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"
31 /* MPB == Metadata Parameter Block */
32 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
33 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
34 #define MPB_VERSION_RAID0 "1.0.00"
35 #define MPB_VERSION_RAID1 "1.1.00"
36 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
37 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
38 #define MPB_VERSION_RAID5 "1.2.02"
39 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
40 #define MPB_VERSION_CNG "1.2.06"
41 #define MPB_VERSION_ATTRIBS "1.3.00"
42 #define MAX_SIGNATURE_LENGTH 32
43 #define MAX_RAID_SERIAL_LEN 16
46 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
50 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
52 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
54 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
55 /* supports RAID CNG */
56 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
57 /* supports expanded stripe sizes of 256K, 512K and 1MB */
58 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
60 /* The OROM Support RST Caching of Volumes */
61 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
62 /* The OROM supports creating disks greater than 2TB */
63 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
64 /* The OROM supports Bad Block Management */
65 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
67 /* THe OROM Supports NVM Caching of Volumes */
68 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
69 /* The OROM supports creating volumes greater than 2TB */
70 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
71 /* originally for PMP, now it's wasted b/c. Never use this bit! */
72 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
73 /* Verify MPB contents against checksum after reading MPB */
74 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
76 /* Define all supported attributes that have to be accepted by mdadm
78 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
80 MPB_ATTRIB_2TB_DISK | \
85 MPB_ATTRIB_EXP_STRIPE_SIZE | \
88 /* Define attributes that are unused but not harmful */
89 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
91 #define MPB_SECTOR_CNT 2210
92 #define IMSM_RESERVED_SECTORS 8192
93 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2048
94 #define SECT_PER_MB_SHIFT 11
95 #define MAX_SECTOR_SIZE 4096
96 #define MULTIPLE_PPL_AREA_SIZE_IMSM (1024 * 1024) /* Size of the whole
101 * Internal Write-intent bitmap is stored in the same area where PPL.
102 * Both features are mutually exclusive, so it is not an issue.
103 * The first 8KiB of the area are reserved and shall not be used.
105 #define IMSM_BITMAP_AREA_RESERVED_SIZE 8192
107 #define IMSM_BITMAP_HEADER_OFFSET (IMSM_BITMAP_AREA_RESERVED_SIZE)
108 #define IMSM_BITMAP_HEADER_SIZE MAX_SECTOR_SIZE
110 #define IMSM_BITMAP_START_OFFSET (IMSM_BITMAP_HEADER_OFFSET + IMSM_BITMAP_HEADER_SIZE)
111 #define IMSM_BITMAP_AREA_SIZE (MULTIPLE_PPL_AREA_SIZE_IMSM - IMSM_BITMAP_START_OFFSET)
112 #define IMSM_BITMAP_AND_HEADER_SIZE (IMSM_BITMAP_AREA_SIZE + IMSM_BITMAP_HEADER_SIZE)
114 #define IMSM_DEFAULT_BITMAP_CHUNKSIZE (64 * 1024 * 1024)
115 #define IMSM_DEFAULT_BITMAP_DAEMON_SLEEP 5
118 * This macro let's us ensure that no-one accidentally
119 * changes the size of a struct
121 #define ASSERT_SIZE(_struct, size) \
122 static inline void __assert_size_##_struct(void) \
126 case (sizeof(struct _struct) == size): break; \
130 /* Disk configuration info. */
131 #define IMSM_MAX_DEVICES 255
133 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
134 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
135 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
136 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
137 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
138 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
139 #define JOURNAL_DISK __cpu_to_le32(0x2000000) /* Device marked as Journaling Drive */
140 __u32 status
; /* 0xF0 - 0xF3 */
141 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
142 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
143 #define IMSM_DISK_FILLERS 3
144 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
146 ASSERT_SIZE(imsm_disk
, 48)
148 /* map selector for map managment
154 /* RAID map configuration infos. */
156 __u32 pba_of_lba0_lo
; /* start address of partition */
157 __u32 blocks_per_member_lo
;/* blocks per member */
158 __u32 num_data_stripes_lo
; /* number of data stripes */
159 __u16 blocks_per_strip
;
160 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
161 #define IMSM_T_STATE_NORMAL 0
162 #define IMSM_T_STATE_UNINITIALIZED 1
163 #define IMSM_T_STATE_DEGRADED 2
164 #define IMSM_T_STATE_FAILED 3
166 #define IMSM_T_RAID0 0
167 #define IMSM_T_RAID1 1
168 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
169 __u8 num_members
; /* number of member disks */
170 __u8 num_domains
; /* number of parity domains */
171 __u8 failed_disk_num
; /* valid only when state is degraded */
173 __u32 pba_of_lba0_hi
;
174 __u32 blocks_per_member_hi
;
175 __u32 num_data_stripes_hi
;
176 __u32 filler
[4]; /* expansion area */
177 #define IMSM_ORD_REBUILD (1 << 24)
178 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
179 * top byte contains some flags
182 ASSERT_SIZE(imsm_map
, 52)
185 __u32 curr_migr_unit_lo
;
186 __u32 checkpoint_id
; /* id to access curr_migr_unit */
187 __u8 migr_state
; /* Normal or Migrating */
189 #define MIGR_REBUILD 1
190 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
191 #define MIGR_GEN_MIGR 3
192 #define MIGR_STATE_CHANGE 4
193 #define MIGR_REPAIR 5
194 __u8 migr_type
; /* Initializing, Rebuilding, ... */
195 #define RAIDVOL_CLEAN 0
196 #define RAIDVOL_DIRTY 1
197 #define RAIDVOL_DSRECORD_VALID 2
199 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
200 __u16 verify_errors
; /* number of mismatches */
201 __u16 bad_blocks
; /* number of bad blocks during verify */
202 __u32 curr_migr_unit_hi
;
204 struct imsm_map map
[1];
205 /* here comes another one if migr_state */
207 ASSERT_SIZE(imsm_vol
, 84)
210 __u8 volume
[MAX_RAID_SERIAL_LEN
];
213 #define DEV_BOOTABLE __cpu_to_le32(0x01)
214 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
215 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
216 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
217 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
218 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
219 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
220 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
221 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
222 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
223 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
224 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
225 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
226 __u32 status
; /* Persistent RaidDev status */
227 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
231 __u8 cng_master_disk
;
235 __u16 my_vol_raid_dev_num
; /* Used in Unique volume Id for this RaidDev */
241 /* Unique Volume Id of the NvCache Volume associated with this volume */
242 __u32 nvc_vol_orig_family_num
;
243 __u16 nvc_vol_raid_dev_num
;
246 #define RWH_DISTRIBUTED 1
247 #define RWH_JOURNALING_DRIVE 2
248 #define RWH_MULTIPLE_DISTRIBUTED 3
249 #define RWH_MULTIPLE_PPLS_JOURNALING_DRIVE 4
250 #define RWH_MULTIPLE_OFF 5
252 __u8 rwh_policy
; /* Raid Write Hole Policy */
253 __u8 jd_serial
[MAX_RAID_SERIAL_LEN
]; /* Journal Drive serial number */
256 #define IMSM_DEV_FILLERS 3
257 __u32 filler
[IMSM_DEV_FILLERS
];
260 ASSERT_SIZE(imsm_dev
, 164)
263 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
264 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
265 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
266 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
267 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
268 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
269 __u32 attributes
; /* 0x34 - 0x37 */
270 __u8 num_disks
; /* 0x38 Number of configured disks */
271 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
272 __u8 error_log_pos
; /* 0x3A */
273 __u8 fill
[1]; /* 0x3B */
274 __u32 cache_size
; /* 0x3c - 0x40 in mb */
275 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
276 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
277 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
278 __u16 num_raid_devs_created
; /* 0x4C - 0x4D Used for generating unique
279 * volume IDs for raid_dev created in this array
282 __u16 filler1
; /* 0x4E - 0x4F */
283 __u64 creation_time
; /* 0x50 - 0x57 Array creation time */
284 #define IMSM_FILLERS 32
285 __u32 filler
[IMSM_FILLERS
]; /* 0x58 - 0xD7 RAID_MPB_FILLERS */
286 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
287 /* here comes imsm_dev[num_raid_devs] */
288 /* here comes BBM logs */
290 ASSERT_SIZE(imsm_super
, 264)
292 #define BBM_LOG_MAX_ENTRIES 254
293 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
294 #define BBM_LOG_SIGNATURE 0xabadb10c
296 struct bbm_log_block_addr
{
299 } __attribute__ ((__packed__
));
301 struct bbm_log_entry
{
302 __u8 marked_count
; /* Number of blocks marked - 1 */
303 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
304 struct bbm_log_block_addr defective_block_start
;
305 } __attribute__ ((__packed__
));
308 __u32 signature
; /* 0xABADB10C */
310 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
312 ASSERT_SIZE(bbm_log
, 2040)
314 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
316 #define BLOCKS_PER_KB (1024/512)
318 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
320 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
322 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
323 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
324 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
327 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
328 * be recovered using srcMap */
329 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
330 * already been migrated and must
331 * be recovered from checkpoint area */
333 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of single PPL, without the header */
336 __u32 rec_status
; /* Status used to determine how to restart
337 * migration in case it aborts
339 __u32 curr_migr_unit_lo
; /* 0..numMigrUnits-1 */
340 __u32 family_num
; /* Family number of MPB
341 * containing the RaidDev
342 * that is migrating */
343 __u32 ascending_migr
; /* True if migrating in increasing
345 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
346 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
348 * advances per unit-of-operation */
349 __u32 ckpt_area_pba_lo
; /* Pba of first block of ckpt copy area */
350 __u32 dest_1st_member_lba_lo
; /* First member lba on first
351 * stripe of destination */
352 __u32 num_migr_units_lo
; /* Total num migration units-of-op */
353 __u32 post_migr_vol_cap
; /* Size of volume after
354 * migration completes */
355 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
356 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
357 * migration ckpt record was read from
358 * (for recovered migrations) */
359 __u32 curr_migr_unit_hi
; /* 0..numMigrUnits-1 high order 32 bits */
360 __u32 ckpt_area_pba_hi
; /* Pba of first block of ckpt copy area
361 * high order 32 bits */
362 __u32 dest_1st_member_lba_hi
; /* First member lba on first stripe of
363 * destination - high order 32 bits */
364 __u32 num_migr_units_hi
; /* Total num migration units-of-op
365 * high order 32 bits */
368 ASSERT_SIZE(migr_record
, 128)
371 * enum imsm_status - internal IMSM return values representation.
372 * @STATUS_OK: function succeeded.
373 * @STATUS_ERROR: General error ocurred (not specified).
375 * Typedefed to imsm_status_t.
377 typedef enum imsm_status
{
378 IMSM_STATUS_ERROR
= -1,
385 * 2: metadata does not match
393 struct md_list
*next
;
396 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
398 static __u8
migr_type(struct imsm_dev
*dev
)
400 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
401 dev
->status
& DEV_VERIFY_AND_FIX
)
404 return dev
->vol
.migr_type
;
407 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
409 /* for compatibility with older oroms convert MIGR_REPAIR, into
410 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
412 if (migr_type
== MIGR_REPAIR
) {
413 dev
->vol
.migr_type
= MIGR_VERIFY
;
414 dev
->status
|= DEV_VERIFY_AND_FIX
;
416 dev
->vol
.migr_type
= migr_type
;
417 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
421 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
423 return ROUND_UP(bytes
, sector_size
) / sector_size
;
426 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
427 unsigned int sector_size
)
429 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
433 struct imsm_dev
*dev
;
434 struct intel_dev
*next
;
439 enum sys_dev_type type
;
442 struct intel_hba
*next
;
449 /* internal representation of IMSM metadata */
452 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
453 struct imsm_super
*anchor
; /* immovable parameters */
456 void *migr_rec_buf
; /* buffer for I/O operations */
457 struct migr_record
*migr_rec
; /* migration record */
459 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
460 array, it indicates that mdmon is allowed to clean migration
462 size_t len
; /* size of the 'buf' allocation */
463 size_t extra_space
; /* extra space in 'buf' that is not used yet */
464 void *next_buf
; /* for realloc'ing buf from the manager */
466 int updates_pending
; /* count of pending updates for mdmon */
467 int current_vol
; /* index of raid device undergoing creation */
468 unsigned long long create_offset
; /* common start for 'current_vol' */
469 __u32 random
; /* random data for seeding new family numbers */
470 struct intel_dev
*devlist
;
471 unsigned int sector_size
; /* sector size of used member drives */
475 __u8 serial
[MAX_RAID_SERIAL_LEN
];
478 struct imsm_disk disk
;
481 struct extent
*e
; /* for determining freespace @ create */
482 int raiddisk
; /* slot to fill in autolayout */
484 } *disks
, *current_disk
;
485 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
487 struct dl
*missing
; /* disks removed while we weren't looking */
488 struct bbm_log
*bbm_log
;
489 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
490 const struct imsm_orom
*orom
; /* platform firmware support */
491 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
492 struct md_bb bb
; /* memory for get_bad_blocks call */
496 struct imsm_disk disk
;
497 #define IMSM_UNKNOWN_OWNER (-1)
499 struct intel_disk
*next
;
503 * struct extent - reserved space details.
504 * @start: start offset.
505 * @size: size of reservation, set to 0 for metadata reservation.
506 * @vol: index of the volume, meaningful if &size is set.
509 unsigned long long start
, size
;
513 /* definitions of reshape process types */
514 enum imsm_reshape_type
{
520 /* definition of messages passed to imsm_process_update */
521 enum imsm_update_type
{
522 update_activate_spare
,
526 update_add_remove_disk
,
527 update_reshape_container_disks
,
528 update_reshape_migration
,
530 update_general_migration_checkpoint
,
532 update_prealloc_badblocks_mem
,
536 struct imsm_update_activate_spare
{
537 enum imsm_update_type type
;
541 struct imsm_update_activate_spare
*next
;
547 unsigned long long size
;
554 enum takeover_direction
{
558 struct imsm_update_takeover
{
559 enum imsm_update_type type
;
561 enum takeover_direction direction
;
564 struct imsm_update_reshape
{
565 enum imsm_update_type type
;
569 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
572 struct imsm_update_reshape_migration
{
573 enum imsm_update_type type
;
576 /* fields for array migration changes
583 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
586 struct imsm_update_size_change
{
587 enum imsm_update_type type
;
592 struct imsm_update_general_migration_checkpoint
{
593 enum imsm_update_type type
;
594 __u64 curr_migr_unit
;
598 __u8 serial
[MAX_RAID_SERIAL_LEN
];
601 struct imsm_update_create_array
{
602 enum imsm_update_type type
;
607 struct imsm_update_kill_array
{
608 enum imsm_update_type type
;
612 struct imsm_update_rename_array
{
613 enum imsm_update_type type
;
614 __u8 name
[MAX_RAID_SERIAL_LEN
];
618 struct imsm_update_add_remove_disk
{
619 enum imsm_update_type type
;
622 struct imsm_update_prealloc_bb_mem
{
623 enum imsm_update_type type
;
626 struct imsm_update_rwh_policy
{
627 enum imsm_update_type type
;
632 static const char *_sys_dev_type
[] = {
633 [SYS_DEV_UNKNOWN
] = "Unknown",
634 [SYS_DEV_SAS
] = "SAS",
635 [SYS_DEV_SATA
] = "SATA",
636 [SYS_DEV_NVME
] = "NVMe",
637 [SYS_DEV_VMD
] = "VMD",
638 [SYS_DEV_SATA_VMD
] = "SATA VMD"
641 static int no_platform
= -1;
643 static int check_no_platform(void)
645 static const char search
[] = "mdadm.imsm.test=1";
648 if (no_platform
>= 0)
651 if (check_env("IMSM_NO_PLATFORM")) {
655 fp
= fopen("/proc/cmdline", "r");
657 char *l
= conf_line(fp
);
666 if (strcmp(w
, search
) == 0)
672 if (no_platform
>= 0)
679 void imsm_set_no_platform(int v
)
684 const char *get_sys_dev_type(enum sys_dev_type type
)
686 if (type
>= SYS_DEV_MAX
)
687 type
= SYS_DEV_UNKNOWN
;
689 return _sys_dev_type
[type
];
692 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
694 struct intel_hba
*result
= xmalloc(sizeof(*result
));
696 result
->type
= device
->type
;
697 result
->path
= xstrdup(device
->path
);
699 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
705 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
707 struct intel_hba
*result
;
709 for (result
= hba
; result
; result
= result
->next
) {
710 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
716 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
718 struct intel_hba
*hba
;
720 /* check if disk attached to Intel HBA */
721 hba
= find_intel_hba(super
->hba
, device
);
724 /* Check if HBA is already attached to super */
725 if (super
->hba
== NULL
) {
726 super
->hba
= alloc_intel_hba(device
);
731 /* Intel metadata allows for all disks attached to the same type HBA.
732 * Do not support HBA types mixing
734 if (device
->type
!= hba
->type
)
737 /* Multiple same type HBAs can be used if they share the same OROM */
738 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
740 if (device_orom
!= super
->orom
)
746 hba
->next
= alloc_intel_hba(device
);
750 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
752 struct sys_dev
*list
, *elem
;
755 if ((list
= find_intel_devices()) == NULL
)
758 if (!is_fd_valid(fd
))
759 disk_path
= (char *) devname
;
761 disk_path
= diskfd_to_devpath(fd
, 1, NULL
);
766 for (elem
= list
; elem
; elem
= elem
->next
)
767 if (path_attached_to_hba(disk_path
, elem
->path
))
770 if (disk_path
!= devname
)
776 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
779 static struct supertype
*match_metadata_desc_imsm(char *arg
)
781 struct supertype
*st
;
783 if (strcmp(arg
, "imsm") != 0 &&
784 strcmp(arg
, "default") != 0
788 st
= xcalloc(1, sizeof(*st
));
789 st
->ss
= &super_imsm
;
790 st
->max_devs
= IMSM_MAX_DEVICES
;
791 st
->minor_version
= 0;
796 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
798 return &mpb
->sig
[MPB_SIG_LEN
];
801 /* retrieve a disk directly from the anchor when the anchor is known to be
802 * up-to-date, currently only at load time
804 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
806 if (index
>= mpb
->num_disks
)
808 return &mpb
->disk
[index
];
811 /* retrieve the disk description based on a index of the disk
814 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
818 for (d
= super
->disks
; d
; d
= d
->next
)
819 if (d
->index
== index
)
824 /* retrieve a disk from the parsed metadata */
825 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
829 dl
= get_imsm_dl_disk(super
, index
);
836 /* generate a checksum directly from the anchor when the anchor is known to be
837 * up-to-date, currently only at load or write_super after coalescing
839 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
841 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
842 __u32
*p
= (__u32
*) mpb
;
846 sum
+= __le32_to_cpu(*p
);
850 return sum
- __le32_to_cpu(mpb
->check_sum
);
853 static size_t sizeof_imsm_map(struct imsm_map
*map
)
855 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
858 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
860 /* A device can have 2 maps if it is in the middle of a migration.
862 * MAP_0 - we return the first map
863 * MAP_1 - we return the second map if it exists, else NULL
864 * MAP_X - we return the second map if it exists, else the first
866 struct imsm_map
*map
= &dev
->vol
.map
[0];
867 struct imsm_map
*map2
= NULL
;
869 if (dev
->vol
.migr_state
)
870 map2
= (void *)map
+ sizeof_imsm_map(map
);
872 switch (second_map
) {
889 /* return the size of the device.
890 * migr_state increases the returned size if map[0] were to be duplicated
892 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
894 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
895 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
897 /* migrating means an additional map */
898 if (dev
->vol
.migr_state
)
899 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
901 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
906 /* retrieve disk serial number list from a metadata update */
907 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
910 struct disk_info
*inf
;
912 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
913 sizeof_imsm_dev(&update
->dev
, 0);
919 * __get_imsm_dev() - Get device with index from imsm_super.
920 * @mpb: &imsm_super pointer, not NULL.
921 * @index: Device index.
923 * Function works as non-NULL, aborting in such a case,
924 * when NULL would be returned.
926 * Device index should be in range 0 up to num_raid_devs.
927 * Function assumes the index was already verified.
928 * Index must be valid, otherwise abort() is called.
930 * Return: Pointer to corresponding imsm_dev.
933 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
939 if (index
>= mpb
->num_raid_devs
)
942 /* devices start after all disks */
943 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
945 for (i
= 0; i
<= index
; i
++, offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0))
947 return _mpb
+ offset
;
949 pr_err("cannot find imsm_dev with index %u in imsm_super\n", index
);
954 * get_imsm_dev() - Get device with index from intel_super.
955 * @super: &intel_super pointer, not NULL.
956 * @index: Device index.
958 * Function works as non-NULL, aborting in such a case,
959 * when NULL would be returned.
961 * Device index should be in range 0 up to num_raid_devs.
962 * Function assumes the index was already verified.
963 * Index must be valid, otherwise abort() is called.
965 * Return: Pointer to corresponding imsm_dev.
968 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
970 struct intel_dev
*dv
;
972 if (index
>= super
->anchor
->num_raid_devs
)
975 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
976 if (dv
->index
== index
)
979 pr_err("cannot find imsm_dev with index %u in intel_super\n", index
);
983 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
986 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
987 __le16_to_cpu(addr
->w1
));
990 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
992 struct bbm_log_block_addr addr
;
994 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
995 addr
.dw1
= __cpu_to_le32((__u32
)(sec
>> 16) & 0xffffffff);
999 /* get size of the bbm log */
1000 static __u32
get_imsm_bbm_log_size(struct bbm_log
*log
)
1002 if (!log
|| log
->entry_count
== 0)
1005 return sizeof(log
->signature
) +
1006 sizeof(log
->entry_count
) +
1007 log
->entry_count
* sizeof(struct bbm_log_entry
);
1010 /* check if bad block is not partially stored in bbm log */
1011 static int is_stored_in_bbm(struct bbm_log
*log
, const __u8 idx
, const unsigned
1012 long long sector
, const int length
, __u32
*pos
)
1016 for (i
= *pos
; i
< log
->entry_count
; i
++) {
1017 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
1018 unsigned long long bb_start
;
1019 unsigned long long bb_end
;
1021 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1022 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1024 if ((entry
->disk_ordinal
== idx
) && (bb_start
>= sector
) &&
1025 (bb_end
<= sector
+ length
)) {
1033 /* record new bad block in bbm log */
1034 static int record_new_badblock(struct bbm_log
*log
, const __u8 idx
, unsigned
1035 long long sector
, int length
)
1039 struct bbm_log_entry
*entry
= NULL
;
1041 while (is_stored_in_bbm(log
, idx
, sector
, length
, &pos
)) {
1042 struct bbm_log_entry
*e
= &log
->marked_block_entries
[pos
];
1044 if ((e
->marked_count
+ 1 == BBM_LOG_MAX_LBA_ENTRY_VAL
) &&
1045 (__le48_to_cpu(&e
->defective_block_start
) == sector
)) {
1046 sector
+= BBM_LOG_MAX_LBA_ENTRY_VAL
;
1047 length
-= BBM_LOG_MAX_LBA_ENTRY_VAL
;
1056 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
1057 BBM_LOG_MAX_LBA_ENTRY_VAL
;
1058 entry
->defective_block_start
= __cpu_to_le48(sector
);
1059 entry
->marked_count
= cnt
- 1;
1066 new_bb
= ROUND_UP(length
, BBM_LOG_MAX_LBA_ENTRY_VAL
) /
1067 BBM_LOG_MAX_LBA_ENTRY_VAL
;
1068 if (log
->entry_count
+ new_bb
> BBM_LOG_MAX_ENTRIES
)
1071 while (length
> 0) {
1072 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
1073 BBM_LOG_MAX_LBA_ENTRY_VAL
;
1074 struct bbm_log_entry
*entry
=
1075 &log
->marked_block_entries
[log
->entry_count
];
1077 entry
->defective_block_start
= __cpu_to_le48(sector
);
1078 entry
->marked_count
= cnt
- 1;
1079 entry
->disk_ordinal
= idx
;
1090 /* clear all bad blocks for given disk */
1091 static void clear_disk_badblocks(struct bbm_log
*log
, const __u8 idx
)
1095 while (i
< log
->entry_count
) {
1096 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
1098 if (entries
[i
].disk_ordinal
== idx
) {
1099 if (i
< log
->entry_count
- 1)
1100 entries
[i
] = entries
[log
->entry_count
- 1];
1108 /* clear given bad block */
1109 static int clear_badblock(struct bbm_log
*log
, const __u8 idx
, const unsigned
1110 long long sector
, const int length
) {
1113 while (i
< log
->entry_count
) {
1114 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
1116 if ((entries
[i
].disk_ordinal
== idx
) &&
1117 (__le48_to_cpu(&entries
[i
].defective_block_start
) ==
1118 sector
) && (entries
[i
].marked_count
+ 1 == length
)) {
1119 if (i
< log
->entry_count
- 1)
1120 entries
[i
] = entries
[log
->entry_count
- 1];
1130 /* allocate and load BBM log from metadata */
1131 static int load_bbm_log(struct intel_super
*super
)
1133 struct imsm_super
*mpb
= super
->anchor
;
1134 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1136 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
1137 if (!super
->bbm_log
)
1141 struct bbm_log
*log
= (void *)mpb
+
1142 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1146 if (bbm_log_size
< sizeof(log
->signature
) +
1147 sizeof(log
->entry_count
))
1150 entry_count
= __le32_to_cpu(log
->entry_count
);
1151 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
1152 (entry_count
> BBM_LOG_MAX_ENTRIES
))
1156 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
1157 entry_count
* sizeof(struct bbm_log_entry
))
1160 memcpy(super
->bbm_log
, log
, bbm_log_size
);
1162 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
1163 super
->bbm_log
->entry_count
= 0;
1169 /* checks if bad block is within volume boundaries */
1170 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
1171 const unsigned long long start_sector
,
1172 const unsigned long long size
)
1174 unsigned long long bb_start
;
1175 unsigned long long bb_end
;
1177 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1178 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1180 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1181 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1187 /* get list of bad blocks on a drive for a volume */
1188 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1189 const unsigned long long start_sector
,
1190 const unsigned long long size
,
1196 for (i
= 0; i
< log
->entry_count
; i
++) {
1197 const struct bbm_log_entry
*ent
=
1198 &log
->marked_block_entries
[i
];
1199 struct md_bb_entry
*bb
;
1201 if ((ent
->disk_ordinal
== idx
) &&
1202 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1204 if (!bbs
->entries
) {
1205 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1211 bb
= &bbs
->entries
[count
++];
1212 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1213 bb
->length
= ent
->marked_count
+ 1;
1221 * == MAP_0 get first map
1222 * == MAP_1 get second map
1223 * == MAP_X than get map according to the current migr_state
1225 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1229 struct imsm_map
*map
;
1231 map
= get_imsm_map(dev
, second_map
);
1233 /* top byte identifies disk under rebuild */
1234 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1237 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1238 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1240 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1242 return ord_to_idx(ord
);
1245 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1247 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1250 static int get_imsm_disk_slot(struct imsm_map
*map
, const unsigned int idx
)
1255 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1256 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1257 if (ord_to_idx(ord
) == idx
)
1261 return IMSM_STATUS_ERROR
;
1264 static int get_imsm_raid_level(struct imsm_map
*map
)
1266 if (map
->raid_level
== 1) {
1267 if (map
->num_members
== 2)
1273 return map
->raid_level
;
1277 * get_disk_slot_in_dev() - retrieve disk slot from &imsm_dev.
1278 * @super: &intel_super pointer, not NULL.
1279 * @dev_idx: imsm device index.
1282 * Return: Slot on success, IMSM_STATUS_ERROR otherwise.
1284 static int get_disk_slot_in_dev(struct intel_super
*super
, const __u8 dev_idx
,
1285 const unsigned int idx
)
1287 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
1288 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1290 return get_imsm_disk_slot(map
, idx
);
1293 static int cmp_extent(const void *av
, const void *bv
)
1295 const struct extent
*a
= av
;
1296 const struct extent
*b
= bv
;
1297 if (a
->start
< b
->start
)
1299 if (a
->start
> b
->start
)
1304 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1306 int memberships
= 0;
1309 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++)
1310 if (get_disk_slot_in_dev(super
, i
, dl
->index
) >= 0)
1316 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1318 static int split_ull(unsigned long long n
, void *lo
, void *hi
)
1320 if (lo
== 0 || hi
== 0)
1322 __put_unaligned32(__cpu_to_le32((__u32
)n
), lo
);
1323 __put_unaligned32(__cpu_to_le32((n
>> 32)), hi
);
1327 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1329 return (unsigned long long)__le32_to_cpu(lo
) |
1330 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1333 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1337 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1341 * imsm_num_data_members() - get data drives count for an array.
1342 * @map: Map to analyze.
1344 * num_data_members value represents minimal count of drives for level.
1345 * The name of the property could be misleading for RAID5 with asymmetric layout
1346 * because some data required to be calculated from parity.
1347 * The property is extracted from level and num_members value.
1349 * Return: num_data_members value on success, zero otherwise.
1351 static __u8
imsm_num_data_members(struct imsm_map
*map
)
1353 switch (get_imsm_raid_level(map
)) {
1355 return map
->num_members
;
1358 return map
->num_members
/ 2;
1360 return map
->num_members
- 1;
1362 dprintf("unsupported raid level\n");
1367 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1371 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1374 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1378 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1381 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1385 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1388 static unsigned long long vol_curr_migr_unit(struct imsm_dev
*dev
)
1393 return join_u32(dev
->vol
.curr_migr_unit_lo
, dev
->vol
.curr_migr_unit_hi
);
1396 static unsigned long long imsm_dev_size(struct imsm_dev
*dev
)
1400 return join_u32(dev
->size_low
, dev
->size_high
);
1403 static unsigned long long migr_chkp_area_pba(struct migr_record
*migr_rec
)
1405 if (migr_rec
== NULL
)
1407 return join_u32(migr_rec
->ckpt_area_pba_lo
,
1408 migr_rec
->ckpt_area_pba_hi
);
1411 static unsigned long long current_migr_unit(struct migr_record
*migr_rec
)
1413 if (migr_rec
== NULL
)
1415 return join_u32(migr_rec
->curr_migr_unit_lo
,
1416 migr_rec
->curr_migr_unit_hi
);
1419 static unsigned long long migr_dest_1st_member_lba(struct migr_record
*migr_rec
)
1421 if (migr_rec
== NULL
)
1423 return join_u32(migr_rec
->dest_1st_member_lba_lo
,
1424 migr_rec
->dest_1st_member_lba_hi
);
1427 static unsigned long long get_num_migr_units(struct migr_record
*migr_rec
)
1429 if (migr_rec
== NULL
)
1431 return join_u32(migr_rec
->num_migr_units_lo
,
1432 migr_rec
->num_migr_units_hi
);
1435 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1437 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1441 * set_num_domains() - Set number of domains for an array.
1442 * @map: Map to be updated.
1444 * num_domains property represents copies count of each data drive, thus make
1445 * it meaningful only for RAID1 and RAID10. IMSM supports two domains for
1448 static void set_num_domains(struct imsm_map
*map
)
1450 int level
= get_imsm_raid_level(map
);
1452 if (level
== 1 || level
== 10)
1453 map
->num_domains
= 2;
1455 map
->num_domains
= 1;
1458 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1460 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1463 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1465 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1468 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1470 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1474 * update_num_data_stripes() - Calculate and update num_data_stripes value.
1475 * @map: map to be updated.
1476 * @dev_size: size of volume.
1478 * num_data_stripes value is addictionally divided by num_domains, therefore for
1479 * levels where num_domains is not 1, nds is a part of real value.
1481 static void update_num_data_stripes(struct imsm_map
*map
,
1482 unsigned long long dev_size
)
1484 unsigned long long nds
= dev_size
/ imsm_num_data_members(map
);
1486 nds
/= map
->num_domains
;
1487 nds
/= map
->blocks_per_strip
;
1488 set_num_data_stripes(map
, nds
);
1491 static void set_vol_curr_migr_unit(struct imsm_dev
*dev
, unsigned long long n
)
1496 split_ull(n
, &dev
->vol
.curr_migr_unit_lo
, &dev
->vol
.curr_migr_unit_hi
);
1499 static void set_imsm_dev_size(struct imsm_dev
*dev
, unsigned long long n
)
1501 split_ull(n
, &dev
->size_low
, &dev
->size_high
);
1504 static void set_migr_chkp_area_pba(struct migr_record
*migr_rec
,
1505 unsigned long long n
)
1507 split_ull(n
, &migr_rec
->ckpt_area_pba_lo
, &migr_rec
->ckpt_area_pba_hi
);
1510 static void set_current_migr_unit(struct migr_record
*migr_rec
,
1511 unsigned long long n
)
1513 split_ull(n
, &migr_rec
->curr_migr_unit_lo
,
1514 &migr_rec
->curr_migr_unit_hi
);
1517 static void set_migr_dest_1st_member_lba(struct migr_record
*migr_rec
,
1518 unsigned long long n
)
1520 split_ull(n
, &migr_rec
->dest_1st_member_lba_lo
,
1521 &migr_rec
->dest_1st_member_lba_hi
);
1524 static void set_num_migr_units(struct migr_record
*migr_rec
,
1525 unsigned long long n
)
1527 split_ull(n
, &migr_rec
->num_migr_units_lo
,
1528 &migr_rec
->num_migr_units_hi
);
1531 static unsigned long long per_dev_array_size(struct imsm_map
*map
)
1533 unsigned long long array_size
= 0;
1538 array_size
= num_data_stripes(map
) * map
->blocks_per_strip
;
1539 if (get_imsm_raid_level(map
) == 1 || get_imsm_raid_level(map
) == 10)
1545 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
,
1546 int get_minimal_reservation
)
1548 /* find a list of used extents on the given physical device */
1549 int memberships
= count_memberships(dl
, super
);
1550 struct extent
*rv
= xcalloc(memberships
+ 1, sizeof(struct extent
));
1551 struct extent
*e
= rv
;
1555 /* trim the reserved area for spares, so they can join any array
1556 * regardless of whether the OROM has assigned sectors from the
1557 * IMSM_RESERVED_SECTORS region
1559 if (dl
->index
== -1 || get_minimal_reservation
)
1560 reservation
= imsm_min_reserved_sectors(super
);
1562 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1564 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1565 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1566 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1568 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1569 e
->start
= pba_of_lba0(map
);
1570 e
->size
= per_dev_array_size(map
);
1575 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1577 /* determine the start of the metadata
1578 * when no raid devices are defined use the default
1579 * ...otherwise allow the metadata to truncate the value
1580 * as is the case with older versions of imsm
1583 struct extent
*last
= &rv
[memberships
- 1];
1584 unsigned long long remainder
;
1586 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1587 /* round down to 1k block to satisfy precision of the kernel
1591 /* make sure remainder is still sane */
1592 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1593 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1594 if (reservation
> remainder
)
1595 reservation
= remainder
;
1597 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1602 /* try to determine how much space is reserved for metadata from
1603 * the last get_extents() entry, otherwise fallback to the
1606 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1612 /* for spares just return a minimal reservation which will grow
1613 * once the spare is picked up by an array
1615 if (dl
->index
== -1)
1616 return MPB_SECTOR_CNT
;
1618 e
= get_extents(super
, dl
, 0);
1620 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1622 /* scroll to last entry */
1623 for (i
= 0; e
[i
].size
; i
++)
1626 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1633 static int is_spare(struct imsm_disk
*disk
)
1635 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1638 static int is_configured(struct imsm_disk
*disk
)
1640 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1643 static int is_failed(struct imsm_disk
*disk
)
1645 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1648 static int is_journal(struct imsm_disk
*disk
)
1650 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1654 * round_member_size_to_mb()- Round given size to closest MiB.
1655 * @size: size to round in sectors.
1657 static inline unsigned long long round_member_size_to_mb(unsigned long long size
)
1659 return (size
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
1663 * round_size_to_mb()- Round given size.
1664 * @array_size: size to round in sectors.
1665 * @disk_count: count of data members.
1667 * Get size per each data member and round it to closest MiB to ensure that data
1668 * splits evenly between members.
1670 * Return: Array size, rounded down.
