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
;
1739 static bool is_gen_migration(struct imsm_dev
*dev
);
1741 #define IMSM_4K_DIV 8
1743 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1744 struct imsm_dev
*dev
);
1746 static void print_imsm_dev(struct intel_super
*super
,
1747 struct imsm_dev
*dev
,
1753 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1754 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1758 printf("[%.16s]:\n", dev
->volume
);
1759 printf(" Subarray : %d\n", super
->current_vol
);
1760 printf(" UUID : %s\n", uuid
);
1761 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1763 printf(" <-- %d", get_imsm_raid_level(map2
));
1765 printf(" Members : %d", map
->num_members
);
1767 printf(" <-- %d", map2
->num_members
);
1769 printf(" Slots : [");
1770 for (i
= 0; i
< map
->num_members
; i
++) {
1771 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1772 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1777 for (i
= 0; i
< map2
->num_members
; i
++) {
1778 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1779 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1784 printf(" Failed disk : ");
1785 if (map
->failed_disk_num
== 0xff)
1786 printf(STR_COMMON_NONE
);
1788 printf("%i", map
->failed_disk_num
);
1790 slot
= get_imsm_disk_slot(map
, disk_idx
);
1792 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1793 printf(" This Slot : %d%s\n", slot
,
1794 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1796 printf(" This Slot : ?\n");
1797 printf(" Sector Size : %u\n", super
->sector_size
);
1798 sz
= imsm_dev_size(dev
);
1799 printf(" Array Size : %llu%s\n",
1800 (unsigned long long)sz
* 512 / super
->sector_size
,
1801 human_size(sz
* 512));
1802 sz
= blocks_per_member(map
);
1803 printf(" Per Dev Size : %llu%s\n",
1804 (unsigned long long)sz
* 512 / super
->sector_size
,
1805 human_size(sz
* 512));
1806 printf(" Sector Offset : %llu\n",
1807 pba_of_lba0(map
) * 512 / super
->sector_size
);
1808 printf(" Num Stripes : %llu\n",
1809 num_data_stripes(map
));
1810 printf(" Chunk Size : %u KiB",
1811 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1813 printf(" <-- %u KiB",
1814 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1816 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1817 printf(" Migrate State : ");
1818 if (dev
->vol
.migr_state
) {
1819 if (migr_type(dev
) == MIGR_INIT
)
1820 printf("initialize\n");
1821 else if (migr_type(dev
) == MIGR_REBUILD
)
1822 printf("rebuild\n");
1823 else if (migr_type(dev
) == MIGR_VERIFY
)
1825 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1826 printf("general migration\n");
1827 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1828 printf("state change\n");
1829 else if (migr_type(dev
) == MIGR_REPAIR
)
1832 printf("<unknown:%d>\n", migr_type(dev
));
1835 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1836 if (dev
->vol
.migr_state
) {
1837 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1839 printf(" <-- %s", map_state_str
[map
->map_state
]);
1840 printf("\n Checkpoint : %llu ", vol_curr_migr_unit(dev
));
1841 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1844 printf("(%llu)", (unsigned long long)
1845 blocks_per_migr_unit(super
, dev
));
1848 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1850 printf(" RWH Policy : ");
1851 if (dev
->rwh_policy
== RWH_OFF
|| dev
->rwh_policy
== RWH_MULTIPLE_OFF
)
1853 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1854 printf("PPL distributed\n");
1855 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1856 printf("PPL journaling drive\n");
1857 else if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
1858 printf("Multiple distributed PPLs\n");
1859 else if (dev
->rwh_policy
== RWH_MULTIPLE_PPLS_JOURNALING_DRIVE
)
1860 printf("Multiple PPLs on journaling drive\n");
1861 else if (dev
->rwh_policy
== RWH_BITMAP
)
1862 printf("Write-intent bitmap\n");
1864 printf("<unknown:%d>\n", dev
->rwh_policy
);
1866 printf(" Volume ID : %u\n", dev
->my_vol_raid_dev_num
);
1869 static void print_imsm_disk(struct imsm_disk
*disk
,
1872 unsigned int sector_size
) {
1873 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1876 if (index
< -1 || !disk
)
1880 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1882 printf(" Disk%02d Serial : %s\n", index
, str
);
1884 printf(" Disk Serial : %s\n", str
);
1885 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1886 is_configured(disk
) ? " active" : "",
1887 is_failed(disk
) ? " failed" : "",
1888 is_journal(disk
) ? " journal" : "");
1889 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1890 sz
= total_blocks(disk
) - reserved
;
1891 printf(" Usable Size : %llu%s\n",
1892 (unsigned long long)sz
* 512 / sector_size
,
1893 human_size(sz
* 512));
1896 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1898 struct migr_record
*migr_rec
= super
->migr_rec
;
1900 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1901 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1902 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1903 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1904 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1905 set_migr_chkp_area_pba(migr_rec
,
1906 migr_chkp_area_pba(migr_rec
) / IMSM_4K_DIV
);
1907 set_migr_dest_1st_member_lba(migr_rec
,
1908 migr_dest_1st_member_lba(migr_rec
) / IMSM_4K_DIV
);
1911 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1913 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1916 void convert_to_4k(struct intel_super
*super
)
1918 struct imsm_super
*mpb
= super
->anchor
;
1919 struct imsm_disk
*disk
;
1921 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1923 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1924 disk
= __get_imsm_disk(mpb
, i
);
1926 convert_to_4k_imsm_disk(disk
);
1928 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1929 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1930 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1932 set_imsm_dev_size(dev
, imsm_dev_size(dev
)/IMSM_4K_DIV
);
1933 set_vol_curr_migr_unit(dev
,
1934 vol_curr_migr_unit(dev
) / IMSM_4K_DIV
);
1937 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1938 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1939 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1941 if (dev
->vol
.migr_state
) {
1943 map
= get_imsm_map(dev
, MAP_1
);
1944 set_blocks_per_member(map
,
1945 blocks_per_member(map
)/IMSM_4K_DIV
);
1946 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1947 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1951 struct bbm_log
*log
= (void *)mpb
+
1952 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1955 for (i
= 0; i
< log
->entry_count
; i
++) {
1956 struct bbm_log_entry
*entry
=
1957 &log
->marked_block_entries
[i
];
1959 __u8 count
= entry
->marked_count
+ 1;
1960 unsigned long long sector
=
1961 __le48_to_cpu(&entry
->defective_block_start
);
1963 entry
->defective_block_start
=
1964 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1965 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1969 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1972 void examine_migr_rec_imsm(struct intel_super
*super
)
1974 struct migr_record
*migr_rec
= super
->migr_rec
;
1975 struct imsm_super
*mpb
= super
->anchor
;
1978 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1979 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1980 struct imsm_map
*map
;
1983 if (is_gen_migration(dev
) == false)
1986 printf("\nMigration Record Information:");
1988 /* first map under migration */
1989 map
= get_imsm_map(dev
, MAP_0
);
1992 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1993 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1994 printf(" Empty\n ");
1995 printf("Examine one of first two disks in array\n");
1998 printf("\n Status : ");
1999 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
2002 printf("Contains Data\n");
2003 printf(" Current Unit : %llu\n",
2004 current_migr_unit(migr_rec
));
2005 printf(" Family : %u\n",
2006 __le32_to_cpu(migr_rec
->family_num
));
2007 printf(" Ascending : %u\n",
2008 __le32_to_cpu(migr_rec
->ascending_migr
));
2009 printf(" Blocks Per Unit : %u\n",
2010 __le32_to_cpu(migr_rec
->blocks_per_unit
));
2011 printf(" Dest. Depth Per Unit : %u\n",
2012 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
2013 printf(" Checkpoint Area pba : %llu\n",
2014 migr_chkp_area_pba(migr_rec
));
2015 printf(" First member lba : %llu\n",
2016 migr_dest_1st_member_lba(migr_rec
));
2017 printf(" Total Number of Units : %llu\n",
2018 get_num_migr_units(migr_rec
));
2019 printf(" Size of volume : %llu\n",
2020 join_u32(migr_rec
->post_migr_vol_cap
,
2021 migr_rec
->post_migr_vol_cap_hi
));
2022 printf(" Record was read from : %u\n",
2023 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
2029 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
2031 struct migr_record
*migr_rec
= super
->migr_rec
;
2033 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
2034 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
2035 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
2036 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
2037 &migr_rec
->post_migr_vol_cap
,
2038 &migr_rec
->post_migr_vol_cap_hi
);
2039 set_migr_chkp_area_pba(migr_rec
,
2040 migr_chkp_area_pba(migr_rec
) * IMSM_4K_DIV
);
2041 set_migr_dest_1st_member_lba(migr_rec
,
2042 migr_dest_1st_member_lba(migr_rec
) * IMSM_4K_DIV
);
2045 void convert_from_4k(struct intel_super
*super
)
2047 struct imsm_super
*mpb
= super
->anchor
;
2048 struct imsm_disk
*disk
;
2050 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
2052 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2053 disk
= __get_imsm_disk(mpb
, i
);
2055 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
2058 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2059 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2060 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2062 set_imsm_dev_size(dev
, imsm_dev_size(dev
)*IMSM_4K_DIV
);
2063 set_vol_curr_migr_unit(dev
,
2064 vol_curr_migr_unit(dev
) * IMSM_4K_DIV
);
2067 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
2068 map
->blocks_per_strip
*= IMSM_4K_DIV
;
2069 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
2071 if (dev
->vol
.migr_state
) {
2073 map
= get_imsm_map(dev
, MAP_1
);
2074 set_blocks_per_member(map
,
2075 blocks_per_member(map
)*IMSM_4K_DIV
);
2076 map
->blocks_per_strip
*= IMSM_4K_DIV
;
2077 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
2081 struct bbm_log
*log
= (void *)mpb
+
2082 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
2085 for (i
= 0; i
< log
->entry_count
; i
++) {
2086 struct bbm_log_entry
*entry
=
2087 &log
->marked_block_entries
[i
];
2089 __u8 count
= entry
->marked_count
+ 1;
2090 unsigned long long sector
=
2091 __le48_to_cpu(&entry
->defective_block_start
);
2093 entry
->defective_block_start
=
2094 __cpu_to_le48(sector
*IMSM_4K_DIV
);
2095 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
2099 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
2102 /*******************************************************************************
2103 * function: imsm_check_attributes
2104 * Description: Function checks if features represented by attributes flags
2105 * are supported by mdadm.
2107 * attributes - Attributes read from metadata
2109 * 0 - passed attributes contains unsupported features flags
2110 * 1 - all features are supported
2111 ******************************************************************************/
2112 static int imsm_check_attributes(__u32 attributes
)
2115 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
2117 not_supported
&= ~MPB_ATTRIB_IGNORED
;
2119 not_supported
&= attributes
;
2120 if (not_supported
) {
2121 pr_err("(IMSM): Unsupported attributes : %x\n",
2122 (unsigned)__le32_to_cpu(not_supported
));
2123 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
2124 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
2125 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
2127 if (not_supported
& MPB_ATTRIB_2TB
) {
2128 dprintf("\t\tMPB_ATTRIB_2TB\n");
2129 not_supported
^= MPB_ATTRIB_2TB
;
2131 if (not_supported
& MPB_ATTRIB_RAID0
) {
2132 dprintf("\t\tMPB_ATTRIB_RAID0\n");
2133 not_supported
^= MPB_ATTRIB_RAID0
;
2135 if (not_supported
& MPB_ATTRIB_RAID1
) {
2136 dprintf("\t\tMPB_ATTRIB_RAID1\n");
2137 not_supported
^= MPB_ATTRIB_RAID1
;
2139 if (not_supported
& MPB_ATTRIB_RAID10
) {
2140 dprintf("\t\tMPB_ATTRIB_RAID10\n");
2141 not_supported
^= MPB_ATTRIB_RAID10
;
2143 if (not_supported
& MPB_ATTRIB_RAID1E
) {
2144 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
2145 not_supported
^= MPB_ATTRIB_RAID1E
;
2147 if (not_supported
& MPB_ATTRIB_RAID5
) {
2148 dprintf("\t\tMPB_ATTRIB_RAID5\n");
2149 not_supported
^= MPB_ATTRIB_RAID5
;
2151 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
2152 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
2153 not_supported
^= MPB_ATTRIB_RAIDCNG
;
2155 if (not_supported
& MPB_ATTRIB_BBM
) {
2156 dprintf("\t\tMPB_ATTRIB_BBM\n");
2157 not_supported
^= MPB_ATTRIB_BBM
;
2159 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
2160 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
2161 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
2163 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
2164 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
2165 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
2167 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
2168 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
2169 not_supported
^= MPB_ATTRIB_2TB_DISK
;
2171 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
2172 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
2173 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
2175 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
2176 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
2177 not_supported
^= MPB_ATTRIB_NEVER_USE
;
2181 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
2189 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
2191 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
2193 struct intel_super
*super
= st
->sb
;
2194 struct imsm_super
*mpb
= super
->anchor
;
2195 char str
[MAX_SIGNATURE_LENGTH
];
2200 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2202 time_t creation_time
;
2204 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
2205 str
[MPB_SIG_LEN
-1] = '\0';
2206 printf(" Magic : %s\n", str
);
2207 printf(" Version : %s\n", get_imsm_version(mpb
));
2208 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
2209 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
2210 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
2211 creation_time
= __le64_to_cpu(mpb
->creation_time
);
2212 printf(" Creation Time : %.24s\n",
2213 creation_time
? ctime(&creation_time
) : "Unknown");
2214 printf(" Attributes : ");
2215 if (imsm_check_attributes(mpb
->attributes
))
2216 printf("All supported\n");
2218 printf("not supported\n");
2219 getinfo_super_imsm(st
, &info
, NULL
);
2220 fname_from_uuid(&info
, nbuf
);
2221 printf(" UUID : %s\n", nbuf
+ 5);
2222 sum
= __le32_to_cpu(mpb
->check_sum
);
2223 printf(" Checksum : %08x %s\n", sum
,
2224 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
2225 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
2226 printf(" Disks : %d\n", mpb
->num_disks
);
2227 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
2228 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
2229 super
->disks
->index
, reserved
, super
->sector_size
);
2230 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
2231 struct bbm_log
*log
= super
->bbm_log
;
2234 printf("Bad Block Management Log:\n");
2235 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
2236 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
2237 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
2239 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2241 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2243 super
->current_vol
= i
;
2244 getinfo_super_imsm(st
, &info
, NULL
);
2245 fname_from_uuid(&info
, nbuf
);
2246 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
2248 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2249 if (i
== super
->disks
->index
)
2251 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
2252 super
->sector_size
);
2255 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2256 if (dl
->index
== -1)
2257 print_imsm_disk(&dl
->disk
, -1, reserved
,
2258 super
->sector_size
);
2260 examine_migr_rec_imsm(super
);
2263 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
2265 /* We just write a generic IMSM ARRAY entry */
2269 getinfo_super_imsm(st
, &info
, NULL
);
2270 fname_from_uuid(&info
, nbuf
);
2271 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
2274 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
2276 /* We just write a generic IMSM ARRAY entry */
2280 struct intel_super
*super
= st
->sb
;
2283 if (!super
->anchor
->num_raid_devs
)
2286 getinfo_super_imsm(st
, &info
, NULL
);
2287 fname_from_uuid(&info
, nbuf
);
2288 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2289 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2291 super
->current_vol
= i
;
2292 getinfo_super_imsm(st
, &info
, NULL
);
2293 fname_from_uuid(&info
, nbuf1
);
2294 printf("ARRAY " DEV_MD_DIR
"%.16s container=%s member=%d UUID=%s\n",
2295 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
2299 static void export_examine_super_imsm(struct supertype
*st
)
2301 struct intel_super
*super
= st
->sb
;
2302 struct imsm_super
*mpb
= super
->anchor
;
2306 getinfo_super_imsm(st
, &info
, NULL
);
2307 fname_from_uuid(&info
, nbuf
);
2308 printf("MD_METADATA=imsm\n");
2309 printf("MD_LEVEL=container\n");
2310 printf("MD_UUID=%s\n", nbuf
+5);
2311 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
2312 printf("MD_CREATION_TIME=%llu\n", __le64_to_cpu(mpb
->creation_time
));
2315 static void detail_super_imsm(struct supertype
*st
, char *homehost
,
2320 struct intel_super
*super
= st
->sb
;
2321 int temp_vol
= super
->current_vol
;
2324 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2326 getinfo_super_imsm(st
, &info
, NULL
);
2327 fname_from_uuid(&info
, nbuf
);
2328 printf("\n UUID : %s\n", nbuf
+ 5);
2330 super
->current_vol
= temp_vol
;
2333 static void brief_detail_super_imsm(struct supertype
*st
, char *subarray
)
2337 struct intel_super
*super
= st
->sb
;
2338 int temp_vol
= super
->current_vol
;
2341 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2343 getinfo_super_imsm(st
, &info
, NULL
);
2344 fname_from_uuid(&info
, nbuf
);
2345 printf(" UUID=%s", nbuf
+ 5);
2347 super
->current_vol
= temp_vol
;
2350 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
,
2351 size_t serial_buf_len
);
2352 static void fd2devname(int fd
, char *name
);
2354 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2356 /* dump an unsorted list of devices attached to AHCI Intel storage
2357 * controller, as well as non-connected ports
2359 int hba_len
= strlen(hba_path
) + 1;
2364 unsigned long port_mask
= (1 << port_count
) - 1;
2366 if (port_count
> (int)sizeof(port_mask
) * 8) {
2368 pr_err("port_count %d out of range\n", port_count
);
2372 /* scroll through /sys/dev/block looking for devices attached to
2375 dir
= opendir("/sys/dev/block");
2379 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2385 char device
[PATH_MAX
];
2390 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2392 path
= devt_to_devpath(makedev(major
, minor
), 1, NULL
);
2395 if (!path_attached_to_hba(path
, hba_path
)) {
2401 /* retrieve the scsi device */
2402 if (!devt_to_devpath(makedev(major
, minor
), 1, device
)) {
2404 pr_err("failed to get device\n");
2408 if (devpath_to_char(device
, "type", buf
, sizeof(buf
), 0)) {
2412 type
= strtoul(buf
, NULL
, 10);
2414 /* if it's not a disk print the vendor and model */
2415 if (!(type
== 0 || type
== 7 || type
== 14)) {
2419 if (devpath_to_char(device
, "vendor", buf
,
2420 sizeof(buf
), 0) == 0) {
2421 strncpy(vendor
, buf
, sizeof(vendor
));
2422 vendor
[sizeof(vendor
) - 1] = '\0';
2423 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2424 while (isspace(*c
) || *c
== '\0')
2429 if (devpath_to_char(device
, "model", buf
,
2430 sizeof(buf
), 0) == 0) {
2431 strncpy(model
, buf
, sizeof(model
));
2432 model
[sizeof(model
) - 1] = '\0';
2433 c
= (char *) &model
[sizeof(model
) - 1];
2434 while (isspace(*c
) || *c
== '\0')
2438 if (vendor
[0] && model
[0])
2439 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2441 switch (type
) { /* numbers from hald/linux/device.c */
2442 case 1: sprintf(buf
, "tape"); break;
2443 case 2: sprintf(buf
, "printer"); break;
2444 case 3: sprintf(buf
, "processor"); break;
2446 case 5: sprintf(buf
, "cdrom"); break;
2447 case 6: sprintf(buf
, "scanner"); break;
2448 case 8: sprintf(buf
, "media_changer"); break;
2449 case 9: sprintf(buf
, "comm"); break;
2450 case 12: sprintf(buf
, "raid"); break;
2451 default: sprintf(buf
, "unknown");
2456 /* chop device path to 'host%d' and calculate the port number */
2457 c
= strchr(&path
[hba_len
], '/');
2460 pr_err("%s - invalid path name\n", path
+ hba_len
);
2465 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2466 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2470 *c
= '/'; /* repair the full string */
2471 pr_err("failed to determine port number for %s\n",
2478 /* mark this port as used */
2479 port_mask
&= ~(1 << port
);
2481 /* print out the device information */
2483 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2487 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2488 if (!is_fd_valid(fd
))
2489 printf(" Port%d : - disk info unavailable -\n", port
);
2491 fd2devname(fd
, buf
);
2492 printf(" Port%d : %s", port
, buf
);
2493 if (imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2495 printf(" (%s)\n", buf
);
2510 for (i
= 0; i
< port_count
; i
++)
2511 if (port_mask
& (1 << i
))
2512 printf(" Port%d : - no device attached -\n", i
);
2518 static int print_nvme_info(struct sys_dev
*hba
)
2523 dir
= opendir("/sys/block/");
2527 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2528 char ns_path
[PATH_MAX
];
2529 char cntrl_path
[PATH_MAX
];
2533 if (!strstr(ent
->d_name
, "nvme"))
2536 fd
= open_dev(ent
->d_name
);
2537 if (!is_fd_valid(fd
))
2540 if (!diskfd_to_devpath(fd
, 0, ns_path
) ||
2541 !diskfd_to_devpath(fd
, 1, cntrl_path
))
2544 if (!path_attached_to_hba(cntrl_path
, hba
->path
))
2547 if (!imsm_is_nvme_namespace_supported(fd
, 0))
2550 fd2devname(fd
, buf
);
2551 if (hba
->type
== SYS_DEV_VMD
)
2552 printf(" NVMe under VMD : %s", buf
);
2553 else if (hba
->type
== SYS_DEV_NVME
)
2554 printf(" NVMe Device : %s", buf
);
2556 if (!imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2558 printf(" (%s)\n", buf
);
2570 static void print_found_intel_controllers(struct sys_dev
*elem
)
2572 for (; elem
; elem
= elem
->next
) {
2573 pr_err("found Intel(R) ");
2574 if (elem
->type
== SYS_DEV_SATA
)
2575 fprintf(stderr
, "SATA ");
2576 else if (elem
->type
== SYS_DEV_SAS
)
2577 fprintf(stderr
, "SAS ");
2578 else if (elem
->type
== SYS_DEV_NVME
)
2579 fprintf(stderr
, "NVMe ");
2581 if (elem
->type
== SYS_DEV_VMD
)
2582 fprintf(stderr
, "VMD domain");
2583 else if (elem
->type
== SYS_DEV_SATA_VMD
)
2584 fprintf(stderr
, "SATA VMD domain");
2586 fprintf(stderr
, "RAID controller");
2589 fprintf(stderr
, " at %s", elem
->pci_id
);
2590 fprintf(stderr
, ".\n");
2595 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2602 if ((dir
= opendir(hba_path
)) == NULL
)
2605 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2608 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2609 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2611 if (*port_count
== 0)
2613 else if (host
< host_base
)
2616 if (host
+ 1 > *port_count
+ host_base
)
2617 *port_count
= host
+ 1 - host_base
;
2623 static void print_imsm_capability(const struct imsm_orom
*orom
)
2625 printf(" Platform : Intel(R) ");
2626 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2627 printf("Matrix Storage Manager\n");
2628 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2629 printf("Virtual RAID on CPU\n");
2631 printf("Rapid Storage Technology%s\n",
2632 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2633 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
) {
2634 if (imsm_orom_is_vmd_without_efi(orom
))
2635 printf(" Version : %d.%d\n", orom
->major_ver
,
2638 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2639 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2641 printf(" RAID Levels :%s%s%s%s%s\n",
2642 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2643 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2644 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2645 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2646 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2647 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2648 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2649 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2650 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2651 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2652 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2653 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2654 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2655 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2656 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2657 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2658 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2659 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2660 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2661 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2662 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2663 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2664 printf(" 2TB volumes :%s supported\n",
2665 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2666 printf(" 2TB disks :%s supported\n",
2667 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2668 printf(" Max Disks : %d\n", orom
->tds
);
2669 printf(" Max Volumes : %d per array, %d per %s\n",
2670 orom
->vpa
, orom
->vphba
,
2671 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2675 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2677 printf("MD_FIRMWARE_TYPE=imsm\n");
2678 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2679 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2680 orom
->hotfix_ver
, orom
->build
);
2681 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2682 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2683 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2684 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2685 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2686 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2687 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2688 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2689 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2690 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2691 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2692 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2693 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2694 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2695 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2696 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2697 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2698 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2699 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2700 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2701 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2702 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2703 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2704 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2705 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2706 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2707 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2708 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2711 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2713 /* There are two components to imsm platform support, the ahci SATA
2714 * controller and the option-rom. To find the SATA controller we
2715 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2716 * controller with the Intel vendor id is present. This approach
2717 * allows mdadm to leverage the kernel's ahci detection logic, with the
2718 * caveat that if ahci.ko is not loaded mdadm will not be able to
2719 * detect platform raid capabilities. The option-rom resides in a
2720 * platform "Adapter ROM". We scan for its signature to retrieve the
2721 * platform capabilities. If raid support is disabled in the BIOS the
2722 * option-rom capability structure will not be available.
