: if set, we are expanding &super->current_vol.
6827 * Build a composite disk with all known extents and generate a size given the
6828 * "all disks in an array must share a common start offset" constraint.
6829 * If a volume is expanded, then return free space after the volume.
6831 * Return: Free space or 0 on failure.
6833 static unsigned long long merge_extents(struct intel_super
*super
, const bool expanding
)
6837 int i
, j
, pos_vol_idx
= -1;
6839 int sum_extents
= 0;
6840 unsigned long long pos
= 0;
6841 unsigned long long start
= 0;
6842 unsigned long long free_size
= 0;
6844 unsigned long pre_reservation
= 0;
6845 unsigned long post_reservation
= IMSM_RESERVED_SECTORS
;
6846 unsigned long reservation_size
;
6848 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6850 sum_extents
+= dl
->extent_cnt
;
6851 e
= xcalloc(sum_extents
, sizeof(struct extent
));
6853 /* coalesce and sort all extents. also, check to see if we need to
6854 * reserve space between member arrays
6857 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6860 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6863 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6868 while (i
< sum_extents
) {
6869 e
[j
].start
= e
[i
].start
;
6870 e
[j
].vol
= e
[i
].vol
;
6871 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6873 if (e
[j
-1].size
== 0)
6879 unsigned long long esize
= e
[i
].start
- pos
;
6881 if (expanding
? pos_vol_idx
== super
->current_vol
: esize
>= free_size
) {
6887 pos
= e
[i
].start
+ e
[i
].size
;
6888 pos_vol_idx
= e
[i
].vol
;
6891 } while (e
[i
-1].size
);
6893 if (free_size
== 0) {
6894 dprintf("imsm: Cannot find free size.\n");
6899 if (!expanding
&& extent_idx
!= 0)
6901 * Not a real first volume in a container is created, pre_reservation is needed.
6903 pre_reservation
= IMSM_RESERVED_SECTORS
;
6905 if (e
[extent_idx
].size
== 0)
6907 * extent_idx points to the metadata, post_reservation is allready done.
6909 post_reservation
= 0;
6912 reservation_size
= pre_reservation
+ post_reservation
;
6914 if (free_size
< reservation_size
) {
6915 dprintf("imsm: Reservation size is greater than free space.\n");
6919 super
->create_offset
= start
+ pre_reservation
;
6920 return free_size
- reservation_size
;
6924 * is_raid_level_supported() - check if this count of drives and level is supported by platform.
6925 * @orom: hardware properties, could be NULL.
6926 * @level: requested raid level.
6927 * @raiddisks: requested disk count.
6929 * IMSM UEFI/OROM does not provide information about supported count of raid disks
6930 * for particular level. That is why it is hardcoded.
6931 * It is recommended to not allow of usage other levels than supported,
6932 * IMSM code is not tested against different level implementations.
6934 * Return: true if supported, false otherwise.
6936 static bool is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6940 for (idx
= 0; imsm_level_ops
[idx
].name
; idx
++) {
6941 if (imsm_level_ops
[idx
].level
== level
)
6945 if (!imsm_level_ops
[idx
].name
)
6948 if (!imsm_level_ops
[idx
].is_raiddisks_count_supported(raiddisks
))
6954 if (imsm_level_ops
[idx
].is_level_supported(orom
))
6961 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6962 int dpa
, int verbose
)
6964 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6965 struct mdstat_ent
*memb
;
6971 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6972 if (memb
->metadata_version
&&
6973 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6974 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6975 !is_subarray(memb
->metadata_version
+9) &&
6977 struct dev_member
*dev
= memb
->members
;
6979 while (dev
&& !is_fd_valid(fd
)) {
6980 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6981 num
= snprintf(path
, PATH_MAX
, "%s%s", "/dev/", dev
->name
);
6983 fd
= open(path
, O_RDONLY
, 0);
6984 if (num
<= 0 || !is_fd_valid(fd
)) {
6985 pr_vrb("Cannot open %s: %s\n",
6986 dev
->name
, strerror(errno
));
6992 if (is_fd_valid(fd
) && disk_attached_to_hba(fd
, hba
)) {
6993 struct mdstat_ent
*vol
;
6994 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6995 if (vol
->active
> 0 &&
6996 vol
->metadata_version
&&
6997 is_container_member(vol
, memb
->devnm
)) {
7002 if (*devlist
&& (found
< dpa
)) {
7003 dv
= xcalloc(1, sizeof(*dv
));
7004 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
7005 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
7008 dv
->next
= *devlist
;
7015 free_mdstat(mdstat
);
7020 static struct md_list
*
7021 get_loop_devices(void)
7024 struct md_list
*devlist
= NULL
;
7027 for(i
= 0; i
< 12; i
++) {
7028 dv
= xcalloc(1, sizeof(*dv
));
7029 dv
->devname
= xmalloc(40);
7030 sprintf(dv
->devname
, "/dev/loop%d", i
);
7038 static struct md_list
*
7039 get_devices(const char *hba_path
)
7041 struct md_list
*devlist
= NULL
;
7048 devlist
= get_loop_devices();
7051 /* scroll through /sys/dev/block looking for devices attached to
7054 dir
= opendir("/sys/dev/block");
7055 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
7060 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
7062 path
= devt_to_devpath(makedev(major
, minor
), 1, NULL
);
7065 if (!path_attached_to_hba(path
, hba_path
)) {
7072 fd
= dev_open(ent
->d_name
, O_RDONLY
);
7073 if (is_fd_valid(fd
)) {
7074 fd2devname(fd
, buf
);
7077 pr_err("cannot open device: %s\n",
7082 dv
= xcalloc(1, sizeof(*dv
));
7083 dv
->devname
= xstrdup(buf
);
7090 devlist
= devlist
->next
;
7100 count_volumes_list(struct md_list
*devlist
, char *homehost
,
7101 int verbose
, int *found
)
7103 struct md_list
*tmpdev
;
7105 struct supertype
*st
;
7107 /* first walk the list of devices to find a consistent set
7108 * that match the criterea, if that is possible.
7109 * We flag the ones we like with 'used'.
7112 st
= match_metadata_desc_imsm("imsm");
7114 pr_vrb("cannot allocate memory for imsm supertype\n");
7118 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7119 char *devname
= tmpdev
->devname
;
7121 struct supertype
*tst
;
7123 if (tmpdev
->used
> 1)
7125 tst
= dup_super(st
);
7127 pr_vrb("cannot allocate memory for imsm supertype\n");
7130 tmpdev
->container
= 0;
7131 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
7132 if (!is_fd_valid(dfd
)) {
7133 dprintf("cannot open device %s: %s\n",
7134 devname
, strerror(errno
));
7136 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
7138 } else if (must_be_container(dfd
)) {
7139 struct supertype
*cst
;
7140 cst
= super_by_fd(dfd
, NULL
);
7142 dprintf("cannot recognize container type %s\n",
7145 } else if (tst
->ss
!= st
->ss
) {
7146 dprintf("non-imsm container - ignore it: %s\n",
7149 } else if (!tst
->ss
->load_container
||
7150 tst
->ss
->load_container(tst
, dfd
, NULL
))
7153 tmpdev
->container
= 1;
7156 cst
->ss
->free_super(cst
);
7158 tmpdev
->st_rdev
= rdev
;
7159 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
7160 dprintf("no RAID superblock on %s\n",
7163 } else if (tst
->ss
->compare_super
== NULL
) {
7164 dprintf("Cannot assemble %s metadata on %s\n",
7165 tst
->ss
->name
, devname
);
7171 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
7172 /* Ignore unrecognised devices during auto-assembly */
7177 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
7179 if (st
->minor_version
== -1)
7180 st
->minor_version
= tst
->minor_version
;
7182 if (memcmp(info
.uuid
, uuid_zero
,
7183 sizeof(int[4])) == 0) {
7184 /* this is a floating spare. It cannot define
7185 * an array unless there are no more arrays of
7186 * this type to be found. It can be included
7187 * in an array of this type though.
7193 if (st
->ss
!= tst
->ss
||
7194 st
->minor_version
!= tst
->minor_version
||
7195 st
->ss
->compare_super(st
, tst
, 1) != 0) {
7196 /* Some mismatch. If exactly one array matches this host,
7197 * we can resolve on that one.
7198 * Or, if we are auto assembling, we just ignore the second
7201 dprintf("superblock on %s doesn't match others - assembly aborted\n",
7207 dprintf("found: devname: %s\n", devname
);
7211 tst
->ss
->free_super(tst
);
7215 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
7216 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
7217 for (iter
= head
; iter
; iter
= iter
->next
) {
7218 dprintf("content->text_version: %s vol\n",
7219 iter
->text_version
);
7220 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
7221 /* do not assemble arrays with unsupported
7223 dprintf("Cannot activate member %s.\n",
7224 iter
->text_version
);
7231 dprintf("No valid super block on device list: err: %d %p\n",
7235 dprintf("no more devices to examine\n");
7238 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7239 if (tmpdev
->used
== 1 && tmpdev
->found
) {
7241 if (count
< tmpdev
->found
)
7244 count
-= tmpdev
->found
;
7247 if (tmpdev
->used
== 1)
7252 st
->ss
->free_super(st
);
7256 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
7259 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
7261 const struct orom_entry
*entry
;
7262 struct devid_list
*dv
, *devid_list
;
7267 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7268 if (strstr(idev
->path
, hba_path
))
7272 if (!idev
|| !idev
->dev_id
)
7275 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7277 if (!entry
|| !entry
->devid_list
)
7280 devid_list
= entry
->devid_list
;
7281 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7282 struct md_list
*devlist
;
7283 struct sys_dev
*device
= NULL
;
7288 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7290 device
= device_by_id(dv
->devid
);
7293 hpath
= device
->path
;
7297 devlist
= get_devices(hpath
);
7298 /* if no intel devices return zero volumes */
7299 if (devlist
== NULL
)
7302 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7304 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7305 if (devlist
== NULL
)
7309 count
+= count_volumes_list(devlist
,
7313 dprintf("found %d count: %d\n", found
, count
);
7316 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7319 struct md_list
*dv
= devlist
;
7320 devlist
= devlist
->next
;
7328 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7332 if (hba
->type
== SYS_DEV_VMD
) {
7333 struct sys_dev
*dev
;
7336 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7337 if (dev
->type
== SYS_DEV_VMD
)
7338 count
+= __count_volumes(dev
->path
, dpa
,
7343 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7346 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7348 /* up to 512 if the plaform supports it, otherwise the platform max.
7349 * 128 if no platform detected
7351 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7353 return min(512, (1 << fs
));
7357 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7358 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7360 /* check/set platform and metadata limits/defaults */
7361 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7362 pr_vrb("platform supports a maximum of %d disks per array\n",
7367 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7368 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7369 pr_vrb("platform does not support raid%d with %d disk%s\n",
7370 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7374 if (*chunk
== 0 || *chunk
== UnSet
)
7375 *chunk
= imsm_default_chunk(super
->orom
);
7377 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7378 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7382 if (layout
!= imsm_level_to_layout(level
)) {
7384 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7385 else if (level
== 10)
7386 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7388 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7393 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7394 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7395 pr_vrb("platform does not support a volume size over 2TB\n");
7402 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7403 * FIX ME add ahci details
7405 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7406 int layout
, int raiddisks
, int *chunk
,
7407 unsigned long long size
,
7408 unsigned long long data_offset
,
7410 unsigned long long *freesize
,
7414 struct intel_super
*super
= st
->sb
;
7415 struct imsm_super
*mpb
;
7417 unsigned long long pos
= 0;
7418 unsigned long long maxsize
;
7422 /* We must have the container info already read in. */
7426 mpb
= super
->anchor
;
7428 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7429 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7433 /* General test: make sure there is space for
7434 * 'raiddisks' device extents of size 'size' at a given
7437 unsigned long long minsize
= size
;
7438 unsigned long long start_offset
= MaxSector
;
7441 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7442 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7447 e
= get_extents(super
, dl
, 0);
7450 unsigned long long esize
;
7451 esize
= e
[i
].start
- pos
;
7452 if (esize
>= minsize
)
7454 if (found
&& start_offset
== MaxSector
) {
7457 } else if (found
&& pos
!= start_offset
) {
7461 pos
= e
[i
].start
+ e
[i
].size
;
7463 } while (e
[i
-1].size
);
7468 if (dcnt
< raiddisks
) {
7470 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7477 /* This device must be a member of the set */
7478 if (!stat_is_blkdev(dev
, &rdev
))
7480 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7481 if (dl
->major
== (int)major(rdev
) &&
7482 dl
->minor
== (int)minor(rdev
))
7487 pr_err("%s is not in the same imsm set\n", dev
);
7489 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7490 /* If a volume is present then the current creation attempt
7491 * cannot incorporate new spares because the orom may not
7492 * understand this configuration (all member disks must be
7493 * members of each array in the container).
7495 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7496 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7498 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7499 mpb
->num_disks
!= raiddisks
) {
7500 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7504 /* retrieve the largest free space block */
7505 e
= get_extents(super
, dl
, 0);
7510 unsigned long long esize
;
7512 esize
= e
[i
].start
- pos
;
7513 if (esize
>= maxsize
)
7515 pos
= e
[i
].start
+ e
[i
].size
;
7517 } while (e
[i
-1].size
);
7522 pr_err("unable to determine free space for: %s\n",
7526 if (maxsize
< size
) {
7528 pr_err("%s not enough space (%llu < %llu)\n",
7529 dev
, maxsize
, size
);
7533 maxsize
= merge_extents(super
, false);
7535 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7536 pr_err("attempting to create a second volume with size less then remaining space.\n");
7538 if (maxsize
< size
|| maxsize
== 0) {
7541 pr_err("no free space left on device. Aborting...\n");
7543 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7549 *freesize
= maxsize
;
7552 int count
= count_volumes(super
->hba
,
7553 super
->orom
->dpa
, verbose
);
7554 if (super
->orom
->vphba
<= count
) {
7555 pr_vrb("platform does not support more than %d raid volumes.\n",
7556 super
->orom
->vphba
);
7564 * imsm_get_free_size() - get the biggest, common free space from members.
7565 * @super: &intel_super pointer, not NULL.
7566 * @raiddisks: number of raid disks.
7567 * @size: requested size, could be 0 (means max size).
7568 * @chunk: requested chunk size in KiB.
7569 * @freesize: pointer for returned size value.
7571 * Return: &IMSM_STATUS_OK or &IMSM_STATUS_ERROR.
7573 * @freesize is set to meaningful value, this can be @size, or calculated
7575 * super->create_offset value is modified and set appropriately in
7576 * merge_extends() for further creation.
7578 static imsm_status_t
imsm_get_free_size(struct intel_super
*super
,
7579 const int raiddisks
,
7580 unsigned long long size
,
7582 unsigned long long *freesize
,
7585 struct imsm_super
*mpb
= super
->anchor
;
7591 unsigned long long maxsize
;
7592 unsigned long long minsize
= size
;
7595 minsize
= chunk
* 2;
7597 /* find the largest common start free region of the possible disks */
7598 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7604 /* don't activate new spares if we are orom constrained
7605 * and there is already a volume active in the container
7607 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7610 e
= get_extents(super
, dl
, 0);
7613 for (i
= 1; e
[i
-1].size
; i
++)
7620 maxsize
= merge_extents(super
, expanding
);
7621 if (maxsize
< minsize
) {
7622 pr_err("imsm: Free space is %llu but must be equal or larger than %llu.\n",
7624 return IMSM_STATUS_ERROR
;
7627 if (cnt
< raiddisks
|| (super
->orom
&& used
&& used
!= raiddisks
)) {
7628 pr_err("imsm: Not enough devices with space to create array.\n");
7629 return IMSM_STATUS_ERROR
;
7640 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7641 pr_err("attempting to create a second volume with size less then remaining space.\n");
7644 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7646 return IMSM_STATUS_OK
;
7650 * autolayout_imsm() - automatically layout a new volume.
7651 * @super: &intel_super pointer, not NULL.
7652 * @raiddisks: number of raid disks.
7653 * @size: requested size, could be 0 (means max size).
7654 * @chunk: requested chunk.
7655 * @freesize: pointer for returned size value.
7657 * We are being asked to automatically layout a new volume based on the current
7658 * contents of the container. If the parameters can be satisfied autolayout_imsm
7659 * will record the disks, start offset, and will return size of the volume to
7660 * be created. See imsm_get_free_size() for details.
7661 * add_to_super() and getinfo_super() detect when autolayout is in progress.
7662 * If first volume exists, slots are set consistently to it.
7664 * Return: &IMSM_STATUS_OK on success, &IMSM_STATUS_ERROR otherwise.
7666 * Disks are marked for creation via dl->raiddisk.
7668 static imsm_status_t
autolayout_imsm(struct intel_super
*super
,
7669 const int raiddisks
,
7670 unsigned long long size
, const int chunk
,
7671 unsigned long long *freesize
)
7675 int vol_cnt
= super
->anchor
->num_raid_devs
;
7678 rv
= imsm_get_free_size(super
, raiddisks
, size
, chunk
, freesize
, false);
7679 if (rv
!= IMSM_STATUS_OK
)
7680 return IMSM_STATUS_ERROR
;
7682 for (disk
= super
->disks
; disk
; disk
= disk
->next
) {
7686 if (curr_slot
== raiddisks
)
7690 disk
->raiddisk
= curr_slot
;
7692 int _slot
= get_disk_slot_in_dev(super
, 0, disk
->index
);
7695 pr_err("Disk %s is not used in first volume, aborting\n",
7697 return IMSM_STATUS_ERROR
;
7699 disk
->raiddisk
= _slot
;
7704 return IMSM_STATUS_OK
;
7707 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7708 int raiddisks
, int *chunk
, unsigned long long size
,
7709 unsigned long long data_offset
,
7710 char *dev
, unsigned long long *freesize
,
7711 int consistency_policy
, int verbose
)
7718 * if given unused devices create a container
7719 * if given given devices in a container create a member volume
7721 if (is_container(level
))
7722 /* Must be a fresh device to add to a container */
7723 return validate_geometry_imsm_container(st
, level
, raiddisks
,
7728 * Size is given in sectors.
7730 if (size
&& (size
< 2048)) {
7731 pr_err("Given size must be greater than 1M.\n");
7732 /* Depends on algorithm in Create.c :
7733 * if container was given (dev == NULL) return -1,
7734 * if block device was given ( dev != NULL) return 0.
7736 return dev
? -1 : 0;
7740 struct intel_super
*super
= st
->sb
;
7743 * Autolayout mode, st->sb must be set.