1672 static inline unsigned long long round_size_to_mb(unsigned long long array_size
,
1673 unsigned int disk_count
)
1675 return round_member_size_to_mb(array_size
/ disk_count
) * disk_count
;
1678 static int able_to_resync(int raid_level
, int missing_disks
)
1680 int max_missing_disks
= 0;
1682 switch (raid_level
) {
1684 max_missing_disks
= 1;
1687 max_missing_disks
= 0;
1689 return missing_disks
<= max_missing_disks
;
1692 /* try to determine how much space is reserved for metadata from
1693 * the last get_extents() entry on the smallest active disk,
1694 * otherwise fallback to the default
1696 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1700 unsigned long long min_active
;
1702 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1703 struct dl
*dl
, *dl_min
= NULL
;
1709 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1712 unsigned long long blocks
= total_blocks(&dl
->disk
);
1713 if (blocks
< min_active
|| min_active
== 0) {
1715 min_active
= blocks
;
1721 /* find last lba used by subarrays on the smallest active disk */
1722 e
= get_extents(super
, dl_min
, 0);
1725 for (i
= 0; e
[i
].size
; i
++)
1728 remainder
= min_active
- e
[i
].start
;
1731 /* to give priority to recovery we should not require full
1732 IMSM_RESERVED_SECTORS from the spare */
1733 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1735 /* if real reservation is smaller use that value */
1736 return (remainder
< rv
) ? remainder
: rv
;
1740 * Return minimum size of a spare and sector size
1741 * that can be used in this array
1743 int get_spare_criteria_imsm(struct supertype
*st
, struct spare_criteria
*c
)
1745 struct intel_super
*super
= st
->sb
;
1749 unsigned long long size
= 0;
1756 /* find first active disk in array */
1758 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1762 /* find last lba used by subarrays */
1763 e
= get_extents(super
, dl
, 0);
1766 for (i
= 0; e
[i
].size
; i
++)
1769 size
= e
[i
-1].start
+ e
[i
-1].size
;
1772 /* add the amount of space needed for metadata */
1773 size
+= imsm_min_reserved_sectors(super
);
1775 c
->min_size
= size
* 512;
1776 c
->sector_size
= super
->sector_size
;
1781 static bool is_gen_migration(struct imsm_dev
*dev
);
1783 #define IMSM_4K_DIV 8
1785 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1786 struct imsm_dev
*dev
);
1788 static void print_imsm_dev(struct intel_super
*super
,
1789 struct imsm_dev
*dev
,
1795 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1796 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1800 printf("[%.16s]:\n", dev
->volume
);
1801 printf(" Subarray : %d\n", super
->current_vol
);
1802 printf(" UUID : %s\n", uuid
);
1803 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1805 printf(" <-- %d", get_imsm_raid_level(map2
));
1807 printf(" Members : %d", map
->num_members
);
1809 printf(" <-- %d", map2
->num_members
);
1811 printf(" Slots : [");
1812 for (i
= 0; i
< map
->num_members
; i
++) {
1813 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1814 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1819 for (i
= 0; i
< map2
->num_members
; i
++) {
1820 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1821 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1826 printf(" Failed disk : ");
1827 if (map
->failed_disk_num
== 0xff)
1830 printf("%i", map
->failed_disk_num
);
1832 slot
= get_imsm_disk_slot(map
, disk_idx
);
1834 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1835 printf(" This Slot : %d%s\n", slot
,
1836 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1838 printf(" This Slot : ?\n");
1839 printf(" Sector Size : %u\n", super
->sector_size
);
1840 sz
= imsm_dev_size(dev
);
1841 printf(" Array Size : %llu%s\n",
1842 (unsigned long long)sz
* 512 / super
->sector_size
,
1843 human_size(sz
* 512));
1844 sz
= blocks_per_member(map
);
1845 printf(" Per Dev Size : %llu%s\n",
1846 (unsigned long long)sz
* 512 / super
->sector_size
,
1847 human_size(sz
* 512));
1848 printf(" Sector Offset : %llu\n",
1849 pba_of_lba0(map
) * 512 / super
->sector_size
);
1850 printf(" Num Stripes : %llu\n",
1851 num_data_stripes(map
));
1852 printf(" Chunk Size : %u KiB",
1853 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1855 printf(" <-- %u KiB",
1856 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1858 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1859 printf(" Migrate State : ");
1860 if (dev
->vol
.migr_state
) {
1861 if (migr_type(dev
) == MIGR_INIT
)
1862 printf("initialize\n");
1863 else if (migr_type(dev
) == MIGR_REBUILD
)
1864 printf("rebuild\n");
1865 else if (migr_type(dev
) == MIGR_VERIFY
)
1867 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1868 printf("general migration\n");
1869 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1870 printf("state change\n");
1871 else if (migr_type(dev
) == MIGR_REPAIR
)
1874 printf("<unknown:%d>\n", migr_type(dev
));
1877 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1878 if (dev
->vol
.migr_state
) {
1879 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1881 printf(" <-- %s", map_state_str
[map
->map_state
]);
1882 printf("\n Checkpoint : %llu ", vol_curr_migr_unit(dev
));
1883 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1886 printf("(%llu)", (unsigned long long)
1887 blocks_per_migr_unit(super
, dev
));
1890 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1892 printf(" RWH Policy : ");
1893 if (dev
->rwh_policy
== RWH_OFF
|| dev
->rwh_policy
== RWH_MULTIPLE_OFF
)
1895 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1896 printf("PPL distributed\n");
1897 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1898 printf("PPL journaling drive\n");
1899 else if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
1900 printf("Multiple distributed PPLs\n");
1901 else if (dev
->rwh_policy
== RWH_MULTIPLE_PPLS_JOURNALING_DRIVE
)
1902 printf("Multiple PPLs on journaling drive\n");
1903 else if (dev
->rwh_policy
== RWH_BITMAP
)
1904 printf("Write-intent bitmap\n");
1906 printf("<unknown:%d>\n", dev
->rwh_policy
);
1908 printf(" Volume ID : %u\n", dev
->my_vol_raid_dev_num
);
1911 static void print_imsm_disk(struct imsm_disk
*disk
,
1914 unsigned int sector_size
) {
1915 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1918 if (index
< -1 || !disk
)
1922 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1924 printf(" Disk%02d Serial : %s\n", index
, str
);
1926 printf(" Disk Serial : %s\n", str
);
1927 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1928 is_configured(disk
) ? " active" : "",
1929 is_failed(disk
) ? " failed" : "",
1930 is_journal(disk
) ? " journal" : "");
1931 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1932 sz
= total_blocks(disk
) - reserved
;
1933 printf(" Usable Size : %llu%s\n",
1934 (unsigned long long)sz
* 512 / sector_size
,
1935 human_size(sz
* 512));
1938 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1940 struct migr_record
*migr_rec
= super
->migr_rec
;
1942 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1943 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1944 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1945 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1946 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1947 set_migr_chkp_area_pba(migr_rec
,
1948 migr_chkp_area_pba(migr_rec
) / IMSM_4K_DIV
);
1949 set_migr_dest_1st_member_lba(migr_rec
,
1950 migr_dest_1st_member_lba(migr_rec
) / IMSM_4K_DIV
);
1953 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1955 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1958 void convert_to_4k(struct intel_super
*super
)
1960 struct imsm_super
*mpb
= super
->anchor
;
1961 struct imsm_disk
*disk
;
1963 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1965 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1966 disk
= __get_imsm_disk(mpb
, i
);
1968 convert_to_4k_imsm_disk(disk
);
1970 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1971 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1972 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1974 set_imsm_dev_size(dev
, imsm_dev_size(dev
)/IMSM_4K_DIV
);
1975 set_vol_curr_migr_unit(dev
,
1976 vol_curr_migr_unit(dev
) / IMSM_4K_DIV
);
1979 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1980 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1981 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1983 if (dev
->vol
.migr_state
) {
1985 map
= get_imsm_map(dev
, MAP_1
);
1986 set_blocks_per_member(map
,
1987 blocks_per_member(map
)/IMSM_4K_DIV
);
1988 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1989 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1993 struct bbm_log
*log
= (void *)mpb
+
1994 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1997 for (i
= 0; i
< log
->entry_count
; i
++) {
1998 struct bbm_log_entry
*entry
=
1999 &log
->marked_block_entries
[i
];
2001 __u8 count
= entry
->marked_count
+ 1;
2002 unsigned long long sector
=
2003 __le48_to_cpu(&entry
->defective_block_start
);
2005 entry
->defective_block_start
=
2006 __cpu_to_le48(sector
/IMSM_4K_DIV
);
2007 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
2011 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
2014 void examine_migr_rec_imsm(struct intel_super
*super
)
2016 struct migr_record
*migr_rec
= super
->migr_rec
;
2017 struct imsm_super
*mpb
= super
->anchor
;
2020 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2021 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2022 struct imsm_map
*map
;
2025 if (is_gen_migration(dev
) == false)
2028 printf("\nMigration Record Information:");
2030 /* first map under migration */
2031 map
= get_imsm_map(dev
, MAP_0
);
2034 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
2035 if (map
== NULL
|| slot
> 1 || slot
< 0) {
2036 printf(" Empty\n ");
2037 printf("Examine one of first two disks in array\n");
2040 printf("\n Status : ");
2041 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
2044 printf("Contains Data\n");
2045 printf(" Current Unit : %llu\n",
2046 current_migr_unit(migr_rec
));
2047 printf(" Family : %u\n",
2048 __le32_to_cpu(migr_rec
->family_num
));
2049 printf(" Ascending : %u\n",
2050 __le32_to_cpu(migr_rec
->ascending_migr
));
2051 printf(" Blocks Per Unit : %u\n",
2052 __le32_to_cpu(migr_rec
->blocks_per_unit
));
2053 printf(" Dest. Depth Per Unit : %u\n",
2054 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
2055 printf(" Checkpoint Area pba : %llu\n",
2056 migr_chkp_area_pba(migr_rec
));
2057 printf(" First member lba : %llu\n",
2058 migr_dest_1st_member_lba(migr_rec
));
2059 printf(" Total Number of Units : %llu\n",
2060 get_num_migr_units(migr_rec
));
2061 printf(" Size of volume : %llu\n",
2062 join_u32(migr_rec
->post_migr_vol_cap
,
2063 migr_rec
->post_migr_vol_cap_hi
));
2064 printf(" Record was read from : %u\n",
2065 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
2071 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
2073 struct migr_record
*migr_rec
= super
->migr_rec
;
2075 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
2076 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
2077 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
2078 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
2079 &migr_rec
->post_migr_vol_cap
,
2080 &migr_rec
->post_migr_vol_cap_hi
);
2081 set_migr_chkp_area_pba(migr_rec
,
2082 migr_chkp_area_pba(migr_rec
) * IMSM_4K_DIV
);
2083 set_migr_dest_1st_member_lba(migr_rec
,
2084 migr_dest_1st_member_lba(migr_rec
) * IMSM_4K_DIV
);
2087 void convert_from_4k(struct intel_super
*super
)
2089 struct imsm_super
*mpb
= super
->anchor
;
2090 struct imsm_disk
*disk
;
2092 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
2094 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2095 disk
= __get_imsm_disk(mpb
, i
);
2097 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
2100 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2101 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2102 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2104 set_imsm_dev_size(dev
, imsm_dev_size(dev
)*IMSM_4K_DIV
);
2105 set_vol_curr_migr_unit(dev
,
2106 vol_curr_migr_unit(dev
) * IMSM_4K_DIV
);
2109 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
2110 map
->blocks_per_strip
*= IMSM_4K_DIV
;
2111 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
2113 if (dev
->vol
.migr_state
) {
2115 map
= get_imsm_map(dev
, MAP_1
);
2116 set_blocks_per_member(map
,
2117 blocks_per_member(map
)*IMSM_4K_DIV
);
2118 map
->blocks_per_strip
*= IMSM_4K_DIV
;
2119 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
2123 struct bbm_log
*log
= (void *)mpb
+
2124 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
2127 for (i
= 0; i
< log
->entry_count
; i
++) {
2128 struct bbm_log_entry
*entry
=
2129 &log
->marked_block_entries
[i
];
2131 __u8 count
= entry
->marked_count
+ 1;
2132 unsigned long long sector
=
2133 __le48_to_cpu(&entry
->defective_block_start
);
2135 entry
->defective_block_start
=
2136 __cpu_to_le48(sector
*IMSM_4K_DIV
);
2137 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
2141 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
2144 /*******************************************************************************
2145 * function: imsm_check_attributes
2146 * Description: Function checks if features represented by attributes flags
2147 * are supported by mdadm.
2149 * attributes - Attributes read from metadata
2151 * 0 - passed attributes contains unsupported features flags
2152 * 1 - all features are supported
2153 ******************************************************************************/
2154 static int imsm_check_attributes(__u32 attributes
)
2157 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
2159 not_supported
&= ~MPB_ATTRIB_IGNORED
;
2161 not_supported
&= attributes
;
2162 if (not_supported
) {
2163 pr_err("(IMSM): Unsupported attributes : %x\n",
2164 (unsigned)__le32_to_cpu(not_supported
));
2165 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
2166 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
2167 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
2169 if (not_supported
& MPB_ATTRIB_2TB
) {
2170 dprintf("\t\tMPB_ATTRIB_2TB\n");
2171 not_supported
^= MPB_ATTRIB_2TB
;
2173 if (not_supported
& MPB_ATTRIB_RAID0
) {
2174 dprintf("\t\tMPB_ATTRIB_RAID0\n");
2175 not_supported
^= MPB_ATTRIB_RAID0
;
2177 if (not_supported
& MPB_ATTRIB_RAID1
) {
2178 dprintf("\t\tMPB_ATTRIB_RAID1\n");
2179 not_supported
^= MPB_ATTRIB_RAID1
;
2181 if (not_supported
& MPB_ATTRIB_RAID10
) {
2182 dprintf("\t\tMPB_ATTRIB_RAID10\n");
2183 not_supported
^= MPB_ATTRIB_RAID10
;
2185 if (not_supported
& MPB_ATTRIB_RAID1E
) {
2186 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
2187 not_supported
^= MPB_ATTRIB_RAID1E
;
2189 if (not_supported
& MPB_ATTRIB_RAID5
) {
2190 dprintf("\t\tMPB_ATTRIB_RAID5\n");
2191 not_supported
^= MPB_ATTRIB_RAID5
;
2193 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
2194 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
2195 not_supported
^= MPB_ATTRIB_RAIDCNG
;
2197 if (not_supported
& MPB_ATTRIB_BBM
) {
2198 dprintf("\t\tMPB_ATTRIB_BBM\n");
2199 not_supported
^= MPB_ATTRIB_BBM
;
2201 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
2202 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
2203 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
2205 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
2206 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
2207 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
2209 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
2210 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
2211 not_supported
^= MPB_ATTRIB_2TB_DISK
;
2213 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
2214 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
2215 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
2217 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
2218 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
2219 not_supported
^= MPB_ATTRIB_NEVER_USE
;
2223 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
2231 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
2233 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
2235 struct intel_super
*super
= st
->sb
;
2236 struct imsm_super
*mpb
= super
->anchor
;
2237 char str
[MAX_SIGNATURE_LENGTH
];
2242 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2244 time_t creation_time
;
2246 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
2247 str
[MPB_SIG_LEN
-1] = '\0';
2248 printf(" Magic : %s\n", str
);
2249 printf(" Version : %s\n", get_imsm_version(mpb
));
2250 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
2251 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
2252 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
2253 creation_time
= __le64_to_cpu(mpb
->creation_time
);
2254 printf(" Creation Time : %.24s\n",
2255 creation_time
? ctime(&creation_time
) : "Unknown");
2256 printf(" Attributes : ");
2257 if (imsm_check_attributes(mpb
->attributes
))
2258 printf("All supported\n");
2260 printf("not supported\n");
2261 getinfo_super_imsm(st
, &info
, NULL
);
2262 fname_from_uuid(st
, &info
, nbuf
, ':');
2263 printf(" UUID : %s\n", nbuf
+ 5);
2264 sum
= __le32_to_cpu(mpb
->check_sum
);
2265 printf(" Checksum : %08x %s\n", sum
,
2266 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
2267 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
2268 printf(" Disks : %d\n", mpb
->num_disks
);
2269 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
2270 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
2271 super
->disks
->index
, reserved
, super
->sector_size
);
2272 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
2273 struct bbm_log
*log
= super
->bbm_log
;
2276 printf("Bad Block Management Log:\n");
2277 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
2278 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
2279 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
2281 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2283 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2285 super
->current_vol
= i
;
2286 getinfo_super_imsm(st
, &info
, NULL
);
2287 fname_from_uuid(st
, &info
, nbuf
, ':');
2288 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
2290 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2291 if (i
== super
->disks
->index
)
2293 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
2294 super
->sector_size
);
2297 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2298 if (dl
->index
== -1)
2299 print_imsm_disk(&dl
->disk
, -1, reserved
,
2300 super
->sector_size
);
2302 examine_migr_rec_imsm(super
);
2305 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
2307 /* We just write a generic IMSM ARRAY entry */
2311 getinfo_super_imsm(st
, &info
, NULL
);
2312 fname_from_uuid(st
, &info
, nbuf
, ':');
2313 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
2316 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
2318 /* We just write a generic IMSM ARRAY entry */
2322 struct intel_super
*super
= st
->sb
;
2325 if (!super
->anchor
->num_raid_devs
)
2328 getinfo_super_imsm(st
, &info
, NULL
);
2329 fname_from_uuid(st
, &info
, nbuf
, ':');
2330 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2331 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2333 super
->current_vol
= i
;
2334 getinfo_super_imsm(st
, &info
, NULL
);
2335 fname_from_uuid(st
, &info
, nbuf1
, ':');
2336 printf("ARRAY " DEV_MD_DIR
"%.16s container=%s member=%d UUID=%s\n",
2337 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
2341 static void export_examine_super_imsm(struct supertype
*st
)
2343 struct intel_super
*super
= st
->sb
;
2344 struct imsm_super
*mpb
= super
->anchor
;
2348 getinfo_super_imsm(st
, &info
, NULL
);
2349 fname_from_uuid(st
, &info
, nbuf
, ':');
2350 printf("MD_METADATA=imsm\n");
2351 printf("MD_LEVEL=container\n");
2352 printf("MD_UUID=%s\n", nbuf
+5);
2353 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
2354 printf("MD_CREATION_TIME=%llu\n", __le64_to_cpu(mpb
->creation_time
));
2357 static void detail_super_imsm(struct supertype
*st
, char *homehost
,
2362 struct intel_super
*super
= st
->sb
;
2363 int temp_vol
= super
->current_vol
;
2366 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2368 getinfo_super_imsm(st
, &info
, NULL
);
2369 fname_from_uuid(st
, &info
, nbuf
, ':');
2370 printf("\n UUID : %s\n", nbuf
+ 5);
2372 super
->current_vol
= temp_vol
;
2375 static void brief_detail_super_imsm(struct supertype
*st
, char *subarray
)
2379 struct intel_super
*super
= st
->sb
;
2380 int temp_vol
= super
->current_vol
;
2383 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2385 getinfo_super_imsm(st
, &info
, NULL
);
2386 fname_from_uuid(st
, &info
, nbuf
, ':');
2387 printf(" UUID=%s", nbuf
+ 5);
2389 super
->current_vol
= temp_vol
;
2392 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
,
2393 size_t serial_buf_len
);
2394 static void fd2devname(int fd
, char *name
);
2396 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2398 /* dump an unsorted list of devices attached to AHCI Intel storage
2399 * controller, as well as non-connected ports
2401 int hba_len
= strlen(hba_path
) + 1;
2406 unsigned long port_mask
= (1 << port_count
) - 1;
2408 if (port_count
> (int)sizeof(port_mask
) * 8) {
2410 pr_err("port_count %d out of range\n", port_count
);
2414 /* scroll through /sys/dev/block looking for devices attached to
2417 dir
= opendir("/sys/dev/block");
2421 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2427 char device
[PATH_MAX
];
2432 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2434 path
= devt_to_devpath(makedev(major
, minor
), 1, NULL
);
2437 if (!path_attached_to_hba(path
, hba_path
)) {
2443 /* retrieve the scsi device */
2444 if (!devt_to_devpath(makedev(major
, minor
), 1, device
)) {
2446 pr_err("failed to get device\n");
2450 if (devpath_to_char(device
, "type", buf
, sizeof(buf
), 0)) {
2454 type
= strtoul(buf
, NULL
, 10);
2456 /* if it's not a disk print the vendor and model */
2457 if (!(type
== 0 || type
== 7 || type
== 14)) {
2461 if (devpath_to_char(device
, "vendor", buf
,
2462 sizeof(buf
), 0) == 0) {
2463 strncpy(vendor
, buf
, sizeof(vendor
));
2464 vendor
[sizeof(vendor
) - 1] = '\0';
2465 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2466 while (isspace(*c
) || *c
== '\0')
2471 if (devpath_to_char(device
, "model", buf
,
2472 sizeof(buf
), 0) == 0) {
2473 strncpy(model
, buf
, sizeof(model
));
2474 model
[sizeof(model
) - 1] = '\0';
2475 c
= (char *) &model
[sizeof(model
) - 1];
2476 while (isspace(*c
) || *c
== '\0')
2480 if (vendor
[0] && model
[0])
2481 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2483 switch (type
) { /* numbers from hald/linux/device.c */
2484 case 1: sprintf(buf
, "tape"); break;
2485 case 2: sprintf(buf
, "printer"); break;
2486 case 3: sprintf(buf
, "processor"); break;
2488 case 5: sprintf(buf
, "cdrom"); break;
2489 case 6: sprintf(buf
, "scanner"); break;
2490 case 8: sprintf(buf
, "media_changer"); break;
2491 case 9: sprintf(buf
, "comm"); break;
2492 case 12: sprintf(buf
, "raid"); break;
2493 default: sprintf(buf
, "unknown");
2498 /* chop device path to 'host%d' and calculate the port number */
2499 c
= strchr(&path
[hba_len
], '/');
2502 pr_err("%s - invalid path name\n", path
+ hba_len
);
2507 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2508 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2512 *c
= '/'; /* repair the full string */
2513 pr_err("failed to determine port number for %s\n",
2520 /* mark this port as used */
2521 port_mask
&= ~(1 << port
);
2523 /* print out the device information */
2525 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2529 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2530 if (!is_fd_valid(fd
))
2531 printf(" Port%d : - disk info unavailable -\n", port
);
2533 fd2devname(fd
, buf
);
2534 printf(" Port%d : %s", port
, buf
);
2535 if (imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2537 printf(" (%s)\n", buf
);
2552 for (i
= 0; i
< port_count
; i
++)
2553 if (port_mask
& (1 << i
))
2554 printf(" Port%d : - no device attached -\n", i
);
2560 static int print_nvme_info(struct sys_dev
*hba
)
2565 dir
= opendir("/sys/block/");
2569 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2570 char ns_path
[PATH_MAX
];
2571 char cntrl_path
[PATH_MAX
];
2575 if (!strstr(ent
->d_name
, "nvme"))
2578 fd
= open_dev(ent
->d_name
);
2579 if (!is_fd_valid(fd
))
2582 if (!diskfd_to_devpath(fd
, 0, ns_path
) ||
2583 !diskfd_to_devpath(fd
, 1, cntrl_path
))
2586 if (!path_attached_to_hba(cntrl_path
, hba
->path
))
2589 if (!imsm_is_nvme_namespace_supported(fd
, 0))
2592 fd2devname(fd
, buf
);
2593 if (hba
->type
== SYS_DEV_VMD
)
2594 printf(" NVMe under VMD : %s", buf
);
2595 else if (hba
->type
== SYS_DEV_NVME
)
2596 printf(" NVMe Device : %s", buf
);
2598 if (!imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2600 printf(" (%s)\n", buf
);
2612 static void print_found_intel_controllers(struct sys_dev
*elem
)
2614 for (; elem
; elem
= elem
->next
) {
2615 pr_err("found Intel(R) ");
2616 if (elem
->type
== SYS_DEV_SATA
)
2617 fprintf(stderr
, "SATA ");
2618 else if (elem
->type
== SYS_DEV_SAS
)
2619 fprintf(stderr
, "SAS ");
2620 else if (elem
->type
== SYS_DEV_NVME
)
2621 fprintf(stderr
, "NVMe ");
2623 if (elem
->type
== SYS_DEV_VMD
)
2624 fprintf(stderr
, "VMD domain");
2625 else if (elem
->type
== SYS_DEV_SATA_VMD
)
2626 fprintf(stderr
, "SATA VMD domain");
2628 fprintf(stderr
, "RAID controller");
2631 fprintf(stderr
, " at %s", elem
->pci_id
);
2632 fprintf(stderr
, ".\n");
2637 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2644 if ((dir
= opendir(hba_path
)) == NULL
)
2647 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2650 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2651 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2653 if (*port_count
== 0)
2655 else if (host
< host_base
)
2658 if (host
+ 1 > *port_count
+ host_base
)
2659 *port_count
= host
+ 1 - host_base
;
2665 static void print_imsm_capability(const struct imsm_orom
*orom
)
2667 printf(" Platform : Intel(R) ");
2668 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2669 printf("Matrix Storage Manager\n");
2670 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2671 printf("Virtual RAID on CPU\n");
2673 printf("Rapid Storage Technology%s\n",
2674 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2675 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
) {
2676 if (imsm_orom_is_vmd_without_efi(orom
))
2677 printf(" Version : %d.%d\n", orom
->major_ver
,
2680 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2681 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2683 printf(" RAID Levels :%s%s%s%s%s\n",
2684 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2685 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2686 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2687 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2688 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2689 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2690 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2691 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2692 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2693 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2694 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2695 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2696 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2697 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2698 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2699 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2700 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2701 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2702 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2703 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2704 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2705 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2706 printf(" 2TB volumes :%s supported\n",
2707 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2708 printf(" 2TB disks :%s supported\n",
2709 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2710 printf(" Max Disks : %d\n", orom
->tds
);
2711 printf(" Max Volumes : %d per array, %d per %s\n",
2712 orom
->vpa
, orom
->vphba
,
2713 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2717 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2719 printf("MD_FIRMWARE_TYPE=imsm\n");
2720 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2721 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2722 orom
->hotfix_ver
, orom
->build
);
2723 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2724 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2725 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2726 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2727 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2728 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2729 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2730 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2731 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2732 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2733 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2734 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2735 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2736 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2737 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2738 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2739 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2740 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2741 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2742 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2743 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2744 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2745 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2746 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2747 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2748 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2749 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2750 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2753 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2755 /* There are two components to imsm platform support, the ahci SATA
2756 * controller and the option-rom. To find the SATA controller we
2757 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2758 * controller with the Intel vendor id is present. This approach
2759 * allows mdadm to leverage the kernel's ahci detection logic, with the
2760 * caveat that if ahci.ko is not loaded mdadm will not be able to
2761 * detect platform raid capabilities. The option-rom resides in a
2762 * platform "Adapter ROM". We scan for its signature to retrieve the
2763 * platform capabilities. If raid support is disabled in the BIOS the
2764 * option-rom capability structure will not be available.
2766 struct sys_dev
*list
, *hba
;
2771 if (enumerate_only
) {
2772 if (check_no_platform())
2774 list
= find_intel_devices();
2777 for (hba
= list
; hba
; hba
= hba
->next
) {
2778 if (find_imsm_capability(hba
)) {
2788 list
= find_intel_devices();
2791 pr_err("no active Intel(R) RAID controller found.\n");
2793 } else if (verbose
> 0)
2794 print_found_intel_controllers(list
);
2796 for (hba
= list
; hba
; hba
= hba
->next
) {
2797 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2799 if (!find_imsm_capability(hba
)) {
2801 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2802 hba
->type
== SYS_DEV_VMD
|| hba
->type
== SYS_DEV_SATA_VMD
?
2803 vmd_domain_to_controller(hba
, buf
) :
2804 hba
->path
, get_sys_dev_type(hba
->type
));
2810 if (controller_path
&& result
== 1) {
2811 pr_err("no active Intel(R) RAID controller found under %s\n",
2816 const struct orom_entry
*entry
;
2818 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2819 if (entry
->type
== SYS_DEV_VMD
) {
2820 print_imsm_capability(&entry
->orom
);
2821 printf(" 3rd party NVMe :%s supported\n",
2822 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2823 for (hba
= list
; hba
; hba
= hba
->next
) {
2824 if (hba
->type
== SYS_DEV_VMD
) {
2826 printf(" I/O Controller : %s (%s)\n",
2827 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2828 if (print_nvme_info(hba
)) {
2830 pr_err("failed to get devices attached to VMD domain.\n");
2839 print_imsm_capability(&entry
->orom
);
2840 if (entry
->type
== SYS_DEV_NVME
) {
2841 for (hba
= list
; hba
; hba
= hba
->next
) {
2842 if (hba
->type
== SYS_DEV_NVME
)
2843 print_nvme_info(hba
);
2849 struct devid_list
*devid
;
2850 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2851 hba
= device_by_id(devid
->devid
);
2855 printf(" I/O Controller : %s (%s)\n",
2856 hba
->path
, get_sys_dev_type(hba
->type
));
2857 if (hba
->type
== SYS_DEV_SATA
|| hba
->type
== SYS_DEV_SATA_VMD
) {
2858 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2859 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2861 pr_err("failed to enumerate ports on %s controller at %s.\n",
2862 get_sys_dev_type(hba
->type
), hba
->pci_id
);
2873 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2875 struct sys_dev
*list
, *hba
;
2878 list
= find_intel_devices();
2881 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2886 for (hba
= list
; hba
; hba
= hba
->next
) {
2887 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2889 if (!find_imsm_capability(hba
) && verbose
> 0) {
2891 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2892 hba
->type
== SYS_DEV_VMD
|| hba
->type
== SYS_DEV_SATA_VMD
?
2893 vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2899 const struct orom_entry
*entry
;
2901 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2902 if (entry
->type
== SYS_DEV_VMD
|| entry
->type
== SYS_DEV_SATA_VMD
) {
2903 for (hba
= list
; hba
; hba
= hba
->next
)
2904 print_imsm_capability_export(&entry
->orom
);
2907 print_imsm_capability_export(&entry
->orom
);
2913 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2915 /* the imsm metadata format does not specify any host
2916 * identification information. We return -1 since we can never
2917 * confirm nor deny whether a given array is "meant" for this
2918 * host. We rely on compare_super and the 'family_num' fields to
2919 * exclude member disks that do not belong, and we rely on
2920 * mdadm.conf to specify the arrays that should be assembled.
2921 * Auto-assembly may still pick up "foreign" arrays.
2927 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2929 /* The uuid returned here is used for:
2930 * uuid to put into bitmap file (Create, Grow)
2931 * uuid for backup header when saving critical section (Grow)
2932 * comparing uuids when re-adding a device into an array
2933 * In these cases the uuid required is that of the data-array,
2934 * not the device-set.
2935 * uuid to recognise same set when adding a missing device back
2936 * to an array. This is a uuid for the device-set.
2938 * For each of these we can make do with a truncated
2939 * or hashed uuid rather than the original, as long as
2941 * In each case the uuid required is that of the data-array,
2942 * not the device-set.
2944 /* imsm does not track uuid's so we synthesis one using sha1 on
2945 * - The signature (Which is constant for all imsm array, but no matter)
2946 * - the orig_family_num of the container
2947 * - the index number of the volume
2948 * - the 'serial' number of the volume.
2949 * Hopefully these are all constant.
2951 struct intel_super
*super
= st
->sb
;
2954 struct sha1_ctx ctx
;
2955 struct imsm_dev
*dev
= NULL
;
2958 /* some mdadm versions failed to set ->orig_family_num, in which
2959 * case fall back to ->family_num. orig_family_num will be
2960 * fixed up with the first metadata update.
2962 family_num
= super
->anchor
->orig_family_num
;
2963 if (family_num
== 0)
2964 family_num
= super
->anchor
->family_num
;
2965 sha1_init_ctx(&ctx
);
2966 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2967 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2968 if (super
->current_vol
>= 0)
2969 dev
= get_imsm_dev(super
, super
->current_vol
);
2971 __u32 vol
= super
->current_vol
;
2972 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2973 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2975 sha1_finish_ctx(&ctx
, buf
);
2976 memcpy(uuid
, buf
, 4*4);
2979 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2981 /* migr_strip_size when repairing or initializing parity */
2982 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2983 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2985 switch (get_imsm_raid_level(map
)) {
2990 return 128*1024 >> 9;
2994 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2996 /* migr_strip_size when rebuilding a degraded disk, no idea why
2997 * this is different than migr_strip_size_resync(), but it's good
3000 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
3001 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
3003 switch (get_imsm_raid_level(map
)) {
3006 if (map
->num_members
% map
->num_domains
== 0)
3007 return 128*1024 >> 9;
3011 return max((__u32
) 64*1024 >> 9, chunk
);
3013 return 128*1024 >> 9;
3017 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
3019 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
3020 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
3021 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
3022 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
3024 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
3027 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
3029 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
3030 int level
= get_imsm_raid_level(lo
);
3032 if (level
== 1 || level
== 10) {
3033 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
3035 return hi
->num_domains
;
3037 return num_stripes_per_unit_resync(dev
);
3040 static unsigned long long calc_component_size(struct imsm_map
*map
,
3041 struct imsm_dev
*dev
)
3043 unsigned long long component_size
;
3044 unsigned long long dev_size
= imsm_dev_size(dev
);
3045 long long calc_dev_size
= 0;
3046 unsigned int member_disks
= imsm_num_data_members(map
);
3048 if (member_disks
== 0)
3051 component_size
= per_dev_array_size(map
);
3052 calc_dev_size
= component_size
* member_disks
;
3054 /* Component size is rounded to 1MB so difference between size from
3055 * metadata and size calculated from num_data_stripes equals up to
3056 * 2048 blocks per each device. If the difference is higher it means
3057 * that array size was expanded and num_data_stripes was not updated.
3059 if (llabs(calc_dev_size
- (long long)dev_size
) >
3060 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
3061 component_size
= dev_size
/ member_disks
;
3062 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
3063 component_size
/ map
->blocks_per_strip
,
3064 num_data_stripes(map
));
3067 return component_size
;
3070 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
3072 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3073 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
3075 switch(get_imsm_raid_level(map
)) {
3078 return chunk
* map
->num_domains
;
3080 return chunk
* map
->num_members
;
3086 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
3088 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
3089 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
3090 __u32 strip
= block
/ chunk
;
3092 switch (get_imsm_raid_level(map
)) {
3095 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
3096 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
3098 return vol_stripe
* chunk
+ block
% chunk
;
3100 __u32 stripe
= strip
/ (map
->num_members
- 1);
3102 return stripe
* chunk
+ block
% chunk
;
3109 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
3110 struct imsm_dev
*dev
)
3112 /* calculate the conversion factor between per member 'blocks'
3113 * (md/{resync,rebuild}_start) and imsm migration units, return
3114 * 0 for the 'not migrating' and 'unsupported migration' cases
3116 if (!dev
->vol
.migr_state
)
3119 switch (migr_type(dev
)) {
3120 case MIGR_GEN_MIGR
: {
3121 struct migr_record
*migr_rec
= super
->migr_rec
;
3122 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
3127 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3128 __u32 stripes_per_unit
;
3129 __u32 blocks_per_unit
;
3138 /* yes, this is really the translation of migr_units to
3139 * per-member blocks in the 'resync' case
3141 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
3142 migr_chunk
= migr_strip_blocks_resync(dev
);
3143 disks
= imsm_num_data_members(map
);
3144 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
3145 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
3146 segment
= blocks_per_unit
/ stripe
;
3147 block_rel
= blocks_per_unit
- segment
* stripe
;
3148 parity_depth
= parity_segment_depth(dev
);
3149 block_map
= map_migr_block(dev
, block_rel
);
3150 return block_map
+ parity_depth
* segment
;
3152 case MIGR_REBUILD
: {
3153 __u32 stripes_per_unit
;
3156 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
3157 migr_chunk
= migr_strip_blocks_rebuild(dev
);
3158 return migr_chunk
* stripes_per_unit
;
3160 case MIGR_STATE_CHANGE
:
3166 static int imsm_level_to_layout(int level
)
3174 return ALGORITHM_LEFT_ASYMMETRIC
;
3181 /*******************************************************************************
3182 * Function: read_imsm_migr_rec
3183 * Description: Function reads imsm migration record from last sector of disk
3185 * fd : disk descriptor
3186 * super : metadata info
3190 ******************************************************************************/
3191 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
3194 unsigned int sector_size
= super
->sector_size
;
3195 unsigned long long dsize
;
3197 get_dev_size(fd
, NULL
, &dsize
);
3198 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
3200 pr_err("Cannot seek to anchor block: %s\n",
3204 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
3205 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3206 MIGR_REC_BUF_SECTORS
*sector_size
) {
3207 pr_err("Cannot read migr record block: %s\n",
3212 if (sector_size
== 4096)
3213 convert_from_4k_imsm_migr_rec(super
);
3219 static struct imsm_dev
*imsm_get_device_during_migration(
3220 struct intel_super
*super
)
3223 struct intel_dev
*dv
;
3225 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
3226 if (is_gen_migration(dv
->dev
))
3232 /*******************************************************************************
3233 * Function: load_imsm_migr_rec
3234 * Description: Function reads imsm migration record (it is stored at the last
3237 * super : imsm internal array info
3241 * -2 : no migration in progress
3242 ******************************************************************************/
3243 static int load_imsm_migr_rec(struct intel_super
*super
)
3249 struct imsm_dev
*dev
;
3250 struct imsm_map
*map
;
3254 /* find map under migration */
3255 dev
= imsm_get_device_during_migration(super
);
3256 /* nothing to load,no migration in progress?