2724 struct sys_dev
*list
, *hba
;
2729 if (enumerate_only
) {
2730 if (check_no_platform())
2732 list
= find_intel_devices();
2735 for (hba
= list
; hba
; hba
= hba
->next
) {
2736 if (find_imsm_capability(hba
)) {
2746 list
= find_intel_devices();
2749 pr_err("no active Intel(R) RAID controller found.\n");
2751 } else if (verbose
> 0)
2752 print_found_intel_controllers(list
);
2754 for (hba
= list
; hba
; hba
= hba
->next
) {
2755 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2757 if (!find_imsm_capability(hba
)) {
2759 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2760 hba
->type
== SYS_DEV_VMD
|| hba
->type
== SYS_DEV_SATA_VMD
?
2761 vmd_domain_to_controller(hba
, buf
) :
2762 hba
->path
, get_sys_dev_type(hba
->type
));
2768 if (controller_path
&& result
== 1) {
2769 pr_err("no active Intel(R) RAID controller found under %s\n",
2774 const struct orom_entry
*entry
;
2776 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2777 if (entry
->type
== SYS_DEV_VMD
) {
2778 print_imsm_capability(&entry
->orom
);
2779 printf(" 3rd party NVMe :%s supported\n",
2780 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2781 for (hba
= list
; hba
; hba
= hba
->next
) {
2782 if (hba
->type
== SYS_DEV_VMD
) {
2784 printf(" I/O Controller : %s (%s)\n",
2785 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2786 if (print_nvme_info(hba
)) {
2788 pr_err("failed to get devices attached to VMD domain.\n");
2797 print_imsm_capability(&entry
->orom
);
2798 if (entry
->type
== SYS_DEV_NVME
) {
2799 for (hba
= list
; hba
; hba
= hba
->next
) {
2800 if (hba
->type
== SYS_DEV_NVME
)
2801 print_nvme_info(hba
);
2807 struct devid_list
*devid
;
2808 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2809 hba
= device_by_id(devid
->devid
);
2813 printf(" I/O Controller : %s (%s)\n",
2814 hba
->path
, get_sys_dev_type(hba
->type
));
2815 if (hba
->type
== SYS_DEV_SATA
|| hba
->type
== SYS_DEV_SATA_VMD
) {
2816 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2817 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2819 pr_err("failed to enumerate ports on %s controller at %s.\n",
2820 get_sys_dev_type(hba
->type
), hba
->pci_id
);
2831 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2833 struct sys_dev
*list
, *hba
;
2836 list
= find_intel_devices();
2839 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2844 for (hba
= list
; hba
; hba
= hba
->next
) {
2845 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2847 if (!find_imsm_capability(hba
) && verbose
> 0) {
2849 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2850 hba
->type
== SYS_DEV_VMD
|| hba
->type
== SYS_DEV_SATA_VMD
?
2851 vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2857 const struct orom_entry
*entry
;
2859 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2860 if (entry
->type
== SYS_DEV_VMD
|| entry
->type
== SYS_DEV_SATA_VMD
) {
2861 for (hba
= list
; hba
; hba
= hba
->next
)
2862 print_imsm_capability_export(&entry
->orom
);
2865 print_imsm_capability_export(&entry
->orom
);
2871 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2873 /* the imsm metadata format does not specify any host
2874 * identification information. We return -1 since we can never
2875 * confirm nor deny whether a given array is "meant" for this
2876 * host. We rely on compare_super and the 'family_num' fields to
2877 * exclude member disks that do not belong, and we rely on
2878 * mdadm.conf to specify the arrays that should be assembled.
2879 * Auto-assembly may still pick up "foreign" arrays.
2885 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2887 /* The uuid returned here is used for:
2888 * uuid to put into bitmap file (Create, Grow)
2889 * uuid for backup header when saving critical section (Grow)
2890 * comparing uuids when re-adding a device into an array
2891 * In these cases the uuid required is that of the data-array,
2892 * not the device-set.
2893 * uuid to recognise same set when adding a missing device back
2894 * to an array. This is a uuid for the device-set.
2896 * For each of these we can make do with a truncated
2897 * or hashed uuid rather than the original, as long as
2899 * In each case the uuid required is that of the data-array,
2900 * not the device-set.
2902 /* imsm does not track uuid's so we synthesis one using sha1 on
2903 * - The signature (Which is constant for all imsm array, but no matter)
2904 * - the orig_family_num of the container
2905 * - the index number of the volume
2906 * - the 'serial' number of the volume.
2907 * Hopefully these are all constant.
2909 struct intel_super
*super
= st
->sb
;
2912 struct sha1_ctx ctx
;
2913 struct imsm_dev
*dev
= NULL
;
2916 /* some mdadm versions failed to set ->orig_family_num, in which
2917 * case fall back to ->family_num. orig_family_num will be
2918 * fixed up with the first metadata update.
2920 family_num
= super
->anchor
->orig_family_num
;
2921 if (family_num
== 0)
2922 family_num
= super
->anchor
->family_num
;
2923 sha1_init_ctx(&ctx
);
2924 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2925 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2926 if (super
->current_vol
>= 0)
2927 dev
= get_imsm_dev(super
, super
->current_vol
);
2929 __u32 vol
= super
->current_vol
;
2930 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2931 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2933 sha1_finish_ctx(&ctx
, buf
);
2934 memcpy(uuid
, buf
, 4*4);
2937 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2939 /* migr_strip_size when repairing or initializing parity */
2940 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2941 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2943 switch (get_imsm_raid_level(map
)) {
2948 return 128*1024 >> 9;
2952 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2954 /* migr_strip_size when rebuilding a degraded disk, no idea why
2955 * this is different than migr_strip_size_resync(), but it's good
2958 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2959 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2961 switch (get_imsm_raid_level(map
)) {
2964 if (map
->num_members
% map
->num_domains
== 0)
2965 return 128*1024 >> 9;
2969 return max((__u32
) 64*1024 >> 9, chunk
);
2971 return 128*1024 >> 9;
2975 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2977 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2978 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2979 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2980 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2982 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2985 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2987 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2988 int level
= get_imsm_raid_level(lo
);
2990 if (level
== 1 || level
== 10) {
2991 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2993 return hi
->num_domains
;
2995 return num_stripes_per_unit_resync(dev
);
2998 static unsigned long long calc_component_size(struct imsm_map
*map
,
2999 struct imsm_dev
*dev
)
3001 unsigned long long component_size
;
3002 unsigned long long dev_size
= imsm_dev_size(dev
);
3003 long long calc_dev_size
= 0;
3004 unsigned int member_disks
= imsm_num_data_members(map
);
3006 if (member_disks
== 0)
3009 component_size
= per_dev_array_size(map
);
3010 calc_dev_size
= component_size
* member_disks
;
3012 /* Component size is rounded to 1MB so difference between size from
3013 * metadata and size calculated from num_data_stripes equals up to
3014 * 2048 blocks per each device. If the difference is higher it means
3015 * that array size was expanded and num_data_stripes was not updated.
3017 if (llabs(calc_dev_size
- (long long)dev_size
) >
3018 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
3019 component_size
= dev_size
/ member_disks
;
3020 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
3021 component_size
/ map
->blocks_per_strip
,
3022 num_data_stripes(map
));
3025 return component_size
;
3028 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
3030 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3031 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
3033 switch(get_imsm_raid_level(map
)) {
3036 return chunk
* map
->num_domains
;
3038 return chunk
* map
->num_members
;
3044 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
3046 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
3047 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
3048 __u32 strip
= block
/ chunk
;
3050 switch (get_imsm_raid_level(map
)) {
3053 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
3054 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
3056 return vol_stripe
* chunk
+ block
% chunk
;
3058 __u32 stripe
= strip
/ (map
->num_members
- 1);
3060 return stripe
* chunk
+ block
% chunk
;
3067 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
3068 struct imsm_dev
*dev
)
3070 /* calculate the conversion factor between per member 'blocks'
3071 * (md/{resync,rebuild}_start) and imsm migration units, return
3072 * 0 for the 'not migrating' and 'unsupported migration' cases
3074 if (!dev
->vol
.migr_state
)
3077 switch (migr_type(dev
)) {
3078 case MIGR_GEN_MIGR
: {
3079 struct migr_record
*migr_rec
= super
->migr_rec
;
3080 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
3085 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3086 __u32 stripes_per_unit
;
3087 __u32 blocks_per_unit
;
3096 /* yes, this is really the translation of migr_units to
3097 * per-member blocks in the 'resync' case
3099 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
3100 migr_chunk
= migr_strip_blocks_resync(dev
);
3101 disks
= imsm_num_data_members(map
);
3102 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
3103 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
3104 segment
= blocks_per_unit
/ stripe
;
3105 block_rel
= blocks_per_unit
- segment
* stripe
;
3106 parity_depth
= parity_segment_depth(dev
);
3107 block_map
= map_migr_block(dev
, block_rel
);
3108 return block_map
+ parity_depth
* segment
;
3110 case MIGR_REBUILD
: {
3111 __u32 stripes_per_unit
;
3114 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
3115 migr_chunk
= migr_strip_blocks_rebuild(dev
);
3116 return migr_chunk
* stripes_per_unit
;
3118 case MIGR_STATE_CHANGE
:
3124 static int imsm_level_to_layout(int level
)
3132 return ALGORITHM_LEFT_ASYMMETRIC
;
3139 /*******************************************************************************
3140 * Function: read_imsm_migr_rec
3141 * Description: Function reads imsm migration record from last sector of disk
3143 * fd : disk descriptor
3144 * super : metadata info
3148 ******************************************************************************/
3149 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
3152 unsigned int sector_size
= super
->sector_size
;
3153 unsigned long long dsize
;
3155 get_dev_size(fd
, NULL
, &dsize
);
3156 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
3158 pr_err("Cannot seek to anchor block: %s\n",
3162 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
3163 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3164 MIGR_REC_BUF_SECTORS
*sector_size
) {
3165 pr_err("Cannot read migr record block: %s\n",
3170 if (sector_size
== 4096)
3171 convert_from_4k_imsm_migr_rec(super
);
3177 static struct imsm_dev
*imsm_get_device_during_migration(
3178 struct intel_super
*super
)
3181 struct intel_dev
*dv
;
3183 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
3184 if (is_gen_migration(dv
->dev
))
3190 /*******************************************************************************
3191 * Function: load_imsm_migr_rec
3192 * Description: Function reads imsm migration record (it is stored at the last
3195 * super : imsm internal array info
3199 * -2 : no migration in progress
3200 ******************************************************************************/
3201 static int load_imsm_migr_rec(struct intel_super
*super
)
3207 struct imsm_dev
*dev
;
3208 struct imsm_map
*map
;
3212 /* find map under migration */
3213 dev
= imsm_get_device_during_migration(super
);
3214 /* nothing to load,no migration in progress?
3219 map
= get_imsm_map(dev
, MAP_0
);
3223 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3224 /* skip spare and failed disks
3228 /* read only from one of the first two slots
3230 slot
= get_imsm_disk_slot(map
, dl
->index
);
3231 if (slot
> 1 || slot
< 0)
3234 if (!is_fd_valid(dl
->fd
)) {
3235 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3236 fd
= dev_open(nm
, O_RDONLY
);
3238 if (is_fd_valid(fd
)) {
3248 if (!is_fd_valid(fd
))
3250 retval
= read_imsm_migr_rec(fd
, super
);
3257 /*******************************************************************************
3258 * function: imsm_create_metadata_checkpoint_update
3259 * Description: It creates update for checkpoint change.
3261 * super : imsm internal array info
3262 * u : pointer to prepared update
3265 * If length is equal to 0, input pointer u contains no update
3266 ******************************************************************************/
3267 static int imsm_create_metadata_checkpoint_update(
3268 struct intel_super
*super
,
3269 struct imsm_update_general_migration_checkpoint
**u
)
3272 int update_memory_size
= 0;
3274 dprintf("(enter)\n");
3280 /* size of all update data without anchor */
3281 update_memory_size
=
3282 sizeof(struct imsm_update_general_migration_checkpoint
);
3284 *u
= xcalloc(1, update_memory_size
);
3286 dprintf("error: cannot get memory\n");
3289 (*u
)->type
= update_general_migration_checkpoint
;
3290 (*u
)->curr_migr_unit
= current_migr_unit(super
->migr_rec
);
3291 dprintf("prepared for %llu\n", (unsigned long long)(*u
)->curr_migr_unit
);
3293 return update_memory_size
;
3296 static void imsm_update_metadata_locally(struct supertype
*st
,
3297 void *buf
, int len
);
3299 /*******************************************************************************
3300 * Function: write_imsm_migr_rec
3301 * Description: Function writes imsm migration record
3302 * (at the last sector of disk)
3304 * super : imsm internal array info
3308 ******************************************************************************/
3309 static int write_imsm_migr_rec(struct supertype
*st
)
3311 struct intel_super
*super
= st
->sb
;
3312 unsigned int sector_size
= super
->sector_size
;
3313 unsigned long long dsize
;
3317 struct imsm_update_general_migration_checkpoint
*u
;
3318 struct imsm_dev
*dev
;
3319 struct imsm_map
*map
;
3321 /* find map under migration */
3322 dev
= imsm_get_device_during_migration(super
);
3323 /* if no migration, write buffer anyway to clear migr_record
3324 * on disk based on first available device
3327 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3328 super
->current_vol
);
3330 map
= get_imsm_map(dev
, MAP_0
);
3332 if (sector_size
== 4096)
3333 convert_to_4k_imsm_migr_rec(super
);
3334 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3337 /* skip failed and spare devices */
3340 /* write to 2 first slots only */
3342 slot
= get_imsm_disk_slot(map
, sd
->index
);
3343 if (map
== NULL
|| slot
> 1 || slot
< 0)
3346 get_dev_size(sd
->fd
, NULL
, &dsize
);
3347 if (lseek64(sd
->fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*
3350 pr_err("Cannot seek to anchor block: %s\n",
3354 if ((unsigned int)write(sd
->fd
, super
->migr_rec_buf
,
3355 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3356 MIGR_REC_BUF_SECTORS
*sector_size
) {
3357 pr_err("Cannot write migr record block: %s\n",
3362 if (sector_size
== 4096)
3363 convert_from_4k_imsm_migr_rec(super
);
3364 /* update checkpoint information in metadata */
3365 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3367 dprintf("imsm: Cannot prepare update\n");
3370 /* update metadata locally */
3371 imsm_update_metadata_locally(st
, u
, len
);
3372 /* and possibly remotely */
3373 if (st
->update_tail
) {
3374 append_metadata_update(st
, u
, len
);
3375 /* during reshape we do all work inside metadata handler
3376 * manage_reshape(), so metadata update has to be triggered
3379 flush_metadata_updates(st
);
3380 st
->update_tail
= &st
->updates
;
3389 /* spare/missing disks activations are not allowe when
3390 * array/container performs reshape operation, because
3391 * all arrays in container works on the same disks set
3393 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3396 struct intel_dev
*i_dev
;
3397 struct imsm_dev
*dev
;
3399 /* check whole container
3401 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3403 if (is_gen_migration(dev
)) {
3404 /* No repair during any migration in container
3412 static unsigned long long imsm_component_size_alignment_check(int level
,
3414 unsigned int sector_size
,
3415 unsigned long long component_size
)
3417 unsigned int component_size_alignment
;
3419 /* check component size alignment
3421 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3423 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3424 level
, chunk_size
, component_size
,
3425 component_size_alignment
);
3427 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3428 dprintf("imsm: reported component size aligned from %llu ",
3430 component_size
-= component_size_alignment
;
3431 dprintf_cont("to %llu (%i).\n",
3432 component_size
, component_size_alignment
);
3435 return component_size
;
3438 /*******************************************************************************
3439 * Function: get_bitmap_header_sector
3440 * Description: Returns the sector where the bitmap header is placed.
3442 * st : supertype information
3443 * dev_idx : index of the device with bitmap
3446 * The sector where the bitmap header is placed
3447 ******************************************************************************/
3448 static unsigned long long get_bitmap_header_sector(struct intel_super
*super
,
3451 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3452 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3454 if (!super
->sector_size
) {
3455 dprintf("sector size is not set\n");
3459 return pba_of_lba0(map
) + calc_component_size(map
, dev
) +
3460 (IMSM_BITMAP_HEADER_OFFSET
/ super
->sector_size
);
3463 /*******************************************************************************
3464 * Function: get_bitmap_sector
3465 * Description: Returns the sector where the bitmap is placed.
3467 * st : supertype information
3468 * dev_idx : index of the device with bitmap
3471 * The sector where the bitmap is placed
3472 ******************************************************************************/
3473 static unsigned long long get_bitmap_sector(struct intel_super
*super
,
3476 if (!super
->sector_size
) {
3477 dprintf("sector size is not set\n");
3481 return get_bitmap_header_sector(super
, dev_idx
) +
3482 (IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
);
3485 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3487 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3488 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3490 return pba_of_lba0(map
) +
3491 (num_data_stripes(map
) * map
->blocks_per_strip
);
3494 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3496 struct intel_super
*super
= st
->sb
;
3497 struct migr_record
*migr_rec
= super
->migr_rec
;
3498 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3499 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3500 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3501 struct imsm_map
*map_to_analyse
= map
;
3503 int map_disks
= info
->array
.raid_disks
;
3505 memset(info
, 0, sizeof(*info
));
3507 map_to_analyse
= prev_map
;
3509 dl
= super
->current_disk
;
3511 info
->container_member
= super
->current_vol
;
3512 info
->array
.raid_disks
= map
->num_members
;
3513 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3514 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3515 info
->array
.md_minor
= -1;
3516 info
->array
.ctime
= 0;
3517 info
->array
.utime
= 0;
3518 info
->array
.chunk_size
=
3519 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3520 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3521 info
->custom_array_size
= imsm_dev_size(dev
);
3522 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3524 if (is_gen_migration(dev
)) {
3526 * device prev_map should be added if it is in the middle
3531 info
->reshape_active
= 1;
3532 info
->new_level
= get_imsm_raid_level(map
);
3533 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3534 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3535 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3536 if (info
->delta_disks
) {
3537 /* this needs to be applied to every array
3540 info
->reshape_active
= CONTAINER_RESHAPE
;
3542 /* We shape information that we give to md might have to be
3543 * modify to cope with md's requirement for reshaping arrays.
3544 * For example, when reshaping a RAID0, md requires it to be
3545 * presented as a degraded RAID4.
3546 * Also if a RAID0 is migrating to a RAID5 we need to specify
3547 * the array as already being RAID5, but the 'before' layout
3548 * is a RAID4-like layout.
3550 switch (info
->array
.level
) {
3552 switch(info
->new_level
) {
3554 /* conversion is happening as RAID4 */
3555 info
->array
.level
= 4;
3556 info
->array
.raid_disks
+= 1;
3559 /* conversion is happening as RAID5 */
3560 info
->array
.level
= 5;
3561 info
->array
.layout
= ALGORITHM_PARITY_N
;
3562 info
->delta_disks
-= 1;
3565 /* FIXME error message */
3566 info
->array
.level
= UnSet
;
3572 info
->new_level
= UnSet
;
3573 info
->new_layout
= UnSet
;
3574 info
->new_chunk
= info
->array
.chunk_size
;
3575 info
->delta_disks
= 0;
3579 info
->disk
.major
= dl
->major
;
3580 info
->disk
.minor
= dl
->minor
;
3581 info
->disk
.number
= dl
->index
;
3582 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3586 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3587 info
->component_size
= calc_component_size(map
, dev
);
3588 info
->component_size
= imsm_component_size_alignment_check(
3590 info
->array
.chunk_size
,
3592 info
->component_size
);
3593 info
->bb
.supported
= 1;
3595 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3596 info
->recovery_start
= MaxSector
;
3598 if (info
->array
.level
== 5 &&
3599 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3600 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3601 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3602 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3603 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3604 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3606 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3608 } else if (info
->array
.level
<= 0) {
3609 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3611 if (dev
->rwh_policy
== RWH_BITMAP
) {
3612 info
->bitmap_offset
= get_bitmap_sector(super
, super
->current_vol
);
3613 info
->consistency_policy
= CONSISTENCY_POLICY_BITMAP
;
3615 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3619 info
->reshape_progress
= 0;
3620 info
->resync_start
= MaxSector
;
3621 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3622 !(info
->array
.state
& 1)) &&
3623 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3624 info
->resync_start
= 0;
3626 if (dev
->vol
.migr_state
) {
3627 switch (migr_type(dev
)) {
3630 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3632 __u64 units
= vol_curr_migr_unit(dev
);
3634 info
->resync_start
= blocks_per_unit
* units
;
3637 case MIGR_GEN_MIGR
: {
3638 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3640 __u64 units
= current_migr_unit(migr_rec
);
3643 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3645 (get_num_migr_units(migr_rec
)-1)) &&
3646 (super
->migr_rec
->rec_status
==
3647 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3650 info
->reshape_progress
= blocks_per_unit
* units
;
3652 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3653 (unsigned long long)units
,
3654 (unsigned long long)blocks_per_unit
,
3655 info
->reshape_progress
);
3657 used_disks
= imsm_num_data_members(prev_map
);
3658 if (used_disks
> 0) {
3659 info
->custom_array_size
= per_dev_array_size(map
) *
3664 /* we could emulate the checkpointing of
3665 * 'sync_action=check' migrations, but for now
3666 * we just immediately complete them
3669 /* this is handled by container_content_imsm() */
3670 case MIGR_STATE_CHANGE
:
3671 /* FIXME handle other migrations */
3673 /* we are not dirty, so... */
3674 info
->resync_start
= MaxSector
;
3678 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3679 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3681 info
->array
.major_version
= -1;
3682 info
->array
.minor_version
= -2;
3683 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3684 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3685 uuid_from_super_imsm(st
, info
->uuid
);
3689 for (i
=0; i
<map_disks
; i
++) {
3691 if (i
< info
->array
.raid_disks
) {
3692 struct imsm_disk
*dsk
;
3693 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3694 dsk
= get_imsm_disk(super
, j
);
3695 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3702 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3703 int failed
, int look_in_map
);
3705 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3708 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3710 if (is_gen_migration(dev
)) {
3713 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3715 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3716 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3717 if (map2
->map_state
!= map_state
) {
3718 map2
->map_state
= map_state
;
3719 super
->updates_pending
++;
3724 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3728 for (d
= super
->missing
; d
; d
= d
->next
)
3729 if (d
->index
== index
)
3734 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3736 struct intel_super
*super
= st
->sb
;
3737 struct imsm_disk
*disk
;
3738 int map_disks
= info
->array
.raid_disks
;
3739 int max_enough
= -1;
3741 struct imsm_super
*mpb
;
3743 if (super
->current_vol
>= 0) {
3744 getinfo_super_imsm_volume(st
, info
, map
);
3747 memset(info
, 0, sizeof(*info
));
3749 /* Set raid_disks to zero so that Assemble will always pull in valid
3752 info
->array
.raid_disks
= 0;
3753 info
->array
.level
= LEVEL_CONTAINER
;
3754 info
->array
.layout
= 0;
3755 info
->array
.md_minor
= -1;
3756 info
->array
.ctime
= 0; /* N/A for imsm */
3757 info
->array
.utime
= 0;
3758 info
->array
.chunk_size
= 0;
3760 info
->disk
.major
= 0;
3761 info
->disk
.minor
= 0;
3762 info
->disk
.raid_disk
= -1;
3763 info
->reshape_active
= 0;
3764 info
->array
.major_version
= -1;
3765 info
->array
.minor_version
= -2;
3766 strcpy(info
->text_version
, "imsm");
3767 info
->safe_mode_delay
= 0;
3768 info
->disk
.number
= -1;
3769 info
->disk
.state
= 0;
3771 info
->recovery_start
= MaxSector
;
3772 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3773 info
->bb
.supported
= 1;
3775 /* do we have the all the insync disks that we expect? */
3776 mpb
= super
->anchor
;
3777 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3779 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3780 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3781 int failed
, enough
, j
, missing
= 0;
3782 struct imsm_map
*map
;
3785 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3786 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3787 map
= get_imsm_map(dev
, MAP_0
);
3789 /* any newly missing disks?