7746 pr_vrb("superblock must be set for autolayout, aborting\n");
7750 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7751 raiddisks
, chunk
, size
,
7755 if (super
->orom
&& freesize
) {
7757 int count
= count_volumes(super
->hba
, super
->orom
->dpa
,
7759 if (super
->orom
->vphba
<= count
) {
7760 pr_vrb("platform does not support more than %d raid volumes.\n",
7761 super
->orom
->vphba
);
7765 rv
= autolayout_imsm(super
, raiddisks
, size
, *chunk
,
7767 if (rv
!= IMSM_STATUS_OK
)
7773 /* creating in a given container */
7774 return validate_geometry_imsm_volume(st
, level
, layout
,
7775 raiddisks
, chunk
, size
,
7777 dev
, freesize
, verbose
);
7780 /* This device needs to be a device in an 'imsm' container */
7781 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7783 if (is_fd_valid(fd
)) {
7784 pr_vrb("Cannot create this array on device %s\n", dev
);
7789 fd
= open(dev
, O_RDONLY
, 0);
7791 if (!is_fd_valid(fd
)) {
7792 pr_vrb("Cannot open %s: %s\n", dev
, strerror(errno
));
7796 /* Well, it is in use by someone, maybe an 'imsm' container. */
7797 cfd
= open_container(fd
);
7800 if (!is_fd_valid(cfd
)) {
7801 pr_vrb("Cannot use %s: It is busy\n", dev
);
7804 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7805 if (sra
&& sra
->array
.major_version
== -1 &&
7806 strcmp(sra
->text_version
, "imsm") == 0)
7810 /* This is a member of a imsm container. Load the container
7811 * and try to create a volume
7813 struct intel_super
*super
;
7815 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7817 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7819 return validate_geometry_imsm_volume(st
, level
, layout
,
7821 size
, data_offset
, dev
,
7828 pr_err("failed container membership check\n");
7834 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7836 struct intel_super
*super
= st
->sb
;
7838 if (level
&& *level
== UnSet
)
7839 *level
= LEVEL_CONTAINER
;
7841 if (level
&& layout
&& *layout
== UnSet
)
7842 *layout
= imsm_level_to_layout(*level
);
7844 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7845 *chunk
= imsm_default_chunk(super
->orom
);
7848 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7850 static int kill_subarray_imsm(struct supertype
*st
, char *subarray_id
)
7852 /* remove the subarray currently referenced by subarray_id */
7854 struct intel_dev
**dp
;
7855 struct intel_super
*super
= st
->sb
;
7856 __u8 current_vol
= strtoul(subarray_id
, NULL
, 10);
7857 struct imsm_super
*mpb
= super
->anchor
;
7859 if (mpb
->num_raid_devs
== 0)
7862 /* block deletions that would change the uuid of active subarrays
7864 * FIXME when immutable ids are available, but note that we'll
7865 * also need to fixup the invalidated/active subarray indexes in
7868 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7871 if (i
< current_vol
)
7873 snprintf(subarray
, sizeof(subarray
), "%u", i
);
7874 if (is_subarray_active(subarray
, st
->devnm
)) {
7875 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7882 if (st
->update_tail
) {
7883 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7885 u
->type
= update_kill_array
;
7886 u
->dev_idx
= current_vol
;
7887 append_metadata_update(st
, u
, sizeof(*u
));
7892 for (dp
= &super
->devlist
; *dp
;)
7893 if ((*dp
)->index
== current_vol
) {
7896 handle_missing(super
, (*dp
)->dev
);
7897 if ((*dp
)->index
> current_vol
)
7902 /* no more raid devices, all active components are now spares,
7903 * but of course failed are still failed
7905 if (--mpb
->num_raid_devs
== 0) {
7908 for (d
= super
->disks
; d
; d
= d
->next
)
7913 super
->updates_pending
++;
7919 * get_rwh_policy_from_update() - Get the rwh policy for update option.
7920 * @update: Update option.
7922 static int get_rwh_policy_from_update(enum update_opt update
)
7926 return RWH_MULTIPLE_DISTRIBUTED
;
7928 return RWH_MULTIPLE_OFF
;
7931 case UOPT_NO_BITMAP
:
7936 return UOPT_UNDEFINED
;
7939 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7940 enum update_opt update
, struct mddev_ident
*ident
)
7942 /* update the subarray currently referenced by ->current_vol */
7943 struct intel_super
*super
= st
->sb
;
7944 struct imsm_super
*mpb
= super
->anchor
;
7946 if (update
== UOPT_NAME
) {
7947 char *name
= ident
->name
;
7951 if (imsm_is_name_allowed(super
, name
, 1) == false)
7954 vol
= strtoul(subarray
, &ep
, 10);
7955 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7958 if (st
->update_tail
) {
7959 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7961 u
->type
= update_rename_array
;
7963 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7964 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7965 append_metadata_update(st
, u
, sizeof(*u
));
7967 struct imsm_dev
*dev
;
7970 dev
= get_imsm_dev(super
, vol
);
7971 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7972 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7973 memcpy(dev
->volume
, name
, namelen
);
7974 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7975 dev
= get_imsm_dev(super
, i
);
7976 handle_missing(super
, dev
);
7978 super
->updates_pending
++;
7980 } else if (get_rwh_policy_from_update(update
) != UOPT_UNDEFINED
) {
7983 int vol
= strtoul(subarray
, &ep
, 10);
7985 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7988 new_policy
= get_rwh_policy_from_update(update
);
7990 if (st
->update_tail
) {
7991 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7993 u
->type
= update_rwh_policy
;
7995 u
->new_policy
= new_policy
;
7996 append_metadata_update(st
, u
, sizeof(*u
));
7998 struct imsm_dev
*dev
;
8000 dev
= get_imsm_dev(super
, vol
);
8001 dev
->rwh_policy
= new_policy
;
8002 super
->updates_pending
++;
8004 if (new_policy
== RWH_BITMAP
)
8005 return write_init_bitmap_imsm_vol(st
, vol
);
8012 static bool is_gen_migration(struct imsm_dev
*dev
)
8014 if (dev
&& dev
->vol
.migr_state
&&
8015 migr_type(dev
) == MIGR_GEN_MIGR
)
8021 static int is_rebuilding(struct imsm_dev
*dev
)
8023 struct imsm_map
*migr_map
;
8025 if (!dev
->vol
.migr_state
)
8028 if (migr_type(dev
) != MIGR_REBUILD
)
8031 migr_map
= get_imsm_map(dev
, MAP_1
);
8033 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
8039 static int is_initializing(struct imsm_dev
*dev
)
8041 struct imsm_map
*migr_map
;
8043 if (!dev
->vol
.migr_state
)
8046 if (migr_type(dev
) != MIGR_INIT
)
8049 migr_map
= get_imsm_map(dev
, MAP_1
);
8051 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8057 static void update_recovery_start(struct intel_super
*super
,
8058 struct imsm_dev
*dev
,
8059 struct mdinfo
*array
)
8061 struct mdinfo
*rebuild
= NULL
;
8065 if (!is_rebuilding(dev
))
8068 /* Find the rebuild target, but punt on the dual rebuild case */
8069 for (d
= array
->devs
; d
; d
= d
->next
)
8070 if (d
->recovery_start
== 0) {
8077 /* (?) none of the disks are marked with
8078 * IMSM_ORD_REBUILD, so assume they are missing and the
8079 * disk_ord_tbl was not correctly updated
8081 dprintf("failed to locate out-of-sync disk\n");
8085 units
= vol_curr_migr_unit(dev
);
8086 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
8089 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
8091 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
8093 /* Given a container loaded by load_super_imsm_all,
8094 * extract information about all the arrays into
8096 * If 'subarray' is given, just extract info about that array.
8098 * For each imsm_dev create an mdinfo, fill it in,
8099 * then look for matching devices in super->disks
8100 * and create appropriate device mdinfo.
8102 struct intel_super
*super
= st
->sb
;
8103 struct imsm_super
*mpb
= super
->anchor
;
8104 struct mdinfo
*rest
= NULL
;
8108 int spare_disks
= 0;
8109 int current_vol
= super
->current_vol
;
8111 /* do not assemble arrays when not all attributes are supported */
8112 if (imsm_check_attributes(mpb
->attributes
) == false) {
8114 pr_err("Unsupported attributes in IMSM metadata. Arrays activation is blocked.\n");
8117 /* count spare devices, not used in maps
8119 for (d
= super
->disks
; d
; d
= d
->next
)
8123 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8124 struct imsm_dev
*dev
;
8125 struct imsm_map
*map
;
8126 struct imsm_map
*map2
;
8127 struct mdinfo
*this;
8134 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
8137 dev
= get_imsm_dev(super
, i
);
8138 map
= get_imsm_map(dev
, MAP_0
);
8139 map2
= get_imsm_map(dev
, MAP_1
);
8140 level
= get_imsm_raid_level(map
);
8142 /* do not publish arrays that are in the middle of an
8143 * unsupported migration
8145 if (dev
->vol
.migr_state
&&
8146 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
8147 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
8151 /* do not publish arrays that are not support by controller's
8155 this = xmalloc(sizeof(*this));
8157 super
->current_vol
= i
;
8158 getinfo_super_imsm_volume(st
, this, NULL
);
8160 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
8161 /* mdadm does not support all metadata features- set the bit in all arrays state */
8162 if (!validate_geometry_imsm_orom(super
,
8163 level
, /* RAID level */
8164 imsm_level_to_layout(level
),
8165 map
->num_members
, /* raid disks */
8166 &chunk
, imsm_dev_size(dev
),
8168 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
8170 this->array
.state
|=
8171 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8172 (1<<MD_SB_BLOCK_VOLUME
);
8175 /* if array has bad blocks, set suitable bit in all arrays state */
8177 this->array
.state
|=
8178 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8179 (1<<MD_SB_BLOCK_VOLUME
);
8181 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
8182 unsigned long long recovery_start
;
8183 struct mdinfo
*info_d
;
8191 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
8192 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
8193 for (d
= super
->disks
; d
; d
= d
->next
)
8194 if (d
->index
== idx
)
8197 recovery_start
= MaxSector
;
8200 if (d
&& is_failed(&d
->disk
))
8202 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
8204 if (!(ord
& IMSM_ORD_REBUILD
))
8205 this->array
.working_disks
++;
8207 * if we skip some disks the array will be assmebled degraded;
8208 * reset resync start to avoid a dirty-degraded
8209 * situation when performing the intial sync
8214 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
8215 if ((!able_to_resync(level
, missing
) ||
8216 recovery_start
== 0))
8217 this->resync_start
= MaxSector
;
8223 info_d
= xcalloc(1, sizeof(*info_d
));
8224 info_d
->next
= this->devs
;
8225 this->devs
= info_d
;
8227 info_d
->disk
.number
= d
->index
;
8228 info_d
->disk
.major
= d
->major
;
8229 info_d
->disk
.minor
= d
->minor
;
8230 info_d
->disk
.raid_disk
= slot
;
8231 info_d
->recovery_start
= recovery_start
;
8233 if (slot
< map2
->num_members
)
8234 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8236 this->array
.spare_disks
++;
8238 if (slot
< map
->num_members
)
8239 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8241 this->array
.spare_disks
++;
8244 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
8245 info_d
->data_offset
= pba_of_lba0(map
);
8246 info_d
->component_size
= calc_component_size(map
, dev
);
8248 if (map
->raid_level
== IMSM_T_RAID5
) {
8249 info_d
->ppl_sector
= this->ppl_sector
;
8250 info_d
->ppl_size
= this->ppl_size
;
8251 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
8252 recovery_start
== 0)
8253 this->resync_start
= 0;
8256 info_d
->bb
.supported
= 1;
8257 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
8258 info_d
->data_offset
,
8259 info_d
->component_size
,
8262 /* now that the disk list is up-to-date fixup recovery_start */
8263 update_recovery_start(super
, dev
, this);
8264 this->array
.spare_disks
+= spare_disks
;
8266 /* check for reshape */
8267 if (this->reshape_active
== 1)
8268 recover_backup_imsm(st
, this);
8272 super
->current_vol
= current_vol
;
8276 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
8277 int failed
, int look_in_map
)
8279 struct imsm_map
*map
;
8281 map
= get_imsm_map(dev
, look_in_map
);
8284 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
8285 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
8287 switch (get_imsm_raid_level(map
)) {
8289 return IMSM_T_STATE_FAILED
;
8292 if (failed
< map
->num_members
)
8293 return IMSM_T_STATE_DEGRADED
;
8295 return IMSM_T_STATE_FAILED
;
8300 * check to see if any mirrors have failed, otherwise we
8301 * are degraded. Even numbered slots are mirrored on
8305 /* gcc -Os complains that this is unused */
8306 int insync
= insync
;
8308 for (i
= 0; i
< map
->num_members
; i
++) {
8309 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8310 int idx
= ord_to_idx(ord
);
8311 struct imsm_disk
*disk
;
8313 /* reset the potential in-sync count on even-numbered
8314 * slots. num_copies is always 2 for imsm raid10
8319 disk
= get_imsm_disk(super
, idx
);
8320 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8323 /* no in-sync disks left in this mirror the
8327 return IMSM_T_STATE_FAILED
;
8330 return IMSM_T_STATE_DEGRADED
;
8334 return IMSM_T_STATE_DEGRADED
;
8336 return IMSM_T_STATE_FAILED
;
8342 return map
->map_state
;
8345 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8350 struct imsm_disk
*disk
;
8351 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8352 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8353 struct imsm_map
*map_for_loop
;
8358 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8359 * disks that are being rebuilt. New failures are recorded to
8360 * map[0]. So we look through all the disks we started with and
8361 * see if any failures are still present, or if any new ones
8365 if (prev
&& (map
->num_members
< prev
->num_members
))
8366 map_for_loop
= prev
;
8368 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8370 /* when MAP_X is passed both maps failures are counted
8373 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8374 i
< prev
->num_members
) {
8375 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8376 idx_1
= ord_to_idx(ord
);
8378 disk
= get_imsm_disk(super
, idx_1
);
8379 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8382 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8383 i
< map
->num_members
) {
8384 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8385 idx
= ord_to_idx(ord
);
8388 disk
= get_imsm_disk(super
, idx
);
8389 if (!disk
|| is_failed(disk
) ||
8390 ord
& IMSM_ORD_REBUILD
)
8399 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8402 struct intel_super
*super
= c
->sb
;
8403 struct imsm_super
*mpb
= super
->anchor
;
8404 struct imsm_update_prealloc_bb_mem u
;
8406 if (inst
>= mpb
->num_raid_devs
) {
8407 pr_err("subarry index %d, out of range\n", inst
);
8411 dprintf("imsm: open_new %d\n", inst
);
8412 a
->info
.container_member
= inst
;
8414 u
.type
= update_prealloc_badblocks_mem
;
8415 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8420 static int is_resyncing(struct imsm_dev
*dev
)
8422 struct imsm_map
*migr_map
;
8424 if (!dev
->vol
.migr_state
)
8427 if (migr_type(dev
) == MIGR_INIT
||
8428 migr_type(dev
) == MIGR_REPAIR
)
8431 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8434 migr_map
= get_imsm_map(dev
, MAP_1
);
8436 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8437 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8443 /* return true if we recorded new information */
8444 static int mark_failure(struct intel_super
*super
,
8445 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8449 struct imsm_map
*map
;
8450 char buf
[MAX_RAID_SERIAL_LEN
+3];
8451 unsigned int len
, shift
= 0;
8453 /* new failures are always set in map[0] */
8454 map
= get_imsm_map(dev
, MAP_0
);
8456 slot
= get_imsm_disk_slot(map
, idx
);
8460 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8461 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8464 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8465 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8467 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8468 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8469 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8471 disk
->status
|= FAILED_DISK
;
8472 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8473 /* mark failures in second map if second map exists and this disk
8475 * This is valid for migration, initialization and rebuild
8477 if (dev
->vol
.migr_state
) {
8478 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8479 int slot2
= get_imsm_disk_slot(map2
, idx
);
8481 if (slot2
< map2
->num_members
&& slot2
>= 0)
8482 set_imsm_ord_tbl_ent(map2
, slot2
,
8483 idx
| IMSM_ORD_REBUILD
);
8485 if (map
->failed_disk_num
== 0xff ||
8486 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8487 map
->failed_disk_num
= slot
;
8489 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8494 static void mark_missing(struct intel_super
*super
,
8495 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8497 mark_failure(super
, dev
, disk
, idx
);
8499 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8502 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8503 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8506 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8510 if (!super
->missing
)
8513 /* When orom adds replacement for missing disk it does
8514 * not remove entry of missing disk, but just updates map with
8515 * new added disk. So it is not enough just to test if there is
8516 * any missing disk, we have to look if there are any failed disks
8517 * in map to stop migration */
8519 dprintf("imsm: mark missing\n");
8520 /* end process for initialization and rebuild only
8522 if (is_gen_migration(dev
) == false) {
8523 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8527 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8528 struct imsm_map
*map1
;
8529 int i
, ord
, ord_map1
;
8532 for (i
= 0; i
< map
->num_members
; i
++) {
8533 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8534 if (!(ord
& IMSM_ORD_REBUILD
))
8537 map1
= get_imsm_map(dev
, MAP_1
);
8541 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8542 if (ord_map1
& IMSM_ORD_REBUILD
)
8547 map_state
= imsm_check_degraded(super
, dev
,
8549 end_migration(dev
, super
, map_state
);
8553 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8554 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8555 super
->updates_pending
++;
8558 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8561 unsigned long long array_blocks
;
8562 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8563 int used_disks
= imsm_num_data_members(map
);
8565 if (used_disks
== 0) {
8566 /* when problems occures
8567 * return current array_blocks value
8569 array_blocks
= imsm_dev_size(dev
);
8571 return array_blocks
;
8574 /* set array size in metadata
8577 /* OLCE size change is caused by added disks
8579 array_blocks
= per_dev_array_size(map
) * used_disks
;
8581 /* Online Volume Size Change
8582 * Using available free space
8584 array_blocks
= new_size
;
8586 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8587 set_imsm_dev_size(dev
, array_blocks
);
8589 return array_blocks
;
8592 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8594 static void imsm_progress_container_reshape(struct intel_super
*super
)
8596 /* if no device has a migr_state, but some device has a
8597 * different number of members than the previous device, start
8598 * changing the number of devices in this device to match
8601 struct imsm_super
*mpb
= super
->anchor
;
8602 int prev_disks
= -1;
8606 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8607 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8608 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8609 struct imsm_map
*map2
;
8610 int prev_num_members
;
8612 if (dev
->vol
.migr_state
)
8615 if (prev_disks
== -1)
8616 prev_disks
= map
->num_members
;
8617 if (prev_disks
== map
->num_members
)
8620 /* OK, this array needs to enter reshape mode.
8621 * i.e it needs a migr_state
8624 copy_map_size
= sizeof_imsm_map(map
);
8625 prev_num_members
= map
->num_members
;
8626 map
->num_members
= prev_disks
;
8627 dev
->vol
.migr_state
= 1;
8628 set_vol_curr_migr_unit(dev
, 0);
8629 set_migr_type(dev
, MIGR_GEN_MIGR
);
8630 for (i
= prev_num_members
;
8631 i
< map
->num_members
; i
++)
8632 set_imsm_ord_tbl_ent(map
, i
, i
);
8633 map2
= get_imsm_map(dev
, MAP_1
);
8634 /* Copy the current map */
8635 memcpy(map2
, map
, copy_map_size
);
8636 map2
->num_members
= prev_num_members
;
8638 imsm_set_array_size(dev
, -1);
8639 super
->clean_migration_record_by_mdmon
= 1;
8640 super
->updates_pending
++;
8644 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8645 * states are handled in imsm_set_disk() with one exception, when a
8646 * resync is stopped due to a new failure this routine will set the
8647 * 'degraded' state for the array.