3261 map
= get_imsm_map(dev
, MAP_0
);
3265 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3266 /* skip spare and failed disks
3270 /* read only from one of the first two slots
3272 slot
= get_imsm_disk_slot(map
, dl
->index
);
3273 if (slot
> 1 || slot
< 0)
3276 if (!is_fd_valid(dl
->fd
)) {
3277 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3278 fd
= dev_open(nm
, O_RDONLY
);
3280 if (is_fd_valid(fd
)) {
3290 if (!is_fd_valid(fd
))
3292 retval
= read_imsm_migr_rec(fd
, super
);
3299 /*******************************************************************************
3300 * function: imsm_create_metadata_checkpoint_update
3301 * Description: It creates update for checkpoint change.
3303 * super : imsm internal array info
3304 * u : pointer to prepared update
3307 * If length is equal to 0, input pointer u contains no update
3308 ******************************************************************************/
3309 static int imsm_create_metadata_checkpoint_update(
3310 struct intel_super
*super
,
3311 struct imsm_update_general_migration_checkpoint
**u
)
3314 int update_memory_size
= 0;
3316 dprintf("(enter)\n");
3322 /* size of all update data without anchor */
3323 update_memory_size
=
3324 sizeof(struct imsm_update_general_migration_checkpoint
);
3326 *u
= xcalloc(1, update_memory_size
);
3328 dprintf("error: cannot get memory\n");
3331 (*u
)->type
= update_general_migration_checkpoint
;
3332 (*u
)->curr_migr_unit
= current_migr_unit(super
->migr_rec
);
3333 dprintf("prepared for %llu\n", (unsigned long long)(*u
)->curr_migr_unit
);
3335 return update_memory_size
;
3338 static void imsm_update_metadata_locally(struct supertype
*st
,
3339 void *buf
, int len
);
3341 /*******************************************************************************
3342 * Function: write_imsm_migr_rec
3343 * Description: Function writes imsm migration record
3344 * (at the last sector of disk)
3346 * super : imsm internal array info
3350 ******************************************************************************/
3351 static int write_imsm_migr_rec(struct supertype
*st
)
3353 struct intel_super
*super
= st
->sb
;
3354 unsigned int sector_size
= super
->sector_size
;
3355 unsigned long long dsize
;
3359 struct imsm_update_general_migration_checkpoint
*u
;
3360 struct imsm_dev
*dev
;
3361 struct imsm_map
*map
;
3363 /* find map under migration */
3364 dev
= imsm_get_device_during_migration(super
);
3365 /* if no migration, write buffer anyway to clear migr_record
3366 * on disk based on first available device
3369 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3370 super
->current_vol
);
3372 map
= get_imsm_map(dev
, MAP_0
);
3374 if (sector_size
== 4096)
3375 convert_to_4k_imsm_migr_rec(super
);
3376 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3379 /* skip failed and spare devices */
3382 /* write to 2 first slots only */
3384 slot
= get_imsm_disk_slot(map
, sd
->index
);
3385 if (map
== NULL
|| slot
> 1 || slot
< 0)
3388 get_dev_size(sd
->fd
, NULL
, &dsize
);
3389 if (lseek64(sd
->fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*
3392 pr_err("Cannot seek to anchor block: %s\n",
3396 if ((unsigned int)write(sd
->fd
, super
->migr_rec_buf
,
3397 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3398 MIGR_REC_BUF_SECTORS
*sector_size
) {
3399 pr_err("Cannot write migr record block: %s\n",
3404 if (sector_size
== 4096)
3405 convert_from_4k_imsm_migr_rec(super
);
3406 /* update checkpoint information in metadata */
3407 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3409 dprintf("imsm: Cannot prepare update\n");
3412 /* update metadata locally */
3413 imsm_update_metadata_locally(st
, u
, len
);
3414 /* and possibly remotely */
3415 if (st
->update_tail
) {
3416 append_metadata_update(st
, u
, len
);
3417 /* during reshape we do all work inside metadata handler
3418 * manage_reshape(), so metadata update has to be triggered
3421 flush_metadata_updates(st
);
3422 st
->update_tail
= &st
->updates
;
3431 /* spare/missing disks activations are not allowe when
3432 * array/container performs reshape operation, because
3433 * all arrays in container works on the same disks set
3435 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3438 struct intel_dev
*i_dev
;
3439 struct imsm_dev
*dev
;
3441 /* check whole container
3443 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3445 if (is_gen_migration(dev
)) {
3446 /* No repair during any migration in container
3454 static unsigned long long imsm_component_size_alignment_check(int level
,
3456 unsigned int sector_size
,
3457 unsigned long long component_size
)
3459 unsigned int component_size_alignment
;
3461 /* check component size alignment
3463 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3465 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3466 level
, chunk_size
, component_size
,
3467 component_size_alignment
);
3469 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3470 dprintf("imsm: reported component size aligned from %llu ",
3472 component_size
-= component_size_alignment
;
3473 dprintf_cont("to %llu (%i).\n",
3474 component_size
, component_size_alignment
);
3477 return component_size
;
3480 /*******************************************************************************
3481 * Function: get_bitmap_header_sector
3482 * Description: Returns the sector where the bitmap header is placed.
3484 * st : supertype information
3485 * dev_idx : index of the device with bitmap
3488 * The sector where the bitmap header is placed
3489 ******************************************************************************/
3490 static unsigned long long get_bitmap_header_sector(struct intel_super
*super
,
3493 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3494 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3496 if (!super
->sector_size
) {
3497 dprintf("sector size is not set\n");
3501 return pba_of_lba0(map
) + calc_component_size(map
, dev
) +
3502 (IMSM_BITMAP_HEADER_OFFSET
/ super
->sector_size
);
3505 /*******************************************************************************
3506 * Function: get_bitmap_sector
3507 * Description: Returns the sector where the bitmap is placed.
3509 * st : supertype information
3510 * dev_idx : index of the device with bitmap
3513 * The sector where the bitmap is placed
3514 ******************************************************************************/
3515 static unsigned long long get_bitmap_sector(struct intel_super
*super
,
3518 if (!super
->sector_size
) {
3519 dprintf("sector size is not set\n");
3523 return get_bitmap_header_sector(super
, dev_idx
) +
3524 (IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
);
3527 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3529 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3530 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3532 return pba_of_lba0(map
) +
3533 (num_data_stripes(map
) * map
->blocks_per_strip
);
3536 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3538 struct intel_super
*super
= st
->sb
;
3539 struct migr_record
*migr_rec
= super
->migr_rec
;
3540 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3541 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3542 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3543 struct imsm_map
*map_to_analyse
= map
;
3545 int map_disks
= info
->array
.raid_disks
;
3547 memset(info
, 0, sizeof(*info
));
3549 map_to_analyse
= prev_map
;
3551 dl
= super
->current_disk
;
3553 info
->container_member
= super
->current_vol
;
3554 info
->array
.raid_disks
= map
->num_members
;
3555 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3556 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3557 info
->array
.md_minor
= -1;
3558 info
->array
.ctime
= 0;
3559 info
->array
.utime
= 0;
3560 info
->array
.chunk_size
=
3561 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3562 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3563 info
->custom_array_size
= imsm_dev_size(dev
);
3564 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3566 if (is_gen_migration(dev
)) {
3568 * device prev_map should be added if it is in the middle
3573 info
->reshape_active
= 1;
3574 info
->new_level
= get_imsm_raid_level(map
);
3575 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3576 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3577 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3578 if (info
->delta_disks
) {
3579 /* this needs to be applied to every array
3582 info
->reshape_active
= CONTAINER_RESHAPE
;
3584 /* We shape information that we give to md might have to be
3585 * modify to cope with md's requirement for reshaping arrays.
3586 * For example, when reshaping a RAID0, md requires it to be
3587 * presented as a degraded RAID4.
3588 * Also if a RAID0 is migrating to a RAID5 we need to specify
3589 * the array as already being RAID5, but the 'before' layout
3590 * is a RAID4-like layout.
3592 switch (info
->array
.level
) {
3594 switch(info
->new_level
) {
3596 /* conversion is happening as RAID4 */
3597 info
->array
.level
= 4;
3598 info
->array
.raid_disks
+= 1;
3601 /* conversion is happening as RAID5 */
3602 info
->array
.level
= 5;
3603 info
->array
.layout
= ALGORITHM_PARITY_N
;
3604 info
->delta_disks
-= 1;
3607 /* FIXME error message */
3608 info
->array
.level
= UnSet
;
3614 info
->new_level
= UnSet
;
3615 info
->new_layout
= UnSet
;
3616 info
->new_chunk
= info
->array
.chunk_size
;
3617 info
->delta_disks
= 0;
3621 info
->disk
.major
= dl
->major
;
3622 info
->disk
.minor
= dl
->minor
;
3623 info
->disk
.number
= dl
->index
;
3624 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3628 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3629 info
->component_size
= calc_component_size(map
, dev
);
3630 info
->component_size
= imsm_component_size_alignment_check(
3632 info
->array
.chunk_size
,
3634 info
->component_size
);
3635 info
->bb
.supported
= 1;
3637 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3638 info
->recovery_start
= MaxSector
;
3640 if (info
->array
.level
== 5 &&
3641 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3642 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3643 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3644 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3645 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3646 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3648 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3650 } else if (info
->array
.level
<= 0) {
3651 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3653 if (dev
->rwh_policy
== RWH_BITMAP
) {
3654 info
->bitmap_offset
= get_bitmap_sector(super
, super
->current_vol
);
3655 info
->consistency_policy
= CONSISTENCY_POLICY_BITMAP
;
3657 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3661 info
->reshape_progress
= 0;
3662 info
->resync_start
= MaxSector
;
3663 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3664 !(info
->array
.state
& 1)) &&
3665 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3666 info
->resync_start
= 0;
3668 if (dev
->vol
.migr_state
) {
3669 switch (migr_type(dev
)) {
3672 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3674 __u64 units
= vol_curr_migr_unit(dev
);
3676 info
->resync_start
= blocks_per_unit
* units
;
3679 case MIGR_GEN_MIGR
: {
3680 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3682 __u64 units
= current_migr_unit(migr_rec
);
3685 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3687 (get_num_migr_units(migr_rec
)-1)) &&
3688 (super
->migr_rec
->rec_status
==
3689 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3692 info
->reshape_progress
= blocks_per_unit
* units
;
3694 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3695 (unsigned long long)units
,
3696 (unsigned long long)blocks_per_unit
,
3697 info
->reshape_progress
);
3699 used_disks
= imsm_num_data_members(prev_map
);
3700 if (used_disks
> 0) {
3701 info
->custom_array_size
= per_dev_array_size(map
) *
3706 /* we could emulate the checkpointing of
3707 * 'sync_action=check' migrations, but for now
3708 * we just immediately complete them
3711 /* this is handled by container_content_imsm() */
3712 case MIGR_STATE_CHANGE
:
3713 /* FIXME handle other migrations */
3715 /* we are not dirty, so... */
3716 info
->resync_start
= MaxSector
;
3720 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3721 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3723 info
->array
.major_version
= -1;
3724 info
->array
.minor_version
= -2;
3725 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3726 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3727 uuid_from_super_imsm(st
, info
->uuid
);
3731 for (i
=0; i
<map_disks
; i
++) {
3733 if (i
< info
->array
.raid_disks
) {
3734 struct imsm_disk
*dsk
;
3735 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3736 dsk
= get_imsm_disk(super
, j
);
3737 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3744 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3745 int failed
, int look_in_map
);
3747 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3750 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3752 if (is_gen_migration(dev
)) {
3755 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3757 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3758 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3759 if (map2
->map_state
!= map_state
) {
3760 map2
->map_state
= map_state
;
3761 super
->updates_pending
++;
3766 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3770 for (d
= super
->missing
; d
; d
= d
->next
)
3771 if (d
->index
== index
)
3776 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3778 struct intel_super
*super
= st
->sb
;
3779 struct imsm_disk
*disk
;
3780 int map_disks
= info
->array
.raid_disks
;
3782 struct imsm_super
*mpb
;
3784 if (super
->current_vol
>= 0) {
3785 getinfo_super_imsm_volume(st
, info
, map
);
3788 memset(info
, 0, sizeof(*info
));
3790 /* Set raid_disks to zero so that Assemble will always pull in valid
3793 info
->array
.raid_disks
= 0;
3794 info
->array
.level
= LEVEL_CONTAINER
;
3795 info
->array
.layout
= 0;
3796 info
->array
.md_minor
= -1;
3797 info
->array
.ctime
= 0; /* N/A for imsm */
3798 info
->array
.utime
= 0;
3799 info
->array
.chunk_size
= 0;
3801 info
->disk
.major
= 0;
3802 info
->disk
.minor
= 0;
3803 info
->disk
.raid_disk
= -1;
3804 info
->reshape_active
= 0;
3805 info
->array
.major_version
= -1;
3806 info
->array
.minor_version
= -2;
3807 strcpy(info
->text_version
, "imsm");
3808 info
->safe_mode_delay
= 0;
3809 info
->disk
.number
= -1;
3810 info
->disk
.state
= 0;
3812 info
->recovery_start
= MaxSector
;
3813 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3814 info
->bb
.supported
= 1;
3816 /* do we have the all the insync disks that we expect? */
3817 mpb
= super
->anchor
;
3818 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3820 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3821 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3823 struct imsm_map
*map
;
3825 map
= get_imsm_map(dev
, MAP_0
);
3827 /* any newly missing disks?
3828 * (catches single-degraded vs double-degraded)
3830 for (j
= 0; j
< map
->num_members
; j
++) {
3831 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3832 __u32 idx
= ord_to_idx(ord
);
3834 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3835 info
->disk
.raid_disk
= j
;
3837 if (!(ord
& IMSM_ORD_REBUILD
) &&
3838 get_imsm_missing(super
, idx
)) {
3845 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3847 disk
= &super
->disks
->disk
;
3848 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3849 info
->component_size
= reserved
;
3850 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3851 /* we don't change info->disk.raid_disk here because
3852 * this state will be finalized in mdmon after we have
3853 * found the 'most fresh' version of the metadata
3855 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3856 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3857 0 : (1 << MD_DISK_SYNC
);
3860 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3861 * ->compare_super may have updated the 'num_raid_devs' field for spares
3863 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3864 uuid_from_super_imsm(st
, info
->uuid
);
3866 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3868 /* I don't know how to compute 'map' on imsm, so use safe default */
3871 for (i
= 0; i
< map_disks
; i
++)
3877 /* allocates memory and fills disk in mdinfo structure
3878 * for each disk in array */
3879 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3881 struct mdinfo
*mddev
;
3882 struct intel_super
*super
= st
->sb
;
3883 struct imsm_disk
*disk
;
3886 if (!super
|| !super
->disks
)
3889 mddev
= xcalloc(1, sizeof(*mddev
));
3893 tmp
= xcalloc(1, sizeof(*tmp
));
3895 tmp
->next
= mddev
->devs
;
3897 tmp
->disk
.number
= count
++;
3898 tmp
->disk
.major
= dl
->major
;
3899 tmp
->disk
.minor
= dl
->minor
;
3900 tmp
->disk
.state
= is_configured(disk
) ?
3901 (1 << MD_DISK_ACTIVE
) : 0;
3902 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3903 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3904 tmp
->disk
.raid_disk
= -1;
3910 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3911 enum update_opt update
, char *devname
,
3912 int verbose
, int uuid_set
, char *homehost
)
3914 /* For 'assemble' and 'force' we need to return non-zero if any
3915 * change was made. For others, the return value is ignored.
3916 * Update options are:
3917 * force-one : This device looks a bit old but needs to be included,
3918 * update age info appropriately.
3919 * assemble: clear any 'faulty' flag to allow this device to
3921 * force-array: Array is degraded but being forced, mark it clean
3922 * if that will be needed to assemble it.
3924 * newdev: not used ????
3925 * grow: Array has gained a new device - this is currently for
3927 * resync: mark as dirty so a resync will happen.
3928 * name: update the name - preserving the homehost
3929 * uuid: Change the uuid of the array to match watch is given
3931 * Following are not relevant for this imsm:
3932 * sparc2.2 : update from old dodgey metadata
3933 * super-minor: change the preferred_minor number
3934 * summaries: update redundant counters.
3935 * homehost: update the recorded homehost
3936 * _reshape_progress: record new reshape_progress position.
3939 struct intel_super
*super
= st
->sb
;
3940 struct imsm_super
*mpb
;
3942 /* we can only update container info */
3943 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3946 mpb
= super
->anchor
;
3950 /* We take this to mean that the family_num should be updated.
3951 * However that is much smaller than the uuid so we cannot really
3952 * allow an explicit uuid to be given. And it is hard to reliably
3954 * So if !uuid_set we know the current uuid is random and just used
3955 * the first 'int' and copy it to the other 3 positions.
3956 * Otherwise we require the 4 'int's to be the same as would be the
3957 * case if we are using a random uuid. So an explicit uuid will be
3958 * accepted as long as all for ints are the same... which shouldn't hurt
3961 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3964 if (info
->uuid
[0] != info
->uuid
[1] ||
3965 info
->uuid
[1] != info
->uuid
[2] ||
3966 info
->uuid
[2] != info
->uuid
[3])
3972 mpb
->orig_family_num
= info
->uuid
[0];
3974 case UOPT_SPEC_ASSEMBLE
:
3982 /* successful update? recompute checksum */
3984 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3989 static size_t disks_to_mpb_size(int disks
)
3993 size
= sizeof(struct imsm_super
);
3994 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3995 size
+= 2 * sizeof(struct imsm_dev
);
3996 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3997 size
+= (4 - 2) * sizeof(struct imsm_map
);
3998 /* 4 possible disk_ord_tbl's */
3999 size
+= 4 * (disks
- 1) * sizeof(__u32
);
4000 /* maximum bbm log */
4001 size
+= sizeof(struct bbm_log
);
4006 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
4007 unsigned long long data_offset
)
4009 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
4012 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
4015 static void free_devlist(struct intel_super
*super
)
4017 struct intel_dev
*dv
;
4019 while (super
->devlist
) {
4020 dv
= super
->devlist
->next
;
4021 free(super
->devlist
->dev
);
4022 free(super
->devlist
);
4023 super
->devlist
= dv
;
4027 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
4029 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
4032 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
,
4036 * 0 same, or first was empty, and second was copied
4037 * 1 sb are different
4039 struct intel_super
*first
= st
->sb
;
4040 struct intel_super
*sec
= tst
->sb
;
4048 /* in platform dependent environment test if the disks
4049 * use the same Intel hba
4050 * if not on Intel hba at all, allow anything.
4051 * doesn't check HBAs if num_raid_devs is not set, as it means
4052 * it is a free floating spare, and all spares regardless of HBA type
4053 * will fall into separate container during the assembly
4055 if (first
->hba
&& sec
->hba
&& first
->anchor
->num_raid_devs
!= 0) {
4056 if (first
->hba
->type
!= sec
->hba
->type
) {
4058 pr_err("HBAs of devices do not match %s != %s\n",
4059 get_sys_dev_type(first
->hba
->type
),
4060 get_sys_dev_type(sec
->hba
->type
));
4063 if (first
->orom
!= sec
->orom
) {
4065 pr_err("HBAs of devices do not match %s != %s\n",
4066 first
->hba
->pci_id
, sec
->hba
->pci_id
);
4071 if (first
->anchor
->num_raid_devs
> 0 &&
4072 sec
->anchor
->num_raid_devs
> 0) {
4073 /* Determine if these disks might ever have been
4074 * related. Further disambiguation can only take place
4075 * in load_super_imsm_all
4077 __u32 first_family
= first
->anchor
->orig_family_num
;
4078 __u32 sec_family
= sec
->anchor
->orig_family_num
;
4080 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
4081 MAX_SIGNATURE_LENGTH
) != 0)
4084 if (first_family
== 0)
4085 first_family
= first
->anchor
->family_num
;
4086 if (sec_family
== 0)
4087 sec_family
= sec
->anchor
->family_num
;
4089 if (first_family
!= sec_family
)
4094 /* if an anchor does not have num_raid_devs set then it is a free
4095 * floating spare. don't assosiate spare with any array, as during assembly
4096 * spares shall fall into separate container, from which they can be moved
4099 if (first
->anchor
->num_raid_devs
^ sec
->anchor
->num_raid_devs
)
4105 static void fd2devname(int fd
, char *name
)
4113 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
4116 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
4118 char path
[PATH_MAX
];
4119 char *name
= fd2kname(fd
);
4124 if (strncmp(name
, "nvme", 4) != 0)
4127 if (!diskfd_to_devpath(fd
, 1, path
))
4130 return devpath_to_char(path
, "serial", buf
, buf_len
, 0);
4133 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
4135 static int imsm_read_serial(int fd
, char *devname
,
4136 __u8
*serial
, size_t serial_buf_len
)
4145 memset(buf
, 0, sizeof(buf
));
4147 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
4150 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
4152 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
4153 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
4154 fd2devname(fd
, (char *) serial
);
4160 pr_err("Failed to retrieve serial for %s\n",
4165 /* trim all whitespace and non-printable characters and convert
4168 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4171 /* ':' is reserved for use in placeholder serial
4172 * numbers for missing disks
4183 if (len
> serial_buf_len
) {
4184 /* truncate leading characters */
4185 dest
+= len
- serial_buf_len
;
4186 len
= serial_buf_len
;
4189 memset(serial
, 0, serial_buf_len
);
4190 memcpy(serial
, dest
, len
);
4195 static int serialcmp(__u8
*s1
, __u8
*s2
)
4197 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4200 static void serialcpy(__u8
*dest
, __u8
*src
)
4202 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4205 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4209 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4210 if (serialcmp(dl
->serial
, serial
) == 0)
4216 static struct imsm_disk
*
4217 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4221 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4222 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4224 if (serialcmp(disk
->serial
, serial
) == 0) {
4235 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4237 struct imsm_disk
*disk
;
4242 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4244 rv
= imsm_read_serial(fd
, devname
, serial
, MAX_RAID_SERIAL_LEN
);
4249 dl
= xcalloc(1, sizeof(*dl
));
4252 dl
->major
= major(stb
.st_rdev
);
4253 dl
->minor
= minor(stb
.st_rdev
);
4254 dl
->next
= super
->disks
;
4255 dl
->fd
= keep_fd
? fd
: -1;
4256 assert(super
->disks
== NULL
);
4258 serialcpy(dl
->serial
, serial
);
4261 fd2devname(fd
, name
);
4263 dl
->devname
= xstrdup(devname
);
4265 dl
->devname
= xstrdup(name
);
4267 /* look up this disk's index in the current anchor */
4268 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4271 /* only set index on disks that are a member of a
4272 * populated contianer, i.e. one with raid_devs
4274 if (is_failed(&dl
->disk
))
4276 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4283 /* When migrating map0 contains the 'destination' state while map1
4284 * contains the current state. When not migrating map0 contains the
4285 * current state. This routine assumes that map[0].map_state is set to
4286 * the current array state before being called.
4288 * Migration is indicated by one of the following states
4289 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4290 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4291 * map1state=unitialized)
4292 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4294 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4295 * map1state=degraded)
4296 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4299 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4300 __u8 to_state
, int migr_type
)
4302 struct imsm_map
*dest
;
4303 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4305 dev
->vol
.migr_state
= 1;
4306 set_migr_type(dev
, migr_type
);
4307 set_vol_curr_migr_unit(dev
, 0);
4308 dest
= get_imsm_map(dev
, MAP_1
);
4310 /* duplicate and then set the target end state in map[0] */
4311 memcpy(dest
, src
, sizeof_imsm_map(src
));
4312 if (migr_type
== MIGR_GEN_MIGR
) {
4316 for (i
= 0; i
< src
->num_members
; i
++) {
4317 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4318 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4322 if (migr_type
== MIGR_GEN_MIGR
)
4323 /* Clear migration record */
4324 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4326 src
->map_state
= to_state
;
4329 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4332 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4333 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4337 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4338 * completed in the last migration.
4340 * FIXME add support for raid-level-migration
4342 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == false) &&
4343 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4344 /* when final map state is other than expected
4345 * merge maps (not for migration)
4349 for (i
= 0; i
< prev
->num_members
; i
++)
4350 for (j
= 0; j
< map
->num_members
; j
++)
4351 /* during online capacity expansion
4352 * disks position can be changed
4353 * if takeover is used
4355 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4356 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4357 map
->disk_ord_tbl
[j
] |=
4358 prev
->disk_ord_tbl
[i
];
4361 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4362 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4365 dev
->vol
.migr_state
= 0;
4366 set_migr_type(dev
, 0);
4367 set_vol_curr_migr_unit(dev
, 0);
4368 map
->map_state
= map_state
;
4371 static int parse_raid_devices(struct intel_super
*super
)
4374 struct imsm_dev
*dev_new
;
4375 size_t len
, len_migr
;
4377 size_t space_needed
= 0;
4378 struct imsm_super
*mpb
= super
->anchor
;
4380 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4381 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4382 struct intel_dev
*dv
;
4384 len
= sizeof_imsm_dev(dev_iter
, 0);
4385 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4387 space_needed
+= len_migr
- len
;
4389 dv
= xmalloc(sizeof(*dv
));
4390 if (max_len
< len_migr
)
4392 if (max_len
> len_migr
)
4393 space_needed
+= max_len
- len_migr
;
4394 dev_new
= xmalloc(max_len
);
4395 imsm_copy_dev(dev_new
, dev_iter
);
4398 dv
->next
= super
->devlist
;
4399 super
->devlist
= dv
;
4402 /* ensure that super->buf is large enough when all raid devices
4405 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4408 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4409 super
->sector_size
);
4410 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4413 memcpy(buf
, super
->buf
, super
->len
);
4414 memset(buf
+ super
->len
, 0, len
- super
->len
);
4420 super
->extra_space
+= space_needed
;
4425 /*******************************************************************************
4426 * Function: check_mpb_migr_compatibility
4427 * Description: Function checks for unsupported migration features:
4428 * - migration optimization area (pba_of_lba0)
4429 * - descending reshape (ascending_migr)
4431 * super : imsm metadata information
4433 * 0 : migration is compatible
4434 * -1 : migration is not compatible
4435 ******************************************************************************/
4436 int check_mpb_migr_compatibility(struct intel_super
*super
)
4438 struct imsm_map
*map0
, *map1
;
4439 struct migr_record
*migr_rec
= super
->migr_rec
;
4442 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4443 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4445 if (dev_iter
->vol
.migr_state
== 1 &&
4446 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4447 /* This device is migrating */
4448 map0
= get_imsm_map(dev_iter
, MAP_0
);
4449 map1
= get_imsm_map(dev_iter
, MAP_1
);
4450 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4451 /* migration optimization area was used */
4453 if (migr_rec
->ascending_migr
== 0 &&
4454 migr_rec
->dest_depth_per_unit
> 0)
4455 /* descending reshape not supported yet */
4462 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4464 /* load_imsm_mpb - read matrix metadata
4465 * allocates super->mpb to be freed by free_imsm
4467 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4469 unsigned long long dsize
;
4470 unsigned long long sectors
;
4471 unsigned int sector_size
= super
->sector_size
;
4473 struct imsm_super
*anchor
;
4476 get_dev_size(fd
, NULL
, &dsize
);
4477 if (dsize
< 2*sector_size
) {
4479 pr_err("%s: device to small for imsm\n",
4484 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4486 pr_err("Cannot seek to anchor block on %s: %s\n",
4487 devname
, strerror(errno
));
4491 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4493 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4496 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4498 pr_err("Cannot read anchor block on %s: %s\n",
4499 devname
, strerror(errno
));
4504 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4506 pr_err("no IMSM anchor on %s\n", devname
);
4511 __free_imsm(super
, 0);
4512 /* reload capability and hba */
4514 /* capability and hba must be updated with new super allocation */
4515 find_intel_hba_capability(fd
, super
, devname
);
4516 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4517 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4519 pr_err("unable to allocate %zu byte mpb buffer\n",
4524 memcpy(super
->buf
, anchor
, sector_size
);
4526 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4529 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4530 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4531 pr_err("could not allocate migr_rec buffer\n");
4536 super
->clean_migration_record_by_mdmon
= 0;
4539 check_sum
= __gen_imsm_checksum(super
->anchor
);
4540 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4542 pr_err("IMSM checksum %x != %x on %s\n",
4544 __le32_to_cpu(super
->anchor
->check_sum
),
4552 /* read the extended mpb */
4553 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4555 pr_err("Cannot seek to extended mpb on %s: %s\n",
4556 devname
, strerror(errno
));
4560 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4561 super
->len
- sector_size
) != super
->len
- sector_size
) {
4563 pr_err("Cannot read extended mpb on %s: %s\n",
4564 devname
, strerror(errno
));
4568 check_sum
= __gen_imsm_checksum(super
->anchor
);
4569 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4571 pr_err("IMSM checksum %x != %x on %s\n",
4572 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4580 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4582 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4583 static void clear_hi(struct intel_super
*super
)
4585 struct imsm_super
*mpb
= super
->anchor
;
4587 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4589 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4590 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4591 disk
->total_blocks_hi
= 0;
4593 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4594 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4595 for (n
= 0; n
< 2; ++n
) {
4596 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4599 map
->pba_of_lba0_hi
= 0;
4600 map
->blocks_per_member_hi
= 0;
4601 map
->num_data_stripes_hi
= 0;
4607 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4611 err
= load_imsm_mpb(fd
, super
, devname
);
4614 if (super
->sector_size
== 4096)
4615 convert_from_4k(super
);
4616 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4619 err
= parse_raid_devices(super
);
4622 err
= load_bbm_log(super
);
4627 static void __free_imsm_disk(struct dl
*d
, int do_close
)
4639 static void free_imsm_disks(struct intel_super
*super
)
4643 while (super
->disks
) {
4645 super
->disks
= d
->next
;
4646 __free_imsm_disk(d
, 1);
4648 while (super
->disk_mgmt_list
) {
4649 d
= super
->disk_mgmt_list
;
4650 super
->disk_mgmt_list
= d
->next
;
4651 __free_imsm_disk(d
, 1);
4653 while (super
->missing
) {
4655 super
->missing
= d
->next
;
4656 __free_imsm_disk(d
, 1);
4661 /* free all the pieces hanging off of a super pointer */
4662 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4664 struct intel_hba
*elem
, *next
;
4670 /* unlink capability description */
4672 if (super
->migr_rec_buf
) {
4673 free(super
->migr_rec_buf
);
4674 super
->migr_rec_buf
= NULL
;
4677 free_imsm_disks(super
);
4678 free_devlist(super
);
4682 free((void *)elem
->path
);
4688 free(super
->bbm_log
);
4692 static void free_imsm(struct intel_super
*super
)
4694 __free_imsm(super
, 1);
4695 free(super
->bb
.entries
);
4699 static void free_super_imsm(struct supertype
*st
)
4701 struct intel_super
*super
= st
->sb
;
4710 static struct intel_super
*alloc_super(void)
4712 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4714 super
->current_vol
= -1;
4715 super
->create_offset
= ~((unsigned long long) 0);
4717 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4718 sizeof(struct md_bb_entry
));
4719 if (!super
->bb
.entries
) {
4728 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4730 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4732 struct sys_dev
*hba_name
;
4735 if (is_fd_valid(fd
) && test_partition(fd
)) {
4736 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4740 if (!is_fd_valid(fd
) || check_no_platform()) {
4745 hba_name
= find_disk_attached_hba(fd
, NULL
);
4748 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4752 rv
= attach_hba_to_super(super
, hba_name
);
4755 struct intel_hba
*hba
= super
->hba
;
4757 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4758 " but the container is assigned to Intel(R) %s %s (",
4760 get_sys_dev_type(hba_name
->type
),
4761 hba_name
->type
== SYS_DEV_VMD
|| hba_name
->type
== SYS_DEV_SATA_VMD
?
4762 "domain" : "RAID controller",
4763 hba_name
->pci_id
? : "Err!",
4764 get_sys_dev_type(super
->hba
->type
),
4765 hba
->type
== SYS_DEV_VMD
|| hba_name
->type
== SYS_DEV_SATA_VMD
?
4766 "domain" : "RAID controller");
4769 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4771 fprintf(stderr
, ", ");
4774 fprintf(stderr
, ").\n"
4775 " Mixing devices attached to different controllers is not allowed.\n");
4779 super
->orom
= find_imsm_capability(hba_name
);
4786 /* find_missing - helper routine for load_super_imsm_all that identifies
4787 * disks that have disappeared from the system. This routine relies on
4788 * the mpb being uptodate, which it is at load time.
4790 static int find_missing(struct intel_super
*super
)
4793 struct imsm_super
*mpb
= super
->anchor
;
4795 struct imsm_disk
*disk
;
4797 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4798 disk
= __get_imsm_disk(mpb
, i
);
4799 dl
= serial_to_dl(disk
->serial
, super
);
4803 dl
= xmalloc(sizeof(*dl
));
4807 dl
->devname
= xstrdup("missing");
4809 serialcpy(dl
->serial
, disk
->serial
);
4812 dl
->next
= super
->missing
;
4813 super
->missing
= dl
;
4819 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4821 struct intel_disk
*idisk
= disk_list
;
4824 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4826 idisk
= idisk
->next
;
4832 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4833 struct intel_super
*super
,
4834 struct intel_disk
**disk_list
)
4836 struct imsm_disk
*d
= &super
->disks
->disk
;
4837 struct imsm_super
*mpb
= super
->anchor
;
4840 for (i
= 0; i
< tbl_size
; i
++) {
4841 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4842 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4844 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4845 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4846 dprintf("mpb from %d:%d matches %d:%d\n",
4847 super
->disks
->major
,
4848 super
->disks
->minor
,
4849 table
[i
]->disks
->major
,
4850 table
[i
]->disks
->minor
);
4854 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4855 is_configured(d
) == is_configured(tbl_d
)) &&
4856 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4857 /* current version of the mpb is a
4858 * better candidate than the one in
4859 * super_table, but copy over "cross
4860 * generational" status
4862 struct intel_disk
*idisk
;
4864 dprintf("mpb from %d:%d replaces %d:%d\n",
4865 super
->disks
->major
,
4866 super
->disks
->minor
,
4867 table
[i
]->disks
->major
,
4868 table
[i
]->disks
->minor
);
4870 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4871 if (idisk
&& is_failed(&idisk
->disk
))
4872 tbl_d
->status
|= FAILED_DISK
;
4875 struct intel_disk
*idisk
;
4876 struct imsm_disk
*disk
;
4878 /* tbl_mpb is more up to date, but copy
4879 * over cross generational status before
4882 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4883 if (disk
&& is_failed(disk
))
4884 d
->status
|= FAILED_DISK
;
4886 idisk
= disk_list_get(d
->serial
, *disk_list
);
4889 if (disk
&& is_configured(disk
))
4890 idisk
->disk
.status
|= CONFIGURED_DISK
;
4893 dprintf("mpb from %d:%d prefer %d:%d\n",
4894 super
->disks
->major
,
4895 super
->disks
->minor
,
4896 table
[i
]->disks
->major
,
4897 table
[i
]->disks
->minor
);
4905 table
[tbl_size
++] = super
;
4909 /* update/extend the merged list of imsm_disk records */
4910 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4911 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4912 struct intel_disk
*idisk
;
4914 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4916 idisk
->disk
.status
|= disk
->status
;
4917 if (is_configured(&idisk
->disk
) ||
4918 is_failed(&idisk
->disk
))
4919 idisk
->disk
.status
&= ~(SPARE_DISK
);
4921 idisk
= xcalloc(1, sizeof(*idisk
));
4922 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4923 idisk
->disk
= *disk
;
4924 idisk
->next
= *disk_list
;
4928 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4935 static struct intel_super
*
4936 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4939 struct imsm_super
*mpb
= super
->anchor
;
4943 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4944 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4945 struct intel_disk
*idisk
;
4947 idisk
= disk_list_get(disk
->serial
, disk_list
);
4949 if (idisk
->owner
== owner
||
4950 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4953 dprintf("'%.16s' owner %d != %d\n",
4954 disk
->serial
, idisk
->owner
,
4957 dprintf("unknown disk %x [%d]: %.16s\n",
4958 __le32_to_cpu(mpb
->family_num
), i
,
4964 if (ok_count
== mpb
->num_disks
)
4969 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4971 struct intel_super
*s
;
4973 for (s
= super_list
; s
; s
= s
->next
) {
4974 if (family_num
!= s
->anchor
->family_num
)
4976 pr_err("Conflict, offlining family %#x on '%s'\n",
4977 __le32_to_cpu(family_num
), s
->disks
->devname
);
4981 static struct intel_super
*
4982 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4984 struct intel_super
*super_table
[len
];
4985 struct intel_disk
*disk_list
= NULL
;
4986 struct intel_super
*champion
, *spare
;
4987 struct intel_super
*s
, **del
;
4992 memset(super_table
, 0, sizeof(super_table
));
4993 for (s
= *super_list
; s
; s
= s
->next
)
4994 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4996 for (i
= 0; i
< tbl_size
; i
++) {
4997 struct imsm_disk
*d
;
4998 struct intel_disk
*idisk
;
4999 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
5002 d
= &s
->disks
->disk
;
5004 /* 'd' must appear in merged disk list for its
5005 * configuration to be valid
5007 idisk
= disk_list_get(d
->serial
, disk_list
);
5008 if (idisk
&& idisk
->owner
== i
)
5009 s
= validate_members(s
, disk_list
, i
);
5014 dprintf("marking family: %#x from %d:%d offline\n",
5016 super_table
[i
]->disks
->major
,
5017 super_table
[i
]->disks
->minor
);
5021 /* This is where the mdadm implementation differs from the Windows
5022 * driver which has no strict concept of a container. We can only
5023 * assemble one family from a container, so when returning a prodigal
5024 * array member to this system the code will not be able to disambiguate
5025 * the container contents that should be assembled ("foreign" versus
5026 * "local"). It requires user intervention to set the orig_family_num
5027 * to a new value to establish a new container. The Windows driver in
5028 * this situation fixes up the volume name in place and manages the
5029 * foreign array as an independent entity.