3790 * (catches single-degraded vs double-degraded)
3792 for (j
= 0; j
< map
->num_members
; j
++) {
3793 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3794 __u32 idx
= ord_to_idx(ord
);
3796 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3797 info
->disk
.raid_disk
= j
;
3799 if (!(ord
& IMSM_ORD_REBUILD
) &&
3800 get_imsm_missing(super
, idx
)) {
3806 if (state
== IMSM_T_STATE_FAILED
)
3808 else if (state
== IMSM_T_STATE_DEGRADED
&&
3809 (state
!= map
->map_state
|| missing
))
3811 else /* we're normal, or already degraded */
3813 if (is_gen_migration(dev
) && missing
) {
3814 /* during general migration we need all disks
3815 * that process is running on.
3816 * No new missing disk is allowed.
3820 /* no more checks necessary
3824 /* in the missing/failed disk case check to see
3825 * if at least one array is runnable
3827 max_enough
= max(max_enough
, enough
);
3830 info
->container_enough
= max_enough
;
3833 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3835 disk
= &super
->disks
->disk
;
3836 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3837 info
->component_size
= reserved
;
3838 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3839 /* we don't change info->disk.raid_disk here because
3840 * this state will be finalized in mdmon after we have
3841 * found the 'most fresh' version of the metadata
3843 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3844 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3845 0 : (1 << MD_DISK_SYNC
);
3848 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3849 * ->compare_super may have updated the 'num_raid_devs' field for spares
3851 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3852 uuid_from_super_imsm(st
, info
->uuid
);
3854 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3856 /* I don't know how to compute 'map' on imsm, so use safe default */
3859 for (i
= 0; i
< map_disks
; i
++)
3865 /* allocates memory and fills disk in mdinfo structure
3866 * for each disk in array */
3867 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3869 struct mdinfo
*mddev
;
3870 struct intel_super
*super
= st
->sb
;
3871 struct imsm_disk
*disk
;
3874 if (!super
|| !super
->disks
)
3877 mddev
= xcalloc(1, sizeof(*mddev
));
3881 tmp
= xcalloc(1, sizeof(*tmp
));
3883 tmp
->next
= mddev
->devs
;
3885 tmp
->disk
.number
= count
++;
3886 tmp
->disk
.major
= dl
->major
;
3887 tmp
->disk
.minor
= dl
->minor
;
3888 tmp
->disk
.state
= is_configured(disk
) ?
3889 (1 << MD_DISK_ACTIVE
) : 0;
3890 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3891 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3892 tmp
->disk
.raid_disk
= -1;
3898 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3899 enum update_opt update
, char *devname
,
3900 int verbose
, int uuid_set
, char *homehost
)
3902 /* For 'assemble' and 'force' we need to return non-zero if any
3903 * change was made. For others, the return value is ignored.
3904 * Update options are:
3905 * force-one : This device looks a bit old but needs to be included,
3906 * update age info appropriately.
3907 * assemble: clear any 'faulty' flag to allow this device to
3909 * force-array: Array is degraded but being forced, mark it clean
3910 * if that will be needed to assemble it.
3912 * newdev: not used ????
3913 * grow: Array has gained a new device - this is currently for
3915 * resync: mark as dirty so a resync will happen.
3916 * name: update the name - preserving the homehost
3917 * uuid: Change the uuid of the array to match watch is given
3919 * Following are not relevant for this imsm:
3920 * sparc2.2 : update from old dodgey metadata
3921 * super-minor: change the preferred_minor number
3922 * summaries: update redundant counters.
3923 * homehost: update the recorded homehost
3924 * _reshape_progress: record new reshape_progress position.
3927 struct intel_super
*super
= st
->sb
;
3928 struct imsm_super
*mpb
;
3930 /* we can only update container info */
3931 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3934 mpb
= super
->anchor
;
3938 /* We take this to mean that the family_num should be updated.
3939 * However that is much smaller than the uuid so we cannot really
3940 * allow an explicit uuid to be given. And it is hard to reliably
3942 * So if !uuid_set we know the current uuid is random and just used
3943 * the first 'int' and copy it to the other 3 positions.
3944 * Otherwise we require the 4 'int's to be the same as would be the
3945 * case if we are using a random uuid. So an explicit uuid will be
3946 * accepted as long as all for ints are the same... which shouldn't hurt
3949 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3952 if (info
->uuid
[0] != info
->uuid
[1] ||
3953 info
->uuid
[1] != info
->uuid
[2] ||
3954 info
->uuid
[2] != info
->uuid
[3])
3960 mpb
->orig_family_num
= info
->uuid
[0];
3962 case UOPT_SPEC_ASSEMBLE
:
3970 /* successful update? recompute checksum */
3972 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3977 static size_t disks_to_mpb_size(int disks
)
3981 size
= sizeof(struct imsm_super
);
3982 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3983 size
+= 2 * sizeof(struct imsm_dev
);
3984 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3985 size
+= (4 - 2) * sizeof(struct imsm_map
);
3986 /* 4 possible disk_ord_tbl's */
3987 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3988 /* maximum bbm log */
3989 size
+= sizeof(struct bbm_log
);
3994 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3995 unsigned long long data_offset
)
3997 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
4000 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
4003 static void free_devlist(struct intel_super
*super
)
4005 struct intel_dev
*dv
;
4007 while (super
->devlist
) {
4008 dv
= super
->devlist
->next
;
4009 free(super
->devlist
->dev
);
4010 free(super
->devlist
);
4011 super
->devlist
= dv
;
4015 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
4017 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
4020 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
,
4024 * 0 same, or first was empty, and second was copied
4025 * 1 sb are different
4027 struct intel_super
*first
= st
->sb
;
4028 struct intel_super
*sec
= tst
->sb
;
4036 /* in platform dependent environment test if the disks
4037 * use the same Intel hba
4038 * if not on Intel hba at all, allow anything.
4039 * doesn't check HBAs if num_raid_devs is not set, as it means
4040 * it is a free floating spare, and all spares regardless of HBA type
4041 * will fall into separate container during the assembly
4043 if (first
->hba
&& sec
->hba
&& first
->anchor
->num_raid_devs
!= 0) {
4044 if (first
->hba
->type
!= sec
->hba
->type
) {
4046 pr_err("HBAs of devices do not match %s != %s\n",
4047 get_sys_dev_type(first
->hba
->type
),
4048 get_sys_dev_type(sec
->hba
->type
));
4051 if (first
->orom
!= sec
->orom
) {
4053 pr_err("HBAs of devices do not match %s != %s\n",
4054 first
->hba
->pci_id
, sec
->hba
->pci_id
);
4059 if (first
->anchor
->num_raid_devs
> 0 &&
4060 sec
->anchor
->num_raid_devs
> 0) {
4061 /* Determine if these disks might ever have been
4062 * related. Further disambiguation can only take place
4063 * in load_super_imsm_all
4065 __u32 first_family
= first
->anchor
->orig_family_num
;
4066 __u32 sec_family
= sec
->anchor
->orig_family_num
;
4068 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
4069 MAX_SIGNATURE_LENGTH
) != 0)
4072 if (first_family
== 0)
4073 first_family
= first
->anchor
->family_num
;
4074 if (sec_family
== 0)
4075 sec_family
= sec
->anchor
->family_num
;
4077 if (first_family
!= sec_family
)
4082 /* if an anchor does not have num_raid_devs set then it is a free
4083 * floating spare. don't assosiate spare with any array, as during assembly
4084 * spares shall fall into separate container, from which they can be moved
4087 if (first
->anchor
->num_raid_devs
^ sec
->anchor
->num_raid_devs
)
4093 static void fd2devname(int fd
, char *name
)
4101 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
4104 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
4106 char path
[PATH_MAX
];
4107 char *name
= fd2kname(fd
);
4112 if (strncmp(name
, "nvme", 4) != 0)
4115 if (!diskfd_to_devpath(fd
, 1, path
))
4118 return devpath_to_char(path
, "serial", buf
, buf_len
, 0);
4121 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
4123 static int imsm_read_serial(int fd
, char *devname
,
4124 __u8
*serial
, size_t serial_buf_len
)
4133 memset(buf
, 0, sizeof(buf
));
4135 if (check_env("IMSM_DEVNAME_AS_SERIAL")) {
4136 memset(serial
, 0, serial_buf_len
);
4137 fd2devname(fd
, (char *) serial
);
4141 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
4144 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
4148 pr_err("Failed to retrieve serial for %s\n",
4153 /* trim all whitespace and non-printable characters and convert
4156 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4159 /* ':' is reserved for use in placeholder serial
4160 * numbers for missing disks
4171 if (len
> serial_buf_len
) {
4172 /* truncate leading characters */
4173 dest
+= len
- serial_buf_len
;
4174 len
= serial_buf_len
;
4177 memset(serial
, 0, serial_buf_len
);
4178 memcpy(serial
, dest
, len
);
4183 static int serialcmp(__u8
*s1
, __u8
*s2
)
4185 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4188 static void serialcpy(__u8
*dest
, __u8
*src
)
4190 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4193 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4197 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4198 if (serialcmp(dl
->serial
, serial
) == 0)
4204 static struct imsm_disk
*
4205 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4209 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4210 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4212 if (serialcmp(disk
->serial
, serial
) == 0) {
4223 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4225 struct imsm_disk
*disk
;
4230 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4232 rv
= imsm_read_serial(fd
, devname
, serial
, MAX_RAID_SERIAL_LEN
);
4237 dl
= xcalloc(1, sizeof(*dl
));
4240 dl
->major
= major(stb
.st_rdev
);
4241 dl
->minor
= minor(stb
.st_rdev
);
4242 dl
->next
= super
->disks
;
4243 dl
->fd
= keep_fd
? fd
: -1;
4244 assert(super
->disks
== NULL
);
4246 serialcpy(dl
->serial
, serial
);
4249 fd2devname(fd
, name
);
4251 dl
->devname
= xstrdup(devname
);
4253 dl
->devname
= xstrdup(name
);
4255 /* look up this disk's index in the current anchor */
4256 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4259 /* only set index on disks that are a member of a
4260 * populated contianer, i.e. one with raid_devs
4262 if (is_failed(&dl
->disk
))
4264 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4271 /* When migrating map0 contains the 'destination' state while map1
4272 * contains the current state. When not migrating map0 contains the
4273 * current state. This routine assumes that map[0].map_state is set to
4274 * the current array state before being called.
4276 * Migration is indicated by one of the following states
4277 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4278 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4279 * map1state=unitialized)
4280 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4282 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4283 * map1state=degraded)
4284 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4287 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4288 __u8 to_state
, int migr_type
)
4290 struct imsm_map
*dest
;
4291 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4293 dev
->vol
.migr_state
= 1;
4294 set_migr_type(dev
, migr_type
);
4295 set_vol_curr_migr_unit(dev
, 0);
4296 dest
= get_imsm_map(dev
, MAP_1
);
4298 /* duplicate and then set the target end state in map[0] */
4299 memcpy(dest
, src
, sizeof_imsm_map(src
));
4300 if (migr_type
== MIGR_GEN_MIGR
) {
4304 for (i
= 0; i
< src
->num_members
; i
++) {
4305 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4306 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4310 if (migr_type
== MIGR_GEN_MIGR
)
4311 /* Clear migration record */
4312 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4314 src
->map_state
= to_state
;
4317 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4320 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4321 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4325 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4326 * completed in the last migration.
4328 * FIXME add support for raid-level-migration
4330 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == false) &&
4331 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4332 /* when final map state is other than expected
4333 * merge maps (not for migration)
4337 for (i
= 0; i
< prev
->num_members
; i
++)
4338 for (j
= 0; j
< map
->num_members
; j
++)
4339 /* during online capacity expansion
4340 * disks position can be changed
4341 * if takeover is used
4343 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4344 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4345 map
->disk_ord_tbl
[j
] |=
4346 prev
->disk_ord_tbl
[i
];
4349 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4350 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4353 dev
->vol
.migr_state
= 0;
4354 set_migr_type(dev
, 0);
4355 set_vol_curr_migr_unit(dev
, 0);
4356 map
->map_state
= map_state
;
4359 static int parse_raid_devices(struct intel_super
*super
)
4362 struct imsm_dev
*dev_new
;
4363 size_t len
, len_migr
;
4365 size_t space_needed
= 0;
4366 struct imsm_super
*mpb
= super
->anchor
;
4368 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4369 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4370 struct intel_dev
*dv
;
4372 len
= sizeof_imsm_dev(dev_iter
, 0);
4373 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4375 space_needed
+= len_migr
- len
;
4377 dv
= xmalloc(sizeof(*dv
));
4378 if (max_len
< len_migr
)
4380 if (max_len
> len_migr
)
4381 space_needed
+= max_len
- len_migr
;
4382 dev_new
= xmalloc(max_len
);
4383 imsm_copy_dev(dev_new
, dev_iter
);
4386 dv
->next
= super
->devlist
;
4387 super
->devlist
= dv
;
4390 /* ensure that super->buf is large enough when all raid devices
4393 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4396 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4397 super
->sector_size
);
4398 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4401 memcpy(buf
, super
->buf
, super
->len
);
4402 memset(buf
+ super
->len
, 0, len
- super
->len
);
4408 super
->extra_space
+= space_needed
;
4413 /*******************************************************************************
4414 * Function: check_mpb_migr_compatibility
4415 * Description: Function checks for unsupported migration features:
4416 * - migration optimization area (pba_of_lba0)
4417 * - descending reshape (ascending_migr)
4419 * super : imsm metadata information
4421 * 0 : migration is compatible
4422 * -1 : migration is not compatible
4423 ******************************************************************************/
4424 int check_mpb_migr_compatibility(struct intel_super
*super
)
4426 struct imsm_map
*map0
, *map1
;
4427 struct migr_record
*migr_rec
= super
->migr_rec
;
4430 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4431 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4433 if (dev_iter
->vol
.migr_state
== 1 &&
4434 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4435 /* This device is migrating */
4436 map0
= get_imsm_map(dev_iter
, MAP_0
);
4437 map1
= get_imsm_map(dev_iter
, MAP_1
);
4438 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4439 /* migration optimization area was used */
4441 if (migr_rec
->ascending_migr
== 0 &&
4442 migr_rec
->dest_depth_per_unit
> 0)
4443 /* descending reshape not supported yet */
4450 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4452 /* load_imsm_mpb - read matrix metadata
4453 * allocates super->mpb to be freed by free_imsm
4455 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4457 unsigned long long dsize
;
4458 unsigned long long sectors
;
4459 unsigned int sector_size
= super
->sector_size
;
4461 struct imsm_super
*anchor
;
4464 get_dev_size(fd
, NULL
, &dsize
);
4465 if (dsize
< 2*sector_size
) {
4467 pr_err("%s: device to small for imsm\n",
4472 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4474 pr_err("Cannot seek to anchor block on %s: %s\n",
4475 devname
, strerror(errno
));
4479 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4481 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4484 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4486 pr_err("Cannot read anchor block on %s: %s\n",
4487 devname
, strerror(errno
));
4492 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4494 pr_err("no IMSM anchor on %s\n", devname
);
4499 __free_imsm(super
, 0);
4500 /* reload capability and hba */
4502 /* capability and hba must be updated with new super allocation */
4503 find_intel_hba_capability(fd
, super
, devname
);
4504 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4505 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4507 pr_err("unable to allocate %zu byte mpb buffer\n",
4512 memcpy(super
->buf
, anchor
, sector_size
);
4514 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4517 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4518 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4519 pr_err("could not allocate migr_rec buffer\n");
4524 super
->clean_migration_record_by_mdmon
= 0;
4527 check_sum
= __gen_imsm_checksum(super
->anchor
);
4528 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4530 pr_err("IMSM checksum %x != %x on %s\n",
4532 __le32_to_cpu(super
->anchor
->check_sum
),
4540 /* read the extended mpb */
4541 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4543 pr_err("Cannot seek to extended mpb on %s: %s\n",
4544 devname
, strerror(errno
));
4548 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4549 super
->len
- sector_size
) != super
->len
- sector_size
) {
4551 pr_err("Cannot read extended mpb on %s: %s\n",
4552 devname
, strerror(errno
));
4556 check_sum
= __gen_imsm_checksum(super
->anchor
);
4557 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4559 pr_err("IMSM checksum %x != %x on %s\n",
4560 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4568 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4570 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4571 static void clear_hi(struct intel_super
*super
)
4573 struct imsm_super
*mpb
= super
->anchor
;
4575 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4577 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4578 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4579 disk
->total_blocks_hi
= 0;
4581 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4582 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4583 for (n
= 0; n
< 2; ++n
) {
4584 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4587 map
->pba_of_lba0_hi
= 0;
4588 map
->blocks_per_member_hi
= 0;
4589 map
->num_data_stripes_hi
= 0;
4595 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4599 err
= load_imsm_mpb(fd
, super
, devname
);
4602 if (super
->sector_size
== 4096)
4603 convert_from_4k(super
);
4604 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4607 err
= parse_raid_devices(super
);
4610 err
= load_bbm_log(super
);
4615 static void __free_imsm_disk(struct dl
*d
, int do_close
)
4627 static void free_imsm_disks(struct intel_super
*super
)
4631 while (super
->disks
) {
4633 super
->disks
= d
->next
;
4634 __free_imsm_disk(d
, 1);
4636 while (super
->disk_mgmt_list
) {
4637 d
= super
->disk_mgmt_list
;
4638 super
->disk_mgmt_list
= d
->next
;
4639 __free_imsm_disk(d
, 1);
4641 while (super
->missing
) {
4643 super
->missing
= d
->next
;
4644 __free_imsm_disk(d
, 1);
4649 /* free all the pieces hanging off of a super pointer */
4650 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4652 struct intel_hba
*elem
, *next
;
4658 /* unlink capability description */
4660 if (super
->migr_rec_buf
) {
4661 free(super
->migr_rec_buf
);
4662 super
->migr_rec_buf
= NULL
;
4665 free_imsm_disks(super
);
4666 free_devlist(super
);
4670 free((void *)elem
->path
);
4676 free(super
->bbm_log
);
4680 static void free_imsm(struct intel_super
*super
)
4682 __free_imsm(super
, 1);
4683 free(super
->bb
.entries
);
4687 static void free_super_imsm(struct supertype
*st
)
4689 struct intel_super
*super
= st
->sb
;
4698 static struct intel_super
*alloc_super(void)
4700 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4702 super
->current_vol
= -1;
4703 super
->create_offset
= ~((unsigned long long) 0);
4705 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4706 sizeof(struct md_bb_entry
));
4707 if (!super
->bb
.entries
) {
4716 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4718 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4720 struct sys_dev
*hba_name
;
4723 if (is_fd_valid(fd
) && test_partition(fd
)) {
4724 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4728 if (!is_fd_valid(fd
) || check_no_platform()) {
4733 hba_name
= find_disk_attached_hba(fd
, NULL
);
4736 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4740 rv
= attach_hba_to_super(super
, hba_name
);
4743 struct intel_hba
*hba
= super
->hba
;
4745 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4746 " but the container is assigned to Intel(R) %s %s (",
4748 get_sys_dev_type(hba_name
->type
),
4749 hba_name
->type
== SYS_DEV_VMD
|| hba_name
->type
== SYS_DEV_SATA_VMD
?
4750 "domain" : "RAID controller",
4751 hba_name
->pci_id
? : "Err!",
4752 get_sys_dev_type(super
->hba
->type
),
4753 hba
->type
== SYS_DEV_VMD
|| hba_name
->type
== SYS_DEV_SATA_VMD
?
4754 "domain" : "RAID controller");
4757 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4759 fprintf(stderr
, ", ");
4762 fprintf(stderr
, ").\n"
4763 " Mixing devices attached to different controllers is not allowed.\n");
4767 super
->orom
= find_imsm_capability(hba_name
);
4774 /* find_missing - helper routine for load_super_imsm_all that identifies
4775 * disks that have disappeared from the system. This routine relies on
4776 * the mpb being uptodate, which it is at load time.