8649 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8651 int inst
= a
->info
.container_member
;
8652 struct intel_super
*super
= a
->container
->sb
;
8653 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8654 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8655 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8656 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8657 __u32 blocks_per_unit
;
8659 if (dev
->vol
.migr_state
&&
8660 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8661 /* array state change is blocked due to reshape action
8663 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8664 * - finish the reshape (if last_checkpoint is big and action != reshape)
8665 * - update vol_curr_migr_unit
8667 if (a
->curr_action
== reshape
) {
8668 /* still reshaping, maybe update vol_curr_migr_unit */
8669 goto mark_checkpoint
;
8671 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8672 unsigned long long array_blocks
;
8676 used_disks
= imsm_num_data_members(map
);
8677 if (used_disks
> 0) {
8679 per_dev_array_size(map
) *
8682 round_size_to_mb(array_blocks
,
8684 a
->info
.custom_array_size
= array_blocks
;
8685 /* encourage manager to update array
8689 a
->check_reshape
= 1;
8691 /* finalize online capacity expansion/reshape */
8692 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8694 mdi
->disk
.raid_disk
,
8697 imsm_progress_container_reshape(super
);
8702 /* before we activate this array handle any missing disks */
8703 if (consistent
== 2)
8704 handle_missing(super
, dev
);
8706 if (consistent
== 2 &&
8707 (!is_resync_complete(&a
->info
) ||
8708 map_state
!= IMSM_T_STATE_NORMAL
||
8709 dev
->vol
.migr_state
))
8712 if (is_resync_complete(&a
->info
)) {
8713 /* complete intialization / resync,
8714 * recovery and interrupted recovery is completed in
8717 if (is_resyncing(dev
)) {
8718 dprintf("imsm: mark resync done\n");
8719 end_migration(dev
, super
, map_state
);
8720 super
->updates_pending
++;
8721 a
->last_checkpoint
= 0;
8723 } else if ((!is_resyncing(dev
) && !failed
) &&
8724 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8725 /* mark the start of the init process if nothing is failed */
8726 dprintf("imsm: mark resync start\n");
8727 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8728 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8730 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8731 super
->updates_pending
++;
8734 if (a
->prev_action
== idle
)
8735 goto skip_mark_checkpoint
;
8738 /* skip checkpointing for general migration,
8739 * it is controlled in mdadm
8741 if (is_gen_migration(dev
))
8742 goto skip_mark_checkpoint
;
8744 /* check if we can update vol_curr_migr_unit from resync_start,
8747 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8748 if (blocks_per_unit
) {
8749 set_vol_curr_migr_unit(dev
,
8750 a
->last_checkpoint
/ blocks_per_unit
);
8751 dprintf("imsm: mark checkpoint (%llu)\n",
8752 vol_curr_migr_unit(dev
));
8753 super
->updates_pending
++;
8756 skip_mark_checkpoint
:
8757 /* mark dirty / clean */
8758 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8759 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8760 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8762 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8764 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8765 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8766 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8767 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8769 super
->updates_pending
++;
8775 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8777 int inst
= a
->info
.container_member
;
8778 struct intel_super
*super
= a
->container
->sb
;
8779 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8780 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8782 if (slot
> map
->num_members
) {
8783 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8784 slot
, map
->num_members
- 1);
8791 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8794 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8796 int inst
= a
->info
.container_member
;
8797 struct intel_super
*super
= a
->container
->sb
;
8798 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8799 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8800 struct imsm_disk
*disk
;
8802 int recovery_not_finished
= 0;
8806 int rebuild_done
= 0;
8809 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8813 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8814 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8816 /* check for new failures */
8817 if (disk
&& (state
& DS_FAULTY
)) {
8818 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8819 super
->updates_pending
++;
8822 /* check if in_sync */
8823 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8824 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8826 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8828 super
->updates_pending
++;
8831 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8832 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8834 /* check if recovery complete, newly degraded, or failed */
8835 dprintf("imsm: Detected transition to state ");
8836 switch (map_state
) {
8837 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8838 dprintf("normal: ");
8839 if (is_rebuilding(dev
)) {
8840 dprintf_cont("while rebuilding");
8841 /* check if recovery is really finished */
8842 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8843 if (mdi
->recovery_start
!= MaxSector
) {
8844 recovery_not_finished
= 1;
8847 if (recovery_not_finished
) {
8849 dprintf("Rebuild has not finished yet, state not changed");
8850 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8851 a
->last_checkpoint
= mdi
->recovery_start
;
8852 super
->updates_pending
++;
8856 end_migration(dev
, super
, map_state
);
8857 map
->failed_disk_num
= ~0;
8858 super
->updates_pending
++;
8859 a
->last_checkpoint
= 0;
8862 if (is_gen_migration(dev
)) {
8863 dprintf_cont("while general migration");
8864 if (a
->last_checkpoint
>= a
->info
.component_size
)
8865 end_migration(dev
, super
, map_state
);
8867 map
->map_state
= map_state
;
8868 map
->failed_disk_num
= ~0;
8869 super
->updates_pending
++;
8873 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8874 dprintf_cont("degraded: ");
8875 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8876 dprintf_cont("mark degraded");
8877 map
->map_state
= map_state
;
8878 super
->updates_pending
++;
8879 a
->last_checkpoint
= 0;
8882 if (is_rebuilding(dev
)) {
8883 dprintf_cont("while rebuilding ");
8884 if (state
& DS_FAULTY
) {
8885 dprintf_cont("removing failed drive ");
8886 if (n
== map
->failed_disk_num
) {
8887 dprintf_cont("end migration");
8888 end_migration(dev
, super
, map_state
);
8889 a
->last_checkpoint
= 0;
8891 dprintf_cont("fail detected during rebuild, changing map state");
8892 map
->map_state
= map_state
;
8894 super
->updates_pending
++;
8900 /* check if recovery is really finished */
8901 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8902 if (mdi
->recovery_start
!= MaxSector
) {
8903 recovery_not_finished
= 1;
8906 if (recovery_not_finished
) {
8908 dprintf_cont("Rebuild has not finished yet");
8909 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8910 a
->last_checkpoint
=
8911 mdi
->recovery_start
;
8912 super
->updates_pending
++;
8917 dprintf_cont(" Rebuild done, still degraded");
8918 end_migration(dev
, super
, map_state
);
8919 a
->last_checkpoint
= 0;
8920 super
->updates_pending
++;
8922 for (i
= 0; i
< map
->num_members
; i
++) {
8923 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8925 if (idx
& IMSM_ORD_REBUILD
)
8926 map
->failed_disk_num
= i
;
8928 super
->updates_pending
++;
8931 if (is_gen_migration(dev
)) {
8932 dprintf_cont("while general migration");
8933 if (a
->last_checkpoint
>= a
->info
.component_size
)
8934 end_migration(dev
, super
, map_state
);
8936 map
->map_state
= map_state
;
8937 manage_second_map(super
, dev
);
8939 super
->updates_pending
++;
8942 if (is_initializing(dev
)) {
8943 dprintf_cont("while initialization.");
8944 map
->map_state
= map_state
;
8945 super
->updates_pending
++;
8949 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8950 dprintf_cont("failed: ");
8951 if (is_gen_migration(dev
)) {
8952 dprintf_cont("while general migration");
8953 map
->map_state
= map_state
;
8954 super
->updates_pending
++;
8957 if (map
->map_state
!= map_state
) {
8958 dprintf_cont("mark failed");
8959 end_migration(dev
, super
, map_state
);
8960 super
->updates_pending
++;
8961 a
->last_checkpoint
= 0;
8966 dprintf_cont("state %i\n", map_state
);
8971 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8974 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8975 unsigned long long dsize
;
8976 unsigned long long sectors
;
8977 unsigned int sector_size
;
8979 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
8981 get_dev_size(fd
, NULL
, &dsize
);
8983 if (mpb_size
> sector_size
) {
8984 /* -1 to account for anchor */
8985 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8987 /* write the extended mpb to the sectors preceeding the anchor */
8988 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8992 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8993 sector_size
* sectors
) != sector_size
* sectors
)
8997 /* first block is stored on second to last sector of the disk */
8998 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
9001 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
9007 static void imsm_sync_metadata(struct supertype
*container
)
9009 struct intel_super
*super
= container
->sb
;
9011 dprintf("sync metadata: %d\n", super
->updates_pending
);
9012 if (!super
->updates_pending
)
9015 write_super_imsm(container
, 0);
9017 super
->updates_pending
= 0;
9020 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
9022 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
9023 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
9026 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9030 if (dl
&& is_failed(&dl
->disk
))
9034 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
9039 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
9040 struct active_array
*a
, int activate_new
,
9041 struct mdinfo
*additional_test_list
)
9043 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
9044 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
9045 struct imsm_super
*mpb
= super
->anchor
;
9046 struct imsm_map
*map
;
9047 unsigned long long pos
;
9052 __u32 array_start
= 0;
9053 __u32 array_end
= 0;
9055 struct mdinfo
*test_list
;
9057 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9058 /* If in this array, skip */
9059 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9060 if (is_fd_valid(d
->state_fd
) &&
9061 d
->disk
.major
== dl
->major
&&
9062 d
->disk
.minor
== dl
->minor
) {
9063 dprintf("%x:%x already in array\n",
9064 dl
->major
, dl
->minor
);
9069 test_list
= additional_test_list
;
9071 if (test_list
->disk
.major
== dl
->major
&&
9072 test_list
->disk
.minor
== dl
->minor
) {
9073 dprintf("%x:%x already in additional test list\n",
9074 dl
->major
, dl
->minor
);
9077 test_list
= test_list
->next
;
9082 /* skip in use or failed drives */
9083 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
9085 dprintf("%x:%x status (failed: %d index: %d)\n",
9086 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
9090 /* skip pure spares when we are looking for partially
9091 * assimilated drives
9093 if (dl
->index
== -1 && !activate_new
)
9096 if (!drive_validate_sector_size(super
, dl
))
9099 /* Does this unused device have the requisite free space?
9100 * It needs to be able to cover all member volumes
9102 ex
= get_extents(super
, dl
, 1);
9104 dprintf("cannot get extents\n");
9107 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9108 dev
= get_imsm_dev(super
, i
);
9109 map
= get_imsm_map(dev
, MAP_0
);
9111 /* check if this disk is already a member of
9114 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
9120 array_start
= pba_of_lba0(map
);
9121 array_end
= array_start
+
9122 per_dev_array_size(map
) - 1;
9125 /* check that we can start at pba_of_lba0 with
9126 * num_data_stripes*blocks_per_stripe of space
9128 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
9132 pos
= ex
[j
].start
+ ex
[j
].size
;
9134 } while (ex
[j
-1].size
);
9141 if (i
< mpb
->num_raid_devs
) {
9142 dprintf("%x:%x does not have %u to %u available\n",
9143 dl
->major
, dl
->minor
, array_start
, array_end
);
9153 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
9155 struct imsm_dev
*dev2
;
9156 struct imsm_map
*map
;
9162 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
9164 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
9165 if (state
== IMSM_T_STATE_FAILED
) {
9166 map
= get_imsm_map(dev2
, MAP_0
);
9167 for (slot
= 0; slot
< map
->num_members
; slot
++) {
9169 * Check if failed disks are deleted from intel
9170 * disk list or are marked to be deleted
9172 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
9173 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
9175 * Do not rebuild the array if failed disks
9176 * from failed sub-array are not removed from
9180 is_failed(&idisk
->disk
) &&
9181 (idisk
->action
!= DISK_REMOVE
))
9188 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
9189 struct metadata_update
**updates
)
9192 * Find a device with unused free space and use it to replace a
9193 * failed/vacant region in an array. We replace failed regions one a
9194 * array at a time. The result is that a new spare disk will be added
9195 * to the first failed array and after the monitor has finished
9196 * propagating failures the remainder will be consumed.
9198 * FIXME add a capability for mdmon to request spares from another
9202 struct intel_super
*super
= a
->container
->sb
;
9203 int inst
= a
->info
.container_member
;
9204 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
9205 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9206 int failed
= a
->info
.array
.raid_disks
;
9207 struct mdinfo
*rv
= NULL
;
9210 struct metadata_update
*mu
;
9212 struct imsm_update_activate_spare
*u
;
9217 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
9218 if (!is_fd_valid(d
->state_fd
))
9221 if (d
->curr_state
& DS_FAULTY
)
9222 /* wait for Removal to happen */
9228 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
9229 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
9231 if (imsm_reshape_blocks_arrays_changes(super
))
9234 /* Cannot activate another spare if rebuild is in progress already
9236 if (is_rebuilding(dev
)) {
9237 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
9241 if (a
->info
.array
.level
== 4)
9242 /* No repair for takeovered array
9243 * imsm doesn't support raid4
9247 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
9248 IMSM_T_STATE_DEGRADED
)
9251 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
9252 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
9257 * If there are any failed disks check state of the other volume.
9258 * Block rebuild if the another one is failed until failed disks
9259 * are removed from container.
9262 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
9263 MAX_RAID_SERIAL_LEN
, dev
->volume
);
9264 /* check if states of the other volumes allow for rebuild */
9265 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
9267 allowed
= imsm_rebuild_allowed(a
->container
,
9275 /* For each slot, if it is not working, find a spare */
9276 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
9277 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9278 if (d
->disk
.raid_disk
== i
)
9280 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
9281 if (d
&& is_fd_valid(d
->state_fd
))
9285 * OK, this device needs recovery. Try to re-add the
9286 * previous occupant of this slot, if this fails see if
9287 * we can continue the assimilation of a spare that was
9288 * partially assimilated, finally try to activate a new
9291 dl
= imsm_readd(super
, i
, a
);
9293 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9295 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9299 /* found a usable disk with enough space */
9300 di
= xcalloc(1, sizeof(*di
));
9302 /* dl->index will be -1 in the case we are activating a
9303 * pristine spare. imsm_process_update() will create a
9304 * new index in this case. Once a disk is found to be
9305 * failed in all member arrays it is kicked from the
9308 di
->disk
.number
= dl
->index
;
9310 /* (ab)use di->devs to store a pointer to the device
9313 di
->devs
= (struct mdinfo
*) dl
;
9315 di
->disk
.raid_disk
= i
;
9316 di
->disk
.major
= dl
->major
;
9317 di
->disk
.minor
= dl
->minor
;
9319 di
->recovery_start
= 0;
9320 di
->data_offset
= pba_of_lba0(map
);
9321 di
->component_size
= a
->info
.component_size
;
9322 di
->container_member
= inst
;
9323 di
->bb
.supported
= 1;
9324 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9325 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9326 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9328 super
->random
= random32();
9332 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9333 i
, di
->data_offset
);
9337 /* No spares found */
9339 /* Now 'rv' has a list of devices to return.
9340 * Create a metadata_update record to update the
9341 * disk_ord_tbl for the array
9343 mu
= xmalloc(sizeof(*mu
));
9344 mu
->buf
= xcalloc(num_spares
,
9345 sizeof(struct imsm_update_activate_spare
));
9347 mu
->space_list
= NULL
;
9348 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9349 mu
->next
= *updates
;
9350 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9352 for (di
= rv
; di
; di
= di
->next
) {
9353 u
->type
= update_activate_spare
;
9354 u
->dl
= (struct dl
*) di
->devs
;
9356 u
->slot
= di
->disk
.raid_disk
;
9367 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9369 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9370 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9371 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9372 struct disk_info
*inf
= get_disk_info(u
);
9373 struct imsm_disk
*disk
;
9377 for (i
= 0; i
< map
->num_members
; i
++) {
9378 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9379 for (j
= 0; j
< new_map
->num_members
; j
++)
9380 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9387 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9391 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9392 if (dl
->major
== major
&& dl
->minor
== minor
)
9397 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9403 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9404 if (dl
->major
== major
&& dl
->minor
== minor
) {
9407 prev
->next
= dl
->next
;
9409 super
->disks
= dl
->next
;
9411 __free_imsm_disk(dl
, 1);
9412 dprintf("removed %x:%x\n", major
, minor
);
9420 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9422 static int add_remove_disk_update(struct intel_super
*super
)
9424 int check_degraded
= 0;
9427 /* add/remove some spares to/from the metadata/contrainer */
9428 while (super
->disk_mgmt_list
) {
9429 struct dl
*disk_cfg
;
9431 disk_cfg
= super
->disk_mgmt_list
;
9432 super
->disk_mgmt_list
= disk_cfg
->next
;
9433 disk_cfg
->next
= NULL
;
9435 if (disk_cfg
->action
== DISK_ADD
) {
9436 disk_cfg
->next
= super
->disks
;
9437 super
->disks
= disk_cfg
;
9439 dprintf("added %x:%x\n",
9440 disk_cfg
->major
, disk_cfg
->minor
);
9441 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9442 dprintf("Disk remove action processed: %x.%x\n",
9443 disk_cfg
->major
, disk_cfg
->minor
);
9444 disk
= get_disk_super(super
,
9448 /* store action status */
9449 disk
->action
= DISK_REMOVE
;
9450 /* remove spare disks only */
9451 if (disk
->index
== -1) {
9452 remove_disk_super(super
,
9456 disk_cfg
->fd
= disk
->fd
;
9460 /* release allocate disk structure */
9461 __free_imsm_disk(disk_cfg
, 1);
9464 return check_degraded
;
9467 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9468 struct intel_super
*super
,
9471 struct intel_dev
*id
;
9472 void **tofree
= NULL
;
9475 dprintf("(enter)\n");
9476 if (u
->subdev
< 0 || u
->subdev
> 1) {
9477 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9480 if (space_list
== NULL
|| *space_list
== NULL
) {
9481 dprintf("imsm: Error: Memory is not allocated\n");
9485 for (id
= super
->devlist
; id
; id
= id
->next
) {
9486 if (id
->index
== (unsigned)u
->subdev
) {
9487 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9488 struct imsm_map
*map
;
9489 struct imsm_dev
*new_dev
=
9490 (struct imsm_dev
*)*space_list
;
9491 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9493 struct dl
*new_disk
;
9495 if (new_dev
== NULL
)
9497 *space_list
= **space_list
;
9498 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9499 map
= get_imsm_map(new_dev
, MAP_0
);
9501 dprintf("imsm: Error: migration in progress");
9505 to_state
= map
->map_state
;
9506 if ((u
->new_level
== IMSM_T_RAID5
) && (map
->raid_level
== IMSM_T_RAID0
)) {
9508 /* this should not happen */
9509 if (u
->new_disks
[0] < 0) {
9510 map
->failed_disk_num
=
9511 map
->num_members
- 1;
9512 to_state
= IMSM_T_STATE_DEGRADED
;
9514 to_state
= IMSM_T_STATE_NORMAL
;
9516 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9518 if (u
->new_level
> -1)
9519 update_imsm_raid_level(map
, u
->new_level
);
9521 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9522 if ((u
->new_level
== IMSM_T_RAID5
) &&
9523 (migr_map
->raid_level
== IMSM_T_RAID0
)) {
9524 int ord
= map
->num_members
- 1;
9525 migr_map
->num_members
--;
9526 if (u
->new_disks
[0] < 0)
9527 ord
|= IMSM_ORD_REBUILD
;
9528 set_imsm_ord_tbl_ent(map
,
9529 map
->num_members
- 1,
9533 tofree
= (void **)dev
;
9535 /* update chunk size
9537 if (u
->new_chunksize
> 0) {
9538 struct imsm_map
*dest_map
=
9539 get_imsm_map(dev
, MAP_0
);
9541 imsm_num_data_members(dest_map
);
9543 if (used_disks
== 0)
9546 map
->blocks_per_strip
=
9547 __cpu_to_le16(u
->new_chunksize
* 2);
9548 update_num_data_stripes(map
, imsm_dev_size(dev
));
9551 /* ensure blocks_per_member has valid value
9553 set_blocks_per_member(map
,
9554 per_dev_array_size(map
) +
9555 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9559 if (u
->new_level
!= IMSM_T_RAID5
|| migr_map
->raid_level
!= IMSM_T_RAID0
||
9560 migr_map
->raid_level
== map
->raid_level
)
9563 if (u
->new_disks
[0] >= 0) {
9566 new_disk
= get_disk_super(super
,
9567 major(u
->new_disks
[0]),
9568 minor(u
->new_disks
[0]));
9569 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9570 major(u
->new_disks
[0]),
9571 minor(u
->new_disks
[0]),
9572 new_disk
, new_disk
->index
);
9573 if (new_disk
== NULL
)
9574 goto error_disk_add
;
9576 new_disk
->index
= map
->num_members
- 1;
9577 /* slot to fill in autolayout
9579 new_disk
->raiddisk
= new_disk
->index
;
9580 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9581 new_disk
->disk
.status
&= ~SPARE_DISK
;
9583 goto error_disk_add
;
9586 *tofree
= *space_list
;
9587 /* calculate new size
9589 imsm_set_array_size(new_dev
, -1);
9596 *space_list
= tofree
;
9600 dprintf("Error: imsm: Cannot find disk.\n");
9604 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9605 struct intel_super
*super
)
9607 struct intel_dev
*id
;
9610 dprintf("(enter)\n");
9611 if (u
->subdev
< 0 || u
->subdev
> 1) {
9612 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9616 for (id
= super
->devlist
; id
; id
= id
->next
) {
9617 if (id
->index
== (unsigned)u
->subdev
) {
9618 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9619 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9620 int used_disks
= imsm_num_data_members(map
);
9621 unsigned long long blocks_per_member
;
9622 unsigned long long new_size_per_disk
;
9624 if (used_disks
== 0)
9627 /* calculate new size
9629 new_size_per_disk
= u
->new_size
/ used_disks
;
9630 blocks_per_member
= new_size_per_disk
+
9631 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9633 imsm_set_array_size(dev
, u
->new_size
);
9634 set_blocks_per_member(map
, blocks_per_member
);
9635 update_num_data_stripes(map
, u
->new_size
);
9644 static int prepare_spare_to_activate(struct supertype
*st
,
9645 struct imsm_update_activate_spare
*u
)
9647 struct intel_super
*super
= st
->sb
;
9648 int prev_current_vol
= super
->current_vol
;
9649 struct active_array
*a
;
9652 for (a
= st
->arrays
; a
; a
= a
->next
)
9654 * Additional initialization (adding bitmap header, filling
9655 * the bitmap area with '1's to force initial rebuild for a whole
9656 * data-area) is required when adding the spare to the volume
9657 * with write-intent bitmap.