5034 for (i
= 0; i
< tbl_size
; i
++) {
5035 struct intel_super
*tbl_ent
= super_table
[i
];
5041 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
5046 if (s
&& !is_spare
) {
5047 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
5049 } else if (!s
&& !is_spare
)
5062 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
5063 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
5065 /* collect all dl's onto 'champion', and update them to
5066 * champion's version of the status
5068 for (s
= *super_list
; s
; s
= s
->next
) {
5069 struct imsm_super
*mpb
= champion
->anchor
;
5070 struct dl
*dl
= s
->disks
;
5075 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
5077 for (i
= 0; i
< mpb
->num_disks
; i
++) {
5078 struct imsm_disk
*disk
;
5080 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
5083 /* only set index on disks that are a member of
5084 * a populated contianer, i.e. one with
5087 if (is_failed(&dl
->disk
))
5089 else if (is_spare(&dl
->disk
))
5095 if (i
>= mpb
->num_disks
) {
5096 struct intel_disk
*idisk
;
5098 idisk
= disk_list_get(dl
->serial
, disk_list
);
5099 if (idisk
&& is_spare(&idisk
->disk
) &&
5100 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
5108 dl
->next
= champion
->disks
;
5109 champion
->disks
= dl
;
5113 /* delete 'champion' from super_list */
5114 for (del
= super_list
; *del
; ) {
5115 if (*del
== champion
) {
5116 *del
= (*del
)->next
;
5119 del
= &(*del
)->next
;
5121 champion
->next
= NULL
;
5125 struct intel_disk
*idisk
= disk_list
;
5127 disk_list
= disk_list
->next
;
5135 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
5136 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5137 int major
, int minor
, int keep_fd
);
5139 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5140 int *max
, int keep_fd
);
5142 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
5143 char *devname
, struct md_list
*devlist
,
5146 struct intel_super
*super_list
= NULL
;
5147 struct intel_super
*super
= NULL
;
5151 if (is_fd_valid(fd
))
5152 /* 'fd' is an opened container */
5153 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
5155 /* get super block from devlist devices */
5156 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
5159 /* all mpbs enter, maybe one leaves */
5160 super
= imsm_thunderdome(&super_list
, i
);
5166 if (find_missing(super
) != 0) {
5172 /* load migration record */
5173 err
= load_imsm_migr_rec(super
);
5175 /* migration is in progress,
5176 * but migr_rec cannot be loaded,
5182 /* Check migration compatibility */
5183 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5184 pr_err("Unsupported migration detected");
5186 fprintf(stderr
, " on %s\n", devname
);
5188 fprintf(stderr
, " (IMSM).\n");
5197 while (super_list
) {
5198 struct intel_super
*s
= super_list
;
5200 super_list
= super_list
->next
;
5208 if (is_fd_valid(fd
))
5209 strcpy(st
->container_devnm
, fd2devnm(fd
));
5211 st
->container_devnm
[0] = 0;
5212 if (err
== 0 && st
->ss
== NULL
) {
5213 st
->ss
= &super_imsm
;
5214 st
->minor_version
= 0;
5215 st
->max_devs
= IMSM_MAX_DEVICES
;
5221 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5222 int *max
, int keep_fd
)
5224 struct md_list
*tmpdev
;
5228 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5229 if (tmpdev
->used
!= 1)
5231 if (tmpdev
->container
== 1) {
5233 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5234 if (!is_fd_valid(fd
)) {
5235 pr_err("cannot open device %s: %s\n",
5236 tmpdev
->devname
, strerror(errno
));
5240 err
= get_sra_super_block(fd
, super_list
,
5241 tmpdev
->devname
, &lmax
,
5250 int major
= major(tmpdev
->st_rdev
);
5251 int minor
= minor(tmpdev
->st_rdev
);
5252 err
= get_super_block(super_list
,
5269 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5270 int major
, int minor
, int keep_fd
)
5272 struct intel_super
*s
;
5284 sprintf(nm
, "%d:%d", major
, minor
);
5285 dfd
= dev_open(nm
, O_RDWR
);
5286 if (!is_fd_valid(dfd
)) {
5291 if (!get_dev_sector_size(dfd
, NULL
, &s
->sector_size
)) {
5295 find_intel_hba_capability(dfd
, s
, devname
);
5296 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5298 /* retry the load if we might have raced against mdmon */
5299 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5300 for (retry
= 0; retry
< 3; retry
++) {
5301 sleep_for(0, MSEC_TO_NSEC(3), true);
5302 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5308 s
->next
= *super_list
;
5322 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5329 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5333 if (sra
->array
.major_version
!= -1 ||
5334 sra
->array
.minor_version
!= -2 ||
5335 strcmp(sra
->text_version
, "imsm") != 0) {
5340 devnm
= fd2devnm(fd
);
5341 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5342 if (get_super_block(super_list
, devnm
, devname
,
5343 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5354 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5356 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5359 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5361 struct intel_super
*super
;
5365 if (test_partition(fd
))
5366 /* IMSM not allowed on partitions */
5369 free_super_imsm(st
);
5371 super
= alloc_super();
5375 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
5379 /* Load hba and capabilities if they exist.
5380 * But do not preclude loading metadata in case capabilities or hba are
5381 * non-compliant and ignore_hw_compat is set.
5383 rv
= find_intel_hba_capability(fd
, super
, devname
);
5384 /* no orom/efi or non-intel hba of the disk */
5385 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5387 pr_err("No OROM/EFI properties for %s\n", devname
);
5391 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5393 /* retry the load if we might have raced against mdmon */
5395 struct mdstat_ent
*mdstat
= NULL
;
5396 char *name
= fd2kname(fd
);
5399 mdstat
= mdstat_by_component(name
);
5401 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5402 for (retry
= 0; retry
< 3; retry
++) {
5403 sleep_for(0, MSEC_TO_NSEC(3), true);
5404 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5410 free_mdstat(mdstat
);
5415 pr_err("Failed to load all information sections on %s\n", devname
);
5421 if (st
->ss
== NULL
) {
5422 st
->ss
= &super_imsm
;
5423 st
->minor_version
= 0;
5424 st
->max_devs
= IMSM_MAX_DEVICES
;
5427 /* load migration record */
5428 if (load_imsm_migr_rec(super
) == 0) {
5429 /* Check for unsupported migration features */
5430 if (check_mpb_migr_compatibility(super
) != 0) {
5431 pr_err("Unsupported migration detected");
5433 fprintf(stderr
, " on %s\n", devname
);
5435 fprintf(stderr
, " (IMSM).\n");
5443 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5445 if (info
->level
== 1)
5447 return info
->chunk_size
>> 9;
5450 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5451 unsigned long long size
)
5453 if (info
->level
== 1)
5456 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5459 static void imsm_update_version_info(struct intel_super
*super
)
5461 /* update the version and attributes */
5462 struct imsm_super
*mpb
= super
->anchor
;
5464 struct imsm_dev
*dev
;
5465 struct imsm_map
*map
;
5468 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5469 dev
= get_imsm_dev(super
, i
);
5470 map
= get_imsm_map(dev
, MAP_0
);
5471 if (__le32_to_cpu(dev
->size_high
) > 0)
5472 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5474 /* FIXME detect when an array spans a port multiplier */
5476 mpb
->attributes
|= MPB_ATTRIB_PM
;
5479 if (mpb
->num_raid_devs
> 1 ||
5480 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5481 version
= MPB_VERSION_ATTRIBS
;
5482 switch (get_imsm_raid_level(map
)) {
5483 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5484 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5485 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5486 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5489 if (map
->num_members
>= 5)
5490 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5491 else if (dev
->status
== DEV_CLONE_N_GO
)
5492 version
= MPB_VERSION_CNG
;
5493 else if (get_imsm_raid_level(map
) == 5)
5494 version
= MPB_VERSION_RAID5
;
5495 else if (map
->num_members
>= 3)
5496 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5497 else if (get_imsm_raid_level(map
) == 1)
5498 version
= MPB_VERSION_RAID1
;
5500 version
= MPB_VERSION_RAID0
;
5502 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5507 * imsm_check_name() - check imsm naming criteria.
5508 * @super: &intel_super pointer, not NULL.
5509 * @name: name to check.
5510 * @verbose: verbose level.
5512 * Name must be no longer than &MAX_RAID_SERIAL_LEN and must be unique across volumes.
5514 * Returns: &true if @name matches, &false otherwise.
5516 static bool imsm_is_name_allowed(struct intel_super
*super
, const char * const name
,
5519 struct imsm_super
*mpb
= super
->anchor
;
5522 if (is_string_lq(name
, MAX_RAID_SERIAL_LEN
+ 1) == false) {
5523 pr_vrb("imsm: Name \"%s\" is too long\n", name
);
5527 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5528 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5530 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5531 pr_vrb("imsm: Name \"%s\" already exists\n", name
);
5539 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5540 struct shape
*s
, char *name
,
5541 char *homehost
, int *uuid
,
5542 long long data_offset
)
5544 /* We are creating a volume inside a pre-existing container.
5545 * so st->sb is already set.
5547 struct intel_super
*super
= st
->sb
;
5548 unsigned int sector_size
= super
->sector_size
;
5549 struct imsm_super
*mpb
= super
->anchor
;
5550 struct intel_dev
*dv
;
5551 struct imsm_dev
*dev
;
5552 struct imsm_vol
*vol
;
5553 struct imsm_map
*map
;
5554 int idx
= mpb
->num_raid_devs
;
5557 unsigned long long array_blocks
;
5558 size_t size_old
, size_new
;
5559 unsigned int data_disks
;
5560 unsigned long long size_per_member
;
5562 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5563 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5567 /* ensure the mpb is large enough for the new data */
5568 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5569 size_new
= disks_to_mpb_size(info
->nr_disks
);
5570 if (size_new
> size_old
) {
5572 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5574 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5575 pr_err("could not allocate new mpb\n");
5578 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5579 MIGR_REC_BUF_SECTORS
*
5580 MAX_SECTOR_SIZE
) != 0) {
5581 pr_err("could not allocate migr_rec buffer\n");
5587 memcpy(mpb_new
, mpb
, size_old
);
5590 super
->anchor
= mpb_new
;
5591 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5592 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5593 super
->len
= size_round
;
5595 super
->current_vol
= idx
;
5597 /* handle 'failed_disks' by either:
5598 * a) create dummy disk entries in the table if this the first
5599 * volume in the array. We add them here as this is the only
5600 * opportunity to add them. add_to_super_imsm_volume()
5601 * handles the non-failed disks and continues incrementing
5603 * b) validate that 'failed_disks' matches the current number
5604 * of missing disks if the container is populated
5606 if (super
->current_vol
== 0) {
5608 for (i
= 0; i
< info
->failed_disks
; i
++) {
5609 struct imsm_disk
*disk
;
5612 disk
= __get_imsm_disk(mpb
, i
);
5613 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5614 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5615 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5616 "missing:%d", (__u8
)i
);
5618 find_missing(super
);
5623 for (d
= super
->missing
; d
; d
= d
->next
)
5625 if (info
->failed_disks
> missing
) {
5626 pr_err("unable to add 'missing' disk to container\n");
5631 if (imsm_is_name_allowed(super
, name
, 1) == false)
5634 dv
= xmalloc(sizeof(*dv
));
5635 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5637 * Explicitly allow truncating to not confuse gcc's
5638 * -Werror=stringop-truncation
5640 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5641 memcpy(dev
->volume
, name
, namelen
);
5642 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5643 info
->layout
, info
->chunk_size
,
5644 s
->size
* BLOCKS_PER_KB
);
5645 data_disks
= get_data_disks(info
->level
, info
->layout
,
5647 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5648 size_per_member
= array_blocks
/ data_disks
;
5650 set_imsm_dev_size(dev
, array_blocks
);
5651 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5653 vol
->migr_state
= 0;
5654 set_migr_type(dev
, MIGR_INIT
);
5655 vol
->dirty
= !info
->state
;
5656 set_vol_curr_migr_unit(dev
, 0);
5657 map
= get_imsm_map(dev
, MAP_0
);
5658 set_pba_of_lba0(map
, super
->create_offset
);
5659 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5660 map
->failed_disk_num
= ~0;
5661 if (info
->level
> 0)
5662 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5663 : IMSM_T_STATE_UNINITIALIZED
);
5665 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5666 IMSM_T_STATE_NORMAL
;
5669 if (info
->level
== 1 && info
->raid_disks
> 2) {
5672 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5676 map
->raid_level
= info
->level
;
5677 if (info
->level
== 10)
5678 map
->raid_level
= 1;
5679 set_num_domains(map
);
5681 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5682 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5686 map
->num_members
= info
->raid_disks
;
5687 update_num_data_stripes(map
, array_blocks
);
5688 for (i
= 0; i
< map
->num_members
; i
++) {
5689 /* initialized in add_to_super */
5690 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5692 mpb
->num_raid_devs
++;
5693 mpb
->num_raid_devs_created
++;
5694 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5696 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5697 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5698 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5699 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5703 pr_err("imsm does not support consistency policy %s\n",
5704 map_num_s(consistency_policies
, s
->consistency_policy
));
5709 dv
->index
= super
->current_vol
;
5710 dv
->next
= super
->devlist
;
5711 super
->devlist
= dv
;
5713 imsm_update_version_info(super
);
5718 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5719 struct shape
*s
, char *name
,
5720 char *homehost
, int *uuid
,
5721 unsigned long long data_offset
)
5723 /* This is primarily called by Create when creating a new array.
5724 * We will then get add_to_super called for each component, and then
5725 * write_init_super called to write it out to each device.
5726 * For IMSM, Create can create on fresh devices or on a pre-existing
5728 * To create on a pre-existing array a different method will be called.
5729 * This one is just for fresh drives.
5731 struct intel_super
*super
;
5732 struct imsm_super
*mpb
;
5736 if (data_offset
!= INVALID_SECTORS
) {
5737 pr_err("data-offset not supported by imsm\n");
5742 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5746 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5748 mpb_size
= MAX_SECTOR_SIZE
;
5750 super
= alloc_super();
5752 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5757 pr_err("could not allocate superblock\n");
5760 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5761 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5762 pr_err("could not allocate migr_rec buffer\n");
5767 memset(super
->buf
, 0, mpb_size
);
5769 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5773 /* zeroing superblock */
5777 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5779 version
= (char *) mpb
->sig
;
5780 strcpy(version
, MPB_SIGNATURE
);
5781 version
+= strlen(MPB_SIGNATURE
);
5782 strcpy(version
, MPB_VERSION_RAID0
);
5787 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5789 unsigned int member_sector_size
;
5791 if (!is_fd_valid(dl
->fd
)) {
5792 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5796 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5798 if (member_sector_size
!= super
->sector_size
)
5803 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5804 int fd
, char *devname
)
5806 struct intel_super
*super
= st
->sb
;
5807 struct imsm_super
*mpb
= super
->anchor
;
5808 struct imsm_disk
*_disk
;
5809 struct imsm_dev
*dev
;
5810 struct imsm_map
*map
;
5815 if (!is_fd_valid(fd
))
5818 dev
= get_imsm_dev(super
, super
->current_vol
);
5819 map
= get_imsm_map(dev
, MAP_0
);
5821 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5822 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5827 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5829 if (dl
->raiddisk
== dk
->raid_disk
)
5831 } else if (dl
->major
== dk
->major
&& dl
->minor
== dk
->minor
)
5837 pr_err("%s is not a member of the same container.\n",
5842 if (!autolayout
&& super
->current_vol
> 0) {
5843 int _slot
= get_disk_slot_in_dev(super
, 0, dl
->index
);
5845 if (_slot
!= dk
->raid_disk
) {
5846 pr_err("Member %s is in %d slot for the first volume, but is in %d slot for a new volume.\n",
5847 dl
->devname
, _slot
, dk
->raid_disk
);
5848 pr_err("Raid members are in different order than for the first volume, aborting.\n");
5853 if (mpb
->num_disks
== 0)
5854 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5855 &super
->sector_size
))
5858 if (!drive_validate_sector_size(super
, dl
)) {
5859 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5863 /* add a pristine spare to the metadata */
5864 if (dl
->index
< 0) {
5865 dl
->index
= super
->anchor
->num_disks
;
5866 super
->anchor
->num_disks
++;
5868 /* Check the device has not already been added */
5869 slot
= get_imsm_disk_slot(map
, dl
->index
);
5871 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5872 pr_err("%s has been included in this array twice\n",
5876 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5877 dl
->disk
.status
= CONFIGURED_DISK
;
5879 /* update size of 'missing' disks to be at least as large as the
5880 * largest acitve member (we only have dummy missing disks when
5881 * creating the first volume)
5883 if (super
->current_vol
== 0) {
5884 for (df
= super
->missing
; df
; df
= df
->next
) {
5885 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5886 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5887 _disk
= __get_imsm_disk(mpb
, df
->index
);
5892 /* refresh unset/failed slots to point to valid 'missing' entries */
5893 for (df
= super
->missing
; df
; df
= df
->next
)
5894 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5895 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5897 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5899 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5900 if (is_gen_migration(dev
)) {
5901 struct imsm_map
*map2
= get_imsm_map(dev
,
5903 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5904 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5905 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5908 if ((unsigned)df
->index
==
5910 set_imsm_ord_tbl_ent(map2
,
5916 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5920 /* if we are creating the first raid device update the family number */
5921 if (super
->current_vol
== 0) {
5923 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5925 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5927 pr_err("BUG mpb setup error\n");
5933 sum
+= __gen_imsm_checksum(mpb
);
5934 mpb
->family_num
= __cpu_to_le32(sum
);
5935 mpb
->orig_family_num
= mpb
->family_num
;
5936 mpb
->creation_time
= __cpu_to_le64((__u64
)time(NULL
));
5938 super
->current_disk
= dl
;
5943 * Function marks disk as spare and restores disk serial
5944 * in case it was previously marked as failed by takeover operation
5946 * -1 : critical error
5947 * 0 : disk is marked as spare but serial is not set
5950 int mark_spare(struct dl
*disk
)
5952 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5959 if (!imsm_read_serial(disk
->fd
, NULL
, serial
, MAX_RAID_SERIAL_LEN
)) {
5960 /* Restore disk serial number, because takeover marks disk
5961 * as failed and adds to serial ':0' before it becomes
5964 serialcpy(disk
->serial
, serial
);
5965 serialcpy(disk
->disk
.serial
, serial
);
5968 disk
->disk
.status
= SPARE_DISK
;
5975 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
);
5977 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5978 int fd
, char *devname
,
5979 unsigned long long data_offset
)
5981 struct intel_super
*super
= st
->sb
;
5983 unsigned long long size
;
5984 unsigned int member_sector_size
;
5989 /* If we are on an RAID enabled platform check that the disk is
5990 * attached to the raid controller.
5991 * We do not need to test disks attachment for container based additions,
5992 * they shall be already tested when container was created/assembled.
5994 rv
= find_intel_hba_capability(fd
, super
, devname
);
5995 /* no orom/efi or non-intel hba of the disk */
5997 dprintf("capability: %p fd: %d ret: %d\n",
5998 super
->orom
, fd
, rv
);
6002 if (super
->current_vol
>= 0)
6003 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
6006 dd
= xcalloc(sizeof(*dd
), 1);
6007 dd
->major
= major(stb
.st_rdev
);
6008 dd
->minor
= minor(stb
.st_rdev
);
6009 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
6012 dd
->action
= DISK_ADD
;
6013 rv
= imsm_read_serial(fd
, devname
, dd
->serial
, MAX_RAID_SERIAL_LEN
);
6015 pr_err("failed to retrieve scsi serial, aborting\n");
6016 __free_imsm_disk(dd
, 0);
6020 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
6021 (super
->hba
->type
== SYS_DEV_VMD
))) {
6023 char cntrl_path
[PATH_MAX
];
6025 char pci_dev_path
[PATH_MAX
];
6027 if (!diskfd_to_devpath(fd
, 2, pci_dev_path
) ||
6028 !diskfd_to_devpath(fd
, 1, cntrl_path
)) {
6029 pr_err("failed to get dev paths, aborting\n");
6030 __free_imsm_disk(dd
, 0);
6034 cntrl_name
= basename(cntrl_path
);
6035 if (is_multipath_nvme(fd
))
6036 pr_err("%s controller supports Multi-Path I/O, Intel (R) VROC does not support multipathing\n",
6039 if (devpath_to_vendor(pci_dev_path
) == 0x8086) {
6041 * If Intel's NVMe drive has serial ended with
6042 * "-A","-B","-1" or "-2" it means that this is "x8"
6043 * device (double drive on single PCIe card).
6044 * User should be warned about potential data loss.
6046 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
6047 /* Skip empty character at the end */
6048 if (dd
->serial
[i
] == 0)
6051 if (((dd
->serial
[i
] == 'A') ||
6052 (dd
->serial
[i
] == 'B') ||
6053 (dd
->serial
[i
] == '1') ||
6054 (dd
->serial
[i
] == '2')) &&
6055 (dd
->serial
[i
-1] == '-'))
6056 pr_err("\tThe action you are about to take may put your data at risk.\n"
6057 "\tPlease note that x8 devices may consist of two separate x4 devices "
6058 "located on a single PCIe port.\n"
6059 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
6062 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
6063 !imsm_orom_has_tpv_support(super
->orom
)) {
6064 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
6065 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
6066 __free_imsm_disk(dd
, 0);
6071 get_dev_size(fd
, NULL
, &size
);
6072 if (!get_dev_sector_size(fd
, NULL
, &member_sector_size
)) {
6073 __free_imsm_disk(dd
, 0);
6077 if (super
->sector_size
== 0) {
6078 /* this a first device, so sector_size is not set yet */
6079 super
->sector_size
= member_sector_size
;
6082 /* clear migr_rec when adding disk to container */
6083 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6084 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
6086 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
6087 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
6088 MIGR_REC_BUF_SECTORS
*member_sector_size
)
6089 perror("Write migr_rec failed");
6093 serialcpy(dd
->disk
.serial
, dd
->serial
);
6094 set_total_blocks(&dd
->disk
, size
);
6095 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
6096 struct imsm_super
*mpb
= super
->anchor
;
6097 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6100 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
6101 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
6103 dd
->disk
.scsi_id
= __cpu_to_le32(0);
6105 if (st
->update_tail
) {
6106 dd
->next
= super
->disk_mgmt_list
;
6107 super
->disk_mgmt_list
= dd
;
6109 /* this is called outside of mdmon
6110 * write initial spare metadata
6111 * mdmon will overwrite it.
6113 dd
->next
= super
->disks
;
6115 write_super_imsm_spare(super
, dd
);
6121 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
6123 struct intel_super
*super
= st
->sb
;
6126 /* remove from super works only in mdmon - for communication
6127 * manager - monitor. Check if communication memory buffer
6130 if (!st
->update_tail
) {
6131 pr_err("shall be used in mdmon context only\n");
6134 dd
= xcalloc(1, sizeof(*dd
));
6135 dd
->major
= dk
->major
;
6136 dd
->minor
= dk
->minor
;
6139 dd
->action
= DISK_REMOVE
;
6141 dd
->next
= super
->disk_mgmt_list
;
6142 super
->disk_mgmt_list
= dd
;
6147 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
6150 char buf
[MAX_SECTOR_SIZE
];
6151 struct imsm_super anchor
;
6152 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
6155 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
)
6157 struct imsm_super
*mpb
= super
->anchor
;
6158 struct imsm_super
*spare
= &spare_record
.anchor
;
6164 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
6165 spare
->generation_num
= __cpu_to_le32(1UL);
6166 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
6167 spare
->num_disks
= 1;
6168 spare
->num_raid_devs
= 0;
6169 spare
->cache_size
= mpb
->cache_size
;
6170 spare
->pwr_cycle_count
= __cpu_to_le32(1);
6172 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
6173 MPB_SIGNATURE MPB_VERSION_RAID0
);
6175 spare
->disk
[0] = d
->disk
;
6176 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
6177 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6179 if (super
->sector_size
== 4096)
6180 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6182 sum
= __gen_imsm_checksum(spare
);
6183 spare
->family_num
= __cpu_to_le32(sum
);
6184 spare
->orig_family_num
= 0;
6185 sum
= __gen_imsm_checksum(spare
);
6186 spare
->check_sum
= __cpu_to_le32(sum
);
6188 if (store_imsm_mpb(d
->fd
, spare
)) {
6189 pr_err("failed for device %d:%d %s\n",
6190 d
->major
, d
->minor
, strerror(errno
));
6196 /* spare records have their own family number and do not have any defined raid
6199 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
6203 for (d
= super
->disks
; d
; d
= d
->next
) {
6207 if (write_super_imsm_spare(super
, d
))
6217 static int write_super_imsm(struct supertype
*st
, int doclose
)
6219 struct intel_super
*super
= st
->sb
;
6220 unsigned int sector_size
= super
->sector_size
;
6221 struct imsm_super
*mpb
= super
->anchor
;
6227 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6229 int clear_migration_record
= 1;
6232 /* 'generation' is incremented everytime the metadata is written */
6233 generation
= __le32_to_cpu(mpb
->generation_num
);
6235 mpb
->generation_num
= __cpu_to_le32(generation
);
6237 /* fix up cases where previous mdadm releases failed to set
6240 if (mpb
->orig_family_num
== 0)
6241 mpb
->orig_family_num
= mpb
->family_num
;
6243 for (d
= super
->disks
; d
; d
= d
->next
) {
6247 mpb
->disk
[d
->index
] = d
->disk
;
6251 for (d
= super
->missing
; d
; d
= d
->next
) {
6252 mpb
->disk
[d
->index
] = d
->disk
;
6255 mpb
->num_disks
= num_disks
;
6256 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6258 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6259 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6260 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6262 imsm_copy_dev(dev
, dev2
);
6263 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6265 if (is_gen_migration(dev2
))
6266 clear_migration_record
= 0;
6269 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6272 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6273 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6275 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6277 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6278 mpb_size
+= bbm_log_size
;
6279 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6282 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6285 /* recalculate checksum */
6286 sum
= __gen_imsm_checksum(mpb
);
6287 mpb
->check_sum
= __cpu_to_le32(sum
);
6289 if (super
->clean_migration_record_by_mdmon
) {
6290 clear_migration_record
= 1;
6291 super
->clean_migration_record_by_mdmon
= 0;
6293 if (clear_migration_record
)
6294 memset(super
->migr_rec_buf
, 0,
6295 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6297 if (sector_size
== 4096)
6298 convert_to_4k(super
);
6300 /* write the mpb for disks that compose raid devices */
6301 for (d
= super
->disks
; d
; d
= d
->next
) {
6302 if (d
->index
< 0 || is_failed(&d
->disk
))
6305 if (clear_migration_record
) {
6306 unsigned long long dsize
;
6308 get_dev_size(d
->fd
, NULL
, &dsize
);
6309 if (lseek64(d
->fd
, dsize
- sector_size
,
6311 if ((unsigned int)write(d
->fd
,
6312 super
->migr_rec_buf
,
6313 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6314 MIGR_REC_BUF_SECTORS
*sector_size
)
6315 perror("Write migr_rec failed");
6319 if (store_imsm_mpb(d
->fd
, mpb
))
6321 "failed for device %d:%d (fd: %d)%s\n",
6323 d
->fd
, strerror(errno
));
6330 return write_super_imsm_spares(super
, doclose
);
6335 static int create_array(struct supertype
*st
, int dev_idx
)
6338 struct imsm_update_create_array
*u
;
6339 struct intel_super
*super
= st
->sb
;
6340 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6341 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6342 struct disk_info
*inf
;
6343 struct imsm_disk
*disk
;
6346 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6347 sizeof(*inf
) * map
->num_members
;
6349 u
->type
= update_create_array
;
6350 u
->dev_idx
= dev_idx
;
6351 imsm_copy_dev(&u
->dev
, dev
);
6352 inf
= get_disk_info(u
);
6353 for (i
= 0; i
< map
->num_members
; i
++) {
6354 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6356 disk
= get_imsm_disk(super
, idx
);
6358 disk
= get_imsm_missing(super
, idx
);
6359 serialcpy(inf
[i
].serial
, disk
->serial
);
6361 append_metadata_update(st
, u
, len
);
6366 static int mgmt_disk(struct supertype
*st
)
6368 struct intel_super
*super
= st
->sb
;
6370 struct imsm_update_add_remove_disk
*u
;
6372 if (!super
->disk_mgmt_list
)
6377 u
->type
= update_add_remove_disk
;
6378 append_metadata_update(st
, u
, len
);
6383 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6385 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6387 struct ppl_header
*ppl_hdr
= buf
;
6390 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6392 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6394 perror("Failed to seek to PPL header location");
6398 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6400 perror("Write PPL header failed");
6409 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6411 struct intel_super
*super
= st
->sb
;
6413 struct ppl_header
*ppl_hdr
;
6416 /* first clear entire ppl space */
6417 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6421 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6423 pr_err("Failed to allocate PPL header buffer\n");
6427 memset(buf
, 0, PPL_HEADER_SIZE
);
6429 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6430 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6432 if (info
->mismatch_cnt
) {
6434 * We are overwriting an invalid ppl. Make one entry with wrong
6435 * checksum to prevent the kernel from skipping resync.
6437 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6438 ppl_hdr
->entries
[0].checksum
= ~0;
6441 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6447 static int is_rebuilding(struct imsm_dev
*dev
);
6449 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6450 struct mdinfo
*disk
)
6452 struct intel_super
*super
= st
->sb
;
6454 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6456 struct ppl_header
*ppl_hdr
= NULL
;
6458 struct imsm_dev
*dev
;
6461 unsigned long long ppl_offset
= 0;
6462 unsigned long long prev_gen_num
= 0;
6464 if (disk
->disk
.raid_disk
< 0)
6467 dev
= get_imsm_dev(super
, info
->container_member
);
6468 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6469 d
= get_imsm_dl_disk(super
, idx
);
6471 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6474 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6475 pr_err("Failed to allocate PPL header buffer\n");
6481 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6484 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6486 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6488 perror("Failed to seek to PPL header location");
6493 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6494 perror("Read PPL header failed");
6501 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6502 ppl_hdr
->checksum
= 0;
6504 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6505 dprintf("Wrong PPL header checksum on %s\n",
6510 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6511 /* previous was newest, it was already checked */
6515 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6516 super
->anchor
->orig_family_num
)) {
6517 dprintf("Wrong PPL header signature on %s\n",
6524 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6526 ppl_offset
+= PPL_HEADER_SIZE
;
6527 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6529 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6532 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6544 * Update metadata to use mutliple PPLs area (1MB).
6545 * This is done once for all RAID members
6547 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6548 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6550 struct mdinfo
*member_dev
;
6552 sprintf(subarray
, "%d", info
->container_member
);
6554 if (mdmon_running(st
->container_devnm
))
6555 st
->update_tail
= &st
->updates
;
6557 if (st
->ss
->update_subarray(st
, subarray
, UOPT_PPL
, NULL
)) {
6558 pr_err("Failed to update subarray %s\n",
6561 if (st
->update_tail
)
6562 flush_metadata_updates(st
);
6564 st
->ss
->sync_metadata(st
);
6565 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6566 for (member_dev
= info
->devs
; member_dev
;
6567 member_dev
= member_dev
->next
)
6568 member_dev
->ppl_size
=
6569 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6574 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6576 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6577 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6578 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6579 (is_rebuilding(dev
) &&
6580 vol_curr_migr_unit(dev
) == 0 &&
6581 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6582 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6584 info
->mismatch_cnt
++;
6585 } else if (ret
== 0 &&
6586 ppl_hdr
->entries_count
== 0 &&
6587 is_rebuilding(dev
) &&
6588 info
->resync_start
== 0) {
6590 * The header has no entries - add a single empty entry and
6591 * rewrite the header to prevent the kernel from going into
6592 * resync after an interrupted rebuild.
6594 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6595 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6603 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6605 struct intel_super
*super
= st
->sb
;
6609 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6610 info
->array
.level
!= 5)
6613 for (d
= super
->disks
; d
; d
= d
->next
) {
6614 if (d
->index
< 0 || is_failed(&d
->disk
))
6617 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6625 /*******************************************************************************
6626 * Function: write_init_bitmap_imsm_vol
6627 * Description: Write a bitmap header and prepares the area for the bitmap.
6629 * st : supertype information
6630 * vol_idx : the volume index to use
6635 ******************************************************************************/
6636 static int write_init_bitmap_imsm_vol(struct supertype
*st
, int vol_idx
)
6638 struct intel_super
*super
= st
->sb
;
6639 int prev_current_vol
= super
->current_vol
;
6643 super
->current_vol
= vol_idx
;
6644 for (d
= super
->disks
; d
; d
= d
->next
) {
6645 if (d
->index
< 0 || is_failed(&d
->disk
))
6647 ret
= st
->ss
->write_bitmap(st
, d
->fd
, NoUpdate
);
6651 super
->current_vol
= prev_current_vol
;
6655 /*******************************************************************************
6656 * Function: write_init_bitmap_imsm_all
6657 * Description: Write a bitmap header and prepares the area for the bitmap.
6658 * Operation is executed for volumes with CONSISTENCY_POLICY_BITMAP.
6660 * st : supertype information
6661 * info : info about the volume where the bitmap should be written
6662 * vol_idx : the volume index to use
6667 ******************************************************************************/
6668 static int write_init_bitmap_imsm_all(struct supertype
*st
, struct mdinfo
*info
,
6673 if (info
&& (info
->consistency_policy
== CONSISTENCY_POLICY_BITMAP
))
6674 ret
= write_init_bitmap_imsm_vol(st
, vol_idx
);
6679 static int write_init_super_imsm(struct supertype
*st
)
6681 struct intel_super
*super
= st
->sb
;
6682 int current_vol
= super
->current_vol
;
6686 getinfo_super_imsm(st
, &info
, NULL
);
6688 /* we are done with current_vol reset it to point st at the container */
6689 super
->current_vol
= -1;
6691 if (st
->update_tail
) {
6692 /* queue the recently created array / added disk
6693 * as a metadata update */
6695 /* determine if we are creating a volume or adding a disk */
6696 if (current_vol
< 0) {
6697 /* in the mgmt (add/remove) disk case we are running
6698 * in mdmon context, so don't close fd's
6702 /* adding the second volume to the array */
6703 rv
= write_init_ppl_imsm_all(st
, &info
);
6705 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6707 rv
= create_array(st
, current_vol
);
6711 for (d
= super
->disks
; d
; d
= d
->next
)
6712 Kill(d
->devname
, NULL
, 0, -1, 1);
6713 if (current_vol
>= 0) {
6714 rv
= write_init_ppl_imsm_all(st
, &info
);
6716 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6720 rv
= write_super_imsm(st
, 1);
6726 static int store_super_imsm(struct supertype
*st
, int fd
)
6728 struct intel_super
*super
= st
->sb
;
6729 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6734 if (super
->sector_size
== 4096)
6735 convert_to_4k(super
);
6736 return store_imsm_mpb(fd
, mpb
);
6739 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6741 unsigned long long data_offset
,
6743 unsigned long long *freesize
,
6747 unsigned long long ldsize
;
6748 struct intel_super
*super
= NULL
;
6751 if (!is_container(level
))
6756 fd
= dev_open(dev
, O_RDONLY
|O_EXCL
);
6757 if (!is_fd_valid(fd
)) {
6758 pr_vrb("imsm: Cannot open %s: %s\n", dev
, strerror(errno
));
6761 if (!get_dev_size(fd
, dev
, &ldsize
))
6764 /* capabilities retrieve could be possible
6765 * note that there is no fd for the disks in array.