4778 static int find_missing(struct intel_super
*super
)
4781 struct imsm_super
*mpb
= super
->anchor
;
4783 struct imsm_disk
*disk
;
4785 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4786 disk
= __get_imsm_disk(mpb
, i
);
4787 dl
= serial_to_dl(disk
->serial
, super
);
4791 dl
= xmalloc(sizeof(*dl
));
4795 dl
->devname
= xstrdup("missing");
4797 serialcpy(dl
->serial
, disk
->serial
);
4800 dl
->next
= super
->missing
;
4801 super
->missing
= dl
;
4807 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4809 struct intel_disk
*idisk
= disk_list
;
4812 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4814 idisk
= idisk
->next
;
4820 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4821 struct intel_super
*super
,
4822 struct intel_disk
**disk_list
)
4824 struct imsm_disk
*d
= &super
->disks
->disk
;
4825 struct imsm_super
*mpb
= super
->anchor
;
4828 for (i
= 0; i
< tbl_size
; i
++) {
4829 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4830 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4832 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4833 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4834 dprintf("mpb from %d:%d matches %d:%d\n",
4835 super
->disks
->major
,
4836 super
->disks
->minor
,
4837 table
[i
]->disks
->major
,
4838 table
[i
]->disks
->minor
);
4842 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4843 is_configured(d
) == is_configured(tbl_d
)) &&
4844 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4845 /* current version of the mpb is a
4846 * better candidate than the one in
4847 * super_table, but copy over "cross
4848 * generational" status
4850 struct intel_disk
*idisk
;
4852 dprintf("mpb from %d:%d replaces %d:%d\n",
4853 super
->disks
->major
,
4854 super
->disks
->minor
,
4855 table
[i
]->disks
->major
,
4856 table
[i
]->disks
->minor
);
4858 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4859 if (idisk
&& is_failed(&idisk
->disk
))
4860 tbl_d
->status
|= FAILED_DISK
;
4863 struct intel_disk
*idisk
;
4864 struct imsm_disk
*disk
;
4866 /* tbl_mpb is more up to date, but copy
4867 * over cross generational status before
4870 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4871 if (disk
&& is_failed(disk
))
4872 d
->status
|= FAILED_DISK
;
4874 idisk
= disk_list_get(d
->serial
, *disk_list
);
4877 if (disk
&& is_configured(disk
))
4878 idisk
->disk
.status
|= CONFIGURED_DISK
;
4881 dprintf("mpb from %d:%d prefer %d:%d\n",
4882 super
->disks
->major
,
4883 super
->disks
->minor
,
4884 table
[i
]->disks
->major
,
4885 table
[i
]->disks
->minor
);
4893 table
[tbl_size
++] = super
;
4897 /* update/extend the merged list of imsm_disk records */
4898 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4899 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4900 struct intel_disk
*idisk
;
4902 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4904 idisk
->disk
.status
|= disk
->status
;
4905 if (is_configured(&idisk
->disk
) ||
4906 is_failed(&idisk
->disk
))
4907 idisk
->disk
.status
&= ~(SPARE_DISK
);
4909 idisk
= xcalloc(1, sizeof(*idisk
));
4910 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4911 idisk
->disk
= *disk
;
4912 idisk
->next
= *disk_list
;
4916 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4923 static struct intel_super
*
4924 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4927 struct imsm_super
*mpb
= super
->anchor
;
4931 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4932 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4933 struct intel_disk
*idisk
;
4935 idisk
= disk_list_get(disk
->serial
, disk_list
);
4937 if (idisk
->owner
== owner
||
4938 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4941 dprintf("'%.16s' owner %d != %d\n",
4942 disk
->serial
, idisk
->owner
,
4945 dprintf("unknown disk %x [%d]: %.16s\n",
4946 __le32_to_cpu(mpb
->family_num
), i
,
4952 if (ok_count
== mpb
->num_disks
)
4957 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4959 struct intel_super
*s
;
4961 for (s
= super_list
; s
; s
= s
->next
) {
4962 if (family_num
!= s
->anchor
->family_num
)
4964 pr_err("Conflict, offlining family %#x on '%s'\n",
4965 __le32_to_cpu(family_num
), s
->disks
->devname
);
4969 static struct intel_super
*
4970 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4972 struct intel_super
*super_table
[len
];
4973 struct intel_disk
*disk_list
= NULL
;
4974 struct intel_super
*champion
, *spare
;
4975 struct intel_super
*s
, **del
;
4980 memset(super_table
, 0, sizeof(super_table
));
4981 for (s
= *super_list
; s
; s
= s
->next
)
4982 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4984 for (i
= 0; i
< tbl_size
; i
++) {
4985 struct imsm_disk
*d
;
4986 struct intel_disk
*idisk
;
4987 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4990 d
= &s
->disks
->disk
;
4992 /* 'd' must appear in merged disk list for its
4993 * configuration to be valid
4995 idisk
= disk_list_get(d
->serial
, disk_list
);
4996 if (idisk
&& idisk
->owner
== i
)
4997 s
= validate_members(s
, disk_list
, i
);
5002 dprintf("marking family: %#x from %d:%d offline\n",
5004 super_table
[i
]->disks
->major
,
5005 super_table
[i
]->disks
->minor
);
5009 /* This is where the mdadm implementation differs from the Windows
5010 * driver which has no strict concept of a container. We can only
5011 * assemble one family from a container, so when returning a prodigal
5012 * array member to this system the code will not be able to disambiguate
5013 * the container contents that should be assembled ("foreign" versus
5014 * "local"). It requires user intervention to set the orig_family_num
5015 * to a new value to establish a new container. The Windows driver in
5016 * this situation fixes up the volume name in place and manages the
5017 * foreign array as an independent entity.
5022 for (i
= 0; i
< tbl_size
; i
++) {
5023 struct intel_super
*tbl_ent
= super_table
[i
];
5029 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
5034 if (s
&& !is_spare
) {
5035 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
5037 } else if (!s
&& !is_spare
)
5050 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
5051 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
5053 /* collect all dl's onto 'champion', and update them to
5054 * champion's version of the status
5056 for (s
= *super_list
; s
; s
= s
->next
) {
5057 struct imsm_super
*mpb
= champion
->anchor
;
5058 struct dl
*dl
= s
->disks
;
5063 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
5065 for (i
= 0; i
< mpb
->num_disks
; i
++) {
5066 struct imsm_disk
*disk
;
5068 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
5071 /* only set index on disks that are a member of
5072 * a populated contianer, i.e. one with
5075 if (is_failed(&dl
->disk
))
5077 else if (is_spare(&dl
->disk
))
5083 if (i
>= mpb
->num_disks
) {
5084 struct intel_disk
*idisk
;
5086 idisk
= disk_list_get(dl
->serial
, disk_list
);
5087 if (idisk
&& is_spare(&idisk
->disk
) &&
5088 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
5096 dl
->next
= champion
->disks
;
5097 champion
->disks
= dl
;
5101 /* delete 'champion' from super_list */
5102 for (del
= super_list
; *del
; ) {
5103 if (*del
== champion
) {
5104 *del
= (*del
)->next
;
5107 del
= &(*del
)->next
;
5109 champion
->next
= NULL
;
5113 struct intel_disk
*idisk
= disk_list
;
5115 disk_list
= disk_list
->next
;
5123 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
5124 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5125 int major
, int minor
, int keep_fd
);
5127 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5128 int *max
, int keep_fd
);
5130 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
5131 char *devname
, struct md_list
*devlist
,
5134 struct intel_super
*super_list
= NULL
;
5135 struct intel_super
*super
= NULL
;
5139 if (is_fd_valid(fd
))
5140 /* 'fd' is an opened container */
5141 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
5143 /* get super block from devlist devices */
5144 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
5147 /* all mpbs enter, maybe one leaves */
5148 super
= imsm_thunderdome(&super_list
, i
);
5154 if (find_missing(super
) != 0) {
5160 /* load migration record */
5161 err
= load_imsm_migr_rec(super
);
5163 /* migration is in progress,
5164 * but migr_rec cannot be loaded,
5170 /* Check migration compatibility */
5171 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5172 pr_err("Unsupported migration detected");
5174 fprintf(stderr
, " on %s\n", devname
);
5176 fprintf(stderr
, " (IMSM).\n");
5185 while (super_list
) {
5186 struct intel_super
*s
= super_list
;
5188 super_list
= super_list
->next
;
5196 if (is_fd_valid(fd
))
5197 strcpy(st
->container_devnm
, fd2devnm(fd
));
5199 st
->container_devnm
[0] = 0;
5200 if (err
== 0 && st
->ss
== NULL
) {
5201 st
->ss
= &super_imsm
;
5202 st
->minor_version
= 0;
5203 st
->max_devs
= IMSM_MAX_DEVICES
;
5209 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5210 int *max
, int keep_fd
)
5212 struct md_list
*tmpdev
;
5216 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5217 if (tmpdev
->used
!= 1)
5219 if (tmpdev
->container
== 1) {
5221 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5222 if (!is_fd_valid(fd
)) {
5223 pr_err("cannot open device %s: %s\n",
5224 tmpdev
->devname
, strerror(errno
));
5228 err
= get_sra_super_block(fd
, super_list
,
5229 tmpdev
->devname
, &lmax
,
5238 int major
= major(tmpdev
->st_rdev
);
5239 int minor
= minor(tmpdev
->st_rdev
);
5240 err
= get_super_block(super_list
,
5257 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5258 int major
, int minor
, int keep_fd
)
5260 struct intel_super
*s
;
5272 sprintf(nm
, "%d:%d", major
, minor
);
5273 dfd
= dev_open(nm
, O_RDWR
);
5274 if (!is_fd_valid(dfd
)) {
5279 if (!get_dev_sector_size(dfd
, NULL
, &s
->sector_size
)) {
5283 find_intel_hba_capability(dfd
, s
, devname
);
5284 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5286 /* retry the load if we might have raced against mdmon */
5287 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5288 for (retry
= 0; retry
< 3; retry
++) {
5289 sleep_for(0, MSEC_TO_NSEC(3), true);
5290 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5296 s
->next
= *super_list
;
5310 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5317 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5321 if (sra
->array
.major_version
!= -1 ||
5322 sra
->array
.minor_version
!= -2 ||
5323 strcmp(sra
->text_version
, "imsm") != 0) {
5328 devnm
= fd2devnm(fd
);
5329 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5330 if (get_super_block(super_list
, devnm
, devname
,
5331 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5342 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5344 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5347 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5349 struct intel_super
*super
;
5353 if (test_partition(fd
))
5354 /* IMSM not allowed on partitions */
5357 free_super_imsm(st
);
5359 super
= alloc_super();
5363 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
5367 /* Load hba and capabilities if they exist.
5368 * But do not preclude loading metadata in case capabilities or hba are
5369 * non-compliant and ignore_hw_compat is set.
5371 rv
= find_intel_hba_capability(fd
, super
, devname
);
5372 /* no orom/efi or non-intel hba of the disk */
5373 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5375 pr_err("No OROM/EFI properties for %s\n", devname
);
5379 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5381 /* retry the load if we might have raced against mdmon */
5383 struct mdstat_ent
*mdstat
= NULL
;
5384 char *name
= fd2kname(fd
);
5387 mdstat
= mdstat_by_component(name
);
5389 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5390 for (retry
= 0; retry
< 3; retry
++) {
5391 sleep_for(0, MSEC_TO_NSEC(3), true);
5392 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5398 free_mdstat(mdstat
);
5403 pr_err("Failed to load all information sections on %s\n", devname
);
5409 if (st
->ss
== NULL
) {
5410 st
->ss
= &super_imsm
;
5411 st
->minor_version
= 0;
5412 st
->max_devs
= IMSM_MAX_DEVICES
;
5415 /* load migration record */
5416 if (load_imsm_migr_rec(super
) == 0) {
5417 /* Check for unsupported migration features */
5418 if (check_mpb_migr_compatibility(super
) != 0) {
5419 pr_err("Unsupported migration detected");
5421 fprintf(stderr
, " on %s\n", devname
);
5423 fprintf(stderr
, " (IMSM).\n");
5431 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5433 if (info
->level
== 1)
5435 return info
->chunk_size
>> 9;
5438 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5439 unsigned long long size
)
5441 if (info
->level
== 1)
5444 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5447 static void imsm_update_version_info(struct intel_super
*super
)
5449 /* update the version and attributes */
5450 struct imsm_super
*mpb
= super
->anchor
;
5452 struct imsm_dev
*dev
;
5453 struct imsm_map
*map
;
5456 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5457 dev
= get_imsm_dev(super
, i
);
5458 map
= get_imsm_map(dev
, MAP_0
);
5459 if (__le32_to_cpu(dev
->size_high
) > 0)
5460 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5462 /* FIXME detect when an array spans a port multiplier */
5464 mpb
->attributes
|= MPB_ATTRIB_PM
;
5467 if (mpb
->num_raid_devs
> 1 ||
5468 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5469 version
= MPB_VERSION_ATTRIBS
;
5470 switch (get_imsm_raid_level(map
)) {
5471 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5472 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5473 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5474 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5477 if (map
->num_members
>= 5)
5478 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5479 else if (dev
->status
== DEV_CLONE_N_GO
)
5480 version
= MPB_VERSION_CNG
;
5481 else if (get_imsm_raid_level(map
) == 5)
5482 version
= MPB_VERSION_RAID5
;
5483 else if (map
->num_members
>= 3)
5484 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5485 else if (get_imsm_raid_level(map
) == 1)
5486 version
= MPB_VERSION_RAID1
;
5488 version
= MPB_VERSION_RAID0
;
5490 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5495 * imsm_check_name() - check imsm naming criteria.
5496 * @super: &intel_super pointer, not NULL.
5497 * @name: name to check.
5498 * @verbose: verbose level.
5500 * Name must be no longer than &MAX_RAID_SERIAL_LEN and must be unique across volumes.
5502 * Returns: &true if @name matches, &false otherwise.
5504 static bool imsm_is_name_allowed(struct intel_super
*super
, const char * const name
,
5507 struct imsm_super
*mpb
= super
->anchor
;
5510 if (is_string_lq(name
, MAX_RAID_SERIAL_LEN
+ 1) == false) {
5511 pr_vrb("imsm: Name \"%s\" is too long\n", name
);
5515 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5516 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5518 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5519 pr_vrb("imsm: Name \"%s\" already exists\n", name
);
5527 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5528 struct shape
*s
, char *name
,
5529 char *homehost
, int *uuid
,
5530 long long data_offset
)
5532 /* We are creating a volume inside a pre-existing container.
5533 * so st->sb is already set.
5535 struct intel_super
*super
= st
->sb
;
5536 unsigned int sector_size
= super
->sector_size
;
5537 struct imsm_super
*mpb
= super
->anchor
;
5538 struct intel_dev
*dv
;
5539 struct imsm_dev
*dev
;
5540 struct imsm_vol
*vol
;
5541 struct imsm_map
*map
;
5542 int idx
= mpb
->num_raid_devs
;
5545 unsigned long long array_blocks
;
5546 size_t size_old
, size_new
;
5547 unsigned int data_disks
;
5548 unsigned long long size_per_member
;
5550 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5551 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5555 /* ensure the mpb is large enough for the new data */
5556 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5557 size_new
= disks_to_mpb_size(info
->nr_disks
);
5558 if (size_new
> size_old
) {
5560 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5562 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5563 pr_err("could not allocate new mpb\n");
5566 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5567 MIGR_REC_BUF_SECTORS
*
5568 MAX_SECTOR_SIZE
) != 0) {
5569 pr_err("could not allocate migr_rec buffer\n");
5575 memcpy(mpb_new
, mpb
, size_old
);
5578 super
->anchor
= mpb_new
;
5579 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5580 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5581 super
->len
= size_round
;
5583 super
->current_vol
= idx
;
5585 /* handle 'failed_disks' by either:
5586 * a) create dummy disk entries in the table if this the first
5587 * volume in the array. We add them here as this is the only
5588 * opportunity to add them. add_to_super_imsm_volume()
5589 * handles the non-failed disks and continues incrementing
5591 * b) validate that 'failed_disks' matches the current number
5592 * of missing disks if the container is populated
5594 if (super
->current_vol
== 0) {
5596 for (i
= 0; i
< info
->failed_disks
; i
++) {
5597 struct imsm_disk
*disk
;
5600 disk
= __get_imsm_disk(mpb
, i
);
5601 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5602 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5603 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5604 "missing:%d", (__u8
)i
);
5606 find_missing(super
);
5611 for (d
= super
->missing
; d
; d
= d
->next
)
5613 if (info
->failed_disks
> missing
) {
5614 pr_err("unable to add 'missing' disk to container\n");
5619 if (imsm_is_name_allowed(super
, name
, 1) == false)
5622 dv
= xmalloc(sizeof(*dv
));
5623 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5625 * Explicitly allow truncating to not confuse gcc's
5626 * -Werror=stringop-truncation
5628 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5629 memcpy(dev
->volume
, name
, namelen
);
5630 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5631 info
->layout
, info
->chunk_size
,
5632 s
->size
* BLOCKS_PER_KB
);
5633 data_disks
= get_data_disks(info
->level
, info
->layout
,
5635 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5636 size_per_member
= array_blocks
/ data_disks
;
5638 set_imsm_dev_size(dev
, array_blocks
);
5639 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5641 vol
->migr_state
= 0;
5642 set_migr_type(dev
, MIGR_INIT
);
5643 vol
->dirty
= !info
->state
;
5644 set_vol_curr_migr_unit(dev
, 0);
5645 map
= get_imsm_map(dev
, MAP_0
);
5646 set_pba_of_lba0(map
, super
->create_offset
);
5647 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5648 map
->failed_disk_num
= ~0;
5649 if (info
->level
> 0)
5650 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5651 : IMSM_T_STATE_UNINITIALIZED
);
5653 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5654 IMSM_T_STATE_NORMAL
;
5657 if (info
->level
== 1 && info
->raid_disks
> 2) {
5660 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5664 map
->raid_level
= info
->level
;
5665 if (info
->level
== 10)
5666 map
->raid_level
= 1;
5667 set_num_domains(map
);
5669 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5670 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5674 map
->num_members
= info
->raid_disks
;
5675 update_num_data_stripes(map
, array_blocks
);
5676 for (i
= 0; i
< map
->num_members
; i
++) {
5677 /* initialized in add_to_super */
5678 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5680 mpb
->num_raid_devs
++;
5681 mpb
->num_raid_devs_created
++;
5682 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5684 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5685 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5686 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5687 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5691 pr_err("imsm does not support consistency policy %s\n",
5692 map_num_s(consistency_policies
, s
->consistency_policy
));
5697 dv
->index
= super
->current_vol
;
5698 dv
->next
= super
->devlist
;
5699 super
->devlist
= dv
;
5701 imsm_update_version_info(super
);
5706 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5707 struct shape
*s
, char *name
,
5708 char *homehost
, int *uuid
,
5709 unsigned long long data_offset
)
5711 /* This is primarily called by Create when creating a new array.
5712 * We will then get add_to_super called for each component, and then
5713 * write_init_super called to write it out to each device.
5714 * For IMSM, Create can create on fresh devices or on a pre-existing
5716 * To create on a pre-existing array a different method will be called.
5717 * This one is just for fresh drives.
5719 struct intel_super
*super
;
5720 struct imsm_super
*mpb
;
5724 if (data_offset
!= INVALID_SECTORS
) {
5725 pr_err("data-offset not supported by imsm\n");
5730 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5734 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5736 mpb_size
= MAX_SECTOR_SIZE
;
5738 super
= alloc_super();
5740 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5745 pr_err("could not allocate superblock\n");
5748 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5749 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5750 pr_err("could not allocate migr_rec buffer\n");
5755 memset(super
->buf
, 0, mpb_size
);
5757 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5761 /* zeroing superblock */
5765 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5767 version
= (char *) mpb
->sig
;
5768 strcpy(version
, MPB_SIGNATURE
);
5769 version
+= strlen(MPB_SIGNATURE
);
5770 strcpy(version
, MPB_VERSION_RAID0
);
5775 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5777 unsigned int member_sector_size
;
5779 if (!is_fd_valid(dl
->fd
)) {
5780 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5784 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5786 if (member_sector_size
!= super
->sector_size
)
5791 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5792 int fd
, char *devname
)
5794 struct intel_super
*super
= st
->sb
;
5795 struct imsm_super
*mpb
= super
->anchor
;
5796 struct imsm_disk
*_disk
;
5797 struct imsm_dev
*dev
;
5798 struct imsm_map
*map
;
5803 if (!is_fd_valid(fd
))
5806 dev
= get_imsm_dev(super
, super
->current_vol
);
5807 map
= get_imsm_map(dev
, MAP_0
);
5809 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5810 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5815 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5817 if (dl
->raiddisk
== dk
->raid_disk
)
5819 } else if (dl
->major
== dk
->major
&& dl
->minor
== dk
->minor
)
5825 pr_err("%s is not a member of the same container.\n",
5830 if (!autolayout
&& super
->current_vol
> 0) {
5831 int _slot
= get_disk_slot_in_dev(super
, 0, dl
->index
);
5833 if (_slot
!= dk
->raid_disk
) {
5834 pr_err("Member %s is in %d slot for the first volume, but is in %d slot for a new volume.\n",
5835 dl
->devname
, _slot
, dk
->raid_disk
);
5836 pr_err("Raid members are in different order than for the first volume, aborting.\n");
5841 if (mpb
->num_disks
== 0)
5842 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5843 &super
->sector_size
))
5846 if (!drive_validate_sector_size(super
, dl
)) {
5847 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5851 /* add a pristine spare to the metadata */
5852 if (dl
->index
< 0) {
5853 dl
->index
= super
->anchor
->num_disks
;
5854 super
->anchor
->num_disks
++;
5856 /* Check the device has not already been added */
5857 slot
= get_imsm_disk_slot(map
, dl
->index
);
5859 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5860 pr_err("%s has been included in this array twice\n",
5864 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5865 dl
->disk
.status
= CONFIGURED_DISK
;
5867 /* update size of 'missing' disks to be at least as large as the
5868 * largest acitve member (we only have dummy missing disks when
5869 * creating the first volume)
5871 if (super
->current_vol
== 0) {
5872 for (df
= super
->missing
; df
; df
= df
->next
) {
5873 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5874 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5875 _disk
= __get_imsm_disk(mpb
, df
->index
);
5880 /* refresh unset/failed slots to point to valid 'missing' entries */
5881 for (df
= super
->missing
; df
; df
= df
->next
)
5882 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5883 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5885 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5887 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5888 if (is_gen_migration(dev
)) {
5889 struct imsm_map
*map2
= get_imsm_map(dev
,
5891 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5892 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5893 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5896 if ((unsigned)df
->index
==
5898 set_imsm_ord_tbl_ent(map2
,
5904 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5908 /* if we are creating the first raid device update the family number */
5909 if (super
->current_vol
== 0) {
5911 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5913 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5915 pr_err("BUG mpb setup error\n");
5921 sum
+= __gen_imsm_checksum(mpb
);
5922 mpb
->family_num
= __cpu_to_le32(sum
);
5923 mpb
->orig_family_num
= mpb
->family_num
;
5924 mpb
->creation_time
= __cpu_to_le64((__u64
)time(NULL
));
5926 super
->current_disk
= dl
;
5931 * Function marks disk as spare and restores disk serial
5932 * in case it was previously marked as failed by takeover operation
5934 * -1 : critical error
5935 * 0 : disk is marked as spare but serial is not set
5938 int mark_spare(struct dl
*disk
)
5940 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5947 if (!imsm_read_serial(disk
->fd
, NULL
, serial
, MAX_RAID_SERIAL_LEN
)) {
5948 /* Restore disk serial number, because takeover marks disk
5949 * as failed and adds to serial ':0' before it becomes
5952 serialcpy(disk
->serial
, serial
);
5953 serialcpy(disk
->disk
.serial
, serial
);
5956 disk
->disk
.status
= SPARE_DISK
;
5963 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
);
5965 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5966 int fd
, char *devname
,
5967 unsigned long long data_offset
)
5969 struct intel_super
*super
= st
->sb
;
5971 unsigned long long size
;
5972 unsigned int member_sector_size
;
5977 /* If we are on an RAID enabled platform check that the disk is
5978 * attached to the raid controller.
5979 * We do not need to test disks attachment for container based additions,
5980 * they shall be already tested when container was created/assembled.
5982 rv
= find_intel_hba_capability(fd
, super
, devname
);
5983 /* no orom/efi or non-intel hba of the disk */
5985 dprintf("capability: %p fd: %d ret: %d\n",
5986 super
->orom
, fd
, rv
);
5990 if (super
->current_vol
>= 0)
5991 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5994 dd
= xcalloc(sizeof(*dd
), 1);
5995 dd
->major
= major(stb
.st_rdev
);
5996 dd
->minor
= minor(stb
.st_rdev
);
5997 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
6000 dd
->action
= DISK_ADD
;
6001 rv
= imsm_read_serial(fd
, devname
, dd
->serial
, MAX_RAID_SERIAL_LEN
);
6003 pr_err("failed to retrieve scsi serial, aborting\n");
6004 __free_imsm_disk(dd
, 0);
6008 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
6009 (super
->hba
->type
== SYS_DEV_VMD
))) {
6011 char cntrl_path
[PATH_MAX
];
6013 char pci_dev_path
[PATH_MAX
];
6015 if (!diskfd_to_devpath(fd
, 2, pci_dev_path
) ||
6016 !diskfd_to_devpath(fd
, 1, cntrl_path
)) {
6017 pr_err("failed to get dev paths, aborting\n");
6018 __free_imsm_disk(dd
, 0);
6022 cntrl_name
= basename(cntrl_path
);
6023 if (is_multipath_nvme(fd
))
6024 pr_err("%s controller supports Multi-Path I/O, Intel (R) VROC does not support multipathing\n",
6027 if (devpath_to_vendor(pci_dev_path
) == 0x8086) {
6029 * If Intel's NVMe drive has serial ended with
6030 * "-A","-B","-1" or "-2" it means that this is "x8"
6031 * device (double drive on single PCIe card).