9659 if (a
->info
.container_member
== u
->array
&&
9660 a
->info
.consistency_policy
== CONSISTENCY_POLICY_BITMAP
) {
9663 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9669 super
->current_vol
= u
->array
;
9670 if (st
->ss
->write_bitmap(st
, dl
->fd
, NoUpdate
))
9672 super
->current_vol
= prev_current_vol
;
9677 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9678 struct intel_super
*super
,
9679 struct active_array
*active_array
)
9681 struct imsm_super
*mpb
= super
->anchor
;
9682 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9683 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9684 struct imsm_map
*migr_map
;
9685 struct active_array
*a
;
9686 struct imsm_disk
*disk
;
9693 int second_map_created
= 0;
9695 for (; u
; u
= u
->next
) {
9696 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9701 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9706 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9711 /* count failures (excluding rebuilds and the victim)
9712 * to determine map[0] state
9715 for (i
= 0; i
< map
->num_members
; i
++) {
9718 disk
= get_imsm_disk(super
,
9719 get_imsm_disk_idx(dev
, i
, MAP_X
));
9720 if (!disk
|| is_failed(disk
))
9724 /* adding a pristine spare, assign a new index */
9725 if (dl
->index
< 0) {
9726 dl
->index
= super
->anchor
->num_disks
;
9727 super
->anchor
->num_disks
++;
9730 disk
->status
|= CONFIGURED_DISK
;
9731 disk
->status
&= ~SPARE_DISK
;
9734 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9735 if (!second_map_created
) {
9736 second_map_created
= 1;
9737 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9738 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9740 map
->map_state
= to_state
;
9741 migr_map
= get_imsm_map(dev
, MAP_1
);
9742 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9743 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9744 dl
->index
| IMSM_ORD_REBUILD
);
9746 /* update the family_num to mark a new container
9747 * generation, being careful to record the existing
9748 * family_num in orig_family_num to clean up after
9749 * earlier mdadm versions that neglected to set it.
9751 if (mpb
->orig_family_num
== 0)
9752 mpb
->orig_family_num
= mpb
->family_num
;
9753 mpb
->family_num
+= super
->random
;
9755 /* count arrays using the victim in the metadata */
9757 for (a
= active_array
; a
; a
= a
->next
) {
9758 int dev_idx
= a
->info
.container_member
;
9760 if (get_disk_slot_in_dev(super
, dev_idx
, victim
) >= 0)
9764 /* delete the victim if it is no longer being
9770 /* We know that 'manager' isn't touching anything,
9771 * so it is safe to delete
9773 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9774 if ((*dlp
)->index
== victim
)
9777 /* victim may be on the missing list */
9779 for (dlp
= &super
->missing
; *dlp
;
9780 dlp
= &(*dlp
)->next
)
9781 if ((*dlp
)->index
== victim
)
9783 imsm_delete(super
, dlp
, victim
);
9790 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9791 struct intel_super
*super
,
9794 struct dl
*new_disk
;
9795 struct intel_dev
*id
;
9797 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9798 int disk_count
= u
->old_raid_disks
;
9799 void **tofree
= NULL
;
9800 int devices_to_reshape
= 1;
9801 struct imsm_super
*mpb
= super
->anchor
;
9803 unsigned int dev_id
;
9805 dprintf("(enter)\n");
9807 /* enable spares to use in array */
9808 for (i
= 0; i
< delta_disks
; i
++) {
9809 new_disk
= get_disk_super(super
,
9810 major(u
->new_disks
[i
]),
9811 minor(u
->new_disks
[i
]));
9812 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9813 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9814 new_disk
, new_disk
->index
);
9815 if (new_disk
== NULL
||
9816 (new_disk
->index
>= 0 &&
9817 new_disk
->index
< u
->old_raid_disks
))
9818 goto update_reshape_exit
;
9819 new_disk
->index
= disk_count
++;
9820 /* slot to fill in autolayout
9822 new_disk
->raiddisk
= new_disk
->index
;
9823 new_disk
->disk
.status
|=
9825 new_disk
->disk
.status
&= ~SPARE_DISK
;
9828 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9829 mpb
->num_raid_devs
);
9830 /* manage changes in volume
9832 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9833 void **sp
= *space_list
;
9834 struct imsm_dev
*newdev
;
9835 struct imsm_map
*newmap
, *oldmap
;
9837 for (id
= super
->devlist
; id
; id
= id
->next
) {
9838 if (id
->index
== dev_id
)
9847 /* Copy the dev, but not (all of) the map */
9848 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9849 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9850 newmap
= get_imsm_map(newdev
, MAP_0
);
9851 /* Copy the current map */
9852 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9853 /* update one device only
9855 if (devices_to_reshape
) {
9856 dprintf("imsm: modifying subdev: %i\n",
9858 devices_to_reshape
--;
9859 newdev
->vol
.migr_state
= 1;
9860 set_vol_curr_migr_unit(newdev
, 0);
9861 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9862 newmap
->num_members
= u
->new_raid_disks
;
9863 for (i
= 0; i
< delta_disks
; i
++) {
9864 set_imsm_ord_tbl_ent(newmap
,
9865 u
->old_raid_disks
+ i
,
9866 u
->old_raid_disks
+ i
);
9868 /* New map is correct, now need to save old map
9870 newmap
= get_imsm_map(newdev
, MAP_1
);
9871 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9873 imsm_set_array_size(newdev
, -1);
9876 sp
= (void **)id
->dev
;
9881 /* Clear migration record */
9882 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9885 *space_list
= tofree
;
9888 update_reshape_exit
:
9893 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9894 struct intel_super
*super
,
9897 struct imsm_dev
*dev
= NULL
;
9898 struct intel_dev
*dv
;
9899 struct imsm_dev
*dev_new
;
9900 struct imsm_map
*map
;
9904 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9905 if (dv
->index
== (unsigned int)u
->subarray
) {
9913 map
= get_imsm_map(dev
, MAP_0
);
9915 if (u
->direction
== R10_TO_R0
) {
9916 /* Number of failed disks must be half of initial disk number */
9917 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9918 (map
->num_members
/ 2))
9921 /* iterate through devices to mark removed disks as spare */
9922 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9923 if (dm
->disk
.status
& FAILED_DISK
) {
9924 int idx
= dm
->index
;
9925 /* update indexes on the disk list */
9926 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9927 the index values will end up being correct.... NB */
9928 for (du
= super
->disks
; du
; du
= du
->next
)
9929 if (du
->index
> idx
)
9931 /* mark as spare disk */
9936 map
->num_members
/= map
->num_domains
;
9937 map
->map_state
= IMSM_T_STATE_NORMAL
;
9938 update_imsm_raid_level(map
, IMSM_T_RAID0
);
9939 set_num_domains(map
);
9940 update_num_data_stripes(map
, imsm_dev_size(dev
));
9941 map
->failed_disk_num
= -1;
9944 if (u
->direction
== R0_TO_R10
) {
9947 /* update slots in current disk list */
9948 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9952 /* create new *missing* disks */
9953 for (i
= 0; i
< map
->num_members
; i
++) {
9954 space
= *space_list
;
9957 *space_list
= *space
;
9959 memcpy(du
, super
->disks
, sizeof(*du
));
9963 du
->index
= (i
* 2) + 1;
9964 sprintf((char *)du
->disk
.serial
,
9965 " MISSING_%d", du
->index
);
9966 sprintf((char *)du
->serial
,
9967 "MISSING_%d", du
->index
);
9968 du
->next
= super
->missing
;
9969 super
->missing
= du
;
9971 /* create new dev and map */
9972 space
= *space_list
;
9975 *space_list
= *space
;
9976 dev_new
= (void *)space
;
9977 memcpy(dev_new
, dev
, sizeof(*dev
));
9978 /* update new map */
9979 map
= get_imsm_map(dev_new
, MAP_0
);
9981 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9982 update_imsm_raid_level(map
, IMSM_T_RAID10
);
9983 set_num_domains(map
);
9984 map
->num_members
= map
->num_members
* map
->num_domains
;
9985 update_num_data_stripes(map
, imsm_dev_size(dev
));
9987 /* replace dev<->dev_new */
9990 /* update disk order table */
9991 for (du
= super
->disks
; du
; du
= du
->next
)
9993 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9994 for (du
= super
->missing
; du
; du
= du
->next
)
9995 if (du
->index
>= 0) {
9996 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9997 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
10003 static void imsm_process_update(struct supertype
*st
,
10004 struct metadata_update
*update
)
10007 * crack open the metadata_update envelope to find the update record
10008 * update can be one of:
10009 * update_reshape_container_disks - all the arrays in the container
10010 * are being reshaped to have more devices. We need to mark
10011 * the arrays for general migration and convert selected spares
10012 * into active devices.
10013 * update_activate_spare - a spare device has replaced a failed
10014 * device in an array, update the disk_ord_tbl. If this disk is
10015 * present in all member arrays then also clear the SPARE_DISK
10017 * update_create_array
10018 * update_kill_array
10019 * update_rename_array
10020 * update_add_remove_disk
10022 struct intel_super
*super
= st
->sb
;
10023 struct imsm_super
*mpb
;
10024 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
10026 /* update requires a larger buf but the allocation failed */
10027 if (super
->next_len
&& !super
->next_buf
) {
10028 super
->next_len
= 0;
10032 if (super
->next_buf
) {
10033 memcpy(super
->next_buf
, super
->buf
, super
->len
);
10035 super
->len
= super
->next_len
;
10036 super
->buf
= super
->next_buf
;
10038 super
->next_len
= 0;
10039 super
->next_buf
= NULL
;
10042 mpb
= super
->anchor
;
10045 case update_general_migration_checkpoint
: {
10046 struct intel_dev
*id
;
10047 struct imsm_update_general_migration_checkpoint
*u
=
10048 (void *)update
->buf
;
10050 dprintf("called for update_general_migration_checkpoint\n");
10052 /* find device under general migration */
10053 for (id
= super
->devlist
; id
; id
= id
->next
) {
10054 if (is_gen_migration(id
->dev
)) {
10055 set_vol_curr_migr_unit(id
->dev
,
10056 u
->curr_migr_unit
);
10057 super
->updates_pending
++;
10062 case update_takeover
: {
10063 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10064 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
10065 imsm_update_version_info(super
);
10066 super
->updates_pending
++;
10071 case update_reshape_container_disks
: {
10072 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10073 if (apply_reshape_container_disks_update(
10074 u
, super
, &update
->space_list
))
10075 super
->updates_pending
++;
10078 case update_reshape_migration
: {
10079 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10080 if (apply_reshape_migration_update(
10081 u
, super
, &update
->space_list
))
10082 super
->updates_pending
++;
10085 case update_size_change
: {
10086 struct imsm_update_size_change
*u
= (void *)update
->buf
;
10087 if (apply_size_change_update(u
, super
))
10088 super
->updates_pending
++;
10091 case update_activate_spare
: {
10092 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
10094 if (prepare_spare_to_activate(st
, u
) &&
10095 apply_update_activate_spare(u
, super
, st
->arrays
))
10096 super
->updates_pending
++;
10099 case update_create_array
: {
10100 /* someone wants to create a new array, we need to be aware of
10101 * a few races/collisions:
10102 * 1/ 'Create' called by two separate instances of mdadm
10103 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
10104 * devices that have since been assimilated via
10106 * In the event this update can not be carried out mdadm will
10107 * (FIX ME) notice that its update did not take hold.
10109 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10110 struct intel_dev
*dv
;
10111 struct imsm_dev
*dev
;
10112 struct imsm_map
*map
, *new_map
;
10113 unsigned long long start
, end
;
10114 unsigned long long new_start
, new_end
;
10116 struct disk_info
*inf
;
10119 /* handle racing creates: first come first serve */
10120 if (u
->dev_idx
< mpb
->num_raid_devs
) {
10121 dprintf("subarray %d already defined\n", u
->dev_idx
);
10125 /* check update is next in sequence */
10126 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
10127 dprintf("can not create array %d expected index %d\n",
10128 u
->dev_idx
, mpb
->num_raid_devs
);
10132 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
10133 new_start
= pba_of_lba0(new_map
);
10134 new_end
= new_start
+ per_dev_array_size(new_map
);
10135 inf
= get_disk_info(u
);
10137 /* handle activate_spare versus create race:
10138 * check to make sure that overlapping arrays do not include
10139 * overalpping disks
10141 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10142 dev
= get_imsm_dev(super
, i
);
10143 map
= get_imsm_map(dev
, MAP_0
);
10144 start
= pba_of_lba0(map
);
10145 end
= start
+ per_dev_array_size(map
);
10146 if ((new_start
>= start
&& new_start
<= end
) ||
10147 (start
>= new_start
&& start
<= new_end
))
10152 if (disks_overlap(super
, i
, u
)) {
10153 dprintf("arrays overlap\n");
10158 /* check that prepare update was successful */
10159 if (!update
->space
) {
10160 dprintf("prepare update failed\n");
10164 /* check that all disks are still active before committing
10165 * changes. FIXME: could we instead handle this by creating a
10166 * degraded array? That's probably not what the user expects,
10167 * so better to drop this update on the floor.
10169 for (i
= 0; i
< new_map
->num_members
; i
++) {
10170 dl
= serial_to_dl(inf
[i
].serial
, super
);
10172 dprintf("disk disappeared\n");
10177 super
->updates_pending
++;
10179 /* convert spares to members and fixup ord_tbl */
10180 for (i
= 0; i
< new_map
->num_members
; i
++) {
10181 dl
= serial_to_dl(inf
[i
].serial
, super
);
10182 if (dl
->index
== -1) {
10183 dl
->index
= mpb
->num_disks
;
10185 dl
->disk
.status
|= CONFIGURED_DISK
;
10186 dl
->disk
.status
&= ~SPARE_DISK
;
10188 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
10191 dv
= update
->space
;
10193 update
->space
= NULL
;
10194 imsm_copy_dev(dev
, &u
->dev
);
10195 dv
->index
= u
->dev_idx
;
10196 dv
->next
= super
->devlist
;
10197 super
->devlist
= dv
;
10198 mpb
->num_raid_devs
++;
10200 imsm_update_version_info(super
);
10203 /* mdmon knows how to release update->space, but not
10204 * ((struct intel_dev *) update->space)->dev
10206 if (update
->space
) {
10207 dv
= update
->space
;
10212 case update_kill_array
: {
10213 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
10214 int victim
= u
->dev_idx
;
10215 struct active_array
*a
;
10216 struct intel_dev
**dp
;
10218 /* sanity check that we are not affecting the uuid of
10219 * active arrays, or deleting an active array
10221 * FIXME when immutable ids are available, but note that
10222 * we'll also need to fixup the invalidated/active
10223 * subarray indexes in mdstat
10225 for (a
= st
->arrays
; a
; a
= a
->next
)
10226 if (a
->info
.container_member
>= victim
)
10228 /* by definition if mdmon is running at least one array
10229 * is active in the container, so checking
10230 * mpb->num_raid_devs is just extra paranoia
10232 if (a
|| mpb
->num_raid_devs
== 1 || victim
>= super
->anchor
->num_raid_devs
) {
10233 dprintf("failed to delete subarray-%d\n", victim
);
10237 for (dp
= &super
->devlist
; *dp
;)
10238 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
10241 if ((*dp
)->index
> (unsigned)victim
)
10245 mpb
->num_raid_devs
--;
10246 super
->updates_pending
++;
10249 case update_rename_array
: {
10250 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
10251 char name
[MAX_RAID_SERIAL_LEN
+1];
10252 int target
= u
->dev_idx
;
10253 struct active_array
*a
;
10254 struct imsm_dev
*dev
;
10256 /* sanity check that we are not affecting the uuid of
10259 memset(name
, 0, sizeof(name
));
10260 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
10261 name
[MAX_RAID_SERIAL_LEN
] = '\0';
10262 for (a
= st
->arrays
; a
; a
= a
->next
)
10263 if (a
->info
.container_member
== target
)
10265 dev
= get_imsm_dev(super
, u
->dev_idx
);
10267 if (a
|| !dev
|| imsm_is_name_allowed(super
, name
, 0) == false) {
10268 dprintf("failed to rename subarray-%d\n", target
);
10272 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
10273 super
->updates_pending
++;
10276 case update_add_remove_disk
: {
10277 /* we may be able to repair some arrays if disks are
10278 * being added, check the status of add_remove_disk
10279 * if discs has been added.