6767 super
= alloc_super();
6771 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
))
6774 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6778 fd2devname(fd
, str
);
6779 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6780 fd
, str
, super
->orom
, rv
, raiddisks
);
6782 /* no orom/efi or non-intel hba of the disk */
6787 if (raiddisks
> super
->orom
->tds
) {
6789 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6790 raiddisks
, super
->orom
->tds
);
6793 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6794 (ldsize
>> 9) >> 32 > 0) {
6796 pr_err("%s exceeds maximum platform supported size\n", dev
);
6800 if (super
->hba
->type
== SYS_DEV_VMD
||
6801 super
->hba
->type
== SYS_DEV_NVME
) {
6802 if (!imsm_is_nvme_namespace_supported(fd
, 1)) {
6804 pr_err("NVMe namespace %s is not supported by IMSM\n",
6811 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6821 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6823 const unsigned long long base_start
= e
[*idx
].start
;
6824 unsigned long long end
= base_start
+ e
[*idx
].size
;
6827 if (base_start
== end
)
6831 for (i
= *idx
; i
< num_extents
; i
++) {
6832 /* extend overlapping extents */
6833 if (e
[i
].start
>= base_start
&&
6834 e
[i
].start
<= end
) {
6837 if (e
[i
].start
+ e
[i
].size
> end
)
6838 end
= e
[i
].start
+ e
[i
].size
;
6839 } else if (e
[i
].start
> end
) {
6845 return end
- base_start
;
6848 /** merge_extents() - analyze extents and get free size.
6849 * @super: Intel metadata, not NULL.
6850 * @expanding: if set, we are expanding &super->current_vol.
6852 * Build a composite disk with all known extents and generate a size given the
6853 * "all disks in an array must share a common start offset" constraint.
6854 * If a volume is expanded, then return free space after the volume.
6856 * Return: Free space or 0 on failure.
6858 static unsigned long long merge_extents(struct intel_super
*super
, const bool expanding
)
6862 int i
, j
, pos_vol_idx
= -1;
6864 int sum_extents
= 0;
6865 unsigned long long pos
= 0;
6866 unsigned long long start
= 0;
6867 unsigned long long free_size
= 0;
6869 unsigned long pre_reservation
= 0;
6870 unsigned long post_reservation
= IMSM_RESERVED_SECTORS
;
6871 unsigned long reservation_size
;
6873 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6875 sum_extents
+= dl
->extent_cnt
;
6876 e
= xcalloc(sum_extents
, sizeof(struct extent
));
6878 /* coalesce and sort all extents. also, check to see if we need to
6879 * reserve space between member arrays
6882 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6885 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6888 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6893 while (i
< sum_extents
) {
6894 e
[j
].start
= e
[i
].start
;
6895 e
[j
].vol
= e
[i
].vol
;
6896 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6898 if (e
[j
-1].size
== 0)
6904 unsigned long long esize
= e
[i
].start
- pos
;
6906 if (expanding
? pos_vol_idx
== super
->current_vol
: esize
>= free_size
) {
6912 pos
= e
[i
].start
+ e
[i
].size
;
6913 pos_vol_idx
= e
[i
].vol
;
6916 } while (e
[i
-1].size
);
6918 if (free_size
== 0) {
6919 dprintf("imsm: Cannot find free size.\n");
6924 if (!expanding
&& extent_idx
!= 0)
6926 * Not a real first volume in a container is created, pre_reservation is needed.
6928 pre_reservation
= IMSM_RESERVED_SECTORS
;
6930 if (e
[extent_idx
].size
== 0)
6932 * extent_idx points to the metadata, post_reservation is allready done.
6934 post_reservation
= 0;
6937 reservation_size
= pre_reservation
+ post_reservation
;
6939 if (free_size
< reservation_size
) {
6940 dprintf("imsm: Reservation size is greater than free space.\n");
6944 super
->create_offset
= start
+ pre_reservation
;
6945 return free_size
- reservation_size
;
6948 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6950 if (level
< 0 || level
== 6 || level
== 4)
6953 /* if we have an orom prevent invalid raid levels */
6956 case 0: return imsm_orom_has_raid0(orom
);
6959 return imsm_orom_has_raid1e(orom
);
6960 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6961 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6962 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6965 return 1; /* not on an Intel RAID platform so anything goes */
6971 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6972 int dpa
, int verbose
)
6974 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6975 struct mdstat_ent
*memb
;
6981 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6982 if (memb
->metadata_version
&&
6983 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6984 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6985 !is_subarray(memb
->metadata_version
+9) &&
6987 struct dev_member
*dev
= memb
->members
;
6989 while (dev
&& !is_fd_valid(fd
)) {
6990 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6991 num
= snprintf(path
, PATH_MAX
, "%s%s", "/dev/", dev
->name
);
6993 fd
= open(path
, O_RDONLY
, 0);
6994 if (num
<= 0 || !is_fd_valid(fd
)) {
6995 pr_vrb("Cannot open %s: %s\n",
6996 dev
->name
, strerror(errno
));
7002 if (is_fd_valid(fd
) && disk_attached_to_hba(fd
, hba
)) {
7003 struct mdstat_ent
*vol
;
7004 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
7005 if (vol
->active
> 0 &&
7006 vol
->metadata_version
&&
7007 is_container_member(vol
, memb
->devnm
)) {
7012 if (*devlist
&& (found
< dpa
)) {
7013 dv
= xcalloc(1, sizeof(*dv
));
7014 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
7015 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
7018 dv
->next
= *devlist
;
7025 free_mdstat(mdstat
);
7030 static struct md_list
*
7031 get_loop_devices(void)
7034 struct md_list
*devlist
= NULL
;
7037 for(i
= 0; i
< 12; i
++) {
7038 dv
= xcalloc(1, sizeof(*dv
));
7039 dv
->devname
= xmalloc(40);
7040 sprintf(dv
->devname
, "/dev/loop%d", i
);
7048 static struct md_list
*
7049 get_devices(const char *hba_path
)
7051 struct md_list
*devlist
= NULL
;
7058 devlist
= get_loop_devices();
7061 /* scroll through /sys/dev/block looking for devices attached to
7064 dir
= opendir("/sys/dev/block");
7065 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
7070 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
7072 path
= devt_to_devpath(makedev(major
, minor
), 1, NULL
);
7075 if (!path_attached_to_hba(path
, hba_path
)) {
7082 fd
= dev_open(ent
->d_name
, O_RDONLY
);
7083 if (is_fd_valid(fd
)) {
7084 fd2devname(fd
, buf
);
7087 pr_err("cannot open device: %s\n",
7092 dv
= xcalloc(1, sizeof(*dv
));
7093 dv
->devname
= xstrdup(buf
);
7100 devlist
= devlist
->next
;
7110 count_volumes_list(struct md_list
*devlist
, char *homehost
,
7111 int verbose
, int *found
)
7113 struct md_list
*tmpdev
;
7115 struct supertype
*st
;
7117 /* first walk the list of devices to find a consistent set
7118 * that match the criterea, if that is possible.
7119 * We flag the ones we like with 'used'.
7122 st
= match_metadata_desc_imsm("imsm");
7124 pr_vrb("cannot allocate memory for imsm supertype\n");
7128 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7129 char *devname
= tmpdev
->devname
;
7131 struct supertype
*tst
;
7133 if (tmpdev
->used
> 1)
7135 tst
= dup_super(st
);
7137 pr_vrb("cannot allocate memory for imsm supertype\n");
7140 tmpdev
->container
= 0;
7141 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
7142 if (!is_fd_valid(dfd
)) {
7143 dprintf("cannot open device %s: %s\n",
7144 devname
, strerror(errno
));
7146 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
7148 } else if (must_be_container(dfd
)) {
7149 struct supertype
*cst
;
7150 cst
= super_by_fd(dfd
, NULL
);
7152 dprintf("cannot recognize container type %s\n",
7155 } else if (tst
->ss
!= st
->ss
) {
7156 dprintf("non-imsm container - ignore it: %s\n",
7159 } else if (!tst
->ss
->load_container
||
7160 tst
->ss
->load_container(tst
, dfd
, NULL
))
7163 tmpdev
->container
= 1;
7166 cst
->ss
->free_super(cst
);
7168 tmpdev
->st_rdev
= rdev
;
7169 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
7170 dprintf("no RAID superblock on %s\n",
7173 } else if (tst
->ss
->compare_super
== NULL
) {
7174 dprintf("Cannot assemble %s metadata on %s\n",
7175 tst
->ss
->name
, devname
);
7181 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
7182 /* Ignore unrecognised devices during auto-assembly */
7187 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
7189 if (st
->minor_version
== -1)
7190 st
->minor_version
= tst
->minor_version
;
7192 if (memcmp(info
.uuid
, uuid_zero
,
7193 sizeof(int[4])) == 0) {
7194 /* this is a floating spare. It cannot define
7195 * an array unless there are no more arrays of
7196 * this type to be found. It can be included
7197 * in an array of this type though.
7203 if (st
->ss
!= tst
->ss
||
7204 st
->minor_version
!= tst
->minor_version
||
7205 st
->ss
->compare_super(st
, tst
, 1) != 0) {
7206 /* Some mismatch. If exactly one array matches this host,
7207 * we can resolve on that one.
7208 * Or, if we are auto assembling, we just ignore the second
7211 dprintf("superblock on %s doesn't match others - assembly aborted\n",
7217 dprintf("found: devname: %s\n", devname
);
7221 tst
->ss
->free_super(tst
);
7225 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
7226 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
7227 for (iter
= head
; iter
; iter
= iter
->next
) {
7228 dprintf("content->text_version: %s vol\n",
7229 iter
->text_version
);
7230 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
7231 /* do not assemble arrays with unsupported
7233 dprintf("Cannot activate member %s.\n",
7234 iter
->text_version
);
7241 dprintf("No valid super block on device list: err: %d %p\n",
7245 dprintf("no more devices to examine\n");
7248 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7249 if (tmpdev
->used
== 1 && tmpdev
->found
) {
7251 if (count
< tmpdev
->found
)
7254 count
-= tmpdev
->found
;
7257 if (tmpdev
->used
== 1)
7262 st
->ss
->free_super(st
);
7266 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
7269 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
7271 const struct orom_entry
*entry
;
7272 struct devid_list
*dv
, *devid_list
;
7277 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7278 if (strstr(idev
->path
, hba_path
))
7282 if (!idev
|| !idev
->dev_id
)
7285 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7287 if (!entry
|| !entry
->devid_list
)
7290 devid_list
= entry
->devid_list
;
7291 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7292 struct md_list
*devlist
;
7293 struct sys_dev
*device
= NULL
;
7298 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7300 device
= device_by_id(dv
->devid
);
7303 hpath
= device
->path
;
7307 devlist
= get_devices(hpath
);
7308 /* if no intel devices return zero volumes */
7309 if (devlist
== NULL
)
7312 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7314 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7315 if (devlist
== NULL
)
7319 count
+= count_volumes_list(devlist
,
7323 dprintf("found %d count: %d\n", found
, count
);
7326 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7329 struct md_list
*dv
= devlist
;
7330 devlist
= devlist
->next
;
7338 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7342 if (hba
->type
== SYS_DEV_VMD
) {
7343 struct sys_dev
*dev
;
7346 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7347 if (dev
->type
== SYS_DEV_VMD
)
7348 count
+= __count_volumes(dev
->path
, dpa
,
7353 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7356 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7358 /* up to 512 if the plaform supports it, otherwise the platform max.
7359 * 128 if no platform detected
7361 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7363 return min(512, (1 << fs
));
7367 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7368 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7370 /* check/set platform and metadata limits/defaults */
7371 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7372 pr_vrb("platform supports a maximum of %d disks per array\n",
7377 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7378 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7379 pr_vrb("platform does not support raid%d with %d disk%s\n",
7380 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7384 if (*chunk
== 0 || *chunk
== UnSet
)
7385 *chunk
= imsm_default_chunk(super
->orom
);
7387 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7388 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7392 if (layout
!= imsm_level_to_layout(level
)) {
7394 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7395 else if (level
== 10)
7396 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7398 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7403 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7404 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7405 pr_vrb("platform does not support a volume size over 2TB\n");
7412 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7413 * FIX ME add ahci details
7415 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7416 int layout
, int raiddisks
, int *chunk
,
7417 unsigned long long size
,
7418 unsigned long long data_offset
,
7420 unsigned long long *freesize
,
7424 struct intel_super
*super
= st
->sb
;
7425 struct imsm_super
*mpb
;
7427 unsigned long long pos
= 0;
7428 unsigned long long maxsize
;
7432 /* We must have the container info already read in. */
7436 mpb
= super
->anchor
;
7438 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7439 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7443 /* General test: make sure there is space for
7444 * 'raiddisks' device extents of size 'size' at a given
7447 unsigned long long minsize
= size
;
7448 unsigned long long start_offset
= MaxSector
;
7451 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7452 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7457 e
= get_extents(super
, dl
, 0);
7460 unsigned long long esize
;
7461 esize
= e
[i
].start
- pos
;
7462 if (esize
>= minsize
)
7464 if (found
&& start_offset
== MaxSector
) {
7467 } else if (found
&& pos
!= start_offset
) {
7471 pos
= e
[i
].start
+ e
[i
].size
;
7473 } while (e
[i
-1].size
);
7478 if (dcnt
< raiddisks
) {
7480 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7487 /* This device must be a member of the set */
7488 if (!stat_is_blkdev(dev
, &rdev
))
7490 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7491 if (dl
->major
== (int)major(rdev
) &&
7492 dl
->minor
== (int)minor(rdev
))
7497 pr_err("%s is not in the same imsm set\n", dev
);
7499 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7500 /* If a volume is present then the current creation attempt
7501 * cannot incorporate new spares because the orom may not
7502 * understand this configuration (all member disks must be
7503 * members of each array in the container).
7505 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7506 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7508 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7509 mpb
->num_disks
!= raiddisks
) {
7510 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7514 /* retrieve the largest free space block */
7515 e
= get_extents(super
, dl
, 0);
7520 unsigned long long esize
;
7522 esize
= e
[i
].start
- pos
;
7523 if (esize
>= maxsize
)
7525 pos
= e
[i
].start
+ e
[i
].size
;
7527 } while (e
[i
-1].size
);
7532 pr_err("unable to determine free space for: %s\n",
7536 if (maxsize
< size
) {
7538 pr_err("%s not enough space (%llu < %llu)\n",
7539 dev
, maxsize
, size
);
7543 maxsize
= merge_extents(super
, false);
7545 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7546 pr_err("attempting to create a second volume with size less then remaining space.\n");
7548 if (maxsize
< size
|| maxsize
== 0) {
7551 pr_err("no free space left on device. Aborting...\n");
7553 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7559 *freesize
= maxsize
;
7562 int count
= count_volumes(super
->hba
,
7563 super
->orom
->dpa
, verbose
);
7564 if (super
->orom
->vphba
<= count
) {
7565 pr_vrb("platform does not support more than %d raid volumes.\n",
7566 super
->orom
->vphba
);
7574 * imsm_get_free_size() - get the biggest, common free space from members.
7575 * @super: &intel_super pointer, not NULL.
7576 * @raiddisks: number of raid disks.
7577 * @size: requested size, could be 0 (means max size).
7578 * @chunk: requested chunk size in KiB.
7579 * @freesize: pointer for returned size value.
7581 * Return: &IMSM_STATUS_OK or &IMSM_STATUS_ERROR.
7583 * @freesize is set to meaningful value, this can be @size, or calculated
7585 * super->create_offset value is modified and set appropriately in
7586 * merge_extends() for further creation.
7588 static imsm_status_t
imsm_get_free_size(struct intel_super
*super
,
7589 const int raiddisks
,
7590 unsigned long long size
,
7592 unsigned long long *freesize
,
7595 struct imsm_super
*mpb
= super
->anchor
;
7601 unsigned long long maxsize
;
7602 unsigned long long minsize
= size
;
7605 minsize
= chunk
* 2;
7607 /* find the largest common start free region of the possible disks */
7608 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7614 /* don't activate new spares if we are orom constrained
7615 * and there is already a volume active in the container
7617 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7620 e
= get_extents(super
, dl
, 0);
7623 for (i
= 1; e
[i
-1].size
; i
++)
7630 maxsize
= merge_extents(super
, expanding
);
7631 if (maxsize
< minsize
) {
7632 pr_err("imsm: Free space is %llu but must be equal or larger than %llu.\n",
7634 return IMSM_STATUS_ERROR
;
7637 if (cnt
< raiddisks
|| (super
->orom
&& used
&& used
!= raiddisks
)) {
7638 pr_err("imsm: Not enough devices with space to create array.\n");
7639 return IMSM_STATUS_ERROR
;
7650 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7651 pr_err("attempting to create a second volume with size less then remaining space.\n");
7654 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7656 return IMSM_STATUS_OK
;
7660 * autolayout_imsm() - automatically layout a new volume.
7661 * @super: &intel_super pointer, not NULL.
7662 * @raiddisks: number of raid disks.
7663 * @size: requested size, could be 0 (means max size).
7664 * @chunk: requested chunk.
7665 * @freesize: pointer for returned size value.
7667 * We are being asked to automatically layout a new volume based on the current
7668 * contents of the container. If the parameters can be satisfied autolayout_imsm
7669 * will record the disks, start offset, and will return size of the volume to
7670 * be created. See imsm_get_free_size() for details.
7671 * add_to_super() and getinfo_super() detect when autolayout is in progress.
7672 * If first volume exists, slots are set consistently to it.
7674 * Return: &IMSM_STATUS_OK on success, &IMSM_STATUS_ERROR otherwise.
7676 * Disks are marked for creation via dl->raiddisk.
7678 static imsm_status_t
autolayout_imsm(struct intel_super
*super
,
7679 const int raiddisks
,
7680 unsigned long long size
, const int chunk
,
7681 unsigned long long *freesize
)
7685 int vol_cnt
= super
->anchor
->num_raid_devs
;
7688 rv
= imsm_get_free_size(super
, raiddisks
, size
, chunk
, freesize
, false);
7689 if (rv
!= IMSM_STATUS_OK
)
7690 return IMSM_STATUS_ERROR
;
7692 for (disk
= super
->disks
; disk
; disk
= disk
->next
) {
7696 if (curr_slot
== raiddisks
)
7700 disk
->raiddisk
= curr_slot
;
7702 int _slot
= get_disk_slot_in_dev(super
, 0, disk
->index
);
7705 pr_err("Disk %s is not used in first volume, aborting\n",
7707 return IMSM_STATUS_ERROR
;
7709 disk
->raiddisk
= _slot
;
7714 return IMSM_STATUS_OK
;
7717 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7718 int raiddisks
, int *chunk
, unsigned long long size
,
7719 unsigned long long data_offset
,
7720 char *dev
, unsigned long long *freesize
,
7721 int consistency_policy
, int verbose
)
7728 * if given unused devices create a container
7729 * if given given devices in a container create a member volume
7731 if (is_container(level
))
7732 /* Must be a fresh device to add to a container */
7733 return validate_geometry_imsm_container(st
, level
, raiddisks
,
7738 * Size is given in sectors.
7740 if (size
&& (size
< 2048)) {
7741 pr_err("Given size must be greater than 1M.\n");
7742 /* Depends on algorithm in Create.c :
7743 * if container was given (dev == NULL) return -1,
7744 * if block device was given ( dev != NULL) return 0.
7746 return dev
? -1 : 0;
7750 struct intel_super
*super
= st
->sb
;
7753 * Autolayout mode, st->sb must be set.
7756 pr_vrb("superblock must be set for autolayout, aborting\n");
7760 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7761 raiddisks
, chunk
, size
,
7765 if (super
->orom
&& freesize
) {
7767 int count
= count_volumes(super
->hba
, super
->orom
->dpa
,
7769 if (super
->orom
->vphba
<= count
) {
7770 pr_vrb("platform does not support more than %d raid volumes.\n",
7771 super
->orom
->vphba
);
7775 rv
= autolayout_imsm(super
, raiddisks
, size
, *chunk
,
7777 if (rv
!= IMSM_STATUS_OK
)
7783 /* creating in a given container */
7784 return validate_geometry_imsm_volume(st
, level
, layout
,
7785 raiddisks
, chunk
, size
,
7787 dev
, freesize
, verbose
);
7790 /* This device needs to be a device in an 'imsm' container */
7791 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7793 if (is_fd_valid(fd
)) {
7794 pr_vrb("Cannot create this array on device %s\n", dev
);
7799 fd
= open(dev
, O_RDONLY
, 0);
7801 if (!is_fd_valid(fd
)) {
7802 pr_vrb("Cannot open %s: %s\n", dev
, strerror(errno
));
7806 /* Well, it is in use by someone, maybe an 'imsm' container. */
7807 cfd
= open_container(fd
);
7810 if (!is_fd_valid(cfd
)) {
7811 pr_vrb("Cannot use %s: It is busy\n", dev
);
7814 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7815 if (sra
&& sra
->array
.major_version
== -1 &&
7816 strcmp(sra
->text_version
, "imsm") == 0)
7820 /* This is a member of a imsm container. Load the container
7821 * and try to create a volume
7823 struct intel_super
*super
;
7825 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7827 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7829 return validate_geometry_imsm_volume(st
, level
, layout
,
7831 size
, data_offset
, dev
,
7838 pr_err("failed container membership check\n");
7844 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7846 struct intel_super
*super
= st
->sb
;
7848 if (level
&& *level
== UnSet
)
7849 *level
= LEVEL_CONTAINER
;
7851 if (level
&& layout
&& *layout
== UnSet
)
7852 *layout
= imsm_level_to_layout(*level
);
7854 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7855 *chunk
= imsm_default_chunk(super
->orom
);
7858 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7860 static int kill_subarray_imsm(struct supertype
*st
, char *subarray_id
)
7862 /* remove the subarray currently referenced by subarray_id */
7864 struct intel_dev
**dp
;
7865 struct intel_super
*super
= st
->sb
;
7866 __u8 current_vol
= strtoul(subarray_id
, NULL
, 10);
7867 struct imsm_super
*mpb
= super
->anchor
;
7869 if (mpb
->num_raid_devs
== 0)
7872 /* block deletions that would change the uuid of active subarrays
7874 * FIXME when immutable ids are available, but note that we'll
7875 * also need to fixup the invalidated/active subarray indexes in
7878 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7881 if (i
< current_vol
)
7883 snprintf(subarray
, sizeof(subarray
), "%u", i
);
7884 if (is_subarray_active(subarray
, st
->devnm
)) {
7885 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7892 if (st
->update_tail
) {
7893 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7895 u
->type
= update_kill_array
;
7896 u
->dev_idx
= current_vol
;
7897 append_metadata_update(st
, u
, sizeof(*u
));
7902 for (dp
= &super
->devlist
; *dp
;)
7903 if ((*dp
)->index
== current_vol
) {
7906 handle_missing(super
, (*dp
)->dev
);
7907 if ((*dp
)->index
> current_vol
)
7912 /* no more raid devices, all active components are now spares,
7913 * but of course failed are still failed
7915 if (--mpb
->num_raid_devs
== 0) {
7918 for (d
= super
->disks
; d
; d
= d
->next
)
7923 super
->updates_pending
++;
7929 * get_rwh_policy_from_update() - Get the rwh policy for update option.
7930 * @update: Update option.
7932 static int get_rwh_policy_from_update(enum update_opt update
)
7936 return RWH_MULTIPLE_DISTRIBUTED
;
7938 return RWH_MULTIPLE_OFF
;
7941 case UOPT_NO_BITMAP
:
7946 return UOPT_UNDEFINED
;
7949 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7950 enum update_opt update
, struct mddev_ident
*ident
)
7952 /* update the subarray currently referenced by ->current_vol */
7953 struct intel_super
*super
= st
->sb
;
7954 struct imsm_super
*mpb
= super
->anchor
;
7956 if (update
== UOPT_NAME
) {
7957 char *name
= ident
->name
;
7961 if (imsm_is_name_allowed(super
, name
, 1) == false)
7964 vol
= strtoul(subarray
, &ep
, 10);
7965 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7968 if (st
->update_tail
) {
7969 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7971 u
->type
= update_rename_array
;
7973 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7974 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7975 append_metadata_update(st
, u
, sizeof(*u
));
7977 struct imsm_dev
*dev
;
7980 dev
= get_imsm_dev(super
, vol
);
7981 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7982 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7983 memcpy(dev
->volume
, name
, namelen
);
7984 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7985 dev
= get_imsm_dev(super
, i
);
7986 handle_missing(super
, dev
);
7988 super
->updates_pending
++;
7990 } else if (get_rwh_policy_from_update(update
) != UOPT_UNDEFINED
) {
7993 int vol
= strtoul(subarray
, &ep
, 10);
7995 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7998 new_policy
= get_rwh_policy_from_update(update
);
8000 if (st
->update_tail
) {
8001 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
8003 u
->type
= update_rwh_policy
;
8005 u
->new_policy
= new_policy
;
8006 append_metadata_update(st
, u
, sizeof(*u
));
8008 struct imsm_dev
*dev
;
8010 dev
= get_imsm_dev(super
, vol
);
8011 dev
->rwh_policy
= new_policy
;
8012 super
->updates_pending
++;
8014 if (new_policy
== RWH_BITMAP
)
8015 return write_init_bitmap_imsm_vol(st
, vol
);
8022 static bool is_gen_migration(struct imsm_dev
*dev
)
8024 if (dev
&& dev
->vol
.migr_state
&&
8025 migr_type(dev
) == MIGR_GEN_MIGR
)
8031 static int is_rebuilding(struct imsm_dev
*dev
)
8033 struct imsm_map
*migr_map
;
8035 if (!dev
->vol
.migr_state
)
8038 if (migr_type(dev
) != MIGR_REBUILD
)
8041 migr_map
= get_imsm_map(dev
, MAP_1
);
8043 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
8049 static int is_initializing(struct imsm_dev
*dev
)
8051 struct imsm_map
*migr_map
;
8053 if (!dev
->vol
.migr_state
)
8056 if (migr_type(dev
) != MIGR_INIT
)
8059 migr_map
= get_imsm_map(dev
, MAP_1
);
8061 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8067 static void update_recovery_start(struct intel_super
*super
,
8068 struct imsm_dev
*dev
,
8069 struct mdinfo
*array
)
8071 struct mdinfo
*rebuild
= NULL
;
8075 if (!is_rebuilding(dev
))
8078 /* Find the rebuild target, but punt on the dual rebuild case */
8079 for (d
= array
->devs
; d
; d
= d
->next
)
8080 if (d
->recovery_start
== 0) {
8087 /* (?) none of the disks are marked with
8088 * IMSM_ORD_REBUILD, so assume they are missing and the
8089 * disk_ord_tbl was not correctly updated
8091 dprintf("failed to locate out-of-sync disk\n");
8095 units
= vol_curr_migr_unit(dev
);
8096 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
8099 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
8101 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
8103 /* Given a container loaded by load_super_imsm_all,
8104 * extract information about all the arrays into
8106 * If 'subarray' is given, just extract info about that array.
8108 * For each imsm_dev create an mdinfo, fill it in,
8109 * then look for matching devices in super->disks
8110 * and create appropriate device mdinfo.
8112 struct intel_super
*super
= st
->sb
;
8113 struct imsm_super
*mpb
= super
->anchor
;
8114 struct mdinfo
*rest
= NULL
;
8118 int spare_disks
= 0;
8119 int current_vol
= super
->current_vol
;
8121 /* do not assemble arrays when not all attributes are supported */
8122 if (imsm_check_attributes(mpb
->attributes
) == 0) {
8124 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
8127 /* count spare devices, not used in maps
8129 for (d
= super
->disks
; d
; d
= d
->next
)
8133 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8134 struct imsm_dev
*dev
;
8135 struct imsm_map
*map
;
8136 struct imsm_map
*map2
;
8137 struct mdinfo
*this;
8144 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
8147 dev
= get_imsm_dev(super
, i
);
8148 map
= get_imsm_map(dev
, MAP_0
);
8149 map2
= get_imsm_map(dev
, MAP_1
);
8150 level
= get_imsm_raid_level(map
);
8152 /* do not publish arrays that are in the middle of an
8153 * unsupported migration
8155 if (dev
->vol
.migr_state
&&
8156 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
8157 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
8161 /* do not publish arrays that are not support by controller's
8165 this = xmalloc(sizeof(*this));
8167 super
->current_vol
= i
;
8168 getinfo_super_imsm_volume(st
, this, NULL
);
8170 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
8171 /* mdadm does not support all metadata features- set the bit in all arrays state */
8172 if (!validate_geometry_imsm_orom(super
,
8173 level
, /* RAID level */
8174 imsm_level_to_layout(level
),
8175 map
->num_members
, /* raid disks */
8176 &chunk
, imsm_dev_size(dev
),
8178 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
8180 this->array
.state
|=
8181 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8182 (1<<MD_SB_BLOCK_VOLUME
);
8185 /* if array has bad blocks, set suitable bit in all arrays state */
8187 this->array
.state
|=
8188 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8189 (1<<MD_SB_BLOCK_VOLUME
);
8191 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
8192 unsigned long long recovery_start
;
8193 struct mdinfo
*info_d
;
8201 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
8202 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
8203 for (d
= super
->disks
; d
; d
= d
->next
)
8204 if (d
->index
== idx
)
8207 recovery_start
= MaxSector
;
8210 if (d
&& is_failed(&d
->disk
))
8212 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
8214 if (!(ord
& IMSM_ORD_REBUILD
))
8215 this->array
.working_disks
++;
8217 * if we skip some disks the array will be assmebled degraded;
8218 * reset resync start to avoid a dirty-degraded
8219 * situation when performing the intial sync
8224 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
8225 if ((!able_to_resync(level
, missing
) ||
8226 recovery_start
== 0))
8227 this->resync_start
= MaxSector
;
8233 info_d
= xcalloc(1, sizeof(*info_d
));
8234 info_d
->next
= this->devs
;
8235 this->devs
= info_d
;
8237 info_d
->disk
.number
= d
->index
;
8238 info_d
->disk
.major
= d
->major
;
8239 info_d
->disk
.minor
= d
->minor
;
8240 info_d
->disk
.raid_disk
= slot
;
8241 info_d
->recovery_start
= recovery_start
;
8243 if (slot
< map2
->num_members
)
8244 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8246 this->array
.spare_disks
++;
8248 if (slot
< map
->num_members
)
8249 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8251 this->array
.spare_disks
++;
8254 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
8255 info_d
->data_offset
= pba_of_lba0(map
);
8256 info_d
->component_size
= calc_component_size(map
, dev
);
8258 if (map
->raid_level
== 5) {
8259 info_d
->ppl_sector
= this->ppl_sector
;
8260 info_d
->ppl_size
= this->ppl_size
;
8261 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
8262 recovery_start
== 0)
8263 this->resync_start
= 0;
8266 info_d
->bb
.supported
= 1;
8267 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
8268 info_d
->data_offset
,
8269 info_d
->component_size
,
8272 /* now that the disk list is up-to-date fixup recovery_start */
8273 update_recovery_start(super
, dev
, this);
8274 this->array
.spare_disks
+= spare_disks
;
8276 /* check for reshape */
8277 if (this->reshape_active
== 1)
8278 recover_backup_imsm(st
, this);
8282 super
->current_vol
= current_vol
;
8286 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
8287 int failed
, int look_in_map
)
8289 struct imsm_map
*map
;
8291 map
= get_imsm_map(dev
, look_in_map
);
8294 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
8295 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
8297 switch (get_imsm_raid_level(map
)) {
8299 return IMSM_T_STATE_FAILED
;
8302 if (failed
< map
->num_members
)
8303 return IMSM_T_STATE_DEGRADED
;
8305 return IMSM_T_STATE_FAILED
;
8310 * check to see if any mirrors have failed, otherwise we
8311 * are degraded. Even numbered slots are mirrored on
8315 /* gcc -Os complains that this is unused */
8316 int insync
= insync
;
8318 for (i
= 0; i
< map
->num_members
; i
++) {
8319 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8320 int idx
= ord_to_idx(ord
);
8321 struct imsm_disk
*disk
;
8323 /* reset the potential in-sync count on even-numbered
8324 * slots. num_copies is always 2 for imsm raid10
8329 disk
= get_imsm_disk(super
, idx
);
8330 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8333 /* no in-sync disks left in this mirror the
8337 return IMSM_T_STATE_FAILED
;
8340 return IMSM_T_STATE_DEGRADED
;
8344 return IMSM_T_STATE_DEGRADED
;
8346 return IMSM_T_STATE_FAILED
;
8352 return map
->map_state
;
8355 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8360 struct imsm_disk
*disk
;
8361 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8362 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8363 struct imsm_map
*map_for_loop
;
8368 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8369 * disks that are being rebuilt. New failures are recorded to
8370 * map[0]. So we look through all the disks we started with and
8371 * see if any failures are still present, or if any new ones
8375 if (prev
&& (map
->num_members
< prev
->num_members
))
8376 map_for_loop
= prev
;
8378 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8380 /* when MAP_X is passed both maps failures are counted
8383 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8384 i
< prev
->num_members
) {
8385 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8386 idx_1
= ord_to_idx(ord
);
8388 disk
= get_imsm_disk(super
, idx_1
);
8389 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8392 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8393 i
< map
->num_members
) {
8394 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8395 idx
= ord_to_idx(ord
);
8398 disk
= get_imsm_disk(super
, idx
);
8399 if (!disk
|| is_failed(disk
) ||
8400 ord
& IMSM_ORD_REBUILD
)
8409 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8412 struct intel_super
*super
= c
->sb
;
8413 struct imsm_super
*mpb
= super
->anchor
;
8414 struct imsm_update_prealloc_bb_mem u
;
8416 if (inst
>= mpb
->num_raid_devs
) {
8417 pr_err("subarry index %d, out of range\n", inst
);
8421 dprintf("imsm: open_new %d\n", inst
);
8422 a
->info
.container_member
= inst
;
8424 u
.type
= update_prealloc_badblocks_mem
;
8425 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8430 static int is_resyncing(struct imsm_dev
*dev
)
8432 struct imsm_map
*migr_map
;
8434 if (!dev
->vol
.migr_state
)
8437 if (migr_type(dev
) == MIGR_INIT
||
8438 migr_type(dev
) == MIGR_REPAIR
)
8441 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8444 migr_map
= get_imsm_map(dev
, MAP_1
);
8446 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8447 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8453 /* return true if we recorded new information */
8454 static int mark_failure(struct intel_super
*super
,
8455 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8459 struct imsm_map
*map
;
8460 char buf
[MAX_RAID_SERIAL_LEN
+3];
8461 unsigned int len
, shift
= 0;
8463 /* new failures are always set in map[0] */
8464 map
= get_imsm_map(dev
, MAP_0
);
8466 slot
= get_imsm_disk_slot(map
, idx
);
8470 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8471 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8474 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8475 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8477 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8478 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8479 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8481 disk
->status
|= FAILED_DISK
;
8482 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8483 /* mark failures in second map if second map exists and this disk
8485 * This is valid for migration, initialization and rebuild
8487 if (dev
->vol
.migr_state
) {
8488 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8489 int slot2
= get_imsm_disk_slot(map2
, idx
);
8491 if (slot2
< map2
->num_members
&& slot2
>= 0)
8492 set_imsm_ord_tbl_ent(map2
, slot2
,
8493 idx
| IMSM_ORD_REBUILD
);
8495 if (map
->failed_disk_num
== 0xff ||
8496 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8497 map
->failed_disk_num
= slot
;
8499 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8504 static void mark_missing(struct intel_super
*super
,
8505 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8507 mark_failure(super
, dev
, disk
, idx
);
8509 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8512 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8513 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8516 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8520 if (!super
->missing
)
8523 /* When orom adds replacement for missing disk it does
8524 * not remove entry of missing disk, but just updates map with
8525 * new added disk. So it is not enough just to test if there is
8526 * any missing disk, we have to look if there are any failed disks
8527 * in map to stop migration */
8529 dprintf("imsm: mark missing\n");
8530 /* end process for initialization and rebuild only
8532 if (is_gen_migration(dev
) == false) {
8533 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8537 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8538 struct imsm_map
*map1
;
8539 int i
, ord
, ord_map1
;
8542 for (i
= 0; i
< map
->num_members
; i
++) {
8543 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8544 if (!(ord
& IMSM_ORD_REBUILD
))
8547 map1
= get_imsm_map(dev
, MAP_1
);
8551 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8552 if (ord_map1
& IMSM_ORD_REBUILD
)
8557 map_state
= imsm_check_degraded(super
, dev
,
8559 end_migration(dev
, super
, map_state
);
8563 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8564 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8565 super
->updates_pending
++;
8568 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8571 unsigned long long array_blocks
;
8572 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8573 int used_disks
= imsm_num_data_members(map
);
8575 if (used_disks
== 0) {
8576 /* when problems occures
8577 * return current array_blocks value
8579 array_blocks
= imsm_dev_size(dev
);
8581 return array_blocks
;
8584 /* set array size in metadata
8587 /* OLCE size change is caused by added disks
8589 array_blocks
= per_dev_array_size(map
) * used_disks
;
8591 /* Online Volume Size Change
8592 * Using available free space
8594 array_blocks
= new_size
;
8596 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8597 set_imsm_dev_size(dev
, array_blocks
);
8599 return array_blocks
;
8602 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8604 static void imsm_progress_container_reshape(struct intel_super
*super
)
8606 /* if no device has a migr_state, but some device has a
8607 * different number of members than the previous device, start
8608 * changing the number of devices in this device to match
8611 struct imsm_super
*mpb
= super
->anchor
;
8612 int prev_disks
= -1;
8616 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8617 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8618 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8619 struct imsm_map
*map2
;
8620 int prev_num_members
;
8622 if (dev
->vol
.migr_state
)
8625 if (prev_disks
== -1)
8626 prev_disks
= map
->num_members
;
8627 if (prev_disks
== map
->num_members
)
8630 /* OK, this array needs to enter reshape mode.