6032 * User should be warned about potential data loss.
6034 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
6035 /* Skip empty character at the end */
6036 if (dd
->serial
[i
] == 0)
6039 if (((dd
->serial
[i
] == 'A') ||
6040 (dd
->serial
[i
] == 'B') ||
6041 (dd
->serial
[i
] == '1') ||
6042 (dd
->serial
[i
] == '2')) &&
6043 (dd
->serial
[i
-1] == '-'))
6044 pr_err("\tThe action you are about to take may put your data at risk.\n"
6045 "\tPlease note that x8 devices may consist of two separate x4 devices "
6046 "located on a single PCIe port.\n"
6047 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
6050 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
6051 !imsm_orom_has_tpv_support(super
->orom
)) {
6052 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
6053 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
6054 __free_imsm_disk(dd
, 0);
6059 get_dev_size(fd
, NULL
, &size
);
6060 if (!get_dev_sector_size(fd
, NULL
, &member_sector_size
)) {
6061 __free_imsm_disk(dd
, 0);
6065 if (super
->sector_size
== 0) {
6066 /* this a first device, so sector_size is not set yet */
6067 super
->sector_size
= member_sector_size
;
6070 /* clear migr_rec when adding disk to container */
6071 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6072 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
6074 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
6075 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
6076 MIGR_REC_BUF_SECTORS
*member_sector_size
)
6077 perror("Write migr_rec failed");
6081 serialcpy(dd
->disk
.serial
, dd
->serial
);
6082 set_total_blocks(&dd
->disk
, size
);
6083 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
6084 struct imsm_super
*mpb
= super
->anchor
;
6085 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6088 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
6089 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
6091 dd
->disk
.scsi_id
= __cpu_to_le32(0);
6093 if (st
->update_tail
) {
6094 dd
->next
= super
->disk_mgmt_list
;
6095 super
->disk_mgmt_list
= dd
;
6097 /* this is called outside of mdmon
6098 * write initial spare metadata
6099 * mdmon will overwrite it.
6101 dd
->next
= super
->disks
;
6103 write_super_imsm_spare(super
, dd
);
6109 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
6111 struct intel_super
*super
= st
->sb
;
6114 /* remove from super works only in mdmon - for communication
6115 * manager - monitor. Check if communication memory buffer
6118 if (!st
->update_tail
) {
6119 pr_err("shall be used in mdmon context only\n");
6122 dd
= xcalloc(1, sizeof(*dd
));
6123 dd
->major
= dk
->major
;
6124 dd
->minor
= dk
->minor
;
6127 dd
->action
= DISK_REMOVE
;
6129 dd
->next
= super
->disk_mgmt_list
;
6130 super
->disk_mgmt_list
= dd
;
6135 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
6138 char buf
[MAX_SECTOR_SIZE
];
6139 struct imsm_super anchor
;
6140 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
6143 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
)
6145 struct imsm_super
*mpb
= super
->anchor
;
6146 struct imsm_super
*spare
= &spare_record
.anchor
;
6152 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
6153 spare
->generation_num
= __cpu_to_le32(1UL);
6154 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
6155 spare
->num_disks
= 1;
6156 spare
->num_raid_devs
= 0;
6157 spare
->cache_size
= mpb
->cache_size
;
6158 spare
->pwr_cycle_count
= __cpu_to_le32(1);
6160 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
6161 MPB_SIGNATURE MPB_VERSION_RAID0
);
6163 spare
->disk
[0] = d
->disk
;
6164 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
6165 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6167 if (super
->sector_size
== 4096)
6168 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6170 sum
= __gen_imsm_checksum(spare
);
6171 spare
->family_num
= __cpu_to_le32(sum
);
6172 spare
->orig_family_num
= 0;
6173 sum
= __gen_imsm_checksum(spare
);
6174 spare
->check_sum
= __cpu_to_le32(sum
);
6176 if (store_imsm_mpb(d
->fd
, spare
)) {
6177 pr_err("failed for device %d:%d %s\n",
6178 d
->major
, d
->minor
, strerror(errno
));
6184 /* spare records have their own family number and do not have any defined raid
6187 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
6191 for (d
= super
->disks
; d
; d
= d
->next
) {
6195 if (write_super_imsm_spare(super
, d
))
6205 static int write_super_imsm(struct supertype
*st
, int doclose
)
6207 struct intel_super
*super
= st
->sb
;
6208 unsigned int sector_size
= super
->sector_size
;
6209 struct imsm_super
*mpb
= super
->anchor
;
6215 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6217 int clear_migration_record
= 1;
6220 /* 'generation' is incremented everytime the metadata is written */
6221 generation
= __le32_to_cpu(mpb
->generation_num
);
6223 mpb
->generation_num
= __cpu_to_le32(generation
);
6225 /* fix up cases where previous mdadm releases failed to set
6228 if (mpb
->orig_family_num
== 0)
6229 mpb
->orig_family_num
= mpb
->family_num
;
6231 for (d
= super
->disks
; d
; d
= d
->next
) {
6235 mpb
->disk
[d
->index
] = d
->disk
;
6239 for (d
= super
->missing
; d
; d
= d
->next
) {
6240 mpb
->disk
[d
->index
] = d
->disk
;
6243 mpb
->num_disks
= num_disks
;
6244 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6246 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6247 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6248 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6250 imsm_copy_dev(dev
, dev2
);
6251 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6253 if (is_gen_migration(dev2
))
6254 clear_migration_record
= 0;
6257 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6260 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6261 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6263 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6265 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6266 mpb_size
+= bbm_log_size
;
6267 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6270 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6273 /* recalculate checksum */
6274 sum
= __gen_imsm_checksum(mpb
);
6275 mpb
->check_sum
= __cpu_to_le32(sum
);
6277 if (super
->clean_migration_record_by_mdmon
) {
6278 clear_migration_record
= 1;
6279 super
->clean_migration_record_by_mdmon
= 0;
6281 if (clear_migration_record
)
6282 memset(super
->migr_rec_buf
, 0,
6283 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6285 if (sector_size
== 4096)
6286 convert_to_4k(super
);
6288 /* write the mpb for disks that compose raid devices */
6289 for (d
= super
->disks
; d
; d
= d
->next
) {
6290 if (d
->index
< 0 || is_failed(&d
->disk
))
6293 if (clear_migration_record
) {
6294 unsigned long long dsize
;
6296 get_dev_size(d
->fd
, NULL
, &dsize
);
6297 if (lseek64(d
->fd
, dsize
- sector_size
,
6299 if ((unsigned int)write(d
->fd
,
6300 super
->migr_rec_buf
,
6301 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6302 MIGR_REC_BUF_SECTORS
*sector_size
)
6303 perror("Write migr_rec failed");
6307 if (store_imsm_mpb(d
->fd
, mpb
))
6309 "failed for device %d:%d (fd: %d)%s\n",
6311 d
->fd
, strerror(errno
));
6318 return write_super_imsm_spares(super
, doclose
);
6323 static int create_array(struct supertype
*st
, int dev_idx
)
6326 struct imsm_update_create_array
*u
;
6327 struct intel_super
*super
= st
->sb
;
6328 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6329 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6330 struct disk_info
*inf
;
6331 struct imsm_disk
*disk
;
6334 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6335 sizeof(*inf
) * map
->num_members
;
6337 u
->type
= update_create_array
;
6338 u
->dev_idx
= dev_idx
;
6339 imsm_copy_dev(&u
->dev
, dev
);
6340 inf
= get_disk_info(u
);
6341 for (i
= 0; i
< map
->num_members
; i
++) {
6342 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6344 disk
= get_imsm_disk(super
, idx
);
6346 disk
= get_imsm_missing(super
, idx
);
6347 serialcpy(inf
[i
].serial
, disk
->serial
);
6349 append_metadata_update(st
, u
, len
);
6354 static int mgmt_disk(struct supertype
*st
)
6356 struct intel_super
*super
= st
->sb
;
6358 struct imsm_update_add_remove_disk
*u
;
6360 if (!super
->disk_mgmt_list
)
6365 u
->type
= update_add_remove_disk
;
6366 append_metadata_update(st
, u
, len
);
6371 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6373 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6375 struct ppl_header
*ppl_hdr
= buf
;
6378 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6380 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6382 perror("Failed to seek to PPL header location");
6386 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6388 perror("Write PPL header failed");
6397 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6399 struct intel_super
*super
= st
->sb
;
6401 struct ppl_header
*ppl_hdr
;
6404 /* first clear entire ppl space */
6405 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6409 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6411 pr_err("Failed to allocate PPL header buffer\n");
6415 memset(buf
, 0, PPL_HEADER_SIZE
);
6417 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6418 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6420 if (info
->mismatch_cnt
) {
6422 * We are overwriting an invalid ppl. Make one entry with wrong
6423 * checksum to prevent the kernel from skipping resync.
6425 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6426 ppl_hdr
->entries
[0].checksum
= ~0;
6429 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6435 static int is_rebuilding(struct imsm_dev
*dev
);
6437 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6438 struct mdinfo
*disk
)
6440 struct intel_super
*super
= st
->sb
;
6442 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6444 struct ppl_header
*ppl_hdr
= NULL
;
6446 struct imsm_dev
*dev
;
6449 unsigned long long ppl_offset
= 0;
6450 unsigned long long prev_gen_num
= 0;
6452 if (disk
->disk
.raid_disk
< 0)
6455 dev
= get_imsm_dev(super
, info
->container_member
);
6456 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6457 d
= get_imsm_dl_disk(super
, idx
);
6459 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6462 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6463 pr_err("Failed to allocate PPL header buffer\n");
6469 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6472 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6474 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6476 perror("Failed to seek to PPL header location");
6481 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6482 perror("Read PPL header failed");
6489 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6490 ppl_hdr
->checksum
= 0;
6492 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6493 dprintf("Wrong PPL header checksum on %s\n",
6498 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6499 /* previous was newest, it was already checked */
6503 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6504 super
->anchor
->orig_family_num
)) {
6505 dprintf("Wrong PPL header signature on %s\n",
6512 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6514 ppl_offset
+= PPL_HEADER_SIZE
;
6515 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6517 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6520 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6532 * Update metadata to use mutliple PPLs area (1MB).
6533 * This is done once for all RAID members
6535 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6536 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6538 struct mdinfo
*member_dev
;
6540 sprintf(subarray
, "%d", info
->container_member
);
6542 if (mdmon_running(st
->container_devnm
))
6543 st
->update_tail
= &st
->updates
;
6545 if (st
->ss
->update_subarray(st
, subarray
, UOPT_PPL
, NULL
)) {
6546 pr_err("Failed to update subarray %s\n",
6549 if (st
->update_tail
)
6550 flush_metadata_updates(st
);
6552 st
->ss
->sync_metadata(st
);
6553 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6554 for (member_dev
= info
->devs
; member_dev
;
6555 member_dev
= member_dev
->next
)
6556 member_dev
->ppl_size
=
6557 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6562 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6564 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6565 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6566 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6567 (is_rebuilding(dev
) &&
6568 vol_curr_migr_unit(dev
) == 0 &&
6569 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6570 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6572 info
->mismatch_cnt
++;
6573 } else if (ret
== 0 &&
6574 ppl_hdr
->entries_count
== 0 &&
6575 is_rebuilding(dev
) &&
6576 info
->resync_start
== 0) {
6578 * The header has no entries - add a single empty entry and
6579 * rewrite the header to prevent the kernel from going into
6580 * resync after an interrupted rebuild.
6582 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6583 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6591 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6593 struct intel_super
*super
= st
->sb
;
6597 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6598 info
->array
.level
!= 5)
6601 for (d
= super
->disks
; d
; d
= d
->next
) {
6602 if (d
->index
< 0 || is_failed(&d
->disk
))
6605 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6613 /*******************************************************************************
6614 * Function: write_init_bitmap_imsm_vol
6615 * Description: Write a bitmap header and prepares the area for the bitmap.
6617 * st : supertype information
6618 * vol_idx : the volume index to use
6623 ******************************************************************************/
6624 static int write_init_bitmap_imsm_vol(struct supertype
*st
, int vol_idx
)
6626 struct intel_super
*super
= st
->sb
;
6627 int prev_current_vol
= super
->current_vol
;
6631 super
->current_vol
= vol_idx
;
6632 for (d
= super
->disks
; d
; d
= d
->next
) {
6633 if (d
->index
< 0 || is_failed(&d
->disk
))
6635 ret
= st
->ss
->write_bitmap(st
, d
->fd
, NoUpdate
);
6639 super
->current_vol
= prev_current_vol
;
6643 /*******************************************************************************
6644 * Function: write_init_bitmap_imsm_all
6645 * Description: Write a bitmap header and prepares the area for the bitmap.
6646 * Operation is executed for volumes with CONSISTENCY_POLICY_BITMAP.
6648 * st : supertype information
6649 * info : info about the volume where the bitmap should be written
6650 * vol_idx : the volume index to use
6655 ******************************************************************************/
6656 static int write_init_bitmap_imsm_all(struct supertype
*st
, struct mdinfo
*info
,
6661 if (info
&& (info
->consistency_policy
== CONSISTENCY_POLICY_BITMAP
))
6662 ret
= write_init_bitmap_imsm_vol(st
, vol_idx
);
6667 static int write_init_super_imsm(struct supertype
*st
)
6669 struct intel_super
*super
= st
->sb
;
6670 int current_vol
= super
->current_vol
;
6674 getinfo_super_imsm(st
, &info
, NULL
);
6676 /* we are done with current_vol reset it to point st at the container */
6677 super
->current_vol
= -1;
6679 if (st
->update_tail
) {
6680 /* queue the recently created array / added disk
6681 * as a metadata update */
6683 /* determine if we are creating a volume or adding a disk */
6684 if (current_vol
< 0) {
6685 /* in the mgmt (add/remove) disk case we are running
6686 * in mdmon context, so don't close fd's
6690 /* adding the second volume to the array */
6691 rv
= write_init_ppl_imsm_all(st
, &info
);
6693 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6695 rv
= create_array(st
, current_vol
);
6699 for (d
= super
->disks
; d
; d
= d
->next
)
6700 Kill(d
->devname
, NULL
, 0, -1, 1);
6701 if (current_vol
>= 0) {
6702 rv
= write_init_ppl_imsm_all(st
, &info
);
6704 rv
= write_init_bitmap_imsm_all(st
, &info
, current_vol
);
6708 rv
= write_super_imsm(st
, 1);
6714 static int store_super_imsm(struct supertype
*st
, int fd
)
6716 struct intel_super
*super
= st
->sb
;
6717 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6722 if (super
->sector_size
== 4096)
6723 convert_to_4k(super
);
6724 return store_imsm_mpb(fd
, mpb
);
6727 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6729 unsigned long long data_offset
,
6731 unsigned long long *freesize
,
6735 unsigned long long ldsize
;
6736 struct intel_super
*super
= NULL
;
6739 if (!is_container(level
))
6744 fd
= dev_open(dev
, O_RDONLY
|O_EXCL
);
6745 if (!is_fd_valid(fd
)) {
6746 pr_vrb("imsm: Cannot open %s: %s\n", dev
, strerror(errno
));
6749 if (!get_dev_size(fd
, dev
, &ldsize
))
6752 /* capabilities retrieve could be possible
6753 * note that there is no fd for the disks in array.
6755 super
= alloc_super();
6759 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
))
6762 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6766 fd2devname(fd
, str
);
6767 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6768 fd
, str
, super
->orom
, rv
, raiddisks
);
6770 /* no orom/efi or non-intel hba of the disk */
6775 if (raiddisks
> super
->orom
->tds
) {
6777 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6778 raiddisks
, super
->orom
->tds
);
6781 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6782 (ldsize
>> 9) >> 32 > 0) {
6784 pr_err("%s exceeds maximum platform supported size\n", dev
);
6788 if (super
->hba
->type
== SYS_DEV_VMD
||
6789 super
->hba
->type
== SYS_DEV_NVME
) {
6790 if (!imsm_is_nvme_namespace_supported(fd
, 1)) {
6792 pr_err("NVMe namespace %s is not supported by IMSM\n",
6799 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6809 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6811 const unsigned long long base_start
= e
[*idx
].start
;
6812 unsigned long long end
= base_start
+ e
[*idx
].size
;
6815 if (base_start
== end
)
6819 for (i
= *idx
; i
< num_extents
; i
++) {
6820 /* extend overlapping extents */
6821 if (e
[i
].start
>= base_start
&&
6822 e
[i
].start
<= end
) {
6825 if (e
[i
].start
+ e
[i
].size
> end
)
6826 end
= e
[i
].start
+ e
[i
].size
;
6827 } else if (e
[i
].start
> end
) {
6833 return end
- base_start
;
6836 /** merge_extents() - analyze extents and get free size.
6837 * @super: Intel metadata, not NULL.
6838 * @expanding: if set, we are expanding &super->current_vol.
6840 * Build a composite disk with all known extents and generate a size given the
6841 * "all disks in an array must share a common start offset" constraint.
6842 * If a volume is expanded, then return free space after the volume.
6844 * Return: Free space or 0 on failure.
6846 static unsigned long long merge_extents(struct intel_super
*super
, const bool expanding
)
6850 int i
, j
, pos_vol_idx
= -1;
6852 int sum_extents
= 0;
6853 unsigned long long pos
= 0;
6854 unsigned long long start
= 0;
6855 unsigned long long free_size
= 0;
6857 unsigned long pre_reservation
= 0;
6858 unsigned long post_reservation
= IMSM_RESERVED_SECTORS
;
6859 unsigned long reservation_size
;
6861 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6863 sum_extents
+= dl
->extent_cnt
;
6864 e
= xcalloc(sum_extents
, sizeof(struct extent
));
6866 /* coalesce and sort all extents. also, check to see if we need to
6867 * reserve space between member arrays
6870 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6873 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6876 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6881 while (i
< sum_extents
) {
6882 e
[j
].start
= e
[i
].start
;
6883 e
[j
].vol
= e
[i
].vol
;
6884 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6886 if (e
[j
-1].size
== 0)
6892 unsigned long long esize
= e
[i
].start
- pos
;
6894 if (expanding
? pos_vol_idx
== super
->current_vol
: esize
>= free_size
) {
6900 pos
= e
[i
].start
+ e
[i
].size
;
6901 pos_vol_idx
= e
[i
].vol
;
6904 } while (e
[i
-1].size
);
6906 if (free_size
== 0) {
6907 dprintf("imsm: Cannot find free size.\n");
6912 if (!expanding
&& extent_idx
!= 0)
6914 * Not a real first volume in a container is created, pre_reservation is needed.
6916 pre_reservation
= IMSM_RESERVED_SECTORS
;
6918 if (e
[extent_idx
].size
== 0)
6920 * extent_idx points to the metadata, post_reservation is allready done.
6922 post_reservation
= 0;
6925 reservation_size
= pre_reservation
+ post_reservation
;
6927 if (free_size
< reservation_size
) {
6928 dprintf("imsm: Reservation size is greater than free space.\n");
6932 super
->create_offset
= start
+ pre_reservation
;
6933 return free_size
- reservation_size
;
6936 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6938 if (level
< 0 || level
== 6 || level
== 4)
6941 /* if we have an orom prevent invalid raid levels */
6944 case 0: return imsm_orom_has_raid0(orom
);
6947 return imsm_orom_has_raid1e(orom
);
6948 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6949 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6950 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6953 return 1; /* not on an Intel RAID platform so anything goes */
6959 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6960 int dpa
, int verbose
)
6962 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6963 struct mdstat_ent
*memb
;
6969 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6970 if (memb
->metadata_version
&&
6971 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6972 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6973 !is_subarray(memb
->metadata_version
+9) &&
6975 struct dev_member
*dev
= memb
->members
;
6977 while (dev
&& !is_fd_valid(fd
)) {
6978 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6979 num
= snprintf(path
, PATH_MAX
, "%s%s", "/dev/", dev
->name
);
6981 fd
= open(path
, O_RDONLY
, 0);
6982 if (num
<= 0 || !is_fd_valid(fd
)) {
6983 pr_vrb("Cannot open %s: %s\n",
6984 dev
->name
, strerror(errno
));
6990 if (is_fd_valid(fd
) && disk_attached_to_hba(fd
, hba
)) {
6991 struct mdstat_ent
*vol
;
6992 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6993 if (vol
->active
> 0 &&
6994 vol
->metadata_version
&&
6995 is_container_member(vol
, memb
->devnm
)) {
7000 if (*devlist
&& (found
< dpa
)) {
7001 dv
= xcalloc(1, sizeof(*dv
));
7002 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
7003 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
7006 dv
->next
= *devlist
;
7013 free_mdstat(mdstat
);
7018 static struct md_list
*
7019 get_loop_devices(void)
7022 struct md_list
*devlist
= NULL
;
7025 for(i
= 0; i
< 12; i
++) {
7026 dv
= xcalloc(1, sizeof(*dv
));
7027 dv
->devname
= xmalloc(40);
7028 sprintf(dv
->devname
, "/dev/loop%d", i
);
7036 static struct md_list
*
7037 get_devices(const char *hba_path
)
7039 struct md_list
*devlist
= NULL
;
7046 devlist
= get_loop_devices();
7049 /* scroll through /sys/dev/block looking for devices attached to
7052 dir
= opendir("/sys/dev/block");
7053 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
7058 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
7060 path
= devt_to_devpath(makedev(major
, minor
), 1, NULL
);
7063 if (!path_attached_to_hba(path
, hba_path
)) {
7070 fd
= dev_open(ent
->d_name
, O_RDONLY
);
7071 if (is_fd_valid(fd
)) {
7072 fd2devname(fd
, buf
);
7075 pr_err("cannot open device: %s\n",
7080 dv
= xcalloc(1, sizeof(*dv
));
7081 dv
->devname
= xstrdup(buf
);
7088 devlist
= devlist
->next
;
7098 count_volumes_list(struct md_list
*devlist
, char *homehost
,
7099 int verbose
, int *found
)
7101 struct md_list
*tmpdev
;
7103 struct supertype
*st
;
7105 /* first walk the list of devices to find a consistent set
7106 * that match the criterea, if that is possible.
7107 * We flag the ones we like with 'used'.
7110 st
= match_metadata_desc_imsm("imsm");
7112 pr_vrb("cannot allocate memory for imsm supertype\n");
7116 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7117 char *devname
= tmpdev
->devname
;
7119 struct supertype
*tst
;
7121 if (tmpdev
->used
> 1)
7123 tst
= dup_super(st
);
7125 pr_vrb("cannot allocate memory for imsm supertype\n");
7128 tmpdev
->container
= 0;
7129 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
7130 if (!is_fd_valid(dfd
)) {
7131 dprintf("cannot open device %s: %s\n",
7132 devname
, strerror(errno
));
7134 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
7136 } else if (must_be_container(dfd
)) {
7137 struct supertype
*cst
;
7138 cst
= super_by_fd(dfd
, NULL
);
7140 dprintf("cannot recognize container type %s\n",
7143 } else if (tst
->ss
!= st
->ss
) {
7144 dprintf("non-imsm container - ignore it: %s\n",
7147 } else if (!tst
->ss
->load_container
||
7148 tst
->ss
->load_container(tst
, dfd
, NULL
))
7151 tmpdev
->container
= 1;
7154 cst
->ss
->free_super(cst
);
7156 tmpdev
->st_rdev
= rdev
;
7157 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
7158 dprintf("no RAID superblock on %s\n",
7161 } else if (tst
->ss
->compare_super
== NULL
) {
7162 dprintf("Cannot assemble %s metadata on %s\n",
7163 tst
->ss
->name
, devname
);
7169 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
7170 /* Ignore unrecognised devices during auto-assembly */
7175 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
7177 if (st
->minor_version
== -1)
7178 st
->minor_version
= tst
->minor_version
;
7180 if (memcmp(info
.uuid
, uuid_zero
,
7181 sizeof(int[4])) == 0) {
7182 /* this is a floating spare. It cannot define
7183 * an array unless there are no more arrays of
7184 * this type to be found. It can be included
7185 * in an array of this type though.