10281 if (add_remove_disk_update(super
)) {
10282 struct active_array
*a
;
10284 super
->updates_pending
++;
10285 for (a
= st
->arrays
; a
; a
= a
->next
)
10286 a
->check_degraded
= 1;
10290 case update_prealloc_badblocks_mem
:
10292 case update_rwh_policy
: {
10293 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
10294 int target
= u
->dev_idx
;
10295 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
10297 if (dev
->rwh_policy
!= u
->new_policy
) {
10298 dev
->rwh_policy
= u
->new_policy
;
10299 super
->updates_pending
++;
10304 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10308 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10310 static int imsm_prepare_update(struct supertype
*st
,
10311 struct metadata_update
*update
)
10314 * Allocate space to hold new disk entries, raid-device entries or a new
10315 * mpb if necessary. The manager synchronously waits for updates to
10316 * complete in the monitor, so new mpb buffers allocated here can be
10317 * integrated by the monitor thread without worrying about live pointers
10318 * in the manager thread.
10320 enum imsm_update_type type
;
10321 struct intel_super
*super
= st
->sb
;
10322 unsigned int sector_size
= super
->sector_size
;
10323 struct imsm_super
*mpb
= super
->anchor
;
10327 if (update
->len
< (int)sizeof(type
))
10330 type
= *(enum imsm_update_type
*) update
->buf
;
10333 case update_general_migration_checkpoint
:
10334 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10336 dprintf("called for update_general_migration_checkpoint\n");
10338 case update_takeover
: {
10339 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10340 if (update
->len
< (int)sizeof(*u
))
10342 if (u
->direction
== R0_TO_R10
) {
10343 void **tail
= (void **)&update
->space_list
;
10344 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10345 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10346 int num_members
= map
->num_members
;
10349 /* allocate memory for added disks */
10350 for (i
= 0; i
< num_members
; i
++) {
10351 size
= sizeof(struct dl
);
10352 space
= xmalloc(size
);
10357 /* allocate memory for new device */
10358 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10359 (num_members
* sizeof(__u32
));
10360 space
= xmalloc(size
);
10364 len
= disks_to_mpb_size(num_members
* 2);
10369 case update_reshape_container_disks
: {
10370 /* Every raid device in the container is about to
10371 * gain some more devices, and we will enter a
10373 * So each 'imsm_map' will be bigger, and the imsm_vol
10374 * will now hold 2 of them.
10375 * Thus we need new 'struct imsm_dev' allocations sized
10376 * as sizeof_imsm_dev but with more devices in both maps.
10378 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10379 struct intel_dev
*dl
;
10380 void **space_tail
= (void**)&update
->space_list
;
10382 if (update
->len
< (int)sizeof(*u
))
10385 dprintf("for update_reshape\n");
10387 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10388 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10390 if (u
->new_raid_disks
> u
->old_raid_disks
)
10391 size
+= sizeof(__u32
)*2*
10392 (u
->new_raid_disks
- u
->old_raid_disks
);
10396 *space_tail
= NULL
;
10399 len
= disks_to_mpb_size(u
->new_raid_disks
);
10400 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10403 case update_reshape_migration
: {
10404 /* for migration level 0->5 we need to add disks
10405 * so the same as for container operation we will copy
10406 * device to the bigger location.
10407 * in memory prepared device and new disk area are prepared
10408 * for usage in process update
10410 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10411 struct intel_dev
*id
;
10412 void **space_tail
= (void **)&update
->space_list
;
10415 int current_level
= -1;
10417 if (update
->len
< (int)sizeof(*u
))
10420 dprintf("for update_reshape\n");
10422 /* add space for bigger array in update
10424 for (id
= super
->devlist
; id
; id
= id
->next
) {
10425 if (id
->index
== (unsigned)u
->subdev
) {
10426 size
= sizeof_imsm_dev(id
->dev
, 1);
10427 if (u
->new_raid_disks
> u
->old_raid_disks
)
10428 size
+= sizeof(__u32
)*2*
10429 (u
->new_raid_disks
- u
->old_raid_disks
);
10433 *space_tail
= NULL
;
10437 if (update
->space_list
== NULL
)
10440 /* add space for disk in update
10442 size
= sizeof(struct dl
);
10446 *space_tail
= NULL
;
10448 /* add spare device to update
10450 for (id
= super
->devlist
; id
; id
= id
->next
)
10451 if (id
->index
== (unsigned)u
->subdev
) {
10452 struct imsm_dev
*dev
;
10453 struct imsm_map
*map
;
10455 dev
= get_imsm_dev(super
, u
->subdev
);
10456 map
= get_imsm_map(dev
, MAP_0
);
10457 current_level
= map
->raid_level
;
10460 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10461 struct mdinfo
*spares
;
10463 spares
= get_spares_for_grow(st
);
10466 struct mdinfo
*dev
;
10468 dev
= spares
->devs
;
10471 makedev(dev
->disk
.major
,
10473 dl
= get_disk_super(super
,
10476 dl
->index
= u
->old_raid_disks
;
10479 sysfs_free(spares
);
10482 len
= disks_to_mpb_size(u
->new_raid_disks
);
10483 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10486 case update_size_change
: {
10487 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10491 case update_activate_spare
: {
10492 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10496 case update_create_array
: {
10497 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10498 struct intel_dev
*dv
;
10499 struct imsm_dev
*dev
= &u
->dev
;
10500 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10502 struct disk_info
*inf
;
10506 if (update
->len
< (int)sizeof(*u
))
10509 inf
= get_disk_info(u
);
10510 len
= sizeof_imsm_dev(dev
, 1);
10511 /* allocate a new super->devlist entry */
10512 dv
= xmalloc(sizeof(*dv
));
10513 dv
->dev
= xmalloc(len
);
10514 update
->space
= dv
;
10516 /* count how many spares will be converted to members */
10517 for (i
= 0; i
< map
->num_members
; i
++) {
10518 dl
= serial_to_dl(inf
[i
].serial
, super
);
10520 /* hmm maybe it failed?, nothing we can do about
10525 if (count_memberships(dl
, super
) == 0)
10528 len
+= activate
* sizeof(struct imsm_disk
);
10531 case update_kill_array
: {
10532 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10536 case update_rename_array
: {
10537 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10541 case update_add_remove_disk
:
10542 /* no update->len needed */
10544 case update_prealloc_badblocks_mem
:
10545 super
->extra_space
+= sizeof(struct bbm_log
) -
10546 get_imsm_bbm_log_size(super
->bbm_log
);
10548 case update_rwh_policy
: {
10549 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10557 /* check if we need a larger metadata buffer */
10558 if (super
->next_buf
)
10559 buf_len
= super
->next_len
;
10561 buf_len
= super
->len
;
10563 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10564 /* ok we need a larger buf than what is currently allocated
10565 * if this allocation fails process_update will notice that
10566 * ->next_len is set and ->next_buf is NULL
10568 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10569 super
->extra_space
+ len
, sector_size
);
10570 if (super
->next_buf
)
10571 free(super
->next_buf
);
10573 super
->next_len
= buf_len
;
10574 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10575 memset(super
->next_buf
, 0, buf_len
);
10577 super
->next_buf
= NULL
;
10582 /* must be called while manager is quiesced */
10583 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10585 struct imsm_super
*mpb
= super
->anchor
;
10587 struct imsm_dev
*dev
;
10588 struct imsm_map
*map
;
10589 unsigned int i
, j
, num_members
;
10590 __u32 ord
, ord_map0
;
10591 struct bbm_log
*log
= super
->bbm_log
;
10593 dprintf("deleting device[%d] from imsm_super\n", index
);
10595 /* shift all indexes down one */
10596 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10597 if (iter
->index
> (int)index
)
10599 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10600 if (iter
->index
> (int)index
)
10603 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10604 dev
= get_imsm_dev(super
, i
);
10605 map
= get_imsm_map(dev
, MAP_0
);
10606 num_members
= map
->num_members
;
10607 for (j
= 0; j
< num_members
; j
++) {
10608 /* update ord entries being careful not to propagate
10609 * ord-flags to the first map
10611 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10612 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10614 if (ord_to_idx(ord
) <= index
)
10617 map
= get_imsm_map(dev
, MAP_0
);
10618 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10619 map
= get_imsm_map(dev
, MAP_1
);
10621 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10625 for (i
= 0; i
< log
->entry_count
; i
++) {
10626 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10628 if (entry
->disk_ordinal
<= index
)
10630 entry
->disk_ordinal
--;
10634 super
->updates_pending
++;
10636 struct dl
*dl
= *dlp
;
10638 *dlp
= (*dlp
)->next
;
10639 __free_imsm_disk(dl
, 1);
10643 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10644 struct intel_super
*super
,
10645 struct imsm_dev
*dev
)
10651 struct imsm_map
*map
;
10654 ret_val
= raid_disks
/2;
10655 /* check map if all disks pairs not failed
10658 map
= get_imsm_map(dev
, MAP_0
);
10659 for (i
= 0; i
< ret_val
; i
++) {
10660 int degradation
= 0;
10661 if (get_imsm_disk(super
, i
) == NULL
)
10663 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10665 if (degradation
== 2)
10668 map
= get_imsm_map(dev
, MAP_1
);
10669 /* if there is no second map
10670 * result can be returned
10674 /* check degradation in second map
10676 for (i
= 0; i
< ret_val
; i
++) {
10677 int degradation
= 0;
10678 if (get_imsm_disk(super
, i
) == NULL
)
10680 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10682 if (degradation
== 2)
10696 /*******************************************************************************
10697 * Function: validate_container_imsm
10698 * Description: This routine validates container after assemble,
10699 * eg. if devices in container are under the same controller.
10702 * info : linked list with info about devices used in array
10706 ******************************************************************************/
10707 int validate_container_imsm(struct mdinfo
*info
)
10709 if (check_no_platform())
10712 struct sys_dev
*idev
;
10713 struct sys_dev
*hba
= NULL
;
10714 struct sys_dev
*intel_devices
= find_intel_devices();
10715 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10716 info
->disk
.minor
), 1, NULL
);
10718 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10719 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10728 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10729 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10733 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10734 struct mdinfo
*dev
;
10736 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10737 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
,
10738 dev
->disk
.minor
), 1, NULL
);
10740 struct sys_dev
*hba2
= NULL
;
10741 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10742 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10750 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10751 get_orom_by_device_id(hba2
->dev_id
);
10753 if (hba2
&& hba
->type
!= hba2
->type
) {
10754 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10755 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10759 if (orom
!= orom2
) {
10760 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10761 " This operation is not supported and can lead to data loss.\n");
10766 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10767 " This operation is not supported and can lead to data loss.\n");
10775 /*******************************************************************************
10776 * Function: imsm_record_badblock
10777 * Description: This routine stores new bad block record in BBM log
10780 * a : array containing a bad block
10781 * slot : disk number containing a bad block
10782 * sector : bad block sector
10783 * length : bad block sectors range
10787 ******************************************************************************/
10788 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10789 unsigned long long sector
, int length
)
10791 struct intel_super
*super
= a
->container
->sb
;
10795 ord
= imsm_disk_slot_to_ord(a
, slot
);
10799 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10802 super
->updates_pending
++;
10806 /*******************************************************************************
10807 * Function: imsm_clear_badblock
10808 * Description: This routine clears bad block record from BBM log
10811 * a : array containing a bad block
10812 * slot : disk number containing a bad block
10813 * sector : bad block sector
10814 * length : bad block sectors range
10818 ******************************************************************************/
10819 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10820 unsigned long long sector
, int length
)
10822 struct intel_super
*super
= a
->container
->sb
;
10826 ord
= imsm_disk_slot_to_ord(a
, slot
);
10830 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10832 super
->updates_pending
++;
10836 /*******************************************************************************
10837 * Function: imsm_get_badblocks
10838 * Description: This routine get list of bad blocks for an array
10842 * slot : disk number
10844 * bb : structure containing bad blocks
10846 ******************************************************************************/
10847 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10849 int inst
= a
->info
.container_member
;
10850 struct intel_super
*super
= a
->container
->sb
;
10851 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10852 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10855 ord
= imsm_disk_slot_to_ord(a
, slot
);
10859 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10860 per_dev_array_size(map
), &super
->bb
);
10864 /*******************************************************************************
10865 * Function: examine_badblocks_imsm
10866 * Description: Prints list of bad blocks on a disk to the standard output
10869 * st : metadata handler
10870 * fd : open file descriptor for device
10871 * devname : device name
10875 ******************************************************************************/
10876 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10878 struct intel_super
*super
= st
->sb
;
10879 struct bbm_log
*log
= super
->bbm_log
;
10880 struct dl
*d
= NULL
;
10883 for (d
= super
->disks
; d
; d
= d
->next
) {
10884 if (strcmp(d
->devname
, devname
) == 0)
10888 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10889 pr_err("%s doesn't appear to be part of a raid array\n",
10896 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10898 for (i
= 0; i
< log
->entry_count
; i
++) {
10899 if (entry
[i
].disk_ordinal
== d
->index
) {
10900 unsigned long long sector
= __le48_to_cpu(
10901 &entry
[i
].defective_block_start
);
10902 int cnt
= entry
[i
].marked_count
+ 1;
10905 printf("Bad-blocks on %s:\n", devname
);
10909 printf("%20llu for %d sectors\n", sector
, cnt
);
10915 printf("No bad-blocks list configured on %s\n", devname
);
10919 /*******************************************************************************
10920 * Function: init_migr_record_imsm
10921 * Description: Function inits imsm migration record
10923 * super : imsm internal array info
10924 * dev : device under migration
10925 * info : general array info to find the smallest device
10928 ******************************************************************************/
10929 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10930 struct mdinfo
*info
)
10932 struct intel_super
*super
= st
->sb
;
10933 struct migr_record
*migr_rec
= super
->migr_rec
;
10934 int new_data_disks
;
10935 unsigned long long dsize
, dev_sectors
;
10936 long long unsigned min_dev_sectors
= -1LLU;
10937 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10938 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10939 unsigned long long num_migr_units
;
10940 unsigned long long array_blocks
;
10941 struct dl
*dl_disk
= NULL
;
10943 memset(migr_rec
, 0, sizeof(struct migr_record
));
10944 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10946 /* only ascending reshape supported now */
10947 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10949 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10950 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10951 migr_rec
->dest_depth_per_unit
*=
10952 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10953 new_data_disks
= imsm_num_data_members(map_dest
);
10954 migr_rec
->blocks_per_unit
=
10955 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10956 migr_rec
->dest_depth_per_unit
=
10957 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10958 array_blocks
= info
->component_size
* new_data_disks
;
10960 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10962 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10964 set_num_migr_units(migr_rec
, num_migr_units
);
10966 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10967 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10969 /* Find the smallest dev */
10970 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
10971 /* ignore spares in container */
10972 if (dl_disk
->index
< 0)
10974 get_dev_size(dl_disk
->fd
, NULL
, &dsize
);
10975 dev_sectors
= dsize
/ 512;
10976 if (dev_sectors
< min_dev_sectors
)
10977 min_dev_sectors
= dev_sectors
;
10979 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
10980 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10982 write_imsm_migr_rec(st
);
10987 /*******************************************************************************
10988 * Function: save_backup_imsm
10989 * Description: Function saves critical data stripes to Migration Copy Area
10990 * and updates the current migration unit status.
10991 * Use restore_stripes() to form a destination stripe,
10992 * and to write it to the Copy Area.
10994 * st : supertype information
10995 * dev : imsm device that backup is saved for
10996 * info : general array info
10997 * buf : input buffer
10998 * length : length of data to backup (blocks_per_unit)
11002 ******************************************************************************/
11003 int save_backup_imsm(struct supertype
*st
,
11004 struct imsm_dev
*dev
,
11005 struct mdinfo
*info
,
11010 struct intel_super
*super
= st
->sb
;
11012 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
11013 int new_disks
= map_dest
->num_members
;
11014 int dest_layout
= 0;
11015 int dest_chunk
, targets
[new_disks
];
11016 unsigned long long start
, target_offsets
[new_disks
];
11017 int data_disks
= imsm_num_data_members(map_dest
);
11019 for (i
= 0; i
< new_disks
; i
++) {
11020 struct dl
*dl_disk
= get_imsm_dl_disk(super
, i
);
11021 if (dl_disk
&& is_fd_valid(dl_disk
->fd
))
11022 targets
[i
] = dl_disk
->fd
;
11027 start
= info
->reshape_progress
* 512;
11028 for (i
= 0; i
< new_disks
; i
++) {
11029 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
11030 /* move back copy area adderss, it will be moved forward
11031 * in restore_stripes() using start input variable
11033 target_offsets
[i
] -= start
/data_disks
;
11036 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
11037 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
11039 if (restore_stripes(targets
, /* list of dest devices */
11040 target_offsets
, /* migration record offsets */
11043 map_dest
->raid_level
,
11045 -1, /* source backup file descriptor */
11046 0, /* input buf offset
11047 * always 0 buf is already offseted */
11051 pr_err("Error restoring stripes\n");
11061 /*******************************************************************************
11062 * Function: save_checkpoint_imsm
11063 * Description: Function called for current unit status update
11064 * in the migration record. It writes it to disk.
11066 * super : imsm internal array info
11067 * info : general array info
11071 * 2: failure, means no valid migration record
11072 * / no general migration in progress /
11073 ******************************************************************************/
11074 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
11076 struct intel_super
*super
= st
->sb
;
11077 unsigned long long blocks_per_unit
;
11078 unsigned long long curr_migr_unit
;
11080 if (load_imsm_migr_rec(super
) != 0) {
11081 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
11085 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
11086 if (blocks_per_unit
== 0) {
11087 dprintf("imsm: no migration in progress.\n");
11090 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
11091 /* check if array is alligned to copy area
11092 * if it is not alligned, add one to current migration unit value
11093 * this can happend on array reshape finish only
11095 if (info
->reshape_progress
% blocks_per_unit
)
11098 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
11099 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
11100 set_migr_dest_1st_member_lba(super
->migr_rec
,
11101 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
11103 if (write_imsm_migr_rec(st
) < 0) {
11104 dprintf("imsm: Cannot write migration record outside backup area\n");
11111 /*******************************************************************************
11112 * Function: recover_backup_imsm
11113 * Description: Function recovers critical data from the Migration Copy Area
11114 * while assembling an array.
11116 * super : imsm internal array info
11117 * info : general array info
11119 * 0 : success (or there is no data to recover)
11121 ******************************************************************************/
11122 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
11124 struct intel_super
*super
= st
->sb
;
11125 struct migr_record
*migr_rec
= super
->migr_rec
;
11126 struct imsm_map
*map_dest
;
11127 struct intel_dev
*id
= NULL
;
11128 unsigned long long read_offset
;
11129 unsigned long long write_offset
;
11131 int new_disks
, err
;
11134 unsigned int sector_size
= super
->sector_size
;
11135 unsigned long long curr_migr_unit
= current_migr_unit(migr_rec
);
11136 unsigned long long num_migr_units
= get_num_migr_units(migr_rec
);
11137 char buffer
[SYSFS_MAX_BUF_SIZE
];
11138 int skipped_disks
= 0;
11139 struct dl
*dl_disk
;
11141 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, sizeof(buffer
));
11145 /* recover data only during assemblation */
11146 if (strncmp(buffer
, "inactive", 8) != 0)
11148 /* no data to recover */
11149 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
11151 if (curr_migr_unit
>= num_migr_units
)
11154 /* find device during reshape */
11155 for (id
= super
->devlist
; id
; id
= id
->next
)
11156 if (is_gen_migration(id
->dev
))
11161 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
11162 new_disks
= map_dest
->num_members
;
11164 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
11166 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
11167 pba_of_lba0(map_dest
)) * 512;
11169 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11170 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
11173 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
11174 if (dl_disk
->index
< 0)
11177 if (!is_fd_valid(dl_disk
->fd
)) {
11181 if (lseek64(dl_disk
->fd
, read_offset
, SEEK_SET
) < 0) {
11182 pr_err("Cannot seek to block: %s\n",
11187 if (read(dl_disk
->fd
, buf
, unit_len
) != (ssize_t
)unit_len
) {
11188 pr_err("Cannot read copy area block: %s\n",
11193 if (lseek64(dl_disk
->fd
, write_offset
, SEEK_SET
) < 0) {
11194 pr_err("Cannot seek to block: %s\n",
11199 if (write(dl_disk
->fd
, buf
, unit_len
) != (ssize_t
)unit_len
) {
11200 pr_err("Cannot restore block: %s\n",
11207 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11211 pr_err("Cannot restore data from backup. Too many failed disks\n");
11215 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11216 /* ignore error == 2, this can mean end of reshape here
11218 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11228 * test_and_add_drive_controller_policy_imsm() - add disk controller to policies list.