8631 * i.e it needs a migr_state
8634 copy_map_size
= sizeof_imsm_map(map
);
8635 prev_num_members
= map
->num_members
;
8636 map
->num_members
= prev_disks
;
8637 dev
->vol
.migr_state
= 1;
8638 set_vol_curr_migr_unit(dev
, 0);
8639 set_migr_type(dev
, MIGR_GEN_MIGR
);
8640 for (i
= prev_num_members
;
8641 i
< map
->num_members
; i
++)
8642 set_imsm_ord_tbl_ent(map
, i
, i
);
8643 map2
= get_imsm_map(dev
, MAP_1
);
8644 /* Copy the current map */
8645 memcpy(map2
, map
, copy_map_size
);
8646 map2
->num_members
= prev_num_members
;
8648 imsm_set_array_size(dev
, -1);
8649 super
->clean_migration_record_by_mdmon
= 1;
8650 super
->updates_pending
++;
8654 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8655 * states are handled in imsm_set_disk() with one exception, when a
8656 * resync is stopped due to a new failure this routine will set the
8657 * 'degraded' state for the array.
8659 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8661 int inst
= a
->info
.container_member
;
8662 struct intel_super
*super
= a
->container
->sb
;
8663 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8664 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8665 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8666 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8667 __u32 blocks_per_unit
;
8669 if (dev
->vol
.migr_state
&&
8670 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8671 /* array state change is blocked due to reshape action
8673 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8674 * - finish the reshape (if last_checkpoint is big and action != reshape)
8675 * - update vol_curr_migr_unit
8677 if (a
->curr_action
== reshape
) {
8678 /* still reshaping, maybe update vol_curr_migr_unit */
8679 goto mark_checkpoint
;
8681 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8682 /* for some reason we aborted the reshape.
8684 * disable automatic metadata rollback
8685 * user action is required to recover process
8688 struct imsm_map
*map2
=
8689 get_imsm_map(dev
, MAP_1
);
8690 dev
->vol
.migr_state
= 0;
8691 set_migr_type(dev
, 0);
8692 set_vol_curr_migr_unit(dev
, 0);
8694 sizeof_imsm_map(map2
));
8695 super
->updates_pending
++;
8698 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8699 unsigned long long array_blocks
;
8703 used_disks
= imsm_num_data_members(map
);
8704 if (used_disks
> 0) {
8706 per_dev_array_size(map
) *
8709 round_size_to_mb(array_blocks
,
8711 a
->info
.custom_array_size
= array_blocks
;
8712 /* encourage manager to update array
8716 a
->check_reshape
= 1;
8718 /* finalize online capacity expansion/reshape */
8719 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8721 mdi
->disk
.raid_disk
,
8724 imsm_progress_container_reshape(super
);
8729 /* before we activate this array handle any missing disks */
8730 if (consistent
== 2)
8731 handle_missing(super
, dev
);
8733 if (consistent
== 2 &&
8734 (!is_resync_complete(&a
->info
) ||
8735 map_state
!= IMSM_T_STATE_NORMAL
||
8736 dev
->vol
.migr_state
))
8739 if (is_resync_complete(&a
->info
)) {
8740 /* complete intialization / resync,
8741 * recovery and interrupted recovery is completed in
8744 if (is_resyncing(dev
)) {
8745 dprintf("imsm: mark resync done\n");
8746 end_migration(dev
, super
, map_state
);
8747 super
->updates_pending
++;
8748 a
->last_checkpoint
= 0;
8750 } else if ((!is_resyncing(dev
) && !failed
) &&
8751 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8752 /* mark the start of the init process if nothing is failed */
8753 dprintf("imsm: mark resync start\n");
8754 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8755 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8757 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8758 super
->updates_pending
++;
8762 /* skip checkpointing for general migration,
8763 * it is controlled in mdadm
8765 if (is_gen_migration(dev
))
8766 goto skip_mark_checkpoint
;
8768 /* check if we can update vol_curr_migr_unit from resync_start,
8771 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8772 if (blocks_per_unit
) {
8773 set_vol_curr_migr_unit(dev
,
8774 a
->last_checkpoint
/ blocks_per_unit
);
8775 dprintf("imsm: mark checkpoint (%llu)\n",
8776 vol_curr_migr_unit(dev
));
8777 super
->updates_pending
++;
8780 skip_mark_checkpoint
:
8781 /* mark dirty / clean */
8782 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8783 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8784 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8786 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8788 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8789 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8790 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8791 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8793 super
->updates_pending
++;
8799 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8801 int inst
= a
->info
.container_member
;
8802 struct intel_super
*super
= a
->container
->sb
;
8803 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8804 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8806 if (slot
> map
->num_members
) {
8807 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8808 slot
, map
->num_members
- 1);
8815 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8818 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8820 int inst
= a
->info
.container_member
;
8821 struct intel_super
*super
= a
->container
->sb
;
8822 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8823 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8824 struct imsm_disk
*disk
;
8826 int recovery_not_finished
= 0;
8830 int rebuild_done
= 0;
8833 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8837 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8838 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8840 /* check for new failures */
8841 if (disk
&& (state
& DS_FAULTY
)) {
8842 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8843 super
->updates_pending
++;
8846 /* check if in_sync */
8847 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8848 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8850 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8852 super
->updates_pending
++;
8855 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8856 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8858 /* check if recovery complete, newly degraded, or failed */
8859 dprintf("imsm: Detected transition to state ");
8860 switch (map_state
) {
8861 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8862 dprintf("normal: ");
8863 if (is_rebuilding(dev
)) {
8864 dprintf_cont("while rebuilding");
8865 /* check if recovery is really finished */
8866 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8867 if (mdi
->recovery_start
!= MaxSector
) {
8868 recovery_not_finished
= 1;
8871 if (recovery_not_finished
) {
8873 dprintf("Rebuild has not finished yet, state not changed");
8874 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8875 a
->last_checkpoint
= mdi
->recovery_start
;
8876 super
->updates_pending
++;
8880 end_migration(dev
, super
, map_state
);
8881 map
->failed_disk_num
= ~0;
8882 super
->updates_pending
++;
8883 a
->last_checkpoint
= 0;
8886 if (is_gen_migration(dev
)) {
8887 dprintf_cont("while general migration");
8888 if (a
->last_checkpoint
>= a
->info
.component_size
)
8889 end_migration(dev
, super
, map_state
);
8891 map
->map_state
= map_state
;
8892 map
->failed_disk_num
= ~0;
8893 super
->updates_pending
++;
8897 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8898 dprintf_cont("degraded: ");
8899 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8900 dprintf_cont("mark degraded");
8901 map
->map_state
= map_state
;
8902 super
->updates_pending
++;
8903 a
->last_checkpoint
= 0;
8906 if (is_rebuilding(dev
)) {
8907 dprintf_cont("while rebuilding ");
8908 if (state
& DS_FAULTY
) {
8909 dprintf_cont("removing failed drive ");
8910 if (n
== map
->failed_disk_num
) {
8911 dprintf_cont("end migration");
8912 end_migration(dev
, super
, map_state
);
8913 a
->last_checkpoint
= 0;
8915 dprintf_cont("fail detected during rebuild, changing map state");
8916 map
->map_state
= map_state
;
8918 super
->updates_pending
++;
8924 /* check if recovery is really finished */
8925 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8926 if (mdi
->recovery_start
!= MaxSector
) {
8927 recovery_not_finished
= 1;
8930 if (recovery_not_finished
) {
8932 dprintf_cont("Rebuild has not finished yet");
8933 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8934 a
->last_checkpoint
=
8935 mdi
->recovery_start
;
8936 super
->updates_pending
++;
8941 dprintf_cont(" Rebuild done, still degraded");
8942 end_migration(dev
, super
, map_state
);
8943 a
->last_checkpoint
= 0;
8944 super
->updates_pending
++;
8946 for (i
= 0; i
< map
->num_members
; i
++) {
8947 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8949 if (idx
& IMSM_ORD_REBUILD
)
8950 map
->failed_disk_num
= i
;
8952 super
->updates_pending
++;
8955 if (is_gen_migration(dev
)) {
8956 dprintf_cont("while general migration");
8957 if (a
->last_checkpoint
>= a
->info
.component_size
)
8958 end_migration(dev
, super
, map_state
);
8960 map
->map_state
= map_state
;
8961 manage_second_map(super
, dev
);
8963 super
->updates_pending
++;
8966 if (is_initializing(dev
)) {
8967 dprintf_cont("while initialization.");
8968 map
->map_state
= map_state
;
8969 super
->updates_pending
++;
8973 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8974 dprintf_cont("failed: ");
8975 if (is_gen_migration(dev
)) {
8976 dprintf_cont("while general migration");
8977 map
->map_state
= map_state
;
8978 super
->updates_pending
++;
8981 if (map
->map_state
!= map_state
) {
8982 dprintf_cont("mark failed");
8983 end_migration(dev
, super
, map_state
);
8984 super
->updates_pending
++;
8985 a
->last_checkpoint
= 0;
8990 dprintf_cont("state %i\n", map_state
);
8995 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8998 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8999 unsigned long long dsize
;
9000 unsigned long long sectors
;
9001 unsigned int sector_size
;
9003 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
9005 get_dev_size(fd
, NULL
, &dsize
);
9007 if (mpb_size
> sector_size
) {
9008 /* -1 to account for anchor */
9009 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
9011 /* write the extended mpb to the sectors preceeding the anchor */
9012 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
9016 if ((unsigned long long)write(fd
, buf
+ sector_size
,
9017 sector_size
* sectors
) != sector_size
* sectors
)
9021 /* first block is stored on second to last sector of the disk */
9022 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
9025 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
9031 static void imsm_sync_metadata(struct supertype
*container
)
9033 struct intel_super
*super
= container
->sb
;
9035 dprintf("sync metadata: %d\n", super
->updates_pending
);
9036 if (!super
->updates_pending
)
9039 write_super_imsm(container
, 0);
9041 super
->updates_pending
= 0;
9044 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
9046 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
9047 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
9050 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9054 if (dl
&& is_failed(&dl
->disk
))
9058 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
9063 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
9064 struct active_array
*a
, int activate_new
,
9065 struct mdinfo
*additional_test_list
)
9067 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
9068 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
9069 struct imsm_super
*mpb
= super
->anchor
;
9070 struct imsm_map
*map
;
9071 unsigned long long pos
;
9076 __u32 array_start
= 0;
9077 __u32 array_end
= 0;
9079 struct mdinfo
*test_list
;
9081 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9082 /* If in this array, skip */
9083 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9084 if (is_fd_valid(d
->state_fd
) &&
9085 d
->disk
.major
== dl
->major
&&
9086 d
->disk
.minor
== dl
->minor
) {
9087 dprintf("%x:%x already in array\n",
9088 dl
->major
, dl
->minor
);
9093 test_list
= additional_test_list
;
9095 if (test_list
->disk
.major
== dl
->major
&&
9096 test_list
->disk
.minor
== dl
->minor
) {
9097 dprintf("%x:%x already in additional test list\n",
9098 dl
->major
, dl
->minor
);
9101 test_list
= test_list
->next
;
9106 /* skip in use or failed drives */
9107 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
9109 dprintf("%x:%x status (failed: %d index: %d)\n",
9110 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
9114 /* skip pure spares when we are looking for partially
9115 * assimilated drives
9117 if (dl
->index
== -1 && !activate_new
)
9120 if (!drive_validate_sector_size(super
, dl
))
9123 /* Does this unused device have the requisite free space?
9124 * It needs to be able to cover all member volumes
9126 ex
= get_extents(super
, dl
, 1);
9128 dprintf("cannot get extents\n");
9131 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9132 dev
= get_imsm_dev(super
, i
);
9133 map
= get_imsm_map(dev
, MAP_0
);
9135 /* check if this disk is already a member of
9138 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
9144 array_start
= pba_of_lba0(map
);
9145 array_end
= array_start
+
9146 per_dev_array_size(map
) - 1;
9149 /* check that we can start at pba_of_lba0 with
9150 * num_data_stripes*blocks_per_stripe of space
9152 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
9156 pos
= ex
[j
].start
+ ex
[j
].size
;
9158 } while (ex
[j
-1].size
);
9165 if (i
< mpb
->num_raid_devs
) {
9166 dprintf("%x:%x does not have %u to %u available\n",
9167 dl
->major
, dl
->minor
, array_start
, array_end
);
9177 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
9179 struct imsm_dev
*dev2
;
9180 struct imsm_map
*map
;
9186 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
9188 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
9189 if (state
== IMSM_T_STATE_FAILED
) {
9190 map
= get_imsm_map(dev2
, MAP_0
);
9191 for (slot
= 0; slot
< map
->num_members
; slot
++) {
9193 * Check if failed disks are deleted from intel
9194 * disk list or are marked to be deleted
9196 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
9197 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
9199 * Do not rebuild the array if failed disks
9200 * from failed sub-array are not removed from
9204 is_failed(&idisk
->disk
) &&
9205 (idisk
->action
!= DISK_REMOVE
))
9212 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
9213 struct metadata_update
**updates
)
9216 * Find a device with unused free space and use it to replace a
9217 * failed/vacant region in an array. We replace failed regions one a
9218 * array at a time. The result is that a new spare disk will be added
9219 * to the first failed array and after the monitor has finished
9220 * propagating failures the remainder will be consumed.
9222 * FIXME add a capability for mdmon to request spares from another
9226 struct intel_super
*super
= a
->container
->sb
;
9227 int inst
= a
->info
.container_member
;
9228 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
9229 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9230 int failed
= a
->info
.array
.raid_disks
;
9231 struct mdinfo
*rv
= NULL
;
9234 struct metadata_update
*mu
;
9236 struct imsm_update_activate_spare
*u
;
9241 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
9242 if (!is_fd_valid(d
->state_fd
))
9245 if (d
->curr_state
& DS_FAULTY
)
9246 /* wait for Removal to happen */
9252 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
9253 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
9255 if (imsm_reshape_blocks_arrays_changes(super
))
9258 /* Cannot activate another spare if rebuild is in progress already
9260 if (is_rebuilding(dev
)) {
9261 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
9265 if (a
->info
.array
.level
== 4)
9266 /* No repair for takeovered array
9267 * imsm doesn't support raid4
9271 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
9272 IMSM_T_STATE_DEGRADED
)
9275 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
9276 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
9281 * If there are any failed disks check state of the other volume.
9282 * Block rebuild if the another one is failed until failed disks
9283 * are removed from container.
9286 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
9287 MAX_RAID_SERIAL_LEN
, dev
->volume
);
9288 /* check if states of the other volumes allow for rebuild */
9289 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
9291 allowed
= imsm_rebuild_allowed(a
->container
,
9299 /* For each slot, if it is not working, find a spare */
9300 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
9301 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9302 if (d
->disk
.raid_disk
== i
)
9304 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
9305 if (d
&& is_fd_valid(d
->state_fd
))
9309 * OK, this device needs recovery. Try to re-add the
9310 * previous occupant of this slot, if this fails see if
9311 * we can continue the assimilation of a spare that was
9312 * partially assimilated, finally try to activate a new
9315 dl
= imsm_readd(super
, i
, a
);
9317 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9319 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9323 /* found a usable disk with enough space */
9324 di
= xcalloc(1, sizeof(*di
));
9326 /* dl->index will be -1 in the case we are activating a
9327 * pristine spare. imsm_process_update() will create a
9328 * new index in this case. Once a disk is found to be
9329 * failed in all member arrays it is kicked from the
9332 di
->disk
.number
= dl
->index
;
9334 /* (ab)use di->devs to store a pointer to the device
9337 di
->devs
= (struct mdinfo
*) dl
;
9339 di
->disk
.raid_disk
= i
;
9340 di
->disk
.major
= dl
->major
;
9341 di
->disk
.minor
= dl
->minor
;
9343 di
->recovery_start
= 0;
9344 di
->data_offset
= pba_of_lba0(map
);
9345 di
->component_size
= a
->info
.component_size
;
9346 di
->container_member
= inst
;
9347 di
->bb
.supported
= 1;
9348 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9349 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9350 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9352 super
->random
= random32();
9356 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9357 i
, di
->data_offset
);
9361 /* No spares found */
9363 /* Now 'rv' has a list of devices to return.
9364 * Create a metadata_update record to update the
9365 * disk_ord_tbl for the array
9367 mu
= xmalloc(sizeof(*mu
));
9368 mu
->buf
= xcalloc(num_spares
,
9369 sizeof(struct imsm_update_activate_spare
));
9371 mu
->space_list
= NULL
;
9372 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9373 mu
->next
= *updates
;
9374 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9376 for (di
= rv
; di
; di
= di
->next
) {
9377 u
->type
= update_activate_spare
;
9378 u
->dl
= (struct dl
*) di
->devs
;
9380 u
->slot
= di
->disk
.raid_disk
;
9391 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9393 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9394 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9395 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9396 struct disk_info
*inf
= get_disk_info(u
);
9397 struct imsm_disk
*disk
;
9401 for (i
= 0; i
< map
->num_members
; i
++) {
9402 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9403 for (j
= 0; j
< new_map
->num_members
; j
++)
9404 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9411 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9415 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9416 if (dl
->major
== major
&& dl
->minor
== minor
)
9421 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9427 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9428 if (dl
->major
== major
&& dl
->minor
== minor
) {
9431 prev
->next
= dl
->next
;
9433 super
->disks
= dl
->next
;
9435 __free_imsm_disk(dl
, 1);
9436 dprintf("removed %x:%x\n", major
, minor
);
9444 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9446 static int add_remove_disk_update(struct intel_super
*super
)
9448 int check_degraded
= 0;
9451 /* add/remove some spares to/from the metadata/contrainer */
9452 while (super
->disk_mgmt_list
) {
9453 struct dl
*disk_cfg
;
9455 disk_cfg
= super
->disk_mgmt_list
;
9456 super
->disk_mgmt_list
= disk_cfg
->next
;
9457 disk_cfg
->next
= NULL
;
9459 if (disk_cfg
->action
== DISK_ADD
) {
9460 disk_cfg
->next
= super
->disks
;
9461 super
->disks
= disk_cfg
;
9463 dprintf("added %x:%x\n",
9464 disk_cfg
->major
, disk_cfg
->minor
);
9465 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9466 dprintf("Disk remove action processed: %x.%x\n",
9467 disk_cfg
->major
, disk_cfg
->minor
);
9468 disk
= get_disk_super(super
,
9472 /* store action status */
9473 disk
->action
= DISK_REMOVE
;
9474 /* remove spare disks only */
9475 if (disk
->index
== -1) {
9476 remove_disk_super(super
,
9480 disk_cfg
->fd
= disk
->fd
;
9484 /* release allocate disk structure */
9485 __free_imsm_disk(disk_cfg
, 1);
9488 return check_degraded
;
9491 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9492 struct intel_super
*super
,
9495 struct intel_dev
*id
;
9496 void **tofree
= NULL
;
9499 dprintf("(enter)\n");
9500 if (u
->subdev
< 0 || u
->subdev
> 1) {
9501 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9504 if (space_list
== NULL
|| *space_list
== NULL
) {
9505 dprintf("imsm: Error: Memory is not allocated\n");
9509 for (id
= super
->devlist
; id
; id
= id
->next
) {
9510 if (id
->index
== (unsigned)u
->subdev
) {
9511 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9512 struct imsm_map
*map
;
9513 struct imsm_dev
*new_dev
=
9514 (struct imsm_dev
*)*space_list
;
9515 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9517 struct dl
*new_disk
;
9519 if (new_dev
== NULL
)
9521 *space_list
= **space_list
;
9522 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9523 map
= get_imsm_map(new_dev
, MAP_0
);
9525 dprintf("imsm: Error: migration in progress");
9529 to_state
= map
->map_state
;
9530 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9532 /* this should not happen */
9533 if (u
->new_disks
[0] < 0) {
9534 map
->failed_disk_num
=
9535 map
->num_members
- 1;
9536 to_state
= IMSM_T_STATE_DEGRADED
;
9538 to_state
= IMSM_T_STATE_NORMAL
;
9540 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9541 if (u
->new_level
> -1)
9542 map
->raid_level
= u
->new_level
;
9543 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9544 if ((u
->new_level
== 5) &&
9545 (migr_map
->raid_level
== 0)) {
9546 int ord
= map
->num_members
- 1;
9547 migr_map
->num_members
--;
9548 if (u
->new_disks
[0] < 0)
9549 ord
|= IMSM_ORD_REBUILD
;
9550 set_imsm_ord_tbl_ent(map
,
9551 map
->num_members
- 1,
9555 tofree
= (void **)dev
;
9557 /* update chunk size
9559 if (u
->new_chunksize
> 0) {
9560 struct imsm_map
*dest_map
=
9561 get_imsm_map(dev
, MAP_0
);
9563 imsm_num_data_members(dest_map
);
9565 if (used_disks
== 0)
9568 map
->blocks_per_strip
=
9569 __cpu_to_le16(u
->new_chunksize
* 2);
9570 update_num_data_stripes(map
, imsm_dev_size(dev
));
9573 /* ensure blocks_per_member has valid value
9575 set_blocks_per_member(map
,
9576 per_dev_array_size(map
) +
9577 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9581 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9582 migr_map
->raid_level
== map
->raid_level
)
9585 if (u
->new_disks
[0] >= 0) {
9588 new_disk
= get_disk_super(super
,
9589 major(u
->new_disks
[0]),
9590 minor(u
->new_disks
[0]));
9591 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9592 major(u
->new_disks
[0]),
9593 minor(u
->new_disks
[0]),
9594 new_disk
, new_disk
->index
);
9595 if (new_disk
== NULL
)
9596 goto error_disk_add
;
9598 new_disk
->index
= map
->num_members
- 1;
9599 /* slot to fill in autolayout
9601 new_disk
->raiddisk
= new_disk
->index
;
9602 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9603 new_disk
->disk
.status
&= ~SPARE_DISK
;
9605 goto error_disk_add
;
9608 *tofree
= *space_list
;
9609 /* calculate new size
9611 imsm_set_array_size(new_dev
, -1);
9618 *space_list
= tofree
;
9622 dprintf("Error: imsm: Cannot find disk.\n");
9626 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9627 struct intel_super
*super
)
9629 struct intel_dev
*id
;
9632 dprintf("(enter)\n");
9633 if (u
->subdev
< 0 || u
->subdev
> 1) {
9634 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9638 for (id
= super
->devlist
; id
; id
= id
->next
) {
9639 if (id
->index
== (unsigned)u
->subdev
) {
9640 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9641 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9642 int used_disks
= imsm_num_data_members(map
);
9643 unsigned long long blocks_per_member
;
9644 unsigned long long new_size_per_disk
;
9646 if (used_disks
== 0)
9649 /* calculate new size
9651 new_size_per_disk
= u
->new_size
/ used_disks
;
9652 blocks_per_member
= new_size_per_disk
+
9653 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9655 imsm_set_array_size(dev
, u
->new_size
);
9656 set_blocks_per_member(map
, blocks_per_member
);
9657 update_num_data_stripes(map
, u
->new_size
);
9666 static int prepare_spare_to_activate(struct supertype
*st
,
9667 struct imsm_update_activate_spare
*u
)
9669 struct intel_super
*super
= st
->sb
;
9670 int prev_current_vol
= super
->current_vol
;
9671 struct active_array
*a
;
9674 for (a
= st
->arrays
; a
; a
= a
->next
)
9676 * Additional initialization (adding bitmap header, filling
9677 * the bitmap area with '1's to force initial rebuild for a whole
9678 * data-area) is required when adding the spare to the volume
9679 * with write-intent bitmap.
9681 if (a
->info
.container_member
== u
->array
&&
9682 a
->info
.consistency_policy
== CONSISTENCY_POLICY_BITMAP
) {
9685 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9691 super
->current_vol
= u
->array
;
9692 if (st
->ss
->write_bitmap(st
, dl
->fd
, NoUpdate
))
9694 super
->current_vol
= prev_current_vol
;
9699 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9700 struct intel_super
*super
,
9701 struct active_array
*active_array
)
9703 struct imsm_super
*mpb
= super
->anchor
;
9704 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9705 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9706 struct imsm_map
*migr_map
;
9707 struct active_array
*a
;
9708 struct imsm_disk
*disk
;
9715 int second_map_created
= 0;
9717 for (; u
; u
= u
->next
) {
9718 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9723 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9728 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9733 /* count failures (excluding rebuilds and the victim)
9734 * to determine map[0] state
9737 for (i
= 0; i
< map
->num_members
; i
++) {
9740 disk
= get_imsm_disk(super
,
9741 get_imsm_disk_idx(dev
, i
, MAP_X
));
9742 if (!disk
|| is_failed(disk
))
9746 /* adding a pristine spare, assign a new index */
9747 if (dl
->index
< 0) {
9748 dl
->index
= super
->anchor
->num_disks
;
9749 super
->anchor
->num_disks
++;
9752 disk
->status
|= CONFIGURED_DISK
;
9753 disk
->status
&= ~SPARE_DISK
;
9756 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9757 if (!second_map_created
) {
9758 second_map_created
= 1;
9759 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9760 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9762 map
->map_state
= to_state
;
9763 migr_map
= get_imsm_map(dev
, MAP_1
);
9764 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9765 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9766 dl
->index
| IMSM_ORD_REBUILD
);
9768 /* update the family_num to mark a new container
9769 * generation, being careful to record the existing
9770 * family_num in orig_family_num to clean up after
9771 * earlier mdadm versions that neglected to set it.
9773 if (mpb
->orig_family_num
== 0)
9774 mpb
->orig_family_num
= mpb
->family_num
;
9775 mpb
->family_num
+= super
->random
;
9777 /* count arrays using the victim in the metadata */
9779 for (a
= active_array
; a
; a
= a
->next
) {
9780 int dev_idx
= a
->info
.container_member
;
9782 if (get_disk_slot_in_dev(super
, dev_idx
, victim
) >= 0)
9786 /* delete the victim if it is no longer being
9792 /* We know that 'manager' isn't touching anything,
9793 * so it is safe to delete
9795 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9796 if ((*dlp
)->index
== victim
)
9799 /* victim may be on the missing list */
9801 for (dlp
= &super
->missing
; *dlp
;
9802 dlp
= &(*dlp
)->next
)
9803 if ((*dlp
)->index
== victim
)
9805 imsm_delete(super
, dlp
, victim
);
9812 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9813 struct intel_super
*super
,
9816 struct dl
*new_disk
;
9817 struct intel_dev
*id
;
9819 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9820 int disk_count
= u
->old_raid_disks
;
9821 void **tofree
= NULL
;
9822 int devices_to_reshape
= 1;
9823 struct imsm_super
*mpb
= super
->anchor
;
9825 unsigned int dev_id
;
9827 dprintf("(enter)\n");
9829 /* enable spares to use in array */
9830 for (i
= 0; i
< delta_disks
; i
++) {
9831 new_disk
= get_disk_super(super
,
9832 major(u
->new_disks
[i
]),
9833 minor(u
->new_disks
[i
]));
9834 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9835 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9836 new_disk
, new_disk
->index
);
9837 if (new_disk
== NULL
||
9838 (new_disk
->index
>= 0 &&
9839 new_disk
->index
< u
->old_raid_disks
))
9840 goto update_reshape_exit
;
9841 new_disk
->index
= disk_count
++;
9842 /* slot to fill in autolayout
9844 new_disk
->raiddisk
= new_disk
->index
;
9845 new_disk
->disk
.status
|=
9847 new_disk
->disk
.status
&= ~SPARE_DISK
;
9850 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9851 mpb
->num_raid_devs
);
9852 /* manage changes in volume
9854 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9855 void **sp
= *space_list
;
9856 struct imsm_dev
*newdev
;
9857 struct imsm_map
*newmap
, *oldmap
;
9859 for (id
= super
->devlist
; id
; id
= id
->next
) {
9860 if (id
->index
== dev_id
)
9869 /* Copy the dev, but not (all of) the map */
9870 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9871 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9872 newmap
= get_imsm_map(newdev
, MAP_0
);
9873 /* Copy the current map */
9874 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9875 /* update one device only
9877 if (devices_to_reshape
) {
9878 dprintf("imsm: modifying subdev: %i\n",
9880 devices_to_reshape
--;
9881 newdev
->vol
.migr_state
= 1;
9882 set_vol_curr_migr_unit(newdev
, 0);
9883 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9884 newmap
->num_members
= u
->new_raid_disks
;
9885 for (i
= 0; i
< delta_disks
; i
++) {
9886 set_imsm_ord_tbl_ent(newmap
,
9887 u
->old_raid_disks
+ i
,
9888 u
->old_raid_disks
+ i
);
9890 /* New map is correct, now need to save old map
9892 newmap
= get_imsm_map(newdev
, MAP_1
);
9893 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9895 imsm_set_array_size(newdev
, -1);
9898 sp
= (void **)id
->dev
;
9903 /* Clear migration record */
9904 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9907 *space_list
= tofree
;
9910 update_reshape_exit
:
9915 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9916 struct intel_super
*super
,
9919 struct imsm_dev
*dev
= NULL
;
9920 struct intel_dev
*dv
;
9921 struct imsm_dev
*dev_new
;
9922 struct imsm_map
*map
;
9926 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9927 if (dv
->index
== (unsigned int)u
->subarray
) {
9935 map
= get_imsm_map(dev
, MAP_0
);
9937 if (u
->direction
== R10_TO_R0
) {
9938 /* Number of failed disks must be half of initial disk number */
9939 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9940 (map
->num_members
/ 2))
9943 /* iterate through devices to mark removed disks as spare */
9944 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9945 if (dm
->disk
.status
& FAILED_DISK
) {
9946 int idx
= dm
->index
;
9947 /* update indexes on the disk list */
9948 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9949 the index values will end up being correct.... NB */
9950 for (du
= super
->disks
; du
; du
= du
->next
)
9951 if (du
->index
> idx
)
9953 /* mark as spare disk */
9958 map
->num_members
/= map
->num_domains
;
9959 map
->map_state
= IMSM_T_STATE_NORMAL
;
9960 map
->raid_level
= 0;
9961 set_num_domains(map
);
9962 update_num_data_stripes(map
, imsm_dev_size(dev
));
9963 map
->failed_disk_num
= -1;
9966 if (u
->direction
== R0_TO_R10
) {
9969 /* update slots in current disk list */
9970 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9974 /* create new *missing* disks */
9975 for (i
= 0; i
< map
->num_members
; i
++) {
9976 space
= *space_list
;
9979 *space_list
= *space
;
9981 memcpy(du
, super
->disks
, sizeof(*du
));
9985 du
->index
= (i
* 2) + 1;
9986 sprintf((char *)du
->disk
.serial
,
9987 " MISSING_%d", du
->index
);
9988 sprintf((char *)du
->serial
,
9989 "MISSING_%d", du
->index
);
9990 du
->next
= super
->missing
;
9991 super
->missing
= du
;
9993 /* create new dev and map */
9994 space
= *space_list
;
9997 *space_list
= *space
;
9998 dev_new
= (void *)space
;
9999 memcpy(dev_new
, dev
, sizeof(*dev
));
10000 /* update new map */
10001 map
= get_imsm_map(dev_new
, MAP_0
);
10003 map
->map_state
= IMSM_T_STATE_DEGRADED
;
10004 map
->raid_level
= 1;
10005 set_num_domains(map
);
10006 map
->num_members
= map
->num_members
* map
->num_domains
;
10007 update_num_data_stripes(map
, imsm_dev_size(dev
));
10009 /* replace dev<->dev_new */
10012 /* update disk order table */
10013 for (du
= super
->disks
; du
; du
= du
->next
)
10014 if (du
->index
>= 0)
10015 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
10016 for (du
= super
->missing
; du
; du
= du
->next
)
10017 if (du
->index
>= 0) {
10018 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
10019 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
10025 static void imsm_process_update(struct supertype
*st
,
10026 struct metadata_update
*update
)
10029 * crack open the metadata_update envelope to find the update record
10030 * update can be one of:
10031 * update_reshape_container_disks - all the arrays in the container
10032 * are being reshaped to have more devices. We need to mark
10033 * the arrays for general migration and convert selected spares
10034 * into active devices.
10035 * update_activate_spare - a spare device has replaced a failed
10036 * device in an array, update the disk_ord_tbl. If this disk is
10037 * present in all member arrays then also clear the SPARE_DISK
10039 * update_create_array
10040 * update_kill_array
10041 * update_rename_array
10042 * update_add_remove_disk
10044 struct intel_super
*super
= st
->sb
;
10045 struct imsm_super
*mpb
;
10046 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
10048 /* update requires a larger buf but the allocation failed */
10049 if (super
->next_len
&& !super
->next_buf
) {
10050 super
->next_len
= 0;
10054 if (super
->next_buf
) {
10055 memcpy(super
->next_buf
, super
->buf
, super
->len
);
10057 super
->len
= super
->next_len
;
10058 super
->buf
= super
->next_buf
;
10060 super
->next_len
= 0;
10061 super
->next_buf
= NULL
;
10064 mpb
= super
->anchor
;
10067 case update_general_migration_checkpoint
: {
10068 struct intel_dev
*id
;
10069 struct imsm_update_general_migration_checkpoint
*u
=
10070 (void *)update
->buf
;
10072 dprintf("called for update_general_migration_checkpoint\n");
10074 /* find device under general migration */
10075 for (id
= super
->devlist
; id
; id
= id
->next
) {
10076 if (is_gen_migration(id
->dev
)) {
10077 set_vol_curr_migr_unit(id
->dev
,
10078 u
->curr_migr_unit
);
10079 super
->updates_pending
++;
10084 case update_takeover
: {
10085 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10086 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
10087 imsm_update_version_info(super
);
10088 super
->updates_pending
++;
10093 case update_reshape_container_disks
: {
10094 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10095 if (apply_reshape_container_disks_update(
10096 u
, super
, &update
->space_list
))
10097 super
->updates_pending
++;
10100 case update_reshape_migration
: {
10101 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10102 if (apply_reshape_migration_update(
10103 u
, super
, &update
->space_list
))
10104 super
->updates_pending
++;
10107 case update_size_change
: {
10108 struct imsm_update_size_change
*u
= (void *)update
->buf
;
10109 if (apply_size_change_update(u
, super
))
10110 super
->updates_pending
++;
10113 case update_activate_spare
: {
10114 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
10116 if (prepare_spare_to_activate(st
, u
) &&
10117 apply_update_activate_spare(u
, super
, st
->arrays
))
10118 super
->updates_pending
++;
10121 case update_create_array
: {
10122 /* someone wants to create a new array, we need to be aware of
10123 * a few races/collisions:
10124 * 1/ 'Create' called by two separate instances of mdadm
10125 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
10126 * devices that have since been assimilated via
10128 * In the event this update can not be carried out mdadm will
10129 * (FIX ME) notice that its update did not take hold.