7191 if (st
->ss
!= tst
->ss
||
7192 st
->minor_version
!= tst
->minor_version
||
7193 st
->ss
->compare_super(st
, tst
, 1) != 0) {
7194 /* Some mismatch. If exactly one array matches this host,
7195 * we can resolve on that one.
7196 * Or, if we are auto assembling, we just ignore the second
7199 dprintf("superblock on %s doesn't match others - assembly aborted\n",
7205 dprintf("found: devname: %s\n", devname
);
7209 tst
->ss
->free_super(tst
);
7213 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
7214 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
7215 for (iter
= head
; iter
; iter
= iter
->next
) {
7216 dprintf("content->text_version: %s vol\n",
7217 iter
->text_version
);
7218 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
7219 /* do not assemble arrays with unsupported
7221 dprintf("Cannot activate member %s.\n",
7222 iter
->text_version
);
7229 dprintf("No valid super block on device list: err: %d %p\n",
7233 dprintf("no more devices to examine\n");
7236 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7237 if (tmpdev
->used
== 1 && tmpdev
->found
) {
7239 if (count
< tmpdev
->found
)
7242 count
-= tmpdev
->found
;
7245 if (tmpdev
->used
== 1)
7250 st
->ss
->free_super(st
);
7254 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
7257 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
7259 const struct orom_entry
*entry
;
7260 struct devid_list
*dv
, *devid_list
;
7265 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7266 if (strstr(idev
->path
, hba_path
))
7270 if (!idev
|| !idev
->dev_id
)
7273 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7275 if (!entry
|| !entry
->devid_list
)
7278 devid_list
= entry
->devid_list
;
7279 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7280 struct md_list
*devlist
;
7281 struct sys_dev
*device
= NULL
;
7286 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7288 device
= device_by_id(dv
->devid
);
7291 hpath
= device
->path
;
7295 devlist
= get_devices(hpath
);
7296 /* if no intel devices return zero volumes */
7297 if (devlist
== NULL
)
7300 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7302 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7303 if (devlist
== NULL
)
7307 count
+= count_volumes_list(devlist
,
7311 dprintf("found %d count: %d\n", found
, count
);
7314 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7317 struct md_list
*dv
= devlist
;
7318 devlist
= devlist
->next
;
7326 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7330 if (hba
->type
== SYS_DEV_VMD
) {
7331 struct sys_dev
*dev
;
7334 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7335 if (dev
->type
== SYS_DEV_VMD
)
7336 count
+= __count_volumes(dev
->path
, dpa
,
7341 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7344 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7346 /* up to 512 if the plaform supports it, otherwise the platform max.
7347 * 128 if no platform detected
7349 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7351 return min(512, (1 << fs
));
7355 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7356 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7358 /* check/set platform and metadata limits/defaults */
7359 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7360 pr_vrb("platform supports a maximum of %d disks per array\n",
7365 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7366 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7367 pr_vrb("platform does not support raid%d with %d disk%s\n",
7368 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7372 if (*chunk
== 0 || *chunk
== UnSet
)
7373 *chunk
= imsm_default_chunk(super
->orom
);
7375 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7376 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7380 if (layout
!= imsm_level_to_layout(level
)) {
7382 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7383 else if (level
== 10)
7384 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7386 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7391 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7392 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7393 pr_vrb("platform does not support a volume size over 2TB\n");
7400 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7401 * FIX ME add ahci details
7403 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7404 int layout
, int raiddisks
, int *chunk
,
7405 unsigned long long size
,
7406 unsigned long long data_offset
,
7408 unsigned long long *freesize
,
7412 struct intel_super
*super
= st
->sb
;
7413 struct imsm_super
*mpb
;
7415 unsigned long long pos
= 0;
7416 unsigned long long maxsize
;
7420 /* We must have the container info already read in. */
7424 mpb
= super
->anchor
;
7426 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7427 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7431 /* General test: make sure there is space for
7432 * 'raiddisks' device extents of size 'size' at a given
7435 unsigned long long minsize
= size
;
7436 unsigned long long start_offset
= MaxSector
;
7439 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7440 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7445 e
= get_extents(super
, dl
, 0);
7448 unsigned long long esize
;
7449 esize
= e
[i
].start
- pos
;
7450 if (esize
>= minsize
)
7452 if (found
&& start_offset
== MaxSector
) {
7455 } else if (found
&& pos
!= start_offset
) {
7459 pos
= e
[i
].start
+ e
[i
].size
;
7461 } while (e
[i
-1].size
);
7466 if (dcnt
< raiddisks
) {
7468 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7475 /* This device must be a member of the set */
7476 if (!stat_is_blkdev(dev
, &rdev
))
7478 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7479 if (dl
->major
== (int)major(rdev
) &&
7480 dl
->minor
== (int)minor(rdev
))
7485 pr_err("%s is not in the same imsm set\n", dev
);
7487 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7488 /* If a volume is present then the current creation attempt
7489 * cannot incorporate new spares because the orom may not
7490 * understand this configuration (all member disks must be
7491 * members of each array in the container).
7493 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7494 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7496 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7497 mpb
->num_disks
!= raiddisks
) {
7498 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7502 /* retrieve the largest free space block */
7503 e
= get_extents(super
, dl
, 0);
7508 unsigned long long esize
;
7510 esize
= e
[i
].start
- pos
;
7511 if (esize
>= maxsize
)
7513 pos
= e
[i
].start
+ e
[i
].size
;
7515 } while (e
[i
-1].size
);
7520 pr_err("unable to determine free space for: %s\n",
7524 if (maxsize
< size
) {
7526 pr_err("%s not enough space (%llu < %llu)\n",
7527 dev
, maxsize
, size
);
7531 maxsize
= merge_extents(super
, false);
7533 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7534 pr_err("attempting to create a second volume with size less then remaining space.\n");
7536 if (maxsize
< size
|| maxsize
== 0) {
7539 pr_err("no free space left on device. Aborting...\n");
7541 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7547 *freesize
= maxsize
;
7550 int count
= count_volumes(super
->hba
,
7551 super
->orom
->dpa
, verbose
);
7552 if (super
->orom
->vphba
<= count
) {
7553 pr_vrb("platform does not support more than %d raid volumes.\n",
7554 super
->orom
->vphba
);
7562 * imsm_get_free_size() - get the biggest, common free space from members.
7563 * @super: &intel_super pointer, not NULL.
7564 * @raiddisks: number of raid disks.
7565 * @size: requested size, could be 0 (means max size).
7566 * @chunk: requested chunk size in KiB.
7567 * @freesize: pointer for returned size value.
7569 * Return: &IMSM_STATUS_OK or &IMSM_STATUS_ERROR.
7571 * @freesize is set to meaningful value, this can be @size, or calculated
7573 * super->create_offset value is modified and set appropriately in
7574 * merge_extends() for further creation.
7576 static imsm_status_t
imsm_get_free_size(struct intel_super
*super
,
7577 const int raiddisks
,
7578 unsigned long long size
,
7580 unsigned long long *freesize
,
7583 struct imsm_super
*mpb
= super
->anchor
;
7589 unsigned long long maxsize
;
7590 unsigned long long minsize
= size
;
7593 minsize
= chunk
* 2;
7595 /* find the largest common start free region of the possible disks */
7596 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7602 /* don't activate new spares if we are orom constrained
7603 * and there is already a volume active in the container
7605 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7608 e
= get_extents(super
, dl
, 0);
7611 for (i
= 1; e
[i
-1].size
; i
++)
7618 maxsize
= merge_extents(super
, expanding
);
7619 if (maxsize
< minsize
) {
7620 pr_err("imsm: Free space is %llu but must be equal or larger than %llu.\n",
7622 return IMSM_STATUS_ERROR
;
7625 if (cnt
< raiddisks
|| (super
->orom
&& used
&& used
!= raiddisks
)) {
7626 pr_err("imsm: Not enough devices with space to create array.\n");
7627 return IMSM_STATUS_ERROR
;
7638 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7639 pr_err("attempting to create a second volume with size less then remaining space.\n");
7642 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7644 return IMSM_STATUS_OK
;
7648 * autolayout_imsm() - automatically layout a new volume.
7649 * @super: &intel_super pointer, not NULL.
7650 * @raiddisks: number of raid disks.
7651 * @size: requested size, could be 0 (means max size).
7652 * @chunk: requested chunk.
7653 * @freesize: pointer for returned size value.
7655 * We are being asked to automatically layout a new volume based on the current
7656 * contents of the container. If the parameters can be satisfied autolayout_imsm
7657 * will record the disks, start offset, and will return size of the volume to
7658 * be created. See imsm_get_free_size() for details.
7659 * add_to_super() and getinfo_super() detect when autolayout is in progress.
7660 * If first volume exists, slots are set consistently to it.
7662 * Return: &IMSM_STATUS_OK on success, &IMSM_STATUS_ERROR otherwise.
7664 * Disks are marked for creation via dl->raiddisk.
7666 static imsm_status_t
autolayout_imsm(struct intel_super
*super
,
7667 const int raiddisks
,
7668 unsigned long long size
, const int chunk
,
7669 unsigned long long *freesize
)
7673 int vol_cnt
= super
->anchor
->num_raid_devs
;
7676 rv
= imsm_get_free_size(super
, raiddisks
, size
, chunk
, freesize
, false);
7677 if (rv
!= IMSM_STATUS_OK
)
7678 return IMSM_STATUS_ERROR
;
7680 for (disk
= super
->disks
; disk
; disk
= disk
->next
) {
7684 if (curr_slot
== raiddisks
)
7688 disk
->raiddisk
= curr_slot
;
7690 int _slot
= get_disk_slot_in_dev(super
, 0, disk
->index
);
7693 pr_err("Disk %s is not used in first volume, aborting\n",
7695 return IMSM_STATUS_ERROR
;
7697 disk
->raiddisk
= _slot
;
7702 return IMSM_STATUS_OK
;
7705 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7706 int raiddisks
, int *chunk
, unsigned long long size
,
7707 unsigned long long data_offset
,
7708 char *dev
, unsigned long long *freesize
,
7709 int consistency_policy
, int verbose
)
7716 * if given unused devices create a container
7717 * if given given devices in a container create a member volume
7719 if (is_container(level
))
7720 /* Must be a fresh device to add to a container */
7721 return validate_geometry_imsm_container(st
, level
, raiddisks
,
7726 * Size is given in sectors.
7728 if (size
&& (size
< 2048)) {
7729 pr_err("Given size must be greater than 1M.\n");
7730 /* Depends on algorithm in Create.c :
7731 * if container was given (dev == NULL) return -1,
7732 * if block device was given ( dev != NULL) return 0.
7734 return dev
? -1 : 0;
7738 struct intel_super
*super
= st
->sb
;
7741 * Autolayout mode, st->sb must be set.
7744 pr_vrb("superblock must be set for autolayout, aborting\n");
7748 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7749 raiddisks
, chunk
, size
,
7753 if (super
->orom
&& freesize
) {
7755 int count
= count_volumes(super
->hba
, super
->orom
->dpa
,
7757 if (super
->orom
->vphba
<= count
) {
7758 pr_vrb("platform does not support more than %d raid volumes.\n",
7759 super
->orom
->vphba
);
7763 rv
= autolayout_imsm(super
, raiddisks
, size
, *chunk
,
7765 if (rv
!= IMSM_STATUS_OK
)
7771 /* creating in a given container */
7772 return validate_geometry_imsm_volume(st
, level
, layout
,
7773 raiddisks
, chunk
, size
,
7775 dev
, freesize
, verbose
);
7778 /* This device needs to be a device in an 'imsm' container */
7779 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7781 if (is_fd_valid(fd
)) {
7782 pr_vrb("Cannot create this array on device %s\n", dev
);
7787 fd
= open(dev
, O_RDONLY
, 0);
7789 if (!is_fd_valid(fd
)) {
7790 pr_vrb("Cannot open %s: %s\n", dev
, strerror(errno
));
7794 /* Well, it is in use by someone, maybe an 'imsm' container. */
7795 cfd
= open_container(fd
);
7798 if (!is_fd_valid(cfd
)) {
7799 pr_vrb("Cannot use %s: It is busy\n", dev
);
7802 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7803 if (sra
&& sra
->array
.major_version
== -1 &&
7804 strcmp(sra
->text_version
, "imsm") == 0)
7808 /* This is a member of a imsm container. Load the container
7809 * and try to create a volume
7811 struct intel_super
*super
;
7813 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7815 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7817 return validate_geometry_imsm_volume(st
, level
, layout
,
7819 size
, data_offset
, dev
,
7826 pr_err("failed container membership check\n");
7832 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7834 struct intel_super
*super
= st
->sb
;
7836 if (level
&& *level
== UnSet
)
7837 *level
= LEVEL_CONTAINER
;
7839 if (level
&& layout
&& *layout
== UnSet
)
7840 *layout
= imsm_level_to_layout(*level
);
7842 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7843 *chunk
= imsm_default_chunk(super
->orom
);
7846 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7848 static int kill_subarray_imsm(struct supertype
*st
, char *subarray_id
)
7850 /* remove the subarray currently referenced by subarray_id */
7852 struct intel_dev
**dp
;
7853 struct intel_super
*super
= st
->sb
;
7854 __u8 current_vol
= strtoul(subarray_id
, NULL
, 10);
7855 struct imsm_super
*mpb
= super
->anchor
;
7857 if (mpb
->num_raid_devs
== 0)
7860 /* block deletions that would change the uuid of active subarrays
7862 * FIXME when immutable ids are available, but note that we'll
7863 * also need to fixup the invalidated/active subarray indexes in
7866 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7869 if (i
< current_vol
)
7871 snprintf(subarray
, sizeof(subarray
), "%u", i
);
7872 if (is_subarray_active(subarray
, st
->devnm
)) {
7873 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7880 if (st
->update_tail
) {
7881 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7883 u
->type
= update_kill_array
;
7884 u
->dev_idx
= current_vol
;
7885 append_metadata_update(st
, u
, sizeof(*u
));
7890 for (dp
= &super
->devlist
; *dp
;)
7891 if ((*dp
)->index
== current_vol
) {
7894 handle_missing(super
, (*dp
)->dev
);
7895 if ((*dp
)->index
> current_vol
)
7900 /* no more raid devices, all active components are now spares,
7901 * but of course failed are still failed
7903 if (--mpb
->num_raid_devs
== 0) {
7906 for (d
= super
->disks
; d
; d
= d
->next
)
7911 super
->updates_pending
++;
7917 * get_rwh_policy_from_update() - Get the rwh policy for update option.
7918 * @update: Update option.
7920 static int get_rwh_policy_from_update(enum update_opt update
)
7924 return RWH_MULTIPLE_DISTRIBUTED
;
7926 return RWH_MULTIPLE_OFF
;
7929 case UOPT_NO_BITMAP
:
7934 return UOPT_UNDEFINED
;
7937 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7938 enum update_opt update
, struct mddev_ident
*ident
)
7940 /* update the subarray currently referenced by ->current_vol */
7941 struct intel_super
*super
= st
->sb
;
7942 struct imsm_super
*mpb
= super
->anchor
;
7944 if (update
== UOPT_NAME
) {
7945 char *name
= ident
->name
;
7949 if (imsm_is_name_allowed(super
, name
, 1) == false)
7952 vol
= strtoul(subarray
, &ep
, 10);
7953 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7956 if (st
->update_tail
) {
7957 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7959 u
->type
= update_rename_array
;
7961 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7962 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7963 append_metadata_update(st
, u
, sizeof(*u
));
7965 struct imsm_dev
*dev
;
7968 dev
= get_imsm_dev(super
, vol
);
7969 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7970 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7971 memcpy(dev
->volume
, name
, namelen
);
7972 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7973 dev
= get_imsm_dev(super
, i
);
7974 handle_missing(super
, dev
);
7976 super
->updates_pending
++;
7978 } else if (get_rwh_policy_from_update(update
) != UOPT_UNDEFINED
) {
7981 int vol
= strtoul(subarray
, &ep
, 10);
7983 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7986 new_policy
= get_rwh_policy_from_update(update
);
7988 if (st
->update_tail
) {
7989 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7991 u
->type
= update_rwh_policy
;
7993 u
->new_policy
= new_policy
;
7994 append_metadata_update(st
, u
, sizeof(*u
));
7996 struct imsm_dev
*dev
;
7998 dev
= get_imsm_dev(super
, vol
);
7999 dev
->rwh_policy
= new_policy
;
8000 super
->updates_pending
++;
8002 if (new_policy
== RWH_BITMAP
)
8003 return write_init_bitmap_imsm_vol(st
, vol
);
8010 static bool is_gen_migration(struct imsm_dev
*dev
)
8012 if (dev
&& dev
->vol
.migr_state
&&
8013 migr_type(dev
) == MIGR_GEN_MIGR
)
8019 static int is_rebuilding(struct imsm_dev
*dev
)
8021 struct imsm_map
*migr_map
;
8023 if (!dev
->vol
.migr_state
)
8026 if (migr_type(dev
) != MIGR_REBUILD
)
8029 migr_map
= get_imsm_map(dev
, MAP_1
);
8031 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
8037 static int is_initializing(struct imsm_dev
*dev
)
8039 struct imsm_map
*migr_map
;
8041 if (!dev
->vol
.migr_state
)
8044 if (migr_type(dev
) != MIGR_INIT
)
8047 migr_map
= get_imsm_map(dev
, MAP_1
);
8049 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8055 static void update_recovery_start(struct intel_super
*super
,
8056 struct imsm_dev
*dev
,
8057 struct mdinfo
*array
)
8059 struct mdinfo
*rebuild
= NULL
;
8063 if (!is_rebuilding(dev
))
8066 /* Find the rebuild target, but punt on the dual rebuild case */
8067 for (d
= array
->devs
; d
; d
= d
->next
)
8068 if (d
->recovery_start
== 0) {
8075 /* (?) none of the disks are marked with
8076 * IMSM_ORD_REBUILD, so assume they are missing and the
8077 * disk_ord_tbl was not correctly updated
8079 dprintf("failed to locate out-of-sync disk\n");
8083 units
= vol_curr_migr_unit(dev
);
8084 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
8087 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
8089 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
8091 /* Given a container loaded by load_super_imsm_all,
8092 * extract information about all the arrays into
8094 * If 'subarray' is given, just extract info about that array.
8096 * For each imsm_dev create an mdinfo, fill it in,
8097 * then look for matching devices in super->disks
8098 * and create appropriate device mdinfo.