11229 * @type: Policy type to search on list.
11230 * @pols: List of currently recorded policies.
11231 * @disk_fd: File descriptor of the device to check.
11232 * @hba: The hba disk is attached, could be NULL if verification is disabled.
11233 * @verbose: verbose flag.
11235 * IMSM cares about drive physical placement. If @hba is not set, it adds unknown policy.
11236 * If there is no controller policy on pols we are free to add first one. If there is a policy then,
11237 * new must be the same - no controller mixing allowed.
11239 static mdadm_status_t
11240 test_and_add_drive_controller_policy_imsm(const char * const type
, dev_policy_t
**pols
, int disk_fd
,
11241 struct sys_dev
*hba
, const int verbose
)
11243 const char *controller_policy
= get_sys_dev_type(SYS_DEV_UNKNOWN
);
11244 struct dev_policy
*pol
= pol_find(*pols
, (char *)type
);
11245 char devname
[MAX_RAID_SERIAL_LEN
];
11248 controller_policy
= get_sys_dev_type(hba
->type
);
11251 pol_add(pols
, (char *)type
, (char *)controller_policy
, "imsm");
11252 return MDADM_STATUS_SUCCESS
;
11255 if (strcmp(pol
->value
, controller_policy
) == 0)
11256 return MDADM_STATUS_SUCCESS
;
11258 fd2devname(disk_fd
, devname
);
11259 pr_vrb("Intel(R) raid controller \"%s\" found for %s, but \"%s\" was detected earlier\n",
11260 controller_policy
, devname
, pol
->value
);
11261 pr_vrb("Disks under different controllers cannot be used, aborting\n");
11263 return MDADM_STATUS_ERROR
;
11267 * test_and_add_drive_encryption_policy_imsm() - add disk encryption to policies list.
11268 * @type: policy type to search in the list.
11269 * @pols: list of currently recorded policies.
11270 * @disk_fd: file descriptor of the device to check.
11271 * @hba: The hba to which the drive is attached, could be NULL if verification is disabled.
11272 * @verbose: verbose flag.
11274 * IMSM cares about drive encryption state. It is not allowed to mix disks with different
11275 * encryption state within one md device.
11276 * If there is no encryption policy on pols we are free to add first one.
11277 * If there is a policy then, new must be the same.
11279 static mdadm_status_t
11280 test_and_add_drive_encryption_policy_imsm(const char * const type
, dev_policy_t
**pols
, int disk_fd
,
11281 struct sys_dev
*hba
, const int verbose
)
11283 struct dev_policy
*expected_policy
= pol_find(*pols
, (char *)type
);
11284 struct encryption_information information
= {0};
11285 char *encryption_state
= "Unknown";
11286 int status
= MDADM_STATUS_SUCCESS
;
11287 bool encryption_checked
= true;
11288 char devname
[PATH_MAX
];
11293 switch (hba
->type
) {
11296 status
= get_nvme_opal_encryption_information(disk_fd
, &information
, verbose
);
11299 case SYS_DEV_SATA_VMD
:
11300 status
= get_ata_encryption_information(disk_fd
, &information
, verbose
);
11303 encryption_checked
= false;
11307 fd2devname(disk_fd
, devname
);
11308 pr_vrb("Failed to read encryption information of device %s\n", devname
);
11309 return MDADM_STATUS_ERROR
;
11312 if (encryption_checked
) {
11313 if (information
.status
== ENC_STATUS_LOCKED
) {
11314 fd2devname(disk_fd
, devname
);
11315 pr_vrb("Device %s is in Locked state, cannot use. Aborting.\n", devname
);
11316 return MDADM_STATUS_ERROR
;
11318 encryption_state
= (char *)get_encryption_status_string(information
.status
);
11322 if (expected_policy
) {
11323 if (strcmp(expected_policy
->value
, encryption_state
) == 0)
11324 return MDADM_STATUS_SUCCESS
;
11326 fd2devname(disk_fd
, devname
);
11327 pr_vrb("Encryption status \"%s\" detected for disk %s, but \"%s\" status was detected eariler.\n",
11328 encryption_state
, devname
, expected_policy
->value
);
11329 pr_vrb("Disks with different encryption status cannot be used.\n");
11330 return MDADM_STATUS_ERROR
;
11333 pol_add(pols
, (char *)type
, encryption_state
, "imsm");
11335 return MDADM_STATUS_SUCCESS
;
11338 struct imsm_drive_policy
{
11340 mdadm_status_t (*test_and_add_drive_policy
)(const char * const type
,
11341 struct dev_policy
**pols
, int disk_fd
,
11342 struct sys_dev
*hba
, const int verbose
);
11345 struct imsm_drive_policy imsm_policies
[] = {
11346 {"controller", test_and_add_drive_controller_policy_imsm
},
11347 {"encryption", test_and_add_drive_encryption_policy_imsm
}
11350 mdadm_status_t
test_and_add_drive_policies_imsm(struct dev_policy
**pols
, int disk_fd
,
11353 struct imsm_drive_policy
*imsm_pol
;
11354 struct sys_dev
*hba
= NULL
;
11355 char path
[PATH_MAX
];
11356 mdadm_status_t ret
;
11359 /* If imsm platform verification is disabled, do not search for hba. */
11360 if (check_no_platform() != 1) {
11361 if (!diskfd_to_devpath(disk_fd
, 1, path
)) {
11362 pr_vrb("IMSM: Failed to retrieve device path by file descriptor.\n");
11363 return MDADM_STATUS_ERROR
;
11366 hba
= find_disk_attached_hba(disk_fd
, path
);
11368 pr_vrb("IMSM: Failed to find hba for %s\n", path
);
11369 return MDADM_STATUS_ERROR
;
11373 for (i
= 0; i
< ARRAY_SIZE(imsm_policies
); i
++) {
11374 imsm_pol
= &imsm_policies
[i
];
11376 ret
= imsm_pol
->test_and_add_drive_policy(imsm_pol
->type
, pols
, disk_fd
, hba
,
11378 if (ret
!= MDADM_STATUS_SUCCESS
)
11379 /* Inherit error code */
11383 return MDADM_STATUS_SUCCESS
;
11387 * get_spare_criteria_imsm() - set spare criteria.
11389 * @mddev_path: path to md device devnode, it must be container.
11390 * @c: spare_criteria struct to fill, not NULL.
11392 * If superblock is not loaded, use mddev_path to load_container. It must be given in this case.
11393 * Filles size and sector size accordingly to superblock.
11395 mdadm_status_t
get_spare_criteria_imsm(struct supertype
*st
, char *mddev_path
,
11396 struct spare_criteria
*c
)
11398 mdadm_status_t ret
= MDADM_STATUS_ERROR
;
11399 bool free_superblock
= false;
11400 unsigned long long size
= 0;
11401 struct intel_super
*super
;
11406 /* If no superblock and no mddev_path, we cannot load superblock. */
11407 assert(st
->sb
|| mddev_path
);
11410 int fd
= open(mddev_path
, O_RDONLY
);
11413 if (!is_fd_valid(fd
))
11414 return MDADM_STATUS_ERROR
;
11417 if (load_container_imsm(st
, fd
, st
->devnm
)) {
11419 return MDADM_STATUS_ERROR
;
11421 free_superblock
= true;
11424 rv
= mddev_test_and_add_drive_policies(st
, &c
->pols
, fd
, 0);
11427 if (rv
!= MDADM_STATUS_SUCCESS
)
11433 /* find first active disk in array */
11435 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
11441 /* find last lba used by subarrays */
11442 e
= get_extents(super
, dl
, 0);
11446 for (i
= 0; e
[i
].size
; i
++)
11449 size
= e
[i
- 1].start
+ e
[i
- 1].size
;
11452 /* add the amount of space needed for metadata */
11453 size
+= imsm_min_reserved_sectors(super
);
11455 c
->min_size
= size
* 512;
11456 c
->sector_size
= super
->sector_size
;
11457 c
->criteria_set
= true;
11458 ret
= MDADM_STATUS_SUCCESS
;
11461 if (free_superblock
)
11462 free_super_imsm(st
);
11464 if (ret
!= MDADM_STATUS_SUCCESS
)
11465 c
->criteria_set
= false;
11470 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11472 static char devnm
[32];
11473 char subdev_name
[20];
11474 struct mdstat_ent
*mdstat
;
11476 sprintf(subdev_name
, "%d", subdev
);
11477 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11481 strcpy(devnm
, mdstat
->devnm
);
11482 free_mdstat(mdstat
);
11486 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11487 struct geo_params
*geo
,
11488 int *old_raid_disks
,
11491 /* currently we only support increasing the number of devices
11492 * for a container. This increases the number of device for each
11493 * member array. They must all be RAID0 or RAID5.
11496 struct mdinfo
*info
, *member
;
11497 int devices_that_can_grow
= 0;
11499 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11501 if (geo
->size
> 0 ||
11502 geo
->level
!= UnSet
||
11503 geo
->layout
!= UnSet
||
11504 geo
->chunksize
!= 0 ||
11505 geo
->raid_disks
== UnSet
) {
11506 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11510 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11511 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11515 info
= container_content_imsm(st
, NULL
);
11516 for (member
= info
; member
; member
= member
->next
) {
11519 dprintf("imsm: checking device_num: %i\n",
11520 member
->container_member
);
11522 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11523 /* we work on container for Online Capacity Expansion
11524 * only so raid_disks has to grow
11526 dprintf("imsm: for container operation raid disks increase is required\n");
11530 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11531 /* we cannot use this container with other raid level
11533 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11534 info
->array
.level
);
11537 /* check for platform support
11538 * for this raid level configuration
11540 struct intel_super
*super
= st
->sb
;
11541 if (!is_raid_level_supported(super
->orom
,
11542 member
->array
.level
,
11543 geo
->raid_disks
)) {
11544 dprintf("platform does not support raid%d with %d disk%s\n",
11547 geo
->raid_disks
> 1 ? "s" : "");
11550 /* check if component size is aligned to chunk size
11552 if (info
->component_size
%
11553 (info
->array
.chunk_size
/512)) {
11554 dprintf("Component size is not aligned to chunk size\n");
11559 if (*old_raid_disks
&&
11560 info
->array
.raid_disks
!= *old_raid_disks
)
11562 *old_raid_disks
= info
->array
.raid_disks
;
11564 /* All raid5 and raid0 volumes in container
11565 * have to be ready for Online Capacity Expansion
11566 * so they need to be assembled. We have already
11567 * checked that no recovery etc is happening.
11569 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11570 st
->container_devnm
);
11571 if (result
== NULL
) {
11572 dprintf("imsm: cannot find array\n");
11575 devices_that_can_grow
++;
11578 if (!member
&& devices_that_can_grow
)
11582 dprintf("Container operation allowed\n");
11584 dprintf("Error: %i\n", ret_val
);
11589 /* Function: get_spares_for_grow
11590 * Description: Allocates memory and creates list of spare devices
11591 * avaliable in container. Checks if spare drive size is acceptable.
11592 * Parameters: Pointer to the supertype structure
11593 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11596 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11598 struct spare_criteria sc
= {0};
11599 struct mdinfo
*spares
;
11601 get_spare_criteria_imsm(st
, NULL
, &sc
);
11602 spares
= container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11604 dev_policy_free(sc
.pols
);
11609 /******************************************************************************
11610 * function: imsm_create_metadata_update_for_reshape
11611 * Function creates update for whole IMSM container.
11613 ******************************************************************************/
11614 static int imsm_create_metadata_update_for_reshape(
11615 struct supertype
*st
,
11616 struct geo_params
*geo
,
11617 int old_raid_disks
,
11618 struct imsm_update_reshape
**updatep
)
11620 struct intel_super
*super
= st
->sb
;
11621 struct imsm_super
*mpb
= super
->anchor
;
11622 int update_memory_size
;
11623 struct imsm_update_reshape
*u
;
11624 struct mdinfo
*spares
;
11627 struct mdinfo
*dev
;
11629 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11631 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11633 /* size of all update data without anchor */
11634 update_memory_size
= sizeof(struct imsm_update_reshape
);
11636 /* now add space for spare disks that we need to add. */
11637 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11639 u
= xcalloc(1, update_memory_size
);
11640 u
->type
= update_reshape_container_disks
;
11641 u
->old_raid_disks
= old_raid_disks
;
11642 u
->new_raid_disks
= geo
->raid_disks
;
11644 /* now get spare disks list
11646 spares
= get_spares_for_grow(st
);
11648 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11649 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11654 /* we have got spares
11655 * update disk list in imsm_disk list table in anchor
11657 dprintf("imsm: %i spares are available.\n\n",
11658 spares
->array
.spare_disks
);
11660 dev
= spares
->devs
;
11661 for (i
= 0; i
< delta_disks
; i
++) {
11666 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11668 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11669 dl
->index
= mpb
->num_disks
;
11677 sysfs_free(spares
);
11679 dprintf("imsm: reshape update preparation :");
11680 if (i
== delta_disks
) {
11681 dprintf_cont(" OK\n");
11683 return update_memory_size
;
11686 dprintf_cont(" Error\n");
11691 /******************************************************************************
11692 * function: imsm_create_metadata_update_for_size_change()
11693 * Creates update for IMSM array for array size change.
11695 ******************************************************************************/
11696 static int imsm_create_metadata_update_for_size_change(
11697 struct supertype
*st
,
11698 struct geo_params
*geo
,
11699 struct imsm_update_size_change
**updatep
)
11701 struct intel_super
*super
= st
->sb
;
11702 int update_memory_size
;
11703 struct imsm_update_size_change
*u
;
11705 dprintf("(enter) New size = %llu\n", geo
->size
);
11707 /* size of all update data without anchor */
11708 update_memory_size
= sizeof(struct imsm_update_size_change
);
11710 u
= xcalloc(1, update_memory_size
);
11711 u
->type
= update_size_change
;
11712 u
->subdev
= super
->current_vol
;
11713 u
->new_size
= geo
->size
;
11715 dprintf("imsm: reshape update preparation : OK\n");
11718 return update_memory_size
;
11721 /******************************************************************************
11722 * function: imsm_create_metadata_update_for_migration()
11723 * Creates update for IMSM array.
11725 ******************************************************************************/
11726 static int imsm_create_metadata_update_for_migration(
11727 struct supertype
*st
,
11728 struct geo_params
*geo
,
11729 struct imsm_update_reshape_migration
**updatep
)
11731 struct intel_super
*super
= st
->sb
;
11732 int update_memory_size
;
11733 int current_chunk_size
;
11734 struct imsm_update_reshape_migration
*u
;
11735 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11736 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11737 int previous_level
= -1;
11739 dprintf("(enter) New Level = %i\n", geo
->level
);
11741 /* size of all update data without anchor */
11742 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11744 u
= xcalloc(1, update_memory_size
);
11745 u
->type
= update_reshape_migration
;
11746 u
->subdev
= super
->current_vol
;
11747 u
->new_level
= geo
->level
;
11748 u
->new_layout
= geo
->layout
;
11749 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11750 u
->new_disks
[0] = -1;
11751 u
->new_chunksize
= -1;
11753 current_chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) / 2;
11755 if (geo
->chunksize
!= current_chunk_size
) {
11756 u
->new_chunksize
= geo
->chunksize
/ 1024;
11757 dprintf("imsm: chunk size change from %i to %i\n",
11758 current_chunk_size
, u
->new_chunksize
);
11760 previous_level
= map
->raid_level
;
11762 if (geo
->level
== 5 && previous_level
== 0) {
11763 struct mdinfo
*spares
= NULL
;
11765 u
->new_raid_disks
++;
11766 spares
= get_spares_for_grow(st
);
11767 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11769 sysfs_free(spares
);
11770 update_memory_size
= 0;
11771 pr_err("cannot get spare device for requested migration\n");
11774 sysfs_free(spares
);
11776 dprintf("imsm: reshape update preparation : OK\n");
11779 return update_memory_size
;
11782 static void imsm_update_metadata_locally(struct supertype
*st
,
11783 void *buf
, int len
)
11785 struct metadata_update mu
;
11790 mu
.space_list
= NULL
;
11792 if (imsm_prepare_update(st
, &mu
))
11793 imsm_process_update(st
, &mu
);
11795 while (mu
.space_list
) {
11796 void **space
= mu
.space_list
;
11797 mu
.space_list
= *space
;
11803 * imsm_analyze_expand() - check expand properties and calculate new size.
11804 * @st: imsm supertype.
11805 * @geo: new geometry params.
11806 * @array: array info.
11807 * @direction: reshape direction.
11809 * Obtain free space after the &array and verify if expand to requested size is
11810 * possible. If geo->size is set to %MAX_SIZE, assume that max free size is
11814 * On success %IMSM_STATUS_OK is returned, geo->size and geo->raid_disks are
11816 * On error, %IMSM_STATUS_ERROR is returned.
11818 static imsm_status_t
imsm_analyze_expand(struct supertype
*st
,
11819 struct geo_params
*geo
,
11820 struct mdinfo
*array
,
11823 struct intel_super
*super
= st
->sb
;
11824 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11825 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11826 int data_disks
= imsm_num_data_members(map
);
11828 unsigned long long current_size
;
11829 unsigned long long free_size
;
11830 unsigned long long new_size
;
11831 unsigned long long max_size
;
11833 const int chunk_kib
= geo
->chunksize
/ 1024;
11836 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11838 * Accept size for rollback only.