10131 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10132 struct intel_dev
*dv
;
10133 struct imsm_dev
*dev
;
10134 struct imsm_map
*map
, *new_map
;
10135 unsigned long long start
, end
;
10136 unsigned long long new_start
, new_end
;
10138 struct disk_info
*inf
;
10141 /* handle racing creates: first come first serve */
10142 if (u
->dev_idx
< mpb
->num_raid_devs
) {
10143 dprintf("subarray %d already defined\n", u
->dev_idx
);
10147 /* check update is next in sequence */
10148 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
10149 dprintf("can not create array %d expected index %d\n",
10150 u
->dev_idx
, mpb
->num_raid_devs
);
10154 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
10155 new_start
= pba_of_lba0(new_map
);
10156 new_end
= new_start
+ per_dev_array_size(new_map
);
10157 inf
= get_disk_info(u
);
10159 /* handle activate_spare versus create race:
10160 * check to make sure that overlapping arrays do not include
10161 * overalpping disks
10163 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10164 dev
= get_imsm_dev(super
, i
);
10165 map
= get_imsm_map(dev
, MAP_0
);
10166 start
= pba_of_lba0(map
);
10167 end
= start
+ per_dev_array_size(map
);
10168 if ((new_start
>= start
&& new_start
<= end
) ||
10169 (start
>= new_start
&& start
<= new_end
))
10174 if (disks_overlap(super
, i
, u
)) {
10175 dprintf("arrays overlap\n");
10180 /* check that prepare update was successful */
10181 if (!update
->space
) {
10182 dprintf("prepare update failed\n");
10186 /* check that all disks are still active before committing
10187 * changes. FIXME: could we instead handle this by creating a
10188 * degraded array? That's probably not what the user expects,
10189 * so better to drop this update on the floor.
10191 for (i
= 0; i
< new_map
->num_members
; i
++) {
10192 dl
= serial_to_dl(inf
[i
].serial
, super
);
10194 dprintf("disk disappeared\n");
10199 super
->updates_pending
++;
10201 /* convert spares to members and fixup ord_tbl */
10202 for (i
= 0; i
< new_map
->num_members
; i
++) {
10203 dl
= serial_to_dl(inf
[i
].serial
, super
);
10204 if (dl
->index
== -1) {
10205 dl
->index
= mpb
->num_disks
;
10207 dl
->disk
.status
|= CONFIGURED_DISK
;
10208 dl
->disk
.status
&= ~SPARE_DISK
;
10210 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
10213 dv
= update
->space
;
10215 update
->space
= NULL
;
10216 imsm_copy_dev(dev
, &u
->dev
);
10217 dv
->index
= u
->dev_idx
;
10218 dv
->next
= super
->devlist
;
10219 super
->devlist
= dv
;
10220 mpb
->num_raid_devs
++;
10222 imsm_update_version_info(super
);
10225 /* mdmon knows how to release update->space, but not
10226 * ((struct intel_dev *) update->space)->dev
10228 if (update
->space
) {
10229 dv
= update
->space
;
10234 case update_kill_array
: {
10235 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
10236 int victim
= u
->dev_idx
;
10237 struct active_array
*a
;
10238 struct intel_dev
**dp
;
10240 /* sanity check that we are not affecting the uuid of
10241 * active arrays, or deleting an active array
10243 * FIXME when immutable ids are available, but note that
10244 * we'll also need to fixup the invalidated/active
10245 * subarray indexes in mdstat
10247 for (a
= st
->arrays
; a
; a
= a
->next
)
10248 if (a
->info
.container_member
>= victim
)
10250 /* by definition if mdmon is running at least one array
10251 * is active in the container, so checking
10252 * mpb->num_raid_devs is just extra paranoia
10254 if (a
|| mpb
->num_raid_devs
== 1 || victim
>= super
->anchor
->num_raid_devs
) {
10255 dprintf("failed to delete subarray-%d\n", victim
);
10259 for (dp
= &super
->devlist
; *dp
;)
10260 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
10263 if ((*dp
)->index
> (unsigned)victim
)
10267 mpb
->num_raid_devs
--;
10268 super
->updates_pending
++;
10271 case update_rename_array
: {
10272 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
10273 char name
[MAX_RAID_SERIAL_LEN
+1];
10274 int target
= u
->dev_idx
;
10275 struct active_array
*a
;
10276 struct imsm_dev
*dev
;
10278 /* sanity check that we are not affecting the uuid of
10281 memset(name
, 0, sizeof(name
));
10282 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
10283 name
[MAX_RAID_SERIAL_LEN
] = '\0';
10284 for (a
= st
->arrays
; a
; a
= a
->next
)
10285 if (a
->info
.container_member
== target
)
10287 dev
= get_imsm_dev(super
, u
->dev_idx
);
10289 if (a
|| !dev
|| imsm_is_name_allowed(super
, name
, 0) == false) {
10290 dprintf("failed to rename subarray-%d\n", target
);
10294 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
10295 super
->updates_pending
++;
10298 case update_add_remove_disk
: {
10299 /* we may be able to repair some arrays if disks are
10300 * being added, check the status of add_remove_disk
10301 * if discs has been added.
10303 if (add_remove_disk_update(super
)) {
10304 struct active_array
*a
;
10306 super
->updates_pending
++;
10307 for (a
= st
->arrays
; a
; a
= a
->next
)
10308 a
->check_degraded
= 1;
10312 case update_prealloc_badblocks_mem
:
10314 case update_rwh_policy
: {
10315 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
10316 int target
= u
->dev_idx
;
10317 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
10319 if (dev
->rwh_policy
!= u
->new_policy
) {
10320 dev
->rwh_policy
= u
->new_policy
;
10321 super
->updates_pending
++;
10326 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10330 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10332 static int imsm_prepare_update(struct supertype
*st
,
10333 struct metadata_update
*update
)
10336 * Allocate space to hold new disk entries, raid-device entries or a new
10337 * mpb if necessary. The manager synchronously waits for updates to
10338 * complete in the monitor, so new mpb buffers allocated here can be
10339 * integrated by the monitor thread without worrying about live pointers
10340 * in the manager thread.
10342 enum imsm_update_type type
;
10343 struct intel_super
*super
= st
->sb
;
10344 unsigned int sector_size
= super
->sector_size
;
10345 struct imsm_super
*mpb
= super
->anchor
;
10349 if (update
->len
< (int)sizeof(type
))
10352 type
= *(enum imsm_update_type
*) update
->buf
;
10355 case update_general_migration_checkpoint
:
10356 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10358 dprintf("called for update_general_migration_checkpoint\n");
10360 case update_takeover
: {
10361 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10362 if (update
->len
< (int)sizeof(*u
))
10364 if (u
->direction
== R0_TO_R10
) {
10365 void **tail
= (void **)&update
->space_list
;
10366 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10367 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10368 int num_members
= map
->num_members
;
10371 /* allocate memory for added disks */
10372 for (i
= 0; i
< num_members
; i
++) {
10373 size
= sizeof(struct dl
);
10374 space
= xmalloc(size
);
10379 /* allocate memory for new device */
10380 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10381 (num_members
* sizeof(__u32
));
10382 space
= xmalloc(size
);
10386 len
= disks_to_mpb_size(num_members
* 2);
10391 case update_reshape_container_disks
: {
10392 /* Every raid device in the container is about to
10393 * gain some more devices, and we will enter a
10395 * So each 'imsm_map' will be bigger, and the imsm_vol
10396 * will now hold 2 of them.
10397 * Thus we need new 'struct imsm_dev' allocations sized
10398 * as sizeof_imsm_dev but with more devices in both maps.
10400 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10401 struct intel_dev
*dl
;
10402 void **space_tail
= (void**)&update
->space_list
;
10404 if (update
->len
< (int)sizeof(*u
))
10407 dprintf("for update_reshape\n");
10409 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10410 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10412 if (u
->new_raid_disks
> u
->old_raid_disks
)
10413 size
+= sizeof(__u32
)*2*
10414 (u
->new_raid_disks
- u
->old_raid_disks
);
10418 *space_tail
= NULL
;
10421 len
= disks_to_mpb_size(u
->new_raid_disks
);
10422 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10425 case update_reshape_migration
: {
10426 /* for migration level 0->5 we need to add disks
10427 * so the same as for container operation we will copy
10428 * device to the bigger location.
10429 * in memory prepared device and new disk area are prepared
10430 * for usage in process update
10432 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10433 struct intel_dev
*id
;
10434 void **space_tail
= (void **)&update
->space_list
;
10437 int current_level
= -1;
10439 if (update
->len
< (int)sizeof(*u
))
10442 dprintf("for update_reshape\n");
10444 /* add space for bigger array in update
10446 for (id
= super
->devlist
; id
; id
= id
->next
) {
10447 if (id
->index
== (unsigned)u
->subdev
) {
10448 size
= sizeof_imsm_dev(id
->dev
, 1);
10449 if (u
->new_raid_disks
> u
->old_raid_disks
)
10450 size
+= sizeof(__u32
)*2*
10451 (u
->new_raid_disks
- u
->old_raid_disks
);
10455 *space_tail
= NULL
;
10459 if (update
->space_list
== NULL
)
10462 /* add space for disk in update
10464 size
= sizeof(struct dl
);
10468 *space_tail
= NULL
;
10470 /* add spare device to update
10472 for (id
= super
->devlist
; id
; id
= id
->next
)
10473 if (id
->index
== (unsigned)u
->subdev
) {
10474 struct imsm_dev
*dev
;
10475 struct imsm_map
*map
;
10477 dev
= get_imsm_dev(super
, u
->subdev
);
10478 map
= get_imsm_map(dev
, MAP_0
);
10479 current_level
= map
->raid_level
;
10482 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10483 struct mdinfo
*spares
;
10485 spares
= get_spares_for_grow(st
);
10488 struct mdinfo
*dev
;
10490 dev
= spares
->devs
;
10493 makedev(dev
->disk
.major
,
10495 dl
= get_disk_super(super
,
10498 dl
->index
= u
->old_raid_disks
;
10501 sysfs_free(spares
);
10504 len
= disks_to_mpb_size(u
->new_raid_disks
);
10505 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10508 case update_size_change
: {
10509 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10513 case update_activate_spare
: {
10514 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10518 case update_create_array
: {
10519 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10520 struct intel_dev
*dv
;
10521 struct imsm_dev
*dev
= &u
->dev
;
10522 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10524 struct disk_info
*inf
;
10528 if (update
->len
< (int)sizeof(*u
))
10531 inf
= get_disk_info(u
);
10532 len
= sizeof_imsm_dev(dev
, 1);
10533 /* allocate a new super->devlist entry */
10534 dv
= xmalloc(sizeof(*dv
));
10535 dv
->dev
= xmalloc(len
);
10536 update
->space
= dv
;
10538 /* count how many spares will be converted to members */
10539 for (i
= 0; i
< map
->num_members
; i
++) {
10540 dl
= serial_to_dl(inf
[i
].serial
, super
);
10542 /* hmm maybe it failed?, nothing we can do about
10547 if (count_memberships(dl
, super
) == 0)
10550 len
+= activate
* sizeof(struct imsm_disk
);
10553 case update_kill_array
: {
10554 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10558 case update_rename_array
: {
10559 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10563 case update_add_remove_disk
:
10564 /* no update->len needed */
10566 case update_prealloc_badblocks_mem
:
10567 super
->extra_space
+= sizeof(struct bbm_log
) -
10568 get_imsm_bbm_log_size(super
->bbm_log
);
10570 case update_rwh_policy
: {
10571 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10579 /* check if we need a larger metadata buffer */
10580 if (super
->next_buf
)
10581 buf_len
= super
->next_len
;
10583 buf_len
= super
->len
;
10585 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10586 /* ok we need a larger buf than what is currently allocated
10587 * if this allocation fails process_update will notice that
10588 * ->next_len is set and ->next_buf is NULL
10590 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10591 super
->extra_space
+ len
, sector_size
);
10592 if (super
->next_buf
)
10593 free(super
->next_buf
);
10595 super
->next_len
= buf_len
;
10596 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10597 memset(super
->next_buf
, 0, buf_len
);
10599 super
->next_buf
= NULL
;
10604 /* must be called while manager is quiesced */
10605 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10607 struct imsm_super
*mpb
= super
->anchor
;
10609 struct imsm_dev
*dev
;
10610 struct imsm_map
*map
;
10611 unsigned int i
, j
, num_members
;
10612 __u32 ord
, ord_map0
;
10613 struct bbm_log
*log
= super
->bbm_log
;
10615 dprintf("deleting device[%d] from imsm_super\n", index
);
10617 /* shift all indexes down one */
10618 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10619 if (iter
->index
> (int)index
)
10621 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10622 if (iter
->index
> (int)index
)
10625 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10626 dev
= get_imsm_dev(super
, i
);
10627 map
= get_imsm_map(dev
, MAP_0
);
10628 num_members
= map
->num_members
;
10629 for (j
= 0; j
< num_members
; j
++) {
10630 /* update ord entries being careful not to propagate
10631 * ord-flags to the first map
10633 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10634 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10636 if (ord_to_idx(ord
) <= index
)
10639 map
= get_imsm_map(dev
, MAP_0
);
10640 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10641 map
= get_imsm_map(dev
, MAP_1
);
10643 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10647 for (i
= 0; i
< log
->entry_count
; i
++) {
10648 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10650 if (entry
->disk_ordinal
<= index
)
10652 entry
->disk_ordinal
--;
10656 super
->updates_pending
++;
10658 struct dl
*dl
= *dlp
;
10660 *dlp
= (*dlp
)->next
;
10661 __free_imsm_disk(dl
, 1);
10665 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10666 struct intel_super
*super
,
10667 struct imsm_dev
*dev
)
10673 struct imsm_map
*map
;
10676 ret_val
= raid_disks
/2;
10677 /* check map if all disks pairs not failed
10680 map
= get_imsm_map(dev
, MAP_0
);
10681 for (i
= 0; i
< ret_val
; i
++) {
10682 int degradation
= 0;
10683 if (get_imsm_disk(super
, i
) == NULL
)
10685 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10687 if (degradation
== 2)
10690 map
= get_imsm_map(dev
, MAP_1
);
10691 /* if there is no second map
10692 * result can be returned
10696 /* check degradation in second map
10698 for (i
= 0; i
< ret_val
; i
++) {
10699 int degradation
= 0;
10700 if (get_imsm_disk(super
, i
) == NULL
)
10702 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10704 if (degradation
== 2)
10718 /*******************************************************************************
10719 * Function: validate_container_imsm
10720 * Description: This routine validates container after assemble,
10721 * eg. if devices in container are under the same controller.
10724 * info : linked list with info about devices used in array
10728 ******************************************************************************/
10729 int validate_container_imsm(struct mdinfo
*info
)
10731 if (check_no_platform())
10734 struct sys_dev
*idev
;
10735 struct sys_dev
*hba
= NULL
;
10736 struct sys_dev
*intel_devices
= find_intel_devices();
10737 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10738 info
->disk
.minor
), 1, NULL
);
10740 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10741 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10750 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10751 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10755 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10756 struct mdinfo
*dev
;
10758 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10759 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
,
10760 dev
->disk
.minor
), 1, NULL
);
10762 struct sys_dev
*hba2
= NULL
;
10763 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10764 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10772 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10773 get_orom_by_device_id(hba2
->dev_id
);
10775 if (hba2
&& hba
->type
!= hba2
->type
) {
10776 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10777 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10781 if (orom
!= orom2
) {
10782 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10783 " This operation is not supported and can lead to data loss.\n");
10788 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10789 " This operation is not supported and can lead to data loss.\n");
10797 /*******************************************************************************
10798 * Function: imsm_record_badblock
10799 * Description: This routine stores new bad block record in BBM log
10802 * a : array containing a bad block
10803 * slot : disk number containing a bad block
10804 * sector : bad block sector
10805 * length : bad block sectors range
10809 ******************************************************************************/
10810 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10811 unsigned long long sector
, int length
)
10813 struct intel_super
*super
= a
->container
->sb
;
10817 ord
= imsm_disk_slot_to_ord(a
, slot
);
10821 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10824 super
->updates_pending
++;
10828 /*******************************************************************************
10829 * Function: imsm_clear_badblock
10830 * Description: This routine clears bad block record from BBM log
10833 * a : array containing a bad block
10834 * slot : disk number containing a bad block
10835 * sector : bad block sector
10836 * length : bad block sectors range
10840 ******************************************************************************/
10841 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10842 unsigned long long sector
, int length
)
10844 struct intel_super
*super
= a
->container
->sb
;
10848 ord
= imsm_disk_slot_to_ord(a
, slot
);
10852 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10854 super
->updates_pending
++;
10858 /*******************************************************************************
10859 * Function: imsm_get_badblocks
10860 * Description: This routine get list of bad blocks for an array
10864 * slot : disk number
10866 * bb : structure containing bad blocks
10868 ******************************************************************************/
10869 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10871 int inst
= a
->info
.container_member
;
10872 struct intel_super
*super
= a
->container
->sb
;
10873 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10874 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10877 ord
= imsm_disk_slot_to_ord(a
, slot
);
10881 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10882 per_dev_array_size(map
), &super
->bb
);
10886 /*******************************************************************************
10887 * Function: examine_badblocks_imsm
10888 * Description: Prints list of bad blocks on a disk to the standard output
10891 * st : metadata handler
10892 * fd : open file descriptor for device
10893 * devname : device name
10897 ******************************************************************************/
10898 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10900 struct intel_super
*super
= st
->sb
;
10901 struct bbm_log
*log
= super
->bbm_log
;
10902 struct dl
*d
= NULL
;
10905 for (d
= super
->disks
; d
; d
= d
->next
) {
10906 if (strcmp(d
->devname
, devname
) == 0)
10910 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10911 pr_err("%s doesn't appear to be part of a raid array\n",
10918 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10920 for (i
= 0; i
< log
->entry_count
; i
++) {
10921 if (entry
[i
].disk_ordinal
== d
->index
) {
10922 unsigned long long sector
= __le48_to_cpu(
10923 &entry
[i
].defective_block_start
);
10924 int cnt
= entry
[i
].marked_count
+ 1;
10927 printf("Bad-blocks on %s:\n", devname
);
10931 printf("%20llu for %d sectors\n", sector
, cnt
);
10937 printf("No bad-blocks list configured on %s\n", devname
);
10941 /*******************************************************************************
10942 * Function: init_migr_record_imsm
10943 * Description: Function inits imsm migration record
10945 * super : imsm internal array info
10946 * dev : device under migration
10947 * info : general array info to find the smallest device
10950 ******************************************************************************/
10951 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10952 struct mdinfo
*info
)
10954 struct intel_super
*super
= st
->sb
;
10955 struct migr_record
*migr_rec
= super
->migr_rec
;
10956 int new_data_disks
;
10957 unsigned long long dsize
, dev_sectors
;
10958 long long unsigned min_dev_sectors
= -1LLU;
10959 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10960 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10961 unsigned long long num_migr_units
;
10962 unsigned long long array_blocks
;
10963 struct dl
*dl_disk
= NULL
;
10965 memset(migr_rec
, 0, sizeof(struct migr_record
));
10966 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10968 /* only ascending reshape supported now */
10969 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10971 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10972 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10973 migr_rec
->dest_depth_per_unit
*=
10974 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10975 new_data_disks
= imsm_num_data_members(map_dest
);
10976 migr_rec
->blocks_per_unit
=
10977 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10978 migr_rec
->dest_depth_per_unit
=
10979 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10980 array_blocks
= info
->component_size
* new_data_disks
;
10982 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10984 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10986 set_num_migr_units(migr_rec
, num_migr_units
);
10988 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10989 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10991 /* Find the smallest dev */
10992 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
10993 /* ignore spares in container */
10994 if (dl_disk
->index
< 0)
10996 get_dev_size(dl_disk
->fd
, NULL
, &dsize
);
10997 dev_sectors
= dsize
/ 512;
10998 if (dev_sectors
< min_dev_sectors
)
10999 min_dev_sectors
= dev_sectors
;
11001 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
11002 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
11004 write_imsm_migr_rec(st
);
11009 /*******************************************************************************
11010 * Function: save_backup_imsm
11011 * Description: Function saves critical data stripes to Migration Copy Area
11012 * and updates the current migration unit status.
11013 * Use restore_stripes() to form a destination stripe,
11014 * and to write it to the Copy Area.
11016 * st : supertype information
11017 * dev : imsm device that backup is saved for
11018 * info : general array info
11019 * buf : input buffer
11020 * length : length of data to backup (blocks_per_unit)
11024 ******************************************************************************/
11025 int save_backup_imsm(struct supertype
*st
,
11026 struct imsm_dev
*dev
,
11027 struct mdinfo
*info
,
11032 struct intel_super
*super
= st
->sb
;
11034 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
11035 int new_disks
= map_dest
->num_members
;
11036 int dest_layout
= 0;
11037 int dest_chunk
, targets
[new_disks
];
11038 unsigned long long start
, target_offsets
[new_disks
];
11039 int data_disks
= imsm_num_data_members(map_dest
);
11041 for (i
= 0; i
< new_disks
; i
++) {
11042 struct dl
*dl_disk
= get_imsm_dl_disk(super
, i
);
11043 if (dl_disk
&& is_fd_valid(dl_disk
->fd
))
11044 targets
[i
] = dl_disk
->fd
;
11049 start
= info
->reshape_progress
* 512;
11050 for (i
= 0; i
< new_disks
; i
++) {
11051 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
11052 /* move back copy area adderss, it will be moved forward
11053 * in restore_stripes() using start input variable
11055 target_offsets
[i
] -= start
/data_disks
;
11058 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
11059 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
11061 if (restore_stripes(targets
, /* list of dest devices */
11062 target_offsets
, /* migration record offsets */
11065 map_dest
->raid_level
,
11067 -1, /* source backup file descriptor */
11068 0, /* input buf offset
11069 * always 0 buf is already offseted */
11073 pr_err("Error restoring stripes\n");
11083 /*******************************************************************************
11084 * Function: save_checkpoint_imsm
11085 * Description: Function called for current unit status update
11086 * in the migration record. It writes it to disk.
11088 * super : imsm internal array info
11089 * info : general array info
11093 * 2: failure, means no valid migration record
11094 * / no general migration in progress /
11095 ******************************************************************************/
11096 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
11098 struct intel_super
*super
= st
->sb
;
11099 unsigned long long blocks_per_unit
;
11100 unsigned long long curr_migr_unit
;
11102 if (load_imsm_migr_rec(super
) != 0) {
11103 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
11107 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
11108 if (blocks_per_unit
== 0) {
11109 dprintf("imsm: no migration in progress.\n");
11112 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
11113 /* check if array is alligned to copy area
11114 * if it is not alligned, add one to current migration unit value
11115 * this can happend on array reshape finish only
11117 if (info
->reshape_progress
% blocks_per_unit
)
11120 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
11121 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
11122 set_migr_dest_1st_member_lba(super
->migr_rec
,
11123 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
11125 if (write_imsm_migr_rec(st
) < 0) {
11126 dprintf("imsm: Cannot write migration record outside backup area\n");
11133 /*******************************************************************************
11134 * Function: recover_backup_imsm
11135 * Description: Function recovers critical data from the Migration Copy Area
11136 * while assembling an array.
11138 * super : imsm internal array info
11139 * info : general array info
11141 * 0 : success (or there is no data to recover)
11143 ******************************************************************************/
11144 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
11146 struct intel_super
*super
= st
->sb
;
11147 struct migr_record
*migr_rec
= super
->migr_rec
;
11148 struct imsm_map
*map_dest
;
11149 struct intel_dev
*id
= NULL
;
11150 unsigned long long read_offset
;
11151 unsigned long long write_offset
;
11153 int new_disks
, err
;
11156 unsigned int sector_size
= super
->sector_size
;
11157 unsigned long long curr_migr_unit
= current_migr_unit(migr_rec
);
11158 unsigned long long num_migr_units
= get_num_migr_units(migr_rec
);
11160 int skipped_disks
= 0;
11161 struct dl
*dl_disk
;
11163 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
11167 /* recover data only during assemblation */
11168 if (strncmp(buffer
, "inactive", 8) != 0)
11170 /* no data to recover */
11171 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
11173 if (curr_migr_unit
>= num_migr_units
)
11176 /* find device during reshape */
11177 for (id
= super
->devlist
; id
; id
= id
->next
)
11178 if (is_gen_migration(id
->dev
))
11183 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
11184 new_disks
= map_dest
->num_members
;
11186 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
11188 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
11189 pba_of_lba0(map_dest
)) * 512;
11191 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11192 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
11195 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
11196 if (dl_disk
->index
< 0)
11199 if (!is_fd_valid(dl_disk
->fd
)) {
11203 if (lseek64(dl_disk
->fd
, read_offset
, SEEK_SET
) < 0) {
11204 pr_err("Cannot seek to block: %s\n",
11209 if (read(dl_disk
->fd
, buf
, unit_len
) != (ssize_t
)unit_len
) {
11210 pr_err("Cannot read copy area block: %s\n",
11215 if (lseek64(dl_disk
->fd
, write_offset
, SEEK_SET
) < 0) {
11216 pr_err("Cannot seek to block: %s\n",
11221 if (write(dl_disk
->fd
, buf
, unit_len
) != (ssize_t
)unit_len
) {
11222 pr_err("Cannot restore block: %s\n",
11229 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11233 pr_err("Cannot restore data from backup. Too many failed disks\n");
11237 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11238 /* ignore error == 2, this can mean end of reshape here
11240 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11249 static char disk_by_path
[] = "/dev/disk/by-path/";
11251 static const char *imsm_get_disk_controller_domain(const char *path
)
11253 char disk_path
[PATH_MAX
];
11257 strncpy(disk_path
, disk_by_path
, PATH_MAX
);
11258 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
11259 if (stat(disk_path
, &st
) == 0) {
11260 struct sys_dev
* hba
;
11263 path
= devt_to_devpath(st
.st_rdev
, 1, NULL
);
11266 hba
= find_disk_attached_hba(-1, path
);
11267 if (hba
&& hba
->type
== SYS_DEV_SAS
)
11269 else if (hba
&& (hba
->type
== SYS_DEV_SATA
|| hba
->type
== SYS_DEV_SATA_VMD
))
11271 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
11273 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
11277 dprintf("path: %s hba: %s attached: %s\n",
11278 path
, (hba
) ? hba
->path
: "NULL", drv
);
11284 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11286 static char devnm
[32];
11287 char subdev_name
[20];
11288 struct mdstat_ent
*mdstat
;
11290 sprintf(subdev_name
, "%d", subdev
);
11291 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11295 strcpy(devnm
, mdstat
->devnm
);
11296 free_mdstat(mdstat
);
11300 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11301 struct geo_params
*geo
,
11302 int *old_raid_disks
,
11305 /* currently we only support increasing the number of devices
11306 * for a container. This increases the number of device for each
11307 * member array. They must all be RAID0 or RAID5.
11310 struct mdinfo
*info
, *member
;
11311 int devices_that_can_grow
= 0;
11313 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11315 if (geo
->size
> 0 ||
11316 geo
->level
!= UnSet
||
11317 geo
->layout
!= UnSet
||
11318 geo
->chunksize
!= 0 ||
11319 geo
->raid_disks
== UnSet
) {
11320 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11324 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11325 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11329 info
= container_content_imsm(st
, NULL
);
11330 for (member
= info
; member
; member
= member
->next
) {
11333 dprintf("imsm: checking device_num: %i\n",
11334 member
->container_member
);
11336 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11337 /* we work on container for Online Capacity Expansion
11338 * only so raid_disks has to grow
11340 dprintf("imsm: for container operation raid disks increase is required\n");
11344 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11345 /* we cannot use this container with other raid level
11347 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11348 info
->array
.level
);
11351 /* check for platform support
11352 * for this raid level configuration
11354 struct intel_super
*super
= st
->sb
;
11355 if (!is_raid_level_supported(super
->orom
,
11356 member
->array
.level
,
11357 geo
->raid_disks
)) {
11358 dprintf("platform does not support raid%d with %d disk%s\n",
11361 geo
->raid_disks
> 1 ? "s" : "");
11364 /* check if component size is aligned to chunk size
11366 if (info
->component_size
%
11367 (info
->array
.chunk_size
/512)) {
11368 dprintf("Component size is not aligned to chunk size\n");
11373 if (*old_raid_disks
&&
11374 info
->array
.raid_disks
!= *old_raid_disks
)
11376 *old_raid_disks
= info
->array
.raid_disks
;
11378 /* All raid5 and raid0 volumes in container
11379 * have to be ready for Online Capacity Expansion
11380 * so they need to be assembled. We have already
11381 * checked that no recovery etc is happening.
11383 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11384 st
->container_devnm
);
11385 if (result
== NULL
) {
11386 dprintf("imsm: cannot find array\n");
11389 devices_that_can_grow
++;
11392 if (!member
&& devices_that_can_grow
)
11396 dprintf("Container operation allowed\n");
11398 dprintf("Error: %i\n", ret_val
);
11403 /* Function: get_spares_for_grow
11404 * Description: Allocates memory and creates list of spare devices
11405 * avaliable in container. Checks if spare drive size is acceptable.
11406 * Parameters: Pointer to the supertype structure
11407 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11410 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11412 struct spare_criteria sc
;
11414 get_spare_criteria_imsm(st
, &sc
);
11415 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11418 /******************************************************************************
11419 * function: imsm_create_metadata_update_for_reshape
11420 * Function creates update for whole IMSM container.
11422 ******************************************************************************/
11423 static int imsm_create_metadata_update_for_reshape(
11424 struct supertype
*st
,
11425 struct geo_params
*geo
,
11426 int old_raid_disks
,
11427 struct imsm_update_reshape
**updatep
)
11429 struct intel_super
*super
= st
->sb
;
11430 struct imsm_super
*mpb
= super
->anchor
;
11431 int update_memory_size
;
11432 struct imsm_update_reshape
*u
;
11433 struct mdinfo
*spares
;
11436 struct mdinfo
*dev
;
11438 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11440 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11442 /* size of all update data without anchor */
11443 update_memory_size
= sizeof(struct imsm_update_reshape
);
11445 /* now add space for spare disks that we need to add. */
11446 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11448 u
= xcalloc(1, update_memory_size
);
11449 u
->type
= update_reshape_container_disks
;
11450 u
->old_raid_disks
= old_raid_disks
;
11451 u
->new_raid_disks
= geo
->raid_disks
;
11453 /* now get spare disks list
11455 spares
= get_spares_for_grow(st
);
11457 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11458 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11463 /* we have got spares
11464 * update disk list in imsm_disk list table in anchor
11466 dprintf("imsm: %i spares are available.\n\n",
11467 spares
->array
.spare_disks
);
11469 dev
= spares
->devs
;
11470 for (i
= 0; i
< delta_disks
; i
++) {
11475 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11477 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11478 dl
->index
= mpb
->num_disks
;
11486 sysfs_free(spares
);
11488 dprintf("imsm: reshape update preparation :");
11489 if (i
== delta_disks
) {
11490 dprintf_cont(" OK\n");
11492 return update_memory_size
;
11495 dprintf_cont(" Error\n");
11500 /******************************************************************************
11501 * function: imsm_create_metadata_update_for_size_change()
11502 * Creates update for IMSM array for array size change.
11504 ******************************************************************************/
11505 static int imsm_create_metadata_update_for_size_change(
11506 struct supertype
*st
,
11507 struct geo_params
*geo
,
11508 struct imsm_update_size_change
**updatep
)
11510 struct intel_super
*super
= st
->sb
;
11511 int update_memory_size
;
11512 struct imsm_update_size_change
*u
;
11514 dprintf("(enter) New size = %llu\n", geo
->size
);
11516 /* size of all update data without anchor */
11517 update_memory_size
= sizeof(struct imsm_update_size_change
);
11519 u
= xcalloc(1, update_memory_size
);
11520 u
->type
= update_size_change
;
11521 u
->subdev
= super
->current_vol
;
11522 u
->new_size
= geo
->size
;
11524 dprintf("imsm: reshape update preparation : OK\n");
11527 return update_memory_size
;
11530 /******************************************************************************
11531 * function: imsm_create_metadata_update_for_migration()
11532 * Creates update for IMSM array.
11534 ******************************************************************************/
11535 static int imsm_create_metadata_update_for_migration(
11536 struct supertype
*st
,
11537 struct geo_params
*geo
,
11538 struct imsm_update_reshape_migration
**updatep
)
11540 struct intel_super
*super
= st
->sb
;
11541 int update_memory_size
;
11542 int current_chunk_size
;
11543 struct imsm_update_reshape_migration
*u
;
11544 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11545 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11546 int previous_level
= -1;
11548 dprintf("(enter) New Level = %i\n", geo
->level
);
11550 /* size of all update data without anchor */
11551 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11553 u
= xcalloc(1, update_memory_size
);
11554 u
->type
= update_reshape_migration
;
11555 u
->subdev
= super
->current_vol
;
11556 u
->new_level
= geo
->level
;
11557 u
->new_layout
= geo
->layout
;
11558 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11559 u
->new_disks
[0] = -1;
11560 u
->new_chunksize
= -1;
11562 current_chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) / 2;
11564 if (geo
->chunksize
!= current_chunk_size
) {
11565 u
->new_chunksize
= geo
->chunksize
/ 1024;
11566 dprintf("imsm: chunk size change from %i to %i\n",
11567 current_chunk_size
, u
->new_chunksize
);
11569 previous_level
= map
->raid_level
;
11571 if (geo
->level
== 5 && previous_level
== 0) {
11572 struct mdinfo
*spares
= NULL
;
11574 u
->new_raid_disks
++;
11575 spares
= get_spares_for_grow(st
);
11576 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11578 sysfs_free(spares
);
11579 update_memory_size
= 0;
11580 pr_err("cannot get spare device for requested migration\n");
11583 sysfs_free(spares
);
11585 dprintf("imsm: reshape update preparation : OK\n");
11588 return update_memory_size
;
11591 static void imsm_update_metadata_locally(struct supertype
*st
,
11592 void *buf
, int len
)
11594 struct metadata_update mu
;
11599 mu
.space_list
= NULL
;
11601 if (imsm_prepare_update(st
, &mu
))
11602 imsm_process_update(st
, &mu
);
11604 while (mu
.space_list
) {
11605 void **space
= mu
.space_list
;
11606 mu
.space_list
= *space
;
11612 * imsm_analyze_expand() - check expand properties and calculate new size.
11613 * @st: imsm supertype.
11614 * @geo: new geometry params.
11615 * @array: array info.
11616 * @direction: reshape direction.
11618 * Obtain free space after the &array and verify if expand to requested size is
11619 * possible. If geo->size is set to %MAX_SIZE, assume that max free size is
11623 * On success %IMSM_STATUS_OK is returned, geo->size and geo->raid_disks are
11625 * On error, %IMSM_STATUS_ERROR is returned.
11627 static imsm_status_t
imsm_analyze_expand(struct supertype
*st
,
11628 struct geo_params
*geo
,
11629 struct mdinfo
*array
,
11632 struct intel_super
*super
= st
->sb
;
11633 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11634 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11635 int data_disks
= imsm_num_data_members(map
);
11637 unsigned long long current_size
;
11638 unsigned long long free_size
;
11639 unsigned long long new_size
;
11640 unsigned long long max_size
;
11642 const int chunk_kib
= geo
->chunksize
/ 1024;
11645 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11647 * Accept size for rollback only.