8100 struct intel_super
*super
= st
->sb
;
8101 struct imsm_super
*mpb
= super
->anchor
;
8102 struct mdinfo
*rest
= NULL
;
8106 int spare_disks
= 0;
8107 int current_vol
= super
->current_vol
;
8109 /* do not assemble arrays when not all attributes are supported */
8110 if (imsm_check_attributes(mpb
->attributes
) == 0) {
8112 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
8115 /* count spare devices, not used in maps
8117 for (d
= super
->disks
; d
; d
= d
->next
)
8121 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8122 struct imsm_dev
*dev
;
8123 struct imsm_map
*map
;
8124 struct imsm_map
*map2
;
8125 struct mdinfo
*this;
8132 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
8135 dev
= get_imsm_dev(super
, i
);
8136 map
= get_imsm_map(dev
, MAP_0
);
8137 map2
= get_imsm_map(dev
, MAP_1
);
8138 level
= get_imsm_raid_level(map
);
8140 /* do not publish arrays that are in the middle of an
8141 * unsupported migration
8143 if (dev
->vol
.migr_state
&&
8144 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
8145 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
8149 /* do not publish arrays that are not support by controller's
8153 this = xmalloc(sizeof(*this));
8155 super
->current_vol
= i
;
8156 getinfo_super_imsm_volume(st
, this, NULL
);
8158 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
8159 /* mdadm does not support all metadata features- set the bit in all arrays state */
8160 if (!validate_geometry_imsm_orom(super
,
8161 level
, /* RAID level */
8162 imsm_level_to_layout(level
),
8163 map
->num_members
, /* raid disks */
8164 &chunk
, imsm_dev_size(dev
),
8166 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
8168 this->array
.state
|=
8169 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8170 (1<<MD_SB_BLOCK_VOLUME
);
8173 /* if array has bad blocks, set suitable bit in all arrays state */
8175 this->array
.state
|=
8176 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
8177 (1<<MD_SB_BLOCK_VOLUME
);
8179 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
8180 unsigned long long recovery_start
;
8181 struct mdinfo
*info_d
;
8189 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
8190 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
8191 for (d
= super
->disks
; d
; d
= d
->next
)
8192 if (d
->index
== idx
)
8195 recovery_start
= MaxSector
;
8198 if (d
&& is_failed(&d
->disk
))
8200 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
8202 if (!(ord
& IMSM_ORD_REBUILD
))
8203 this->array
.working_disks
++;
8205 * if we skip some disks the array will be assmebled degraded;
8206 * reset resync start to avoid a dirty-degraded
8207 * situation when performing the intial sync
8212 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
8213 if ((!able_to_resync(level
, missing
) ||
8214 recovery_start
== 0))
8215 this->resync_start
= MaxSector
;
8221 info_d
= xcalloc(1, sizeof(*info_d
));
8222 info_d
->next
= this->devs
;
8223 this->devs
= info_d
;
8225 info_d
->disk
.number
= d
->index
;
8226 info_d
->disk
.major
= d
->major
;
8227 info_d
->disk
.minor
= d
->minor
;
8228 info_d
->disk
.raid_disk
= slot
;
8229 info_d
->recovery_start
= recovery_start
;
8231 if (slot
< map2
->num_members
)
8232 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8234 this->array
.spare_disks
++;
8236 if (slot
< map
->num_members
)
8237 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
8239 this->array
.spare_disks
++;
8242 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
8243 info_d
->data_offset
= pba_of_lba0(map
);
8244 info_d
->component_size
= calc_component_size(map
, dev
);
8246 if (map
->raid_level
== 5) {
8247 info_d
->ppl_sector
= this->ppl_sector
;
8248 info_d
->ppl_size
= this->ppl_size
;
8249 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
8250 recovery_start
== 0)
8251 this->resync_start
= 0;
8254 info_d
->bb
.supported
= 1;
8255 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
8256 info_d
->data_offset
,
8257 info_d
->component_size
,
8260 /* now that the disk list is up-to-date fixup recovery_start */
8261 update_recovery_start(super
, dev
, this);
8262 this->array
.spare_disks
+= spare_disks
;
8264 /* check for reshape */
8265 if (this->reshape_active
== 1)
8266 recover_backup_imsm(st
, this);
8270 super
->current_vol
= current_vol
;
8274 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
8275 int failed
, int look_in_map
)
8277 struct imsm_map
*map
;
8279 map
= get_imsm_map(dev
, look_in_map
);
8282 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
8283 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
8285 switch (get_imsm_raid_level(map
)) {
8287 return IMSM_T_STATE_FAILED
;
8290 if (failed
< map
->num_members
)
8291 return IMSM_T_STATE_DEGRADED
;
8293 return IMSM_T_STATE_FAILED
;
8298 * check to see if any mirrors have failed, otherwise we
8299 * are degraded. Even numbered slots are mirrored on
8303 /* gcc -Os complains that this is unused */
8304 int insync
= insync
;
8306 for (i
= 0; i
< map
->num_members
; i
++) {
8307 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8308 int idx
= ord_to_idx(ord
);
8309 struct imsm_disk
*disk
;
8311 /* reset the potential in-sync count on even-numbered
8312 * slots. num_copies is always 2 for imsm raid10
8317 disk
= get_imsm_disk(super
, idx
);
8318 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8321 /* no in-sync disks left in this mirror the
8325 return IMSM_T_STATE_FAILED
;
8328 return IMSM_T_STATE_DEGRADED
;
8332 return IMSM_T_STATE_DEGRADED
;
8334 return IMSM_T_STATE_FAILED
;
8340 return map
->map_state
;
8343 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8348 struct imsm_disk
*disk
;
8349 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8350 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8351 struct imsm_map
*map_for_loop
;
8356 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8357 * disks that are being rebuilt. New failures are recorded to
8358 * map[0]. So we look through all the disks we started with and
8359 * see if any failures are still present, or if any new ones
8363 if (prev
&& (map
->num_members
< prev
->num_members
))
8364 map_for_loop
= prev
;
8366 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8368 /* when MAP_X is passed both maps failures are counted
8371 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8372 i
< prev
->num_members
) {
8373 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8374 idx_1
= ord_to_idx(ord
);
8376 disk
= get_imsm_disk(super
, idx_1
);
8377 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8380 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8381 i
< map
->num_members
) {
8382 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8383 idx
= ord_to_idx(ord
);
8386 disk
= get_imsm_disk(super
, idx
);
8387 if (!disk
|| is_failed(disk
) ||
8388 ord
& IMSM_ORD_REBUILD
)
8397 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8400 struct intel_super
*super
= c
->sb
;
8401 struct imsm_super
*mpb
= super
->anchor
;
8402 struct imsm_update_prealloc_bb_mem u
;
8404 if (inst
>= mpb
->num_raid_devs
) {
8405 pr_err("subarry index %d, out of range\n", inst
);
8409 dprintf("imsm: open_new %d\n", inst
);
8410 a
->info
.container_member
= inst
;
8412 u
.type
= update_prealloc_badblocks_mem
;
8413 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8418 static int is_resyncing(struct imsm_dev
*dev
)
8420 struct imsm_map
*migr_map
;
8422 if (!dev
->vol
.migr_state
)
8425 if (migr_type(dev
) == MIGR_INIT
||
8426 migr_type(dev
) == MIGR_REPAIR
)
8429 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8432 migr_map
= get_imsm_map(dev
, MAP_1
);
8434 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8435 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8441 /* return true if we recorded new information */
8442 static int mark_failure(struct intel_super
*super
,
8443 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8447 struct imsm_map
*map
;
8448 char buf
[MAX_RAID_SERIAL_LEN
+3];
8449 unsigned int len
, shift
= 0;
8451 /* new failures are always set in map[0] */
8452 map
= get_imsm_map(dev
, MAP_0
);
8454 slot
= get_imsm_disk_slot(map
, idx
);
8458 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8459 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8462 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8463 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8465 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8466 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8467 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8469 disk
->status
|= FAILED_DISK
;
8470 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8471 /* mark failures in second map if second map exists and this disk
8473 * This is valid for migration, initialization and rebuild
8475 if (dev
->vol
.migr_state
) {
8476 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8477 int slot2
= get_imsm_disk_slot(map2
, idx
);
8479 if (slot2
< map2
->num_members
&& slot2
>= 0)
8480 set_imsm_ord_tbl_ent(map2
, slot2
,
8481 idx
| IMSM_ORD_REBUILD
);
8483 if (map
->failed_disk_num
== 0xff ||
8484 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8485 map
->failed_disk_num
= slot
;
8487 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8492 static void mark_missing(struct intel_super
*super
,
8493 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8495 mark_failure(super
, dev
, disk
, idx
);
8497 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8500 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8501 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8504 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8508 if (!super
->missing
)
8511 /* When orom adds replacement for missing disk it does
8512 * not remove entry of missing disk, but just updates map with
8513 * new added disk. So it is not enough just to test if there is
8514 * any missing disk, we have to look if there are any failed disks
8515 * in map to stop migration */
8517 dprintf("imsm: mark missing\n");
8518 /* end process for initialization and rebuild only
8520 if (is_gen_migration(dev
) == false) {
8521 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8525 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8526 struct imsm_map
*map1
;
8527 int i
, ord
, ord_map1
;
8530 for (i
= 0; i
< map
->num_members
; i
++) {
8531 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8532 if (!(ord
& IMSM_ORD_REBUILD
))
8535 map1
= get_imsm_map(dev
, MAP_1
);
8539 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8540 if (ord_map1
& IMSM_ORD_REBUILD
)
8545 map_state
= imsm_check_degraded(super
, dev
,
8547 end_migration(dev
, super
, map_state
);
8551 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8552 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8553 super
->updates_pending
++;
8556 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8559 unsigned long long array_blocks
;
8560 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8561 int used_disks
= imsm_num_data_members(map
);
8563 if (used_disks
== 0) {
8564 /* when problems occures
8565 * return current array_blocks value
8567 array_blocks
= imsm_dev_size(dev
);
8569 return array_blocks
;
8572 /* set array size in metadata
8575 /* OLCE size change is caused by added disks
8577 array_blocks
= per_dev_array_size(map
) * used_disks
;
8579 /* Online Volume Size Change
8580 * Using available free space
8582 array_blocks
= new_size
;
8584 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8585 set_imsm_dev_size(dev
, array_blocks
);
8587 return array_blocks
;
8590 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8592 static void imsm_progress_container_reshape(struct intel_super
*super
)
8594 /* if no device has a migr_state, but some device has a
8595 * different number of members than the previous device, start
8596 * changing the number of devices in this device to match
8599 struct imsm_super
*mpb
= super
->anchor
;
8600 int prev_disks
= -1;
8604 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8605 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8606 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8607 struct imsm_map
*map2
;
8608 int prev_num_members
;
8610 if (dev
->vol
.migr_state
)
8613 if (prev_disks
== -1)
8614 prev_disks
= map
->num_members
;
8615 if (prev_disks
== map
->num_members
)
8618 /* OK, this array needs to enter reshape mode.
8619 * i.e it needs a migr_state
8622 copy_map_size
= sizeof_imsm_map(map
);
8623 prev_num_members
= map
->num_members
;
8624 map
->num_members
= prev_disks
;
8625 dev
->vol
.migr_state
= 1;
8626 set_vol_curr_migr_unit(dev
, 0);
8627 set_migr_type(dev
, MIGR_GEN_MIGR
);
8628 for (i
= prev_num_members
;
8629 i
< map
->num_members
; i
++)
8630 set_imsm_ord_tbl_ent(map
, i
, i
);
8631 map2
= get_imsm_map(dev
, MAP_1
);
8632 /* Copy the current map */
8633 memcpy(map2
, map
, copy_map_size
);
8634 map2
->num_members
= prev_num_members
;
8636 imsm_set_array_size(dev
, -1);
8637 super
->clean_migration_record_by_mdmon
= 1;
8638 super
->updates_pending
++;
8642 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8643 * states are handled in imsm_set_disk() with one exception, when a
8644 * resync is stopped due to a new failure this routine will set the
8645 * 'degraded' state for the array.
8647 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8649 int inst
= a
->info
.container_member
;
8650 struct intel_super
*super
= a
->container
->sb
;
8651 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8652 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8653 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8654 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8655 __u32 blocks_per_unit
;
8657 if (dev
->vol
.migr_state
&&
8658 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8659 /* array state change is blocked due to reshape action
8661 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8662 * - finish the reshape (if last_checkpoint is big and action != reshape)
8663 * - update vol_curr_migr_unit
8665 if (a
->curr_action
== reshape
) {
8666 /* still reshaping, maybe update vol_curr_migr_unit */
8667 goto mark_checkpoint
;
8669 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8670 unsigned long long array_blocks
;
8674 used_disks
= imsm_num_data_members(map
);
8675 if (used_disks
> 0) {
8677 per_dev_array_size(map
) *
8680 round_size_to_mb(array_blocks
,
8682 a
->info
.custom_array_size
= array_blocks
;
8683 /* encourage manager to update array
8687 a
->check_reshape
= 1;
8689 /* finalize online capacity expansion/reshape */
8690 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8692 mdi
->disk
.raid_disk
,
8695 imsm_progress_container_reshape(super
);
8700 /* before we activate this array handle any missing disks */
8701 if (consistent
== 2)
8702 handle_missing(super
, dev
);
8704 if (consistent
== 2 &&
8705 (!is_resync_complete(&a
->info
) ||
8706 map_state
!= IMSM_T_STATE_NORMAL
||
8707 dev
->vol
.migr_state
))
8710 if (is_resync_complete(&a
->info
)) {
8711 /* complete intialization / resync,
8712 * recovery and interrupted recovery is completed in
8715 if (is_resyncing(dev
)) {
8716 dprintf("imsm: mark resync done\n");
8717 end_migration(dev
, super
, map_state
);
8718 super
->updates_pending
++;
8719 a
->last_checkpoint
= 0;
8721 } else if ((!is_resyncing(dev
) && !failed
) &&
8722 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8723 /* mark the start of the init process if nothing is failed */
8724 dprintf("imsm: mark resync start\n");
8725 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8726 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8728 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8729 super
->updates_pending
++;
8732 if (a
->prev_action
== idle
)
8733 goto skip_mark_checkpoint
;
8736 /* skip checkpointing for general migration,
8737 * it is controlled in mdadm
8739 if (is_gen_migration(dev
))
8740 goto skip_mark_checkpoint
;
8742 /* check if we can update vol_curr_migr_unit from resync_start,
8745 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8746 if (blocks_per_unit
) {
8747 set_vol_curr_migr_unit(dev
,
8748 a
->last_checkpoint
/ blocks_per_unit
);
8749 dprintf("imsm: mark checkpoint (%llu)\n",
8750 vol_curr_migr_unit(dev
));
8751 super
->updates_pending
++;
8754 skip_mark_checkpoint
:
8755 /* mark dirty / clean */
8756 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8757 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8758 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8760 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8762 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8763 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8764 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8765 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8767 super
->updates_pending
++;
8773 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8775 int inst
= a
->info
.container_member
;
8776 struct intel_super
*super
= a
->container
->sb
;
8777 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8778 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8780 if (slot
> map
->num_members
) {
8781 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8782 slot
, map
->num_members
- 1);
8789 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8792 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8794 int inst
= a
->info
.container_member
;
8795 struct intel_super
*super
= a
->container
->sb
;
8796 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8797 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8798 struct imsm_disk
*disk
;
8800 int recovery_not_finished
= 0;
8804 int rebuild_done
= 0;
8807 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8811 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8812 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8814 /* check for new failures */
8815 if (disk
&& (state
& DS_FAULTY
)) {
8816 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8817 super
->updates_pending
++;
8820 /* check if in_sync */
8821 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8822 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8824 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8826 super
->updates_pending
++;
8829 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8830 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8832 /* check if recovery complete, newly degraded, or failed */
8833 dprintf("imsm: Detected transition to state ");
8834 switch (map_state
) {
8835 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8836 dprintf("normal: ");
8837 if (is_rebuilding(dev
)) {
8838 dprintf_cont("while rebuilding");
8839 /* check if recovery is really finished */
8840 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8841 if (mdi
->recovery_start
!= MaxSector
) {
8842 recovery_not_finished
= 1;
8845 if (recovery_not_finished
) {
8847 dprintf("Rebuild has not finished yet, state not changed");
8848 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8849 a
->last_checkpoint
= mdi
->recovery_start
;
8850 super
->updates_pending
++;
8854 end_migration(dev
, super
, map_state
);
8855 map
->failed_disk_num
= ~0;
8856 super
->updates_pending
++;
8857 a
->last_checkpoint
= 0;
8860 if (is_gen_migration(dev
)) {
8861 dprintf_cont("while general migration");
8862 if (a
->last_checkpoint
>= a
->info
.component_size
)
8863 end_migration(dev
, super
, map_state
);
8865 map
->map_state
= map_state
;
8866 map
->failed_disk_num
= ~0;
8867 super
->updates_pending
++;
8871 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8872 dprintf_cont("degraded: ");
8873 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8874 dprintf_cont("mark degraded");
8875 map
->map_state
= map_state
;
8876 super
->updates_pending
++;
8877 a
->last_checkpoint
= 0;
8880 if (is_rebuilding(dev
)) {
8881 dprintf_cont("while rebuilding ");
8882 if (state
& DS_FAULTY
) {
8883 dprintf_cont("removing failed drive ");
8884 if (n
== map
->failed_disk_num
) {
8885 dprintf_cont("end migration");
8886 end_migration(dev
, super
, map_state
);
8887 a
->last_checkpoint
= 0;
8889 dprintf_cont("fail detected during rebuild, changing map state");
8890 map
->map_state
= map_state
;
8892 super
->updates_pending
++;
8898 /* check if recovery is really finished */
8899 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8900 if (mdi
->recovery_start
!= MaxSector
) {
8901 recovery_not_finished
= 1;
8904 if (recovery_not_finished
) {
8906 dprintf_cont("Rebuild has not finished yet");
8907 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8908 a
->last_checkpoint
=
8909 mdi
->recovery_start
;
8910 super
->updates_pending
++;
8915 dprintf_cont(" Rebuild done, still degraded");
8916 end_migration(dev
, super
, map_state
);
8917 a
->last_checkpoint
= 0;
8918 super
->updates_pending
++;
8920 for (i
= 0; i
< map
->num_members
; i
++) {
8921 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8923 if (idx
& IMSM_ORD_REBUILD
)
8924 map
->failed_disk_num
= i
;
8926 super
->updates_pending
++;
8929 if (is_gen_migration(dev
)) {
8930 dprintf_cont("while general migration");
8931 if (a
->last_checkpoint
>= a
->info
.component_size
)
8932 end_migration(dev
, super
, map_state
);
8934 map
->map_state
= map_state
;
8935 manage_second_map(super
, dev
);
8937 super
->updates_pending
++;
8940 if (is_initializing(dev
)) {
8941 dprintf_cont("while initialization.");
8942 map
->map_state
= map_state
;
8943 super
->updates_pending
++;
8947 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8948 dprintf_cont("failed: ");
8949 if (is_gen_migration(dev
)) {
8950 dprintf_cont("while general migration");
8951 map
->map_state
= map_state
;
8952 super
->updates_pending
++;
8955 if (map
->map_state
!= map_state
) {
8956 dprintf_cont("mark failed");
8957 end_migration(dev
, super
, map_state
);
8958 super
->updates_pending
++;
8959 a
->last_checkpoint
= 0;
8964 dprintf_cont("state %i\n", map_state
);
8969 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8972 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8973 unsigned long long dsize
;
8974 unsigned long long sectors
;
8975 unsigned int sector_size
;
8977 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
8979 get_dev_size(fd
, NULL
, &dsize
);
8981 if (mpb_size
> sector_size
) {
8982 /* -1 to account for anchor */
8983 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8985 /* write the extended mpb to the sectors preceeding the anchor */
8986 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8990 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8991 sector_size
* sectors
) != sector_size
* sectors
)
8995 /* first block is stored on second to last sector of the disk */
8996 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8999 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
9005 static void imsm_sync_metadata(struct supertype
*container
)
9007 struct intel_super
*super
= container
->sb
;
9009 dprintf("sync metadata: %d\n", super
->updates_pending
);
9010 if (!super
->updates_pending
)
9013 write_super_imsm(container
, 0);
9015 super
->updates_pending
= 0;
9018 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
9020 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
9021 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
9024 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9028 if (dl
&& is_failed(&dl
->disk
))
9032 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
9037 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
9038 struct active_array
*a
, int activate_new
,
9039 struct mdinfo
*additional_test_list
)
9041 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
9042 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
9043 struct imsm_super
*mpb
= super
->anchor
;
9044 struct imsm_map
*map
;
9045 unsigned long long pos
;
9050 __u32 array_start
= 0;
9051 __u32 array_end
= 0;
9053 struct mdinfo
*test_list
;
9055 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9056 /* If in this array, skip */
9057 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9058 if (is_fd_valid(d
->state_fd
) &&
9059 d
->disk
.major
== dl
->major
&&
9060 d
->disk
.minor
== dl
->minor
) {
9061 dprintf("%x:%x already in array\n",
9062 dl
->major
, dl
->minor
);
9067 test_list
= additional_test_list
;
9069 if (test_list
->disk
.major
== dl
->major
&&
9070 test_list
->disk
.minor
== dl
->minor
) {
9071 dprintf("%x:%x already in additional test list\n",
9072 dl
->major
, dl
->minor
);
9075 test_list
= test_list
->next
;
9080 /* skip in use or failed drives */
9081 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
9083 dprintf("%x:%x status (failed: %d index: %d)\n",
9084 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
9088 /* skip pure spares when we are looking for partially
9089 * assimilated drives
9091 if (dl
->index
== -1 && !activate_new
)
9094 if (!drive_validate_sector_size(super
, dl
))
9097 /* Does this unused device have the requisite free space?
9098 * It needs to be able to cover all member volumes
9100 ex
= get_extents(super
, dl
, 1);
9102 dprintf("cannot get extents\n");
9105 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9106 dev
= get_imsm_dev(super
, i
);
9107 map
= get_imsm_map(dev
, MAP_0
);
9109 /* check if this disk is already a member of
9112 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
9118 array_start
= pba_of_lba0(map
);
9119 array_end
= array_start
+
9120 per_dev_array_size(map
) - 1;
9123 /* check that we can start at pba_of_lba0 with
9124 * num_data_stripes*blocks_per_stripe of space
9126 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
9130 pos
= ex
[j
].start
+ ex
[j
].size
;
9132 } while (ex
[j
-1].size
);
9139 if (i
< mpb
->num_raid_devs
) {
9140 dprintf("%x:%x does not have %u to %u available\n",
9141 dl
->major
, dl
->minor
, array_start
, array_end
);
9151 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
9153 struct imsm_dev
*dev2
;
9154 struct imsm_map
*map
;
9160 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
9162 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
9163 if (state
== IMSM_T_STATE_FAILED
) {
9164 map
= get_imsm_map(dev2
, MAP_0
);
9165 for (slot
= 0; slot
< map
->num_members
; slot
++) {
9167 * Check if failed disks are deleted from intel
9168 * disk list or are marked to be deleted
9170 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
9171 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
9173 * Do not rebuild the array if failed disks
9174 * from failed sub-array are not removed from
9178 is_failed(&idisk
->disk
) &&
9179 (idisk
->action
!= DISK_REMOVE
))
9186 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
9187 struct metadata_update
**updates
)
9190 * Find a device with unused free space and use it to replace a
9191 * failed/vacant region in an array. We replace failed regions one a
9192 * array at a time. The result is that a new spare disk will be added
9193 * to the first failed array and after the monitor has finished
9194 * propagating failures the remainder will be consumed.
9196 * FIXME add a capability for mdmon to request spares from another
9200 struct intel_super
*super
= a
->container
->sb
;
9201 int inst
= a
->info
.container_member
;
9202 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
9203 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9204 int failed
= a
->info
.array
.raid_disks
;
9205 struct mdinfo
*rv
= NULL
;
9208 struct metadata_update
*mu
;
9210 struct imsm_update_activate_spare
*u
;
9215 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
9216 if (!is_fd_valid(d
->state_fd
))
9219 if (d
->curr_state
& DS_FAULTY
)
9220 /* wait for Removal to happen */
9226 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
9227 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
9229 if (imsm_reshape_blocks_arrays_changes(super
))
9232 /* Cannot activate another spare if rebuild is in progress already
9234 if (is_rebuilding(dev
)) {
9235 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
9239 if (a
->info
.array
.level
== 4)
9240 /* No repair for takeovered array
9241 * imsm doesn't support raid4
9245 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
9246 IMSM_T_STATE_DEGRADED
)
9249 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
9250 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
9255 * If there are any failed disks check state of the other volume.
9256 * Block rebuild if the another one is failed until failed disks
9257 * are removed from container.
9260 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
9261 MAX_RAID_SERIAL_LEN
, dev
->volume
);
9262 /* check if states of the other volumes allow for rebuild */
9263 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
9265 allowed
= imsm_rebuild_allowed(a
->container
,
9273 /* For each slot, if it is not working, find a spare */
9274 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
9275 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9276 if (d
->disk
.raid_disk
== i
)
9278 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
9279 if (d
&& is_fd_valid(d
->state_fd
))
9283 * OK, this device needs recovery. Try to re-add the
9284 * previous occupant of this slot, if this fails see if
9285 * we can continue the assimilation of a spare that was
9286 * partially assimilated, finally try to activate a new
9289 dl
= imsm_readd(super
, i
, a
);
9291 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9293 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9297 /* found a usable disk with enough space */
9298 di
= xcalloc(1, sizeof(*di
));
9300 /* dl->index will be -1 in the case we are activating a
9301 * pristine spare. imsm_process_update() will create a
9302 * new index in this case. Once a disk is found to be
9303 * failed in all member arrays it is kicked from the
9306 di
->disk
.number
= dl
->index
;
9308 /* (ab)use di->devs to store a pointer to the device
9311 di
->devs
= (struct mdinfo
*) dl
;
9313 di
->disk
.raid_disk
= i
;
9314 di
->disk
.major
= dl
->major
;
9315 di
->disk
.minor
= dl
->minor
;
9317 di
->recovery_start
= 0;
9318 di
->data_offset
= pba_of_lba0(map
);
9319 di
->component_size
= a
->info
.component_size
;
9320 di
->container_member
= inst
;
9321 di
->bb
.supported
= 1;
9322 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9323 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9324 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9326 super
->random
= random32();
9330 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9331 i
, di
->data_offset
);
9335 /* No spares found */
9337 /* Now 'rv' has a list of devices to return.