11840 new_size
= geo
->size
* 2;
11844 if (data_disks
== 0) {
11845 pr_err("imsm: Cannot retrieve data disks.\n");
11846 return IMSM_STATUS_ERROR
;
11848 current_size
= array
->custom_array_size
/ data_disks
;
11850 rv
= imsm_get_free_size(super
, dev
->vol
.map
->num_members
, 0, chunk_kib
, &free_size
, true);
11851 if (rv
!= IMSM_STATUS_OK
) {
11852 pr_err("imsm: Cannot find free space for expand.\n");
11853 return IMSM_STATUS_ERROR
;
11855 max_size
= round_member_size_to_mb(free_size
+ current_size
);
11857 if (geo
->size
== MAX_SIZE
)
11858 new_size
= max_size
;
11860 new_size
= round_member_size_to_mb(geo
->size
* 2);
11862 if (new_size
== 0) {
11863 pr_err("imsm: Rounded requested size is 0.\n");
11864 return IMSM_STATUS_ERROR
;
11867 if (new_size
> max_size
) {
11868 pr_err("imsm: Rounded requested size (%llu) is larger than free space available (%llu).\n",
11869 new_size
, max_size
);
11870 return IMSM_STATUS_ERROR
;
11873 if (new_size
== current_size
) {
11874 pr_err("imsm: Rounded requested size (%llu) is same as current size (%llu).\n",
11875 new_size
, current_size
);
11876 return IMSM_STATUS_ERROR
;
11879 if (new_size
< current_size
) {
11880 pr_err("imsm: Size reduction is not supported, rounded requested size (%llu) is smaller than current (%llu).\n",
11881 new_size
, current_size
);
11882 return IMSM_STATUS_ERROR
;
11886 dprintf("imsm: New size per member is %llu.\n", new_size
);
11887 geo
->size
= data_disks
* new_size
;
11888 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11889 return IMSM_STATUS_OK
;
11892 /***************************************************************************
11893 * Function: imsm_analyze_change
11894 * Description: Function analyze change for single volume
11895 * and validate if transition is supported
11896 * Parameters: Geometry parameters, supertype structure,
11897 * metadata change direction (apply/rollback)
11898 * Returns: Operation type code on success, -1 if fail
11899 ****************************************************************************/
11900 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11901 struct geo_params
*geo
,
11902 int direction
, struct context
*c
)
11904 struct mdinfo info
;
11906 int check_devs
= 0;
11908 /* imsm compatible layout value for array geometry verification */
11909 int imsm_layout
= -1;
11910 int raid_disks
= geo
->raid_disks
;
11913 getinfo_super_imsm_volume(st
, &info
, NULL
);
11914 if (geo
->level
!= info
.array
.level
&& geo
->level
>= IMSM_T_RAID0
&&
11915 geo
->level
!= UnSet
) {
11916 switch (info
.array
.level
) {
11918 if (geo
->level
== IMSM_T_RAID5
) {
11919 change
= CH_MIGRATION
;
11920 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11921 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11923 goto analyse_change_exit
;
11925 imsm_layout
= geo
->layout
;
11927 raid_disks
+= 1; /* parity disk added */
11928 } else if (geo
->level
== IMSM_T_RAID10
) {
11929 if (geo
->level
== IMSM_T_RAID10
&& geo
->raid_disks
> 2 &&
11931 pr_err("Warning! VROC UEFI driver does not support RAID10 in requested layout.\n");
11932 pr_err("Array won't be suitable as boot device.\n");
11933 pr_err("Note: You can omit this check with \"--force\"\n");
11934 if (ask("Do you want to continue") < 1)
11937 change
= CH_TAKEOVER
;
11939 raid_disks
*= 2; /* mirrors added */
11940 imsm_layout
= 0x102; /* imsm supported layout */
11944 case IMSM_T_RAID10
:
11945 if (geo
->level
== 0) {
11946 change
= CH_TAKEOVER
;
11949 imsm_layout
= 0; /* imsm raid0 layout */
11953 if (change
== -1) {
11954 pr_err("Error. Level Migration from %d to %d not supported!\n",
11955 info
.array
.level
, geo
->level
);
11956 goto analyse_change_exit
;
11959 geo
->level
= info
.array
.level
;
11961 if (geo
->layout
!= info
.array
.layout
&&
11962 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11963 change
= CH_MIGRATION
;
11964 if (info
.array
.layout
== 0 && info
.array
.level
== IMSM_T_RAID5
&&
11965 geo
->layout
== 5) {
11966 /* reshape 5 -> 4 */
11967 } else if (info
.array
.layout
== 5 && info
.array
.level
== IMSM_T_RAID5
&&
11968 geo
->layout
== 0) {
11969 /* reshape 4 -> 5 */
11973 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11974 info
.array
.layout
, geo
->layout
);
11976 goto analyse_change_exit
;
11979 geo
->layout
= info
.array
.layout
;
11980 if (imsm_layout
== -1)
11981 imsm_layout
= info
.array
.layout
;
11984 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11985 geo
->chunksize
!= info
.array
.chunk_size
) {
11986 if (info
.array
.level
== IMSM_T_RAID10
) {
11987 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11989 goto analyse_change_exit
;
11990 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11991 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11992 geo
->chunksize
/1024, info
.component_size
/2);
11994 goto analyse_change_exit
;
11996 change
= CH_MIGRATION
;
11998 geo
->chunksize
= info
.array
.chunk_size
;
12001 if (geo
->size
> 0) {
12002 if (change
!= -1) {
12003 pr_err("Error. Size change should be the only one at a time.\n");
12005 goto analyse_change_exit
;
12008 rv
= imsm_analyze_expand(st
, geo
, &info
, direction
);
12009 if (rv
!= IMSM_STATUS_OK
)
12010 goto analyse_change_exit
;
12011 raid_disks
= geo
->raid_disks
;
12012 change
= CH_ARRAY_SIZE
;
12015 chunk
= geo
->chunksize
/ 1024;
12017 if (!validate_geometry_imsm(st
,
12022 geo
->size
, INVALID_SECTORS
,
12023 0, 0, info
.consistency_policy
, 1))
12027 struct intel_super
*super
= st
->sb
;
12028 struct imsm_super
*mpb
= super
->anchor
;
12030 if (mpb
->num_raid_devs
> 1) {
12031 pr_err("Error. Cannot perform operation on %s- for this operation "
12032 "it MUST be single array in container\n", geo
->dev_name
);
12037 analyse_change_exit
:
12038 if (direction
== ROLLBACK_METADATA_CHANGES
&&
12039 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
12040 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
12046 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
12048 struct intel_super
*super
= st
->sb
;
12049 struct imsm_update_takeover
*u
;
12051 u
= xmalloc(sizeof(struct imsm_update_takeover
));
12053 u
->type
= update_takeover
;
12054 u
->subarray
= super
->current_vol
;
12056 /* 10->0 transition */
12057 if (geo
->level
== 0)
12058 u
->direction
= R10_TO_R0
;
12060 /* 0->10 transition */
12061 if (geo
->level
== 10)
12062 u
->direction
= R0_TO_R10
;
12064 /* update metadata locally */
12065 imsm_update_metadata_locally(st
, u
,
12066 sizeof(struct imsm_update_takeover
));
12067 /* and possibly remotely */
12068 if (st
->update_tail
)
12069 append_metadata_update(st
, u
,
12070 sizeof(struct imsm_update_takeover
));
12077 /* Flush size update if size calculated by num_data_stripes is higher than
12078 * imsm_dev_size to eliminate differences during reshape.
12079 * Mdmon will recalculate them correctly.
12080 * If subarray index is not set then check whole container.
12082 * 0 - no error occurred
12083 * 1 - error detected
12085 static int imsm_fix_size_mismatch(struct supertype
*st
, int subarray_index
)
12087 struct intel_super
*super
= st
->sb
;
12088 int tmp
= super
->current_vol
;
12092 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
12093 if (subarray_index
>= 0 && i
!= subarray_index
)
12095 super
->current_vol
= i
;
12096 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
12097 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
12098 unsigned int disc_count
= imsm_num_data_members(map
);
12099 struct geo_params geo
;
12100 struct imsm_update_size_change
*update
;
12101 unsigned long long calc_size
= per_dev_array_size(map
) * disc_count
;
12102 unsigned long long d_size
= imsm_dev_size(dev
);
12105 if (calc_size
== d_size
)
12108 /* There is a difference, confirm that imsm_dev_size is
12109 * smaller and push update.
12111 if (d_size
> calc_size
) {
12112 pr_err("imsm: dev size of subarray %d is incorrect\n",
12116 memset(&geo
, 0, sizeof(struct geo_params
));
12118 u_size
= imsm_create_metadata_update_for_size_change(st
, &geo
,
12120 imsm_update_metadata_locally(st
, update
, u_size
);
12121 if (st
->update_tail
) {
12122 append_metadata_update(st
, update
, u_size
);
12123 flush_metadata_updates(st
);
12124 st
->update_tail
= &st
->updates
;
12126 imsm_sync_metadata(st
);
12132 super
->current_vol
= tmp
;
12137 * shape_to_geo() - fill geo_params from shape.
12139 * @shape: array details.
12140 * @geo: new geometry params.
12141 * Returns: 0 on success, 1 otherwise.
12143 static void shape_to_geo(struct shape
*shape
, struct geo_params
*geo
)
12148 geo
->dev_name
= shape
->dev
;
12149 geo
->size
= shape
->size
;
12150 geo
->level
= shape
->level
;
12151 geo
->layout
= shape
->layout
;
12152 geo
->chunksize
= shape
->chunk
;
12153 geo
->raid_disks
= shape
->raiddisks
;
12156 static int imsm_reshape_super(struct supertype
*st
, struct shape
*shape
, struct context
*c
)
12159 struct geo_params geo
= {0};
12161 dprintf("(enter)\n");
12163 shape_to_geo(shape
, &geo
);
12164 strcpy(geo
.devnm
, st
->devnm
);
12165 if (shape
->delta_disks
!= UnSet
)
12166 geo
.raid_disks
+= shape
->delta_disks
;
12168 dprintf("for level : %i\n", geo
.level
);
12169 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
12171 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
12172 /* On container level we can only increase number of devices. */
12173 dprintf("imsm: info: Container operation\n");
12174 int old_raid_disks
= 0;
12176 if (imsm_reshape_is_allowed_on_container(
12177 st
, &geo
, &old_raid_disks
, shape
->direction
)) {
12178 struct imsm_update_reshape
*u
= NULL
;
12181 if (imsm_fix_size_mismatch(st
, -1)) {
12182 dprintf("imsm: Cannot fix size mismatch\n");
12183 goto exit_imsm_reshape_super
;
12186 len
= imsm_create_metadata_update_for_reshape(
12187 st
, &geo
, old_raid_disks
, &u
);
12190 dprintf("imsm: Cannot prepare update\n");
12191 goto exit_imsm_reshape_super
;
12195 /* update metadata locally */
12196 imsm_update_metadata_locally(st
, u
, len
);
12197 /* and possibly remotely */
12198 if (st
->update_tail
)
12199 append_metadata_update(st
, u
, len
);
12204 pr_err("(imsm) Operation is not allowed on this container\n");
12207 /* On volume level we support following operations
12208 * - takeover: raid10 -> raid0; raid0 -> raid10
12209 * - chunk size migration
12210 * - migration: raid5 -> raid0; raid0 -> raid5
12212 struct intel_super
*super
= st
->sb
;
12213 struct intel_dev
*dev
= super
->devlist
;
12215 dprintf("imsm: info: Volume operation\n");
12216 /* find requested device */
12219 imsm_find_array_devnm_by_subdev(
12220 dev
->index
, st
->container_devnm
);
12221 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
12226 pr_err("Cannot find %s (%s) subarray\n",
12227 geo
.dev_name
, geo
.devnm
);
12228 goto exit_imsm_reshape_super
;
12230 super
->current_vol
= dev
->index
;
12231 change
= imsm_analyze_change(st
, &geo
, shape
->direction
, c
);
12234 ret_val
= imsm_takeover(st
, &geo
);
12236 case CH_MIGRATION
: {
12237 struct imsm_update_reshape_migration
*u
= NULL
;
12239 imsm_create_metadata_update_for_migration(
12242 dprintf("imsm: Cannot prepare update\n");
12246 /* update metadata locally */
12247 imsm_update_metadata_locally(st
, u
, len
);
12248 /* and possibly remotely */
12249 if (st
->update_tail
)
12250 append_metadata_update(st
, u
, len
);
12255 case CH_ARRAY_SIZE
: {
12256 struct imsm_update_size_change
*u
= NULL
;
12258 imsm_create_metadata_update_for_size_change(
12261 dprintf("imsm: Cannot prepare update\n");
12265 /* update metadata locally */
12266 imsm_update_metadata_locally(st
, u
, len
);
12267 /* and possibly remotely */
12268 if (st
->update_tail
)
12269 append_metadata_update(st
, u
, len
);
12280 exit_imsm_reshape_super
:
12281 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
12285 #define COMPLETED_OK 0
12286 #define COMPLETED_NONE 1
12287 #define COMPLETED_DELAYED 2
12289 static int read_completed(int fd
, unsigned long long *val
)
12292 char buf
[SYSFS_MAX_BUF_SIZE
];
12294 ret
= sysfs_fd_get_str(fd
, buf
, sizeof(buf
));
12298 ret
= COMPLETED_OK
;
12299 if (str_is_none(buf
) == true) {
12300 ret
= COMPLETED_NONE
;
12301 } else if (strncmp(buf
, "delayed", 7) == 0) {
12302 ret
= COMPLETED_DELAYED
;
12305 *val
= strtoull(buf
, &ep
, 0);
12306 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
12312 /*******************************************************************************
12313 * Function: wait_for_reshape_imsm
12314 * Description: Function writes new sync_max value and waits until
12315 * reshape process reach new position
12317 * sra : general array info
12318 * ndata : number of disks in new array's layout
12321 * 1 : there is no reshape in progress,
12323 ******************************************************************************/
12324 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
12326 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
12328 unsigned long long completed
;
12329 /* to_complete : new sync_max position */
12330 unsigned long long to_complete
= sra
->reshape_progress
;
12331 unsigned long long position_to_set
= to_complete
/ ndata
;
12333 if (!is_fd_valid(fd
)) {
12334 dprintf("cannot open reshape_position\n");
12339 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
12341 dprintf("cannot read reshape_position (no reshape in progres)\n");
12345 sleep_for(0, MSEC_TO_NSEC(30), true);
12350 if (completed
> position_to_set
) {
12351 dprintf("wrong next position to set %llu (%llu)\n",
12352 to_complete
, position_to_set
);
12356 dprintf("Position set: %llu\n", position_to_set
);
12357 if (sysfs_set_num(sra
, NULL
, "sync_max",
12358 position_to_set
) != 0) {
12359 dprintf("cannot set reshape position to %llu\n",
12367 char action
[SYSFS_MAX_BUF_SIZE
];
12368 int timeout
= 3000;
12370 sysfs_wait(fd
, &timeout
);
12371 if (sysfs_get_str(sra
, NULL
, "sync_action",
12372 action
, sizeof(action
)) > 0 &&
12373 strncmp(action
, "reshape", 7) != 0) {
12374 if (strncmp(action
, "idle", 4) == 0)
12380 rc
= read_completed(fd
, &completed
);
12382 dprintf("cannot read reshape_position (in loop)\n");
12385 } else if (rc
== COMPLETED_NONE
)
12387 } while (completed
< position_to_set
);
12393 /*******************************************************************************
12394 * Function: check_degradation_change
12395 * Description: Check that array hasn't become failed.
12397 * info : for sysfs access
12398 * sources : source disks descriptors
12399 * degraded: previous degradation level
12401 * degradation level
12402 ******************************************************************************/
12403 int check_degradation_change(struct mdinfo
*info
,
12407 unsigned long long new_degraded
;
12410 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
12411 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
12412 /* check each device to ensure it is still working */
12415 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
12416 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
12418 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
12419 char sbuf
[SYSFS_MAX_BUF_SIZE
];
12420 int raid_disk
= sd
->disk
.raid_disk
;
12422 if (sysfs_get_str(info
,
12423 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
12424 strstr(sbuf
, "faulty") ||
12425 strstr(sbuf
, "in_sync") == NULL
) {
12426 /* this device is dead */
12427 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
12428 if (raid_disk
>= 0)
12429 close_fd(&sources
[raid_disk
]);
12436 return new_degraded
;
12439 /*******************************************************************************
12440 * Function: imsm_manage_reshape
12441 * Description: Function finds array under reshape and it manages reshape
12442 * process. It creates stripes backups (if required) and sets
12445 * afd : Backup handle (nattive) - not used
12446 * sra : general array info
12447 * reshape : reshape parameters - not used
12448 * st : supertype structure
12449 * blocks : size of critical section [blocks]
12450 * fds : table of source device descriptor
12451 * offsets : start of array (offest per devices)
12453 * destfd : table of destination device descriptor
12454 * destoffsets : table of destination offsets (per device)
12456 * 1 : success, reshape is done
12458 ******************************************************************************/
12459 static int imsm_manage_reshape(
12460 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
12461 struct supertype
*st
, unsigned long backup_blocks
,
12462 int *fds
, unsigned long long *offsets
,
12463 int dests
, int *destfd
, unsigned long long *destoffsets
)
12466 struct intel_super
*super
= st
->sb
;
12467 struct intel_dev
*dv
;
12468 unsigned int sector_size
= super
->sector_size
;
12469 struct imsm_dev
*dev
= NULL
;
12470 struct imsm_map
*map_src
, *map_dest
;
12471 int migr_vol_qan
= 0;
12472 int ndata
, odata
; /* [bytes] */
12473 int chunk
; /* [bytes] */
12474 struct migr_record
*migr_rec
;
12476 unsigned int buf_size
; /* [bytes] */
12477 unsigned long long max_position
; /* array size [bytes] */
12478 unsigned long long next_step
; /* [blocks]/[bytes] */
12479 unsigned long long old_data_stripe_length
;
12480 unsigned long long start_src
; /* [bytes] */
12481 unsigned long long start
; /* [bytes] */
12482 unsigned long long start_buf_shift
; /* [bytes] */
12484 int source_layout
= 0;
12485 int subarray_index
= -1;
12490 if (!fds
|| !offsets
)
12493 /* Find volume during the reshape */
12494 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
12495 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12496 dv
->dev
->vol
.migr_state
== 1) {
12499 subarray_index
= dv
->index
;
12502 /* Only one volume can migrate at the same time */
12503 if (migr_vol_qan
!= 1) {
12504 pr_err("%s", migr_vol_qan
?
12505 "Number of migrating volumes greater than 1\n" :
12506 "There is no volume during migrationg\n");
12510 map_dest
= get_imsm_map(dev
, MAP_0
);
12511 map_src
= get_imsm_map(dev
, MAP_1
);
12512 if (map_src
== NULL
)
12515 ndata
= imsm_num_data_members(map_dest
);
12516 odata
= imsm_num_data_members(map_src
);
12518 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12519 old_data_stripe_length
= odata
* chunk
;
12521 migr_rec
= super
->migr_rec
;
12523 /* initialize migration record for start condition */
12524 if (sra
->reshape_progress
== 0)
12525 init_migr_record_imsm(st
, dev
, sra
);
12527 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12528 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12531 /* Save checkpoint to update migration record for current
12532 * reshape position (in md). It can be farther than current
12533 * reshape position in metadata.
12535 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12536 /* ignore error == 2, this can mean end of reshape here
12538 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12543 /* size for data */
12544 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12545 /* extend buffer size for parity disk */
12546 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12547 /* add space for stripe alignment */
12548 buf_size
+= old_data_stripe_length
;
12549 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12550 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12554 max_position
= sra
->component_size
* ndata
;
12555 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12557 while (current_migr_unit(migr_rec
) <
12558 get_num_migr_units(migr_rec
)) {
12559 /* current reshape position [blocks] */
12560 unsigned long long current_position
=
12561 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12562 * current_migr_unit(migr_rec
);
12563 unsigned long long border
;
12565 /* Check that array hasn't become failed.