11649 new_size
= geo
->size
* 2;
11653 if (data_disks
== 0) {
11654 pr_err("imsm: Cannot retrieve data disks.\n");
11655 return IMSM_STATUS_ERROR
;
11657 current_size
= array
->custom_array_size
/ data_disks
;
11659 rv
= imsm_get_free_size(super
, dev
->vol
.map
->num_members
, 0, chunk_kib
, &free_size
, true);
11660 if (rv
!= IMSM_STATUS_OK
) {
11661 pr_err("imsm: Cannot find free space for expand.\n");
11662 return IMSM_STATUS_ERROR
;
11664 max_size
= round_member_size_to_mb(free_size
+ current_size
);
11666 if (geo
->size
== MAX_SIZE
)
11667 new_size
= max_size
;
11669 new_size
= round_member_size_to_mb(geo
->size
* 2);
11671 if (new_size
== 0) {
11672 pr_err("imsm: Rounded requested size is 0.\n");
11673 return IMSM_STATUS_ERROR
;
11676 if (new_size
> max_size
) {
11677 pr_err("imsm: Rounded requested size (%llu) is larger than free space available (%llu).\n",
11678 new_size
, max_size
);
11679 return IMSM_STATUS_ERROR
;
11682 if (new_size
== current_size
) {
11683 pr_err("imsm: Rounded requested size (%llu) is same as current size (%llu).\n",
11684 new_size
, current_size
);
11685 return IMSM_STATUS_ERROR
;
11688 if (new_size
< current_size
) {
11689 pr_err("imsm: Size reduction is not supported, rounded requested size (%llu) is smaller than current (%llu).\n",
11690 new_size
, current_size
);
11691 return IMSM_STATUS_ERROR
;
11695 dprintf("imsm: New size per member is %llu.\n", new_size
);
11696 geo
->size
= data_disks
* new_size
;
11697 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11698 return IMSM_STATUS_OK
;
11701 /***************************************************************************
11702 * Function: imsm_analyze_change
11703 * Description: Function analyze change for single volume
11704 * and validate if transition is supported
11705 * Parameters: Geometry parameters, supertype structure,
11706 * metadata change direction (apply/rollback)
11707 * Returns: Operation type code on success, -1 if fail
11708 ****************************************************************************/
11709 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11710 struct geo_params
*geo
,
11713 struct mdinfo info
;
11715 int check_devs
= 0;
11717 /* number of added/removed disks in operation result */
11718 int devNumChange
= 0;
11719 /* imsm compatible layout value for array geometry verification */
11720 int imsm_layout
= -1;
11723 getinfo_super_imsm_volume(st
, &info
, NULL
);
11724 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11725 geo
->level
!= UnSet
) {
11726 switch (info
.array
.level
) {
11728 if (geo
->level
== 5) {
11729 change
= CH_MIGRATION
;
11730 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11731 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11733 goto analyse_change_exit
;
11735 imsm_layout
= geo
->layout
;
11737 devNumChange
= 1; /* parity disk added */
11738 } else if (geo
->level
== 10) {
11739 change
= CH_TAKEOVER
;
11741 devNumChange
= 2; /* two mirrors added */
11742 imsm_layout
= 0x102; /* imsm supported layout */
11747 if (geo
->level
== 0) {
11748 change
= CH_TAKEOVER
;
11750 devNumChange
= -(geo
->raid_disks
/2);
11751 imsm_layout
= 0; /* imsm raid0 layout */
11755 if (change
== -1) {
11756 pr_err("Error. Level Migration from %d to %d not supported!\n",
11757 info
.array
.level
, geo
->level
);
11758 goto analyse_change_exit
;
11761 geo
->level
= info
.array
.level
;
11763 if (geo
->layout
!= info
.array
.layout
&&
11764 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11765 change
= CH_MIGRATION
;
11766 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11767 geo
->layout
== 5) {
11768 /* reshape 5 -> 4 */
11769 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11770 geo
->layout
== 0) {
11771 /* reshape 4 -> 5 */
11775 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11776 info
.array
.layout
, geo
->layout
);
11778 goto analyse_change_exit
;
11781 geo
->layout
= info
.array
.layout
;
11782 if (imsm_layout
== -1)
11783 imsm_layout
= info
.array
.layout
;
11786 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11787 geo
->chunksize
!= info
.array
.chunk_size
) {
11788 if (info
.array
.level
== 10) {
11789 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11791 goto analyse_change_exit
;
11792 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11793 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11794 geo
->chunksize
/1024, info
.component_size
/2);
11796 goto analyse_change_exit
;
11798 change
= CH_MIGRATION
;
11800 geo
->chunksize
= info
.array
.chunk_size
;
11803 if (geo
->size
> 0) {
11804 if (change
!= -1) {
11805 pr_err("Error. Size change should be the only one at a time.\n");
11807 goto analyse_change_exit
;
11810 rv
= imsm_analyze_expand(st
, geo
, &info
, direction
);
11811 if (rv
!= IMSM_STATUS_OK
)
11812 goto analyse_change_exit
;
11813 change
= CH_ARRAY_SIZE
;
11816 chunk
= geo
->chunksize
/ 1024;
11817 if (!validate_geometry_imsm(st
,
11820 geo
->raid_disks
+ devNumChange
,
11822 geo
->size
, INVALID_SECTORS
,
11823 0, 0, info
.consistency_policy
, 1))
11827 struct intel_super
*super
= st
->sb
;
11828 struct imsm_super
*mpb
= super
->anchor
;
11830 if (mpb
->num_raid_devs
> 1) {
11831 pr_err("Error. Cannot perform operation on %s- for this operation "
11832 "it MUST be single array in container\n", geo
->dev_name
);
11837 analyse_change_exit
:
11838 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11839 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11840 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11846 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11848 struct intel_super
*super
= st
->sb
;
11849 struct imsm_update_takeover
*u
;
11851 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11853 u
->type
= update_takeover
;
11854 u
->subarray
= super
->current_vol
;
11856 /* 10->0 transition */
11857 if (geo
->level
== 0)
11858 u
->direction
= R10_TO_R0
;
11860 /* 0->10 transition */
11861 if (geo
->level
== 10)
11862 u
->direction
= R0_TO_R10
;
11864 /* update metadata locally */
11865 imsm_update_metadata_locally(st
, u
,
11866 sizeof(struct imsm_update_takeover
));
11867 /* and possibly remotely */
11868 if (st
->update_tail
)
11869 append_metadata_update(st
, u
,
11870 sizeof(struct imsm_update_takeover
));
11877 /* Flush size update if size calculated by num_data_stripes is higher than
11878 * imsm_dev_size to eliminate differences during reshape.
11879 * Mdmon will recalculate them correctly.
11880 * If subarray index is not set then check whole container.
11882 * 0 - no error occurred
11883 * 1 - error detected
11885 static int imsm_fix_size_mismatch(struct supertype
*st
, int subarray_index
)
11887 struct intel_super
*super
= st
->sb
;
11888 int tmp
= super
->current_vol
;
11892 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
11893 if (subarray_index
>= 0 && i
!= subarray_index
)
11895 super
->current_vol
= i
;
11896 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11897 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11898 unsigned int disc_count
= imsm_num_data_members(map
);
11899 struct geo_params geo
;
11900 struct imsm_update_size_change
*update
;
11901 unsigned long long calc_size
= per_dev_array_size(map
) * disc_count
;
11902 unsigned long long d_size
= imsm_dev_size(dev
);
11905 if (calc_size
== d_size
)
11908 /* There is a difference, confirm that imsm_dev_size is
11909 * smaller and push update.
11911 if (d_size
> calc_size
) {
11912 pr_err("imsm: dev size of subarray %d is incorrect\n",
11916 memset(&geo
, 0, sizeof(struct geo_params
));
11918 u_size
= imsm_create_metadata_update_for_size_change(st
, &geo
,
11920 imsm_update_metadata_locally(st
, update
, u_size
);
11921 if (st
->update_tail
) {
11922 append_metadata_update(st
, update
, u_size
);
11923 flush_metadata_updates(st
);
11924 st
->update_tail
= &st
->updates
;
11926 imsm_sync_metadata(st
);
11932 super
->current_vol
= tmp
;
11936 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11938 int layout
, int chunksize
, int raid_disks
,
11939 int delta_disks
, char *backup
, char *dev
,
11940 int direction
, int verbose
)
11943 struct geo_params geo
;
11945 dprintf("(enter)\n");
11947 memset(&geo
, 0, sizeof(struct geo_params
));
11949 geo
.dev_name
= dev
;
11950 strcpy(geo
.devnm
, st
->devnm
);
11953 geo
.layout
= layout
;
11954 geo
.chunksize
= chunksize
;
11955 geo
.raid_disks
= raid_disks
;
11956 if (delta_disks
!= UnSet
)
11957 geo
.raid_disks
+= delta_disks
;
11959 dprintf("for level : %i\n", geo
.level
);
11960 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11962 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11963 /* On container level we can only increase number of devices. */
11964 dprintf("imsm: info: Container operation\n");
11965 int old_raid_disks
= 0;
11967 if (imsm_reshape_is_allowed_on_container(
11968 st
, &geo
, &old_raid_disks
, direction
)) {
11969 struct imsm_update_reshape
*u
= NULL
;
11972 if (imsm_fix_size_mismatch(st
, -1)) {
11973 dprintf("imsm: Cannot fix size mismatch\n");
11974 goto exit_imsm_reshape_super
;
11977 len
= imsm_create_metadata_update_for_reshape(
11978 st
, &geo
, old_raid_disks
, &u
);
11981 dprintf("imsm: Cannot prepare update\n");
11982 goto exit_imsm_reshape_super
;
11986 /* update metadata locally */
11987 imsm_update_metadata_locally(st
, u
, len
);
11988 /* and possibly remotely */
11989 if (st
->update_tail
)
11990 append_metadata_update(st
, u
, len
);
11995 pr_err("(imsm) Operation is not allowed on this container\n");
11998 /* On volume level we support following operations
11999 * - takeover: raid10 -> raid0; raid0 -> raid10
12000 * - chunk size migration
12001 * - migration: raid5 -> raid0; raid0 -> raid5
12003 struct intel_super
*super
= st
->sb
;
12004 struct intel_dev
*dev
= super
->devlist
;
12006 dprintf("imsm: info: Volume operation\n");
12007 /* find requested device */
12010 imsm_find_array_devnm_by_subdev(
12011 dev
->index
, st
->container_devnm
);
12012 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
12017 pr_err("Cannot find %s (%s) subarray\n",
12018 geo
.dev_name
, geo
.devnm
);
12019 goto exit_imsm_reshape_super
;
12021 super
->current_vol
= dev
->index
;
12022 change
= imsm_analyze_change(st
, &geo
, direction
);
12025 ret_val
= imsm_takeover(st
, &geo
);
12027 case CH_MIGRATION
: {
12028 struct imsm_update_reshape_migration
*u
= NULL
;
12030 imsm_create_metadata_update_for_migration(
12033 dprintf("imsm: Cannot prepare update\n");
12037 /* update metadata locally */
12038 imsm_update_metadata_locally(st
, u
, len
);
12039 /* and possibly remotely */
12040 if (st
->update_tail
)
12041 append_metadata_update(st
, u
, len
);
12046 case CH_ARRAY_SIZE
: {
12047 struct imsm_update_size_change
*u
= NULL
;
12049 imsm_create_metadata_update_for_size_change(
12052 dprintf("imsm: Cannot prepare update\n");
12056 /* update metadata locally */
12057 imsm_update_metadata_locally(st
, u
, len
);
12058 /* and possibly remotely */
12059 if (st
->update_tail
)
12060 append_metadata_update(st
, u
, len
);
12070 exit_imsm_reshape_super
:
12071 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
12075 #define COMPLETED_OK 0
12076 #define COMPLETED_NONE 1
12077 #define COMPLETED_DELAYED 2
12079 static int read_completed(int fd
, unsigned long long *val
)
12084 ret
= sysfs_fd_get_str(fd
, buf
, 50);
12088 ret
= COMPLETED_OK
;
12089 if (strncmp(buf
, "none", 4) == 0) {
12090 ret
= COMPLETED_NONE
;
12091 } else if (strncmp(buf
, "delayed", 7) == 0) {
12092 ret
= COMPLETED_DELAYED
;
12095 *val
= strtoull(buf
, &ep
, 0);
12096 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
12102 /*******************************************************************************
12103 * Function: wait_for_reshape_imsm
12104 * Description: Function writes new sync_max value and waits until
12105 * reshape process reach new position
12107 * sra : general array info
12108 * ndata : number of disks in new array's layout
12111 * 1 : there is no reshape in progress,
12113 ******************************************************************************/
12114 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
12116 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
12118 unsigned long long completed
;
12119 /* to_complete : new sync_max position */
12120 unsigned long long to_complete
= sra
->reshape_progress
;
12121 unsigned long long position_to_set
= to_complete
/ ndata
;
12123 if (!is_fd_valid(fd
)) {
12124 dprintf("cannot open reshape_position\n");
12129 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
12131 dprintf("cannot read reshape_position (no reshape in progres)\n");
12135 sleep_for(0, MSEC_TO_NSEC(30), true);
12140 if (completed
> position_to_set
) {
12141 dprintf("wrong next position to set %llu (%llu)\n",
12142 to_complete
, position_to_set
);
12146 dprintf("Position set: %llu\n", position_to_set
);
12147 if (sysfs_set_num(sra
, NULL
, "sync_max",
12148 position_to_set
) != 0) {
12149 dprintf("cannot set reshape position to %llu\n",
12158 int timeout
= 3000;
12160 sysfs_wait(fd
, &timeout
);
12161 if (sysfs_get_str(sra
, NULL
, "sync_action",
12163 strncmp(action
, "reshape", 7) != 0) {
12164 if (strncmp(action
, "idle", 4) == 0)
12170 rc
= read_completed(fd
, &completed
);
12172 dprintf("cannot read reshape_position (in loop)\n");
12175 } else if (rc
== COMPLETED_NONE
)
12177 } while (completed
< position_to_set
);
12183 /*******************************************************************************
12184 * Function: check_degradation_change
12185 * Description: Check that array hasn't become failed.
12187 * info : for sysfs access
12188 * sources : source disks descriptors
12189 * degraded: previous degradation level
12191 * degradation level
12192 ******************************************************************************/
12193 int check_degradation_change(struct mdinfo
*info
,
12197 unsigned long long new_degraded
;
12200 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
12201 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
12202 /* check each device to ensure it is still working */
12205 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
12206 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
12208 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
12210 int raid_disk
= sd
->disk
.raid_disk
;
12212 if (sysfs_get_str(info
,
12213 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
12214 strstr(sbuf
, "faulty") ||
12215 strstr(sbuf
, "in_sync") == NULL
) {
12216 /* this device is dead */
12217 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
12218 if (raid_disk
>= 0)
12219 close_fd(&sources
[raid_disk
]);
12226 return new_degraded
;
12229 /*******************************************************************************
12230 * Function: imsm_manage_reshape
12231 * Description: Function finds array under reshape and it manages reshape
12232 * process. It creates stripes backups (if required) and sets
12235 * afd : Backup handle (nattive) - not used
12236 * sra : general array info
12237 * reshape : reshape parameters - not used
12238 * st : supertype structure
12239 * blocks : size of critical section [blocks]
12240 * fds : table of source device descriptor
12241 * offsets : start of array (offest per devices)
12243 * destfd : table of destination device descriptor
12244 * destoffsets : table of destination offsets (per device)
12246 * 1 : success, reshape is done
12248 ******************************************************************************/
12249 static int imsm_manage_reshape(
12250 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
12251 struct supertype
*st
, unsigned long backup_blocks
,
12252 int *fds
, unsigned long long *offsets
,
12253 int dests
, int *destfd
, unsigned long long *destoffsets
)
12256 struct intel_super
*super
= st
->sb
;
12257 struct intel_dev
*dv
;
12258 unsigned int sector_size
= super
->sector_size
;
12259 struct imsm_dev
*dev
= NULL
;
12260 struct imsm_map
*map_src
, *map_dest
;
12261 int migr_vol_qan
= 0;
12262 int ndata
, odata
; /* [bytes] */
12263 int chunk
; /* [bytes] */
12264 struct migr_record
*migr_rec
;
12266 unsigned int buf_size
; /* [bytes] */
12267 unsigned long long max_position
; /* array size [bytes] */
12268 unsigned long long next_step
; /* [blocks]/[bytes] */
12269 unsigned long long old_data_stripe_length
;
12270 unsigned long long start_src
; /* [bytes] */
12271 unsigned long long start
; /* [bytes] */
12272 unsigned long long start_buf_shift
; /* [bytes] */
12274 int source_layout
= 0;
12275 int subarray_index
= -1;
12280 if (!fds
|| !offsets
)
12283 /* Find volume during the reshape */
12284 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
12285 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12286 dv
->dev
->vol
.migr_state
== 1) {
12289 subarray_index
= dv
->index
;
12292 /* Only one volume can migrate at the same time */
12293 if (migr_vol_qan
!= 1) {
12294 pr_err("%s", migr_vol_qan
?
12295 "Number of migrating volumes greater than 1\n" :
12296 "There is no volume during migrationg\n");
12300 map_dest
= get_imsm_map(dev
, MAP_0
);
12301 map_src
= get_imsm_map(dev
, MAP_1
);
12302 if (map_src
== NULL
)
12305 ndata
= imsm_num_data_members(map_dest
);
12306 odata
= imsm_num_data_members(map_src
);
12308 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12309 old_data_stripe_length
= odata
* chunk
;
12311 migr_rec
= super
->migr_rec
;
12313 /* initialize migration record for start condition */
12314 if (sra
->reshape_progress
== 0)
12315 init_migr_record_imsm(st
, dev
, sra
);
12317 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12318 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12321 /* Save checkpoint to update migration record for current
12322 * reshape position (in md). It can be farther than current
12323 * reshape position in metadata.
12325 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12326 /* ignore error == 2, this can mean end of reshape here
12328 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12333 /* size for data */
12334 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12335 /* extend buffer size for parity disk */
12336 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12337 /* add space for stripe alignment */
12338 buf_size
+= old_data_stripe_length
;
12339 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12340 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12344 max_position
= sra
->component_size
* ndata
;
12345 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12347 while (current_migr_unit(migr_rec
) <
12348 get_num_migr_units(migr_rec
)) {
12349 /* current reshape position [blocks] */
12350 unsigned long long current_position
=
12351 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12352 * current_migr_unit(migr_rec
);
12353 unsigned long long border
;
12355 /* Check that array hasn't become failed.
12357 degraded
= check_degradation_change(sra
, fds
, degraded
);
12358 if (degraded
> 1) {
12359 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12363 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12365 if ((current_position
+ next_step
) > max_position
)
12366 next_step
= max_position
- current_position
;
12368 start
= current_position
* 512;
12370 /* align reading start to old geometry */
12371 start_buf_shift
= start
% old_data_stripe_length
;
12372 start_src
= start
- start_buf_shift
;
12374 border
= (start_src
/ odata
) - (start
/ ndata
);
12376 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12377 /* save critical stripes to buf
12378 * start - start address of current unit
12379 * to backup [bytes]
12380 * start_src - start address of current unit
12381 * to backup alligned to source array
12384 unsigned long long next_step_filler
;
12385 unsigned long long copy_length
= next_step
* 512;
12387 /* allign copy area length to stripe in old geometry */
12388 next_step_filler
= ((copy_length
+ start_buf_shift
)
12389 % old_data_stripe_length
);
12390 if (next_step_filler
)
12391 next_step_filler
= (old_data_stripe_length
12392 - next_step_filler
);
12393 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12394 start
, start_src
, copy_length
,
12395 start_buf_shift
, next_step_filler
);
12397 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12398 chunk
, map_src
->raid_level
,
12399 source_layout
, 0, NULL
, start_src
,
12401 next_step_filler
+ start_buf_shift
,
12403 dprintf("imsm: Cannot save stripes to buffer\n");
12406 /* Convert data to destination format and store it
12407 * in backup general migration area
12409 if (save_backup_imsm(st
, dev
, sra
,
12410 buf
+ start_buf_shift
, copy_length
)) {
12411 dprintf("imsm: Cannot save stripes to target devices\n");
12414 if (save_checkpoint_imsm(st
, sra
,
12415 UNIT_SRC_IN_CP_AREA
)) {
12416 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12420 /* set next step to use whole border area */
12421 border
/= next_step
;
12423 next_step
*= border
;
12425 /* When data backed up, checkpoint stored,
12426 * kick the kernel to reshape unit of data
12428 next_step
= next_step
+ sra
->reshape_progress
;
12429 /* limit next step to array max position */
12430 if (next_step
> max_position
)
12431 next_step
= max_position
;
12432 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12433 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12434 sra
->reshape_progress
= next_step
;
12436 /* wait until reshape finish */
12437 if (wait_for_reshape_imsm(sra
, ndata
)) {
12438 dprintf("wait_for_reshape_imsm returned error!\n");
12444 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12445 /* ignore error == 2, this can mean end of reshape here
12447 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12453 /* clear migr_rec on disks after successful migration */
12456 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12457 for (d
= super
->disks
; d
; d
= d
->next
) {
12458 if (d
->index
< 0 || is_failed(&d
->disk
))
12460 unsigned long long dsize
;
12462 get_dev_size(d
->fd
, NULL
, &dsize
);
12463 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12465 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12466 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12467 MIGR_REC_BUF_SECTORS
*sector_size
)
12468 perror("Write migr_rec failed");
12472 /* return '1' if done */
12475 /* After the reshape eliminate size mismatch in metadata.
12476 * Don't update md/component_size here, volume hasn't
12477 * to take whole space. It is allowed by kernel.
12478 * md/component_size will be set propoperly after next assembly.
12480 imsm_fix_size_mismatch(st
, subarray_index
);
12484 /* See Grow.c: abort_reshape() for further explanation */
12485 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12486 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12487 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12492 /*******************************************************************************
12493 * Function: calculate_bitmap_min_chunksize
12494 * Description: Calculates the minimal valid bitmap chunk size
12496 * max_bits : indicate how many bits can be used for the bitmap
12497 * data_area_size : the size of the data area covered by the bitmap
12500 * The bitmap chunk size
12501 ******************************************************************************/
12502 static unsigned long long
12503 calculate_bitmap_min_chunksize(unsigned long long max_bits
,
12504 unsigned long long data_area_size
)
12506 unsigned long long min_chunk
=
12507 4096; /* sub-page chunks don't work yet.. */
12508 unsigned long long bits
= data_area_size
/ min_chunk
+ 1;
12510 while (bits
> max_bits
) {
12512 bits
= (bits
+ 1) / 2;
12517 /*******************************************************************************
12518 * Function: calculate_bitmap_chunksize
12519 * Description: Calculates the bitmap chunk size for the given device
12521 * st : supertype information
12522 * dev : device for the bitmap
12525 * The bitmap chunk size
12526 ******************************************************************************/
12527 static unsigned long long calculate_bitmap_chunksize(struct supertype
*st
,
12528 struct imsm_dev
*dev
)
12530 struct intel_super
*super
= st
->sb
;
12531 unsigned long long min_chunksize
;
12532 unsigned long long result
= IMSM_DEFAULT_BITMAP_CHUNKSIZE
;
12533 size_t dev_size
= imsm_dev_size(dev
);
12535 min_chunksize
= calculate_bitmap_min_chunksize(
12536 IMSM_BITMAP_AREA_SIZE
* super
->sector_size
, dev_size
);
12538 if (result
< min_chunksize
)
12539 result
= min_chunksize
;
12544 /*******************************************************************************
12545 * Function: init_bitmap_header
12546 * Description: Initialize the bitmap header structure
12548 * st : supertype information
12549 * bms : bitmap header struct to initialize
12550 * dev : device for the bitmap
12555 ******************************************************************************/
12556 static int init_bitmap_header(struct supertype
*st
, struct bitmap_super_s
*bms
,
12557 struct imsm_dev
*dev
)
12564 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
12565 bms
->version
= __cpu_to_le32(BITMAP_MAJOR_HI
);
12566 bms
->daemon_sleep
= __cpu_to_le32(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
);
12567 bms
->sync_size
= __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
);
12568 bms
->write_behind
= __cpu_to_le32(0);
12570 uuid_from_super_imsm(st
, vol_uuid
);
12571 memcpy(bms
->uuid
, vol_uuid
, 16);
12573 bms
->chunksize
= calculate_bitmap_chunksize(st
, dev
);
12578 /*******************************************************************************
12579 * Function: validate_internal_bitmap_for_drive
12580 * Description: Verify if the bitmap header for a given drive.
12582 * st : supertype information
12583 * offset : The offset from the beginning of the drive where to look for
12584 * the bitmap header.
12585 * d : the drive info
12590 ******************************************************************************/
12591 static int validate_internal_bitmap_for_drive(struct supertype
*st
,
12592 unsigned long long offset
,
12595 struct intel_super
*super
= st
->sb
;
12598 bitmap_super_t
*bms
;
12606 if (posix_memalign(&read_buf
, MAX_SECTOR_SIZE
, IMSM_BITMAP_HEADER_SIZE
))
12610 if (!is_fd_valid(fd
)) {
12611 fd
= open(d
->devname
, O_RDONLY
, 0);
12613 if (!is_fd_valid(fd
)) {
12614 dprintf("cannot open the device %s\n", d
->devname
);
12619 if (lseek64(fd
, offset
* super
->sector_size
, SEEK_SET
) < 0)
12621 if (read(fd
, read_buf
, IMSM_BITMAP_HEADER_SIZE
) !=
12622 IMSM_BITMAP_HEADER_SIZE
)
12625 uuid_from_super_imsm(st
, vol_uuid
);
12628 if ((bms
->magic
!= __cpu_to_le32(BITMAP_MAGIC
)) ||
12629 (bms
->version
!= __cpu_to_le32(BITMAP_MAJOR_HI
)) ||
12630 (!same_uuid((int *)bms
->uuid
, vol_uuid
, st
->ss
->swapuuid
))) {
12631 dprintf("wrong bitmap header detected\n");
12637 if (!is_fd_valid(d
->fd
))
12646 /*******************************************************************************
12647 * Function: validate_internal_bitmap_imsm
12648 * Description: Verify if the bitmap header is in place and with proper data.
12650 * st : supertype information
12653 * 0 : success or device w/o RWH_BITMAP
12655 ******************************************************************************/
12656 static int validate_internal_bitmap_imsm(struct supertype
*st
)
12658 struct intel_super
*super
= st
->sb
;
12659 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
12660 unsigned long long offset
;
12663 if (dev
->rwh_policy
!= RWH_BITMAP
)
12666 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12667 for (d
= super
->disks
; d
; d
= d
->next
) {
12668 if (d
->index
< 0 || is_failed(&d
->disk
))
12671 if (validate_internal_bitmap_for_drive(st
, offset
, d
)) {
12672 pr_err("imsm: bitmap validation failed\n");
12679 /*******************************************************************************
12680 * Function: add_internal_bitmap_imsm
12681 * Description: Mark the volume to use the bitmap and updates the chunk size value.
12683 * st : supertype information
12684 * chunkp : bitmap chunk size
12685 * delay : not used for imsm
12686 * write_behind : not used for imsm
12687 * size : not used for imsm
12688 * may_change : not used for imsm
12689 * amajor : not used for imsm
12694 ******************************************************************************/
12695 static int add_internal_bitmap_imsm(struct supertype
*st
, int *chunkp
,
12696 int delay
, int write_behind
,
12697 unsigned long long size
, int may_change
,
12700 struct intel_super
*super
= st
->sb
;
12701 int vol_idx
= super
->current_vol
;
12702 struct imsm_dev
*dev
;
12704 if (!super
->devlist
|| vol_idx
== -1 || !chunkp
)
12707 dev
= get_imsm_dev(super
, vol_idx
);
12708 dev
->rwh_policy
= RWH_BITMAP
;
12709 *chunkp
= calculate_bitmap_chunksize(st
, dev
);
12713 /*******************************************************************************
12714 * Function: locate_bitmap_imsm
12715 * Description: Seek 'fd' to start of write-intent-bitmap.
12717 * st : supertype information
12718 * fd : file descriptor for the device
12719 * node_num : not used for imsm
12724 ******************************************************************************/
12725 static int locate_bitmap_imsm(struct supertype
*st
, int fd
, int node_num
)
12727 struct intel_super
*super
= st
->sb
;
12728 unsigned long long offset
;
12729 int vol_idx
= super
->current_vol
;
12731 if (!super
->devlist
|| vol_idx
== -1)
12734 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12735 dprintf("bitmap header offset is %llu\n", offset
);
12737 lseek64(fd
, offset
<< 9, 0);
12742 /*******************************************************************************
12743 * Function: write_init_bitmap_imsm
12744 * Description: Write a bitmap header and prepares the area for the bitmap.
12746 * st : supertype information
12747 * fd : file descriptor for the device
12748 * update : not used for imsm
12753 ******************************************************************************/
12754 static int write_init_bitmap_imsm(struct supertype
*st
, int fd
,
12755 enum bitmap_update update
)
12757 struct intel_super
*super
= st
->sb
;
12758 int vol_idx
= super
->current_vol
;
12760 unsigned long long offset
;
12761 bitmap_super_t bms
= { 0 };
12762 size_t written
= 0;
12767 if (!super
->devlist
|| !super
->sector_size
|| vol_idx
== -1)
12770 struct imsm_dev
*dev
= get_imsm_dev(super
, vol_idx
);
12772 /* first clear the space for bitmap header */
12773 unsigned long long bitmap_area_start
=
12774 get_bitmap_header_sector(super
, vol_idx
);
12776 dprintf("zeroing area start (%llu) and size (%u)\n", bitmap_area_start
,
12777 IMSM_BITMAP_AND_HEADER_SIZE
/ super
->sector_size
);
12778 if (zero_disk_range(fd
, bitmap_area_start
,
12779 IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
)) {
12780 pr_err("imsm: cannot zeroing the space for the bitmap\n");
12784 /* The bitmap area should be filled with "1"s to perform initial
12787 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
))
12789 memset(buf
, 0xFF, MAX_SECTOR_SIZE
);
12790 offset
= get_bitmap_sector(super
, vol_idx
);
12791 lseek64(fd
, offset
<< 9, 0);
12792 while (written
< IMSM_BITMAP_AREA_SIZE
) {
12793 to_write
= IMSM_BITMAP_AREA_SIZE
- written
;
12794 if (to_write
> MAX_SECTOR_SIZE
)
12795 to_write
= MAX_SECTOR_SIZE
;
12796 rv_num
= write(fd
, buf
, MAX_SECTOR_SIZE
);
12797 if (rv_num
!= MAX_SECTOR_SIZE
) {
12799 dprintf("cannot initialize bitmap area\n");
12805 /* write a bitmap header */
12806 init_bitmap_header(st
, &bms
, dev
);
12807 memset(buf
, 0, MAX_SECTOR_SIZE
);
12808 memcpy(buf
, &bms
, sizeof(bitmap_super_t
));
12809 if (locate_bitmap_imsm(st
, fd
, 0)) {
12811 dprintf("cannot locate the bitmap\n");
12814 if (write(fd
, buf
, MAX_SECTOR_SIZE
) != MAX_SECTOR_SIZE
) {
12816 dprintf("cannot write the bitmap header\n");
12827 /*******************************************************************************
12828 * Function: is_vol_to_setup_bitmap
12829 * Description: Checks if a bitmap should be activated on the dev.
12831 * info : info about the volume to setup the bitmap
12832 * dev : the device to check against bitmap creation
12835 * 0 : bitmap should be set up on the device
12837 ******************************************************************************/
12838 static int is_vol_to_setup_bitmap(struct mdinfo
*info
, struct imsm_dev
*dev
)
12843 if ((strcmp((char *)dev
->volume
, info
->name
) == 0) &&
12844 (dev
->rwh_policy
== RWH_BITMAP
))
12850 /*******************************************************************************
12851 * Function: set_bitmap_sysfs
12852 * Description: Set the sysfs atributes of a given volume to activate the bitmap.
12854 * info : info about the volume where the bitmap should be setup
12855 * chunksize : bitmap chunk size
12856 * location : location of the bitmap
12861 ******************************************************************************/
12862 static int set_bitmap_sysfs(struct mdinfo
*info
, unsigned long long chunksize
,
12865 /* The bitmap/metadata is set to external to allow changing of value for
12866 * bitmap/location. When external is used, the kernel will treat an offset
12867 * related to the device's first lba (in opposition to the "internal" case
12868 * when this value is related to the beginning of the superblock).
12870 if (sysfs_set_str(info
, NULL
, "bitmap/metadata", "external")) {
12871 dprintf("failed to set bitmap/metadata\n");
12875 /* It can only be changed when no bitmap is active.
12876 * Should be bigger than 512 and must be power of 2.
12877 * It is expecting the value in bytes.
12879 if (sysfs_set_num(info
, NULL
, "bitmap/chunksize",
12880 __cpu_to_le32(chunksize
))) {
12881 dprintf("failed to set bitmap/chunksize\n");
12885 /* It is expecting the value in sectors. */
12886 if (sysfs_set_num(info
, NULL
, "bitmap/space",
12887 __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
))) {
12888 dprintf("failed to set bitmap/space\n");
12892 /* Determines the delay between the bitmap updates.
12893 * It is expecting the value in seconds.
12895 if (sysfs_set_num(info
, NULL
, "bitmap/time_base",
12896 __cpu_to_le64(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
))) {
12897 dprintf("failed to set bitmap/time_base\n");
12901 /* It is expecting the value in sectors with a sign at the beginning. */
12902 if (sysfs_set_str(info
, NULL
, "bitmap/location", location
)) {
12903 dprintf("failed to set bitmap/location\n");
12910 /*******************************************************************************
12911 * Function: set_bitmap_imsm
12912 * Description: Setup the bitmap for the given volume
12914 * st : supertype information
12915 * info : info about the volume where the bitmap should be setup
12920 ******************************************************************************/
12921 static int set_bitmap_imsm(struct supertype
*st
, struct mdinfo
*info
)
12923 struct intel_super
*super
= st
->sb
;
12924 int prev_current_vol
= super
->current_vol
;
12925 struct imsm_dev
*dev
;
12927 char location
[16] = "";
12928 unsigned long long chunksize
;
12929 struct intel_dev
*dev_it
;
12931 for (dev_it
= super
->devlist
; dev_it
; dev_it
= dev_it
->next
) {
12932 super
->current_vol
= dev_it
->index
;
12933 dev
= get_imsm_dev(super
, super
->current_vol
);
12935 if (is_vol_to_setup_bitmap(info
, dev
)) {
12936 if (validate_internal_bitmap_imsm(st
)) {
12937 dprintf("bitmap header validation failed\n");
12941 chunksize
= calculate_bitmap_chunksize(st
, dev
);
12942 dprintf("chunk size is %llu\n", chunksize
);
12944 snprintf(location
, sizeof(location
), "+%llu",
12945 get_bitmap_sector(super
, super
->current_vol
));
12946 dprintf("bitmap offset is %s\n", location
);
12948 if (set_bitmap_sysfs(info
, chunksize
, location
)) {
12949 dprintf("cannot setup the bitmap\n");
12956 super
->current_vol
= prev_current_vol
;
12960 struct superswitch super_imsm
= {
12961 .examine_super
= examine_super_imsm
,
12962 .brief_examine_super
= brief_examine_super_imsm
,
12963 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12964 .export_examine_super
= export_examine_super_imsm
,
12965 .detail_super
= detail_super_imsm
,
12966 .brief_detail_super
= brief_detail_super_imsm
,
12967 .write_init_super
= write_init_super_imsm
,
12968 .validate_geometry
= validate_geometry_imsm
,
12969 .add_to_super
= add_to_super_imsm
,
12970 .remove_from_super
= remove_from_super_imsm
,
12971 .detail_platform
= detail_platform_imsm
,
12972 .export_detail_platform
= export_detail_platform_imsm
,
12973 .kill_subarray
= kill_subarray_imsm
,
12974 .update_subarray
= update_subarray_imsm
,
12975 .load_container
= load_container_imsm
,
12976 .default_geometry
= default_geometry_imsm
,
12977 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12978 .reshape_super
= imsm_reshape_super
,
12979 .manage_reshape
= imsm_manage_reshape
,
12980 .recover_backup
= recover_backup_imsm
,
12981 .examine_badblocks
= examine_badblocks_imsm
,
12982 .match_home
= match_home_imsm
,
12983 .uuid_from_super
= uuid_from_super_imsm
,
12984 .getinfo_super
= getinfo_super_imsm
,
12985 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12986 .update_super
= update_super_imsm
,
12988 .avail_size
= avail_size_imsm
,
12989 .get_spare_criteria
= get_spare_criteria_imsm
,
12991 .compare_super
= compare_super_imsm
,
12993 .load_super
= load_super_imsm
,
12994 .init_super
= init_super_imsm
,
12995 .store_super
= store_super_imsm
,
12996 .free_super
= free_super_imsm
,
12997 .match_metadata_desc
= match_metadata_desc_imsm
,
12998 .container_content
= container_content_imsm
,
12999 .validate_container
= validate_container_imsm
,
13001 .add_internal_bitmap
= add_internal_bitmap_imsm
,
13002 .locate_bitmap
= locate_bitmap_imsm
,
13003 .write_bitmap
= write_init_bitmap_imsm
,
13004 .set_bitmap
= set_bitmap_imsm
,
13006 .write_init_ppl
= write_init_ppl_imsm
,
13007 .validate_ppl
= validate_ppl_imsm
,
13013 .open_new
= imsm_open_new
,
13014 .set_array_state
= imsm_set_array_state
,
13015 .set_disk
= imsm_set_disk
,
13016 .sync_metadata
= imsm_sync_metadata
,
13017 .activate_spare
= imsm_activate_spare
,
13018 .process_update
= imsm_process_update
,
13019 .prepare_update
= imsm_prepare_update
,
13020 .record_bad_block
= imsm_record_badblock
,
13021 .clear_bad_block
= imsm_clear_badblock
,
13022 .get_bad_blocks
= imsm_get_badblocks
,