12567 degraded
= check_degradation_change(sra
, fds
, degraded
);
12568 if (degraded
> 1) {
12569 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12573 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12575 if ((current_position
+ next_step
) > max_position
)
12576 next_step
= max_position
- current_position
;
12578 start
= current_position
* 512;
12580 /* align reading start to old geometry */
12581 start_buf_shift
= start
% old_data_stripe_length
;
12582 start_src
= start
- start_buf_shift
;
12584 border
= (start_src
/ odata
) - (start
/ ndata
);
12586 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12587 /* save critical stripes to buf
12588 * start - start address of current unit
12589 * to backup [bytes]
12590 * start_src - start address of current unit
12591 * to backup alligned to source array
12594 unsigned long long next_step_filler
;
12595 unsigned long long copy_length
= next_step
* 512;
12597 /* allign copy area length to stripe in old geometry */
12598 next_step_filler
= ((copy_length
+ start_buf_shift
)
12599 % old_data_stripe_length
);
12600 if (next_step_filler
)
12601 next_step_filler
= (old_data_stripe_length
12602 - next_step_filler
);
12603 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12604 start
, start_src
, copy_length
,
12605 start_buf_shift
, next_step_filler
);
12607 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12608 chunk
, map_src
->raid_level
,
12609 source_layout
, 0, NULL
, start_src
,
12611 next_step_filler
+ start_buf_shift
,
12613 dprintf("imsm: Cannot save stripes to buffer\n");
12616 /* Convert data to destination format and store it
12617 * in backup general migration area
12619 if (save_backup_imsm(st
, dev
, sra
,
12620 buf
+ start_buf_shift
, copy_length
)) {
12621 dprintf("imsm: Cannot save stripes to target devices\n");
12624 if (save_checkpoint_imsm(st
, sra
,
12625 UNIT_SRC_IN_CP_AREA
)) {
12626 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12630 /* set next step to use whole border area */
12631 border
/= next_step
;
12633 next_step
*= border
;
12635 /* When data backed up, checkpoint stored,
12636 * kick the kernel to reshape unit of data
12638 next_step
= next_step
+ sra
->reshape_progress
;
12639 /* limit next step to array max position */
12640 if (next_step
> max_position
)
12641 next_step
= max_position
;
12642 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12643 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12644 sra
->reshape_progress
= next_step
;
12646 /* wait until reshape finish */
12647 if (wait_for_reshape_imsm(sra
, ndata
)) {
12648 dprintf("wait_for_reshape_imsm returned error!\n");
12654 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12655 /* ignore error == 2, this can mean end of reshape here
12657 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12663 /* clear migr_rec on disks after successful migration */
12666 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12667 for (d
= super
->disks
; d
; d
= d
->next
) {
12668 if (d
->index
< 0 || is_failed(&d
->disk
))
12670 unsigned long long dsize
;
12672 get_dev_size(d
->fd
, NULL
, &dsize
);
12673 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12675 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12676 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12677 MIGR_REC_BUF_SECTORS
*sector_size
)
12678 perror("Write migr_rec failed");
12682 /* return '1' if done */
12685 /* After the reshape eliminate size mismatch in metadata.
12686 * Don't update md/component_size here, volume hasn't
12687 * to take whole space. It is allowed by kernel.
12688 * md/component_size will be set propoperly after next assembly.
12690 imsm_fix_size_mismatch(st
, subarray_index
);
12694 /* See Grow.c: abort_reshape() for further explanation */
12695 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12696 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12697 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12702 /*******************************************************************************
12703 * Function: calculate_bitmap_min_chunksize
12704 * Description: Calculates the minimal valid bitmap chunk size
12706 * max_bits : indicate how many bits can be used for the bitmap
12707 * data_area_size : the size of the data area covered by the bitmap
12710 * The bitmap chunk size
12711 ******************************************************************************/
12712 static unsigned long long
12713 calculate_bitmap_min_chunksize(unsigned long long max_bits
,
12714 unsigned long long data_area_size
)
12716 unsigned long long min_chunk
=
12717 4096; /* sub-page chunks don't work yet.. */
12718 unsigned long long bits
= data_area_size
/ min_chunk
+ 1;
12720 while (bits
> max_bits
) {
12722 bits
= (bits
+ 1) / 2;
12727 /*******************************************************************************
12728 * Function: calculate_bitmap_chunksize
12729 * Description: Calculates the bitmap chunk size for the given device
12731 * st : supertype information
12732 * dev : device for the bitmap
12735 * The bitmap chunk size
12736 ******************************************************************************/
12737 static unsigned long long calculate_bitmap_chunksize(struct supertype
*st
,
12738 struct imsm_dev
*dev
)
12740 struct intel_super
*super
= st
->sb
;
12741 unsigned long long min_chunksize
;
12742 unsigned long long result
= IMSM_DEFAULT_BITMAP_CHUNKSIZE
;
12743 size_t dev_size
= imsm_dev_size(dev
);
12745 min_chunksize
= calculate_bitmap_min_chunksize(
12746 IMSM_BITMAP_AREA_SIZE
* super
->sector_size
, dev_size
);
12748 if (result
< min_chunksize
)
12749 result
= min_chunksize
;
12754 /*******************************************************************************
12755 * Function: init_bitmap_header
12756 * Description: Initialize the bitmap header structure
12758 * st : supertype information
12759 * bms : bitmap header struct to initialize
12760 * dev : device for the bitmap
12765 ******************************************************************************/
12766 static int init_bitmap_header(struct supertype
*st
, struct bitmap_super_s
*bms
,
12767 struct imsm_dev
*dev
)
12774 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
12775 bms
->version
= __cpu_to_le32(BITMAP_MAJOR_HI
);
12776 bms
->daemon_sleep
= __cpu_to_le32(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
);
12777 bms
->sync_size
= __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
);
12778 bms
->write_behind
= __cpu_to_le32(0);
12780 uuid_from_super_imsm(st
, vol_uuid
);
12781 memcpy(bms
->uuid
, vol_uuid
, 16);
12783 bms
->chunksize
= calculate_bitmap_chunksize(st
, dev
);
12788 /*******************************************************************************
12789 * Function: validate_internal_bitmap_for_drive
12790 * Description: Verify if the bitmap header for a given drive.
12792 * st : supertype information
12793 * offset : The offset from the beginning of the drive where to look for
12794 * the bitmap header.
12795 * d : the drive info
12800 ******************************************************************************/
12801 static int validate_internal_bitmap_for_drive(struct supertype
*st
,
12802 unsigned long long offset
,
12805 struct intel_super
*super
= st
->sb
;
12808 bitmap_super_t
*bms
;
12816 if (posix_memalign(&read_buf
, MAX_SECTOR_SIZE
, IMSM_BITMAP_HEADER_SIZE
))
12820 if (!is_fd_valid(fd
)) {
12821 fd
= open(d
->devname
, O_RDONLY
, 0);
12823 if (!is_fd_valid(fd
)) {
12824 dprintf("cannot open the device %s\n", d
->devname
);
12829 if (lseek64(fd
, offset
* super
->sector_size
, SEEK_SET
) < 0)
12831 if (read(fd
, read_buf
, IMSM_BITMAP_HEADER_SIZE
) !=
12832 IMSM_BITMAP_HEADER_SIZE
)
12835 uuid_from_super_imsm(st
, vol_uuid
);
12838 if ((bms
->magic
!= __cpu_to_le32(BITMAP_MAGIC
)) ||
12839 (bms
->version
!= __cpu_to_le32(BITMAP_MAJOR_HI
)) ||
12840 (!same_uuid((int *)bms
->uuid
, vol_uuid
, st
->ss
->swapuuid
))) {
12841 dprintf("wrong bitmap header detected\n");
12847 if (!is_fd_valid(d
->fd
))
12856 /*******************************************************************************
12857 * Function: validate_internal_bitmap_imsm
12858 * Description: Verify if the bitmap header is in place and with proper data.
12860 * st : supertype information
12863 * 0 : success or device w/o RWH_BITMAP
12865 ******************************************************************************/
12866 static int validate_internal_bitmap_imsm(struct supertype
*st
)
12868 struct intel_super
*super
= st
->sb
;
12869 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
12870 unsigned long long offset
;
12873 if (dev
->rwh_policy
!= RWH_BITMAP
)
12876 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12877 for (d
= super
->disks
; d
; d
= d
->next
) {
12878 if (d
->index
< 0 || is_failed(&d
->disk
))
12881 if (validate_internal_bitmap_for_drive(st
, offset
, d
)) {
12882 pr_err("imsm: bitmap validation failed\n");
12889 /*******************************************************************************
12890 * Function: add_internal_bitmap_imsm
12891 * Description: Mark the volume to use the bitmap and updates the chunk size value.
12893 * st : supertype information
12894 * chunkp : bitmap chunk size
12895 * delay : not used for imsm
12896 * write_behind : not used for imsm
12897 * size : not used for imsm
12898 * may_change : not used for imsm
12899 * amajor : not used for imsm
12904 ******************************************************************************/
12905 static int add_internal_bitmap_imsm(struct supertype
*st
, int *chunkp
,
12906 int delay
, int write_behind
,
12907 unsigned long long size
, int may_change
,
12910 struct intel_super
*super
= st
->sb
;
12911 int vol_idx
= super
->current_vol
;
12912 struct imsm_dev
*dev
;
12914 if (!super
->devlist
|| vol_idx
== -1 || !chunkp
)
12917 dev
= get_imsm_dev(super
, vol_idx
);
12918 dev
->rwh_policy
= RWH_BITMAP
;
12919 *chunkp
= calculate_bitmap_chunksize(st
, dev
);
12923 /*******************************************************************************
12924 * Function: locate_bitmap_imsm
12925 * Description: Seek 'fd' to start of write-intent-bitmap.
12927 * st : supertype information
12928 * fd : file descriptor for the device
12929 * node_num : not used for imsm
12934 ******************************************************************************/
12935 static int locate_bitmap_imsm(struct supertype
*st
, int fd
, int node_num
)
12937 struct intel_super
*super
= st
->sb
;
12938 unsigned long long offset
;
12939 int vol_idx
= super
->current_vol
;
12941 if (!super
->devlist
|| vol_idx
== -1)
12944 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12945 dprintf("bitmap header offset is %llu\n", offset
);
12947 lseek64(fd
, offset
<< 9, 0);
12952 /*******************************************************************************
12953 * Function: write_init_bitmap_imsm
12954 * Description: Write a bitmap header and prepares the area for the bitmap.
12956 * st : supertype information
12957 * fd : file descriptor for the device
12958 * update : not used for imsm
12963 ******************************************************************************/
12964 static int write_init_bitmap_imsm(struct supertype
*st
, int fd
,
12965 enum bitmap_update update
)
12967 struct intel_super
*super
= st
->sb
;
12968 int vol_idx
= super
->current_vol
;
12970 unsigned long long offset
;
12971 bitmap_super_t bms
= { 0 };
12972 size_t written
= 0;
12977 if (!super
->devlist
|| !super
->sector_size
|| vol_idx
== -1)
12980 struct imsm_dev
*dev
= get_imsm_dev(super
, vol_idx
);
12982 /* first clear the space for bitmap header */
12983 unsigned long long bitmap_area_start
=
12984 get_bitmap_header_sector(super
, vol_idx
);
12986 dprintf("zeroing area start (%llu) and size (%u)\n", bitmap_area_start
,
12987 IMSM_BITMAP_AND_HEADER_SIZE
/ super
->sector_size
);
12988 if (zero_disk_range(fd
, bitmap_area_start
,
12989 IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
)) {
12990 pr_err("imsm: cannot zeroing the space for the bitmap\n");
12994 /* The bitmap area should be filled with "1"s to perform initial
12997 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
))
12999 memset(buf
, 0xFF, MAX_SECTOR_SIZE
);
13000 offset
= get_bitmap_sector(super
, vol_idx
);
13001 lseek64(fd
, offset
<< 9, 0);
13002 while (written
< IMSM_BITMAP_AREA_SIZE
) {
13003 to_write
= IMSM_BITMAP_AREA_SIZE
- written
;
13004 if (to_write
> MAX_SECTOR_SIZE
)
13005 to_write
= MAX_SECTOR_SIZE
;
13006 rv_num
= write(fd
, buf
, MAX_SECTOR_SIZE
);
13007 if (rv_num
!= MAX_SECTOR_SIZE
) {
13009 dprintf("cannot initialize bitmap area\n");
13015 /* write a bitmap header */
13016 init_bitmap_header(st
, &bms
, dev
);
13017 memset(buf
, 0, MAX_SECTOR_SIZE
);
13018 memcpy(buf
, &bms
, sizeof(bitmap_super_t
));
13019 if (locate_bitmap_imsm(st
, fd
, 0)) {
13021 dprintf("cannot locate the bitmap\n");
13024 if (write(fd
, buf
, MAX_SECTOR_SIZE
) != MAX_SECTOR_SIZE
) {
13026 dprintf("cannot write the bitmap header\n");
13037 /*******************************************************************************
13038 * Function: is_vol_to_setup_bitmap
13039 * Description: Checks if a bitmap should be activated on the dev.
13041 * info : info about the volume to setup the bitmap
13042 * dev : the device to check against bitmap creation
13045 * 0 : bitmap should be set up on the device
13047 ******************************************************************************/
13048 static int is_vol_to_setup_bitmap(struct mdinfo
*info
, struct imsm_dev
*dev
)
13053 if ((strcmp((char *)dev
->volume
, info
->name
) == 0) &&
13054 (dev
->rwh_policy
== RWH_BITMAP
))
13060 /*******************************************************************************
13061 * Function: set_bitmap_sysfs
13062 * Description: Set the sysfs atributes of a given volume to activate the bitmap.
13064 * info : info about the volume where the bitmap should be setup
13065 * chunksize : bitmap chunk size
13066 * location : location of the bitmap
13071 ******************************************************************************/
13072 static int set_bitmap_sysfs(struct mdinfo
*info
, unsigned long long chunksize
,
13075 /* The bitmap/metadata is set to external to allow changing of value for
13076 * bitmap/location. When external is used, the kernel will treat an offset
13077 * related to the device's first lba (in opposition to the "internal" case
13078 * when this value is related to the beginning of the superblock).
13080 if (sysfs_set_str(info
, NULL
, "bitmap/metadata", "external")) {
13081 dprintf("failed to set bitmap/metadata\n");
13085 /* It can only be changed when no bitmap is active.
13086 * Should be bigger than 512 and must be power of 2.
13087 * It is expecting the value in bytes.
13089 if (sysfs_set_num(info
, NULL
, "bitmap/chunksize",
13090 __cpu_to_le32(chunksize
))) {
13091 dprintf("failed to set bitmap/chunksize\n");
13095 /* It is expecting the value in sectors. */
13096 if (sysfs_set_num(info
, NULL
, "bitmap/space",
13097 __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
))) {
13098 dprintf("failed to set bitmap/space\n");
13102 /* Determines the delay between the bitmap updates.
13103 * It is expecting the value in seconds.
13105 if (sysfs_set_num(info
, NULL
, "bitmap/time_base",
13106 __cpu_to_le64(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
))) {
13107 dprintf("failed to set bitmap/time_base\n");
13111 /* It is expecting the value in sectors with a sign at the beginning. */
13112 if (sysfs_set_str(info
, NULL
, "bitmap/location", location
)) {
13113 dprintf("failed to set bitmap/location\n");
13120 /*******************************************************************************
13121 * Function: set_bitmap_imsm
13122 * Description: Setup the bitmap for the given volume
13124 * st : supertype information
13125 * info : info about the volume where the bitmap should be setup
13130 ******************************************************************************/
13131 static int set_bitmap_imsm(struct supertype
*st
, struct mdinfo
*info
)
13133 struct intel_super
*super
= st
->sb
;
13134 int prev_current_vol
= super
->current_vol
;
13135 struct imsm_dev
*dev
;
13137 char location
[16] = "";
13138 unsigned long long chunksize
;
13139 struct intel_dev
*dev_it
;
13141 for (dev_it
= super
->devlist
; dev_it
; dev_it
= dev_it
->next
) {
13142 super
->current_vol
= dev_it
->index
;
13143 dev
= get_imsm_dev(super
, super
->current_vol
);
13145 if (is_vol_to_setup_bitmap(info
, dev
)) {
13146 if (validate_internal_bitmap_imsm(st
)) {
13147 dprintf("bitmap header validation failed\n");
13151 chunksize
= calculate_bitmap_chunksize(st
, dev
);
13152 dprintf("chunk size is %llu\n", chunksize
);
13154 snprintf(location
, sizeof(location
), "+%llu",
13155 get_bitmap_sector(super
, super
->current_vol
));
13156 dprintf("bitmap offset is %s\n", location
);
13158 if (set_bitmap_sysfs(info
, chunksize
, location
)) {
13159 dprintf("cannot setup the bitmap\n");
13166 super
->current_vol
= prev_current_vol
;
13170 struct superswitch super_imsm
= {
13171 .examine_super
= examine_super_imsm
,
13172 .brief_examine_super
= brief_examine_super_imsm
,
13173 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
13174 .export_examine_super
= export_examine_super_imsm
,
13175 .detail_super
= detail_super_imsm
,
13176 .brief_detail_super
= brief_detail_super_imsm
,
13177 .write_init_super
= write_init_super_imsm
,
13178 .validate_geometry
= validate_geometry_imsm
,
13179 .add_to_super
= add_to_super_imsm
,
13180 .remove_from_super
= remove_from_super_imsm
,
13181 .detail_platform
= detail_platform_imsm
,
13182 .export_detail_platform
= export_detail_platform_imsm
,
13183 .kill_subarray
= kill_subarray_imsm
,
13184 .update_subarray
= update_subarray_imsm
,
13185 .load_container
= load_container_imsm
,
13186 .default_geometry
= default_geometry_imsm
,
13187 .test_and_add_drive_policies
= test_and_add_drive_policies_imsm
,
13188 .reshape_super
= imsm_reshape_super
,
13189 .manage_reshape
= imsm_manage_reshape
,
13190 .recover_backup
= recover_backup_imsm
,
13191 .examine_badblocks
= examine_badblocks_imsm
,
13192 .match_home
= match_home_imsm
,
13193 .uuid_from_super
= uuid_from_super_imsm
,
13194 .getinfo_super
= getinfo_super_imsm
,
13195 .getinfo_super_disks
= getinfo_super_disks_imsm
,
13196 .update_super
= update_super_imsm
,
13198 .avail_size
= avail_size_imsm
,
13199 .get_spare_criteria
= get_spare_criteria_imsm
,
13201 .compare_super
= compare_super_imsm
,
13203 .load_super
= load_super_imsm
,
13204 .init_super
= init_super_imsm
,
13205 .store_super
= store_super_imsm
,
13206 .free_super
= free_super_imsm
,
13207 .match_metadata_desc
= match_metadata_desc_imsm
,
13208 .container_content
= container_content_imsm
,
13209 .validate_container
= validate_container_imsm
,
13211 .add_internal_bitmap
= add_internal_bitmap_imsm
,
13212 .locate_bitmap
= locate_bitmap_imsm
,
13213 .write_bitmap
= write_init_bitmap_imsm
,
13214 .set_bitmap
= set_bitmap_imsm
,
13216 .write_init_ppl
= write_init_ppl_imsm
,
13217 .validate_ppl
= validate_ppl_imsm
,
13224 .open_new
= imsm_open_new
,
13225 .set_array_state
= imsm_set_array_state
,
13226 .set_disk
= imsm_set_disk
,
13227 .sync_metadata
= imsm_sync_metadata
,
13228 .activate_spare
= imsm_activate_spare
,
13229 .process_update
= imsm_process_update
,
13230 .prepare_update
= imsm_prepare_update
,
13231 .record_bad_block
= imsm_record_badblock
,
13232 .clear_bad_block
= imsm_clear_badblock
,
13233 .get_bad_blocks
= imsm_get_badblocks
,