9338 * Create a metadata_update record to update the
9339 * disk_ord_tbl for the array
9341 mu
= xmalloc(sizeof(*mu
));
9342 mu
->buf
= xcalloc(num_spares
,
9343 sizeof(struct imsm_update_activate_spare
));
9345 mu
->space_list
= NULL
;
9346 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9347 mu
->next
= *updates
;
9348 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9350 for (di
= rv
; di
; di
= di
->next
) {
9351 u
->type
= update_activate_spare
;
9352 u
->dl
= (struct dl
*) di
->devs
;
9354 u
->slot
= di
->disk
.raid_disk
;
9365 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9367 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9368 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9369 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9370 struct disk_info
*inf
= get_disk_info(u
);
9371 struct imsm_disk
*disk
;
9375 for (i
= 0; i
< map
->num_members
; i
++) {
9376 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9377 for (j
= 0; j
< new_map
->num_members
; j
++)
9378 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9385 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9389 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9390 if (dl
->major
== major
&& dl
->minor
== minor
)
9395 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9401 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9402 if (dl
->major
== major
&& dl
->minor
== minor
) {
9405 prev
->next
= dl
->next
;
9407 super
->disks
= dl
->next
;
9409 __free_imsm_disk(dl
, 1);
9410 dprintf("removed %x:%x\n", major
, minor
);
9418 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9420 static int add_remove_disk_update(struct intel_super
*super
)
9422 int check_degraded
= 0;
9425 /* add/remove some spares to/from the metadata/contrainer */
9426 while (super
->disk_mgmt_list
) {
9427 struct dl
*disk_cfg
;
9429 disk_cfg
= super
->disk_mgmt_list
;
9430 super
->disk_mgmt_list
= disk_cfg
->next
;
9431 disk_cfg
->next
= NULL
;
9433 if (disk_cfg
->action
== DISK_ADD
) {
9434 disk_cfg
->next
= super
->disks
;
9435 super
->disks
= disk_cfg
;
9437 dprintf("added %x:%x\n",
9438 disk_cfg
->major
, disk_cfg
->minor
);
9439 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9440 dprintf("Disk remove action processed: %x.%x\n",
9441 disk_cfg
->major
, disk_cfg
->minor
);
9442 disk
= get_disk_super(super
,
9446 /* store action status */
9447 disk
->action
= DISK_REMOVE
;
9448 /* remove spare disks only */
9449 if (disk
->index
== -1) {
9450 remove_disk_super(super
,
9454 disk_cfg
->fd
= disk
->fd
;
9458 /* release allocate disk structure */
9459 __free_imsm_disk(disk_cfg
, 1);
9462 return check_degraded
;
9465 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9466 struct intel_super
*super
,
9469 struct intel_dev
*id
;
9470 void **tofree
= NULL
;
9473 dprintf("(enter)\n");
9474 if (u
->subdev
< 0 || u
->subdev
> 1) {
9475 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9478 if (space_list
== NULL
|| *space_list
== NULL
) {
9479 dprintf("imsm: Error: Memory is not allocated\n");
9483 for (id
= super
->devlist
; id
; id
= id
->next
) {
9484 if (id
->index
== (unsigned)u
->subdev
) {
9485 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9486 struct imsm_map
*map
;
9487 struct imsm_dev
*new_dev
=
9488 (struct imsm_dev
*)*space_list
;
9489 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9491 struct dl
*new_disk
;
9493 if (new_dev
== NULL
)
9495 *space_list
= **space_list
;
9496 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9497 map
= get_imsm_map(new_dev
, MAP_0
);
9499 dprintf("imsm: Error: migration in progress");
9503 to_state
= map
->map_state
;
9504 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9506 /* this should not happen */
9507 if (u
->new_disks
[0] < 0) {
9508 map
->failed_disk_num
=
9509 map
->num_members
- 1;
9510 to_state
= IMSM_T_STATE_DEGRADED
;
9512 to_state
= IMSM_T_STATE_NORMAL
;
9514 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9515 if (u
->new_level
> -1)
9516 map
->raid_level
= u
->new_level
;
9517 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9518 if ((u
->new_level
== 5) &&
9519 (migr_map
->raid_level
== 0)) {
9520 int ord
= map
->num_members
- 1;
9521 migr_map
->num_members
--;
9522 if (u
->new_disks
[0] < 0)
9523 ord
|= IMSM_ORD_REBUILD
;
9524 set_imsm_ord_tbl_ent(map
,
9525 map
->num_members
- 1,
9529 tofree
= (void **)dev
;
9531 /* update chunk size
9533 if (u
->new_chunksize
> 0) {
9534 struct imsm_map
*dest_map
=
9535 get_imsm_map(dev
, MAP_0
);
9537 imsm_num_data_members(dest_map
);
9539 if (used_disks
== 0)
9542 map
->blocks_per_strip
=
9543 __cpu_to_le16(u
->new_chunksize
* 2);
9544 update_num_data_stripes(map
, imsm_dev_size(dev
));
9547 /* ensure blocks_per_member has valid value
9549 set_blocks_per_member(map
,
9550 per_dev_array_size(map
) +
9551 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9555 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9556 migr_map
->raid_level
== map
->raid_level
)
9559 if (u
->new_disks
[0] >= 0) {
9562 new_disk
= get_disk_super(super
,
9563 major(u
->new_disks
[0]),
9564 minor(u
->new_disks
[0]));
9565 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9566 major(u
->new_disks
[0]),
9567 minor(u
->new_disks
[0]),
9568 new_disk
, new_disk
->index
);
9569 if (new_disk
== NULL
)
9570 goto error_disk_add
;
9572 new_disk
->index
= map
->num_members
- 1;
9573 /* slot to fill in autolayout
9575 new_disk
->raiddisk
= new_disk
->index
;
9576 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9577 new_disk
->disk
.status
&= ~SPARE_DISK
;
9579 goto error_disk_add
;
9582 *tofree
= *space_list
;
9583 /* calculate new size
9585 imsm_set_array_size(new_dev
, -1);
9592 *space_list
= tofree
;
9596 dprintf("Error: imsm: Cannot find disk.\n");
9600 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9601 struct intel_super
*super
)
9603 struct intel_dev
*id
;
9606 dprintf("(enter)\n");
9607 if (u
->subdev
< 0 || u
->subdev
> 1) {
9608 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9612 for (id
= super
->devlist
; id
; id
= id
->next
) {
9613 if (id
->index
== (unsigned)u
->subdev
) {
9614 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9615 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9616 int used_disks
= imsm_num_data_members(map
);
9617 unsigned long long blocks_per_member
;
9618 unsigned long long new_size_per_disk
;
9620 if (used_disks
== 0)
9623 /* calculate new size
9625 new_size_per_disk
= u
->new_size
/ used_disks
;
9626 blocks_per_member
= new_size_per_disk
+
9627 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9629 imsm_set_array_size(dev
, u
->new_size
);
9630 set_blocks_per_member(map
, blocks_per_member
);
9631 update_num_data_stripes(map
, u
->new_size
);
9640 static int prepare_spare_to_activate(struct supertype
*st
,
9641 struct imsm_update_activate_spare
*u
)
9643 struct intel_super
*super
= st
->sb
;
9644 int prev_current_vol
= super
->current_vol
;
9645 struct active_array
*a
;
9648 for (a
= st
->arrays
; a
; a
= a
->next
)
9650 * Additional initialization (adding bitmap header, filling
9651 * the bitmap area with '1's to force initial rebuild for a whole
9652 * data-area) is required when adding the spare to the volume
9653 * with write-intent bitmap.
9655 if (a
->info
.container_member
== u
->array
&&
9656 a
->info
.consistency_policy
== CONSISTENCY_POLICY_BITMAP
) {
9659 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9665 super
->current_vol
= u
->array
;
9666 if (st
->ss
->write_bitmap(st
, dl
->fd
, NoUpdate
))
9668 super
->current_vol
= prev_current_vol
;
9673 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9674 struct intel_super
*super
,
9675 struct active_array
*active_array
)
9677 struct imsm_super
*mpb
= super
->anchor
;
9678 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9679 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9680 struct imsm_map
*migr_map
;
9681 struct active_array
*a
;
9682 struct imsm_disk
*disk
;
9689 int second_map_created
= 0;
9691 for (; u
; u
= u
->next
) {
9692 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9697 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9702 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9707 /* count failures (excluding rebuilds and the victim)
9708 * to determine map[0] state
9711 for (i
= 0; i
< map
->num_members
; i
++) {
9714 disk
= get_imsm_disk(super
,
9715 get_imsm_disk_idx(dev
, i
, MAP_X
));
9716 if (!disk
|| is_failed(disk
))
9720 /* adding a pristine spare, assign a new index */
9721 if (dl
->index
< 0) {
9722 dl
->index
= super
->anchor
->num_disks
;
9723 super
->anchor
->num_disks
++;
9726 disk
->status
|= CONFIGURED_DISK
;
9727 disk
->status
&= ~SPARE_DISK
;
9730 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9731 if (!second_map_created
) {
9732 second_map_created
= 1;
9733 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9734 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9736 map
->map_state
= to_state
;
9737 migr_map
= get_imsm_map(dev
, MAP_1
);
9738 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9739 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9740 dl
->index
| IMSM_ORD_REBUILD
);
9742 /* update the family_num to mark a new container
9743 * generation, being careful to record the existing
9744 * family_num in orig_family_num to clean up after
9745 * earlier mdadm versions that neglected to set it.
9747 if (mpb
->orig_family_num
== 0)
9748 mpb
->orig_family_num
= mpb
->family_num
;
9749 mpb
->family_num
+= super
->random
;
9751 /* count arrays using the victim in the metadata */
9753 for (a
= active_array
; a
; a
= a
->next
) {
9754 int dev_idx
= a
->info
.container_member
;
9756 if (get_disk_slot_in_dev(super
, dev_idx
, victim
) >= 0)
9760 /* delete the victim if it is no longer being
9766 /* We know that 'manager' isn't touching anything,
9767 * so it is safe to delete
9769 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9770 if ((*dlp
)->index
== victim
)
9773 /* victim may be on the missing list */
9775 for (dlp
= &super
->missing
; *dlp
;
9776 dlp
= &(*dlp
)->next
)
9777 if ((*dlp
)->index
== victim
)
9779 imsm_delete(super
, dlp
, victim
);
9786 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9787 struct intel_super
*super
,
9790 struct dl
*new_disk
;
9791 struct intel_dev
*id
;
9793 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9794 int disk_count
= u
->old_raid_disks
;
9795 void **tofree
= NULL
;
9796 int devices_to_reshape
= 1;
9797 struct imsm_super
*mpb
= super
->anchor
;
9799 unsigned int dev_id
;
9801 dprintf("(enter)\n");
9803 /* enable spares to use in array */
9804 for (i
= 0; i
< delta_disks
; i
++) {
9805 new_disk
= get_disk_super(super
,
9806 major(u
->new_disks
[i
]),
9807 minor(u
->new_disks
[i
]));
9808 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9809 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9810 new_disk
, new_disk
->index
);
9811 if (new_disk
== NULL
||
9812 (new_disk
->index
>= 0 &&
9813 new_disk
->index
< u
->old_raid_disks
))
9814 goto update_reshape_exit
;
9815 new_disk
->index
= disk_count
++;
9816 /* slot to fill in autolayout
9818 new_disk
->raiddisk
= new_disk
->index
;
9819 new_disk
->disk
.status
|=
9821 new_disk
->disk
.status
&= ~SPARE_DISK
;
9824 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9825 mpb
->num_raid_devs
);
9826 /* manage changes in volume
9828 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9829 void **sp
= *space_list
;
9830 struct imsm_dev
*newdev
;
9831 struct imsm_map
*newmap
, *oldmap
;
9833 for (id
= super
->devlist
; id
; id
= id
->next
) {
9834 if (id
->index
== dev_id
)
9843 /* Copy the dev, but not (all of) the map */
9844 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9845 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9846 newmap
= get_imsm_map(newdev
, MAP_0
);
9847 /* Copy the current map */
9848 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9849 /* update one device only
9851 if (devices_to_reshape
) {
9852 dprintf("imsm: modifying subdev: %i\n",
9854 devices_to_reshape
--;
9855 newdev
->vol
.migr_state
= 1;
9856 set_vol_curr_migr_unit(newdev
, 0);
9857 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9858 newmap
->num_members
= u
->new_raid_disks
;
9859 for (i
= 0; i
< delta_disks
; i
++) {
9860 set_imsm_ord_tbl_ent(newmap
,
9861 u
->old_raid_disks
+ i
,
9862 u
->old_raid_disks
+ i
);
9864 /* New map is correct, now need to save old map
9866 newmap
= get_imsm_map(newdev
, MAP_1
);
9867 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9869 imsm_set_array_size(newdev
, -1);
9872 sp
= (void **)id
->dev
;
9877 /* Clear migration record */
9878 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9881 *space_list
= tofree
;
9884 update_reshape_exit
:
9889 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9890 struct intel_super
*super
,
9893 struct imsm_dev
*dev
= NULL
;
9894 struct intel_dev
*dv
;
9895 struct imsm_dev
*dev_new
;
9896 struct imsm_map
*map
;
9900 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9901 if (dv
->index
== (unsigned int)u
->subarray
) {
9909 map
= get_imsm_map(dev
, MAP_0
);
9911 if (u
->direction
== R10_TO_R0
) {
9912 /* Number of failed disks must be half of initial disk number */
9913 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9914 (map
->num_members
/ 2))
9917 /* iterate through devices to mark removed disks as spare */
9918 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9919 if (dm
->disk
.status
& FAILED_DISK
) {
9920 int idx
= dm
->index
;
9921 /* update indexes on the disk list */
9922 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9923 the index values will end up being correct.... NB */
9924 for (du
= super
->disks
; du
; du
= du
->next
)
9925 if (du
->index
> idx
)
9927 /* mark as spare disk */
9932 map
->num_members
/= map
->num_domains
;
9933 map
->map_state
= IMSM_T_STATE_NORMAL
;
9934 map
->raid_level
= 0;
9935 set_num_domains(map
);
9936 update_num_data_stripes(map
, imsm_dev_size(dev
));
9937 map
->failed_disk_num
= -1;
9940 if (u
->direction
== R0_TO_R10
) {
9943 /* update slots in current disk list */
9944 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9948 /* create new *missing* disks */
9949 for (i
= 0; i
< map
->num_members
; i
++) {
9950 space
= *space_list
;
9953 *space_list
= *space
;
9955 memcpy(du
, super
->disks
, sizeof(*du
));
9959 du
->index
= (i
* 2) + 1;
9960 sprintf((char *)du
->disk
.serial
,
9961 " MISSING_%d", du
->index
);
9962 sprintf((char *)du
->serial
,
9963 "MISSING_%d", du
->index
);
9964 du
->next
= super
->missing
;
9965 super
->missing
= du
;
9967 /* create new dev and map */
9968 space
= *space_list
;
9971 *space_list
= *space
;
9972 dev_new
= (void *)space
;
9973 memcpy(dev_new
, dev
, sizeof(*dev
));
9974 /* update new map */
9975 map
= get_imsm_map(dev_new
, MAP_0
);
9977 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9978 map
->raid_level
= 1;
9979 set_num_domains(map
);
9980 map
->num_members
= map
->num_members
* map
->num_domains
;
9981 update_num_data_stripes(map
, imsm_dev_size(dev
));
9983 /* replace dev<->dev_new */
9986 /* update disk order table */
9987 for (du
= super
->disks
; du
; du
= du
->next
)
9989 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9990 for (du
= super
->missing
; du
; du
= du
->next
)
9991 if (du
->index
>= 0) {
9992 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9993 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9999 static void imsm_process_update(struct supertype
*st
,
10000 struct metadata_update
*update
)
10003 * crack open the metadata_update envelope to find the update record
10004 * update can be one of:
10005 * update_reshape_container_disks - all the arrays in the container
10006 * are being reshaped to have more devices. We need to mark
10007 * the arrays for general migration and convert selected spares
10008 * into active devices.
10009 * update_activate_spare - a spare device has replaced a failed
10010 * device in an array, update the disk_ord_tbl. If this disk is
10011 * present in all member arrays then also clear the SPARE_DISK
10013 * update_create_array
10014 * update_kill_array
10015 * update_rename_array
10016 * update_add_remove_disk
10018 struct intel_super
*super
= st
->sb
;
10019 struct imsm_super
*mpb
;
10020 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
10022 /* update requires a larger buf but the allocation failed */
10023 if (super
->next_len
&& !super
->next_buf
) {
10024 super
->next_len
= 0;
10028 if (super
->next_buf
) {
10029 memcpy(super
->next_buf
, super
->buf
, super
->len
);
10031 super
->len
= super
->next_len
;
10032 super
->buf
= super
->next_buf
;
10034 super
->next_len
= 0;
10035 super
->next_buf
= NULL
;
10038 mpb
= super
->anchor
;
10041 case update_general_migration_checkpoint
: {
10042 struct intel_dev
*id
;
10043 struct imsm_update_general_migration_checkpoint
*u
=
10044 (void *)update
->buf
;
10046 dprintf("called for update_general_migration_checkpoint\n");
10048 /* find device under general migration */
10049 for (id
= super
->devlist
; id
; id
= id
->next
) {
10050 if (is_gen_migration(id
->dev
)) {
10051 set_vol_curr_migr_unit(id
->dev
,
10052 u
->curr_migr_unit
);
10053 super
->updates_pending
++;
10058 case update_takeover
: {
10059 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10060 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
10061 imsm_update_version_info(super
);
10062 super
->updates_pending
++;
10067 case update_reshape_container_disks
: {
10068 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10069 if (apply_reshape_container_disks_update(
10070 u
, super
, &update
->space_list
))
10071 super
->updates_pending
++;
10074 case update_reshape_migration
: {
10075 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10076 if (apply_reshape_migration_update(
10077 u
, super
, &update
->space_list
))
10078 super
->updates_pending
++;
10081 case update_size_change
: {
10082 struct imsm_update_size_change
*u
= (void *)update
->buf
;
10083 if (apply_size_change_update(u
, super
))
10084 super
->updates_pending
++;
10087 case update_activate_spare
: {
10088 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
10090 if (prepare_spare_to_activate(st
, u
) &&
10091 apply_update_activate_spare(u
, super
, st
->arrays
))
10092 super
->updates_pending
++;
10095 case update_create_array
: {
10096 /* someone wants to create a new array, we need to be aware of
10097 * a few races/collisions:
10098 * 1/ 'Create' called by two separate instances of mdadm
10099 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
10100 * devices that have since been assimilated via
10102 * In the event this update can not be carried out mdadm will
10103 * (FIX ME) notice that its update did not take hold.
10105 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10106 struct intel_dev
*dv
;
10107 struct imsm_dev
*dev
;
10108 struct imsm_map
*map
, *new_map
;
10109 unsigned long long start
, end
;
10110 unsigned long long new_start
, new_end
;
10112 struct disk_info
*inf
;
10115 /* handle racing creates: first come first serve */
10116 if (u
->dev_idx
< mpb
->num_raid_devs
) {
10117 dprintf("subarray %d already defined\n", u
->dev_idx
);
10121 /* check update is next in sequence */
10122 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
10123 dprintf("can not create array %d expected index %d\n",
10124 u
->dev_idx
, mpb
->num_raid_devs
);
10128 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
10129 new_start
= pba_of_lba0(new_map
);
10130 new_end
= new_start
+ per_dev_array_size(new_map
);
10131 inf
= get_disk_info(u
);
10133 /* handle activate_spare versus create race:
10134 * check to make sure that overlapping arrays do not include
10135 * overalpping disks
10137 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10138 dev
= get_imsm_dev(super
, i
);
10139 map
= get_imsm_map(dev
, MAP_0
);
10140 start
= pba_of_lba0(map
);
10141 end
= start
+ per_dev_array_size(map
);
10142 if ((new_start
>= start
&& new_start
<= end
) ||
10143 (start
>= new_start
&& start
<= new_end
))
10148 if (disks_overlap(super
, i
, u
)) {
10149 dprintf("arrays overlap\n");
10154 /* check that prepare update was successful */
10155 if (!update
->space
) {
10156 dprintf("prepare update failed\n");
10160 /* check that all disks are still active before committing
10161 * changes. FIXME: could we instead handle this by creating a
10162 * degraded array? That's probably not what the user expects,
10163 * so better to drop this update on the floor.
10165 for (i
= 0; i
< new_map
->num_members
; i
++) {
10166 dl
= serial_to_dl(inf
[i
].serial
, super
);
10168 dprintf("disk disappeared\n");
10173 super
->updates_pending
++;
10175 /* convert spares to members and fixup ord_tbl */
10176 for (i
= 0; i
< new_map
->num_members
; i
++) {
10177 dl
= serial_to_dl(inf
[i
].serial
, super
);
10178 if (dl
->index
== -1) {
10179 dl
->index
= mpb
->num_disks
;
10181 dl
->disk
.status
|= CONFIGURED_DISK
;
10182 dl
->disk
.status
&= ~SPARE_DISK
;
10184 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
10187 dv
= update
->space
;
10189 update
->space
= NULL
;
10190 imsm_copy_dev(dev
, &u
->dev
);
10191 dv
->index
= u
->dev_idx
;
10192 dv
->next
= super
->devlist
;
10193 super
->devlist
= dv
;
10194 mpb
->num_raid_devs
++;
10196 imsm_update_version_info(super
);
10199 /* mdmon knows how to release update->space, but not
10200 * ((struct intel_dev *) update->space)->dev
10202 if (update
->space
) {
10203 dv
= update
->space
;
10208 case update_kill_array
: {
10209 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
10210 int victim
= u
->dev_idx
;
10211 struct active_array
*a
;
10212 struct intel_dev
**dp
;
10214 /* sanity check that we are not affecting the uuid of
10215 * active arrays, or deleting an active array
10217 * FIXME when immutable ids are available, but note that
10218 * we'll also need to fixup the invalidated/active
10219 * subarray indexes in mdstat
10221 for (a
= st
->arrays
; a
; a
= a
->next
)
10222 if (a
->info
.container_member
>= victim
)
10224 /* by definition if mdmon is running at least one array
10225 * is active in the container, so checking
10226 * mpb->num_raid_devs is just extra paranoia
10228 if (a
|| mpb
->num_raid_devs
== 1 || victim
>= super
->anchor
->num_raid_devs
) {
10229 dprintf("failed to delete subarray-%d\n", victim
);
10233 for (dp
= &super
->devlist
; *dp
;)
10234 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
10237 if ((*dp
)->index
> (unsigned)victim
)
10241 mpb
->num_raid_devs
--;
10242 super
->updates_pending
++;
10245 case update_rename_array
: {
10246 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
10247 char name
[MAX_RAID_SERIAL_LEN
+1];
10248 int target
= u
->dev_idx
;
10249 struct active_array
*a
;
10250 struct imsm_dev
*dev
;
10252 /* sanity check that we are not affecting the uuid of
10255 memset(name
, 0, sizeof(name
));
10256 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
10257 name
[MAX_RAID_SERIAL_LEN
] = '\0';
10258 for (a
= st
->arrays
; a
; a
= a
->next
)
10259 if (a
->info
.container_member
== target
)
10261 dev
= get_imsm_dev(super
, u
->dev_idx
);
10263 if (a
|| !dev
|| imsm_is_name_allowed(super
, name
, 0) == false) {
10264 dprintf("failed to rename subarray-%d\n", target
);
10268 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
10269 super
->updates_pending
++;
10272 case update_add_remove_disk
: {
10273 /* we may be able to repair some arrays if disks are
10274 * being added, check the status of add_remove_disk
10275 * if discs has been added.
10277 if (add_remove_disk_update(super
)) {
10278 struct active_array
*a
;
10280 super
->updates_pending
++;
10281 for (a
= st
->arrays
; a
; a
= a
->next
)
10282 a
->check_degraded
= 1;
10286 case update_prealloc_badblocks_mem
:
10288 case update_rwh_policy
: {
10289 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
10290 int target
= u
->dev_idx
;
10291 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
10293 if (dev
->rwh_policy
!= u
->new_policy
) {
10294 dev
->rwh_policy
= u
->new_policy
;
10295 super
->updates_pending
++;
10300 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10304 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10306 static int imsm_prepare_update(struct supertype
*st
,
10307 struct metadata_update
*update
)
10310 * Allocate space to hold new disk entries, raid-device entries or a new
10311 * mpb if necessary. The manager synchronously waits for updates to
10312 * complete in the monitor, so new mpb buffers allocated here can be
10313 * integrated by the monitor thread without worrying about live pointers
10314 * in the manager thread.
10316 enum imsm_update_type type
;
10317 struct intel_super
*super
= st
->sb
;
10318 unsigned int sector_size
= super
->sector_size
;
10319 struct imsm_super
*mpb
= super
->anchor
;
10323 if (update
->len
< (int)sizeof(type
))
10326 type
= *(enum imsm_update_type
*) update
->buf
;
10329 case update_general_migration_checkpoint
:
10330 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10332 dprintf("called for update_general_migration_checkpoint\n");
10334 case update_takeover
: {
10335 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10336 if (update
->len
< (int)sizeof(*u
))
10338 if (u
->direction
== R0_TO_R10
) {
10339 void **tail
= (void **)&update
->space_list
;
10340 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10341 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10342 int num_members
= map
->num_members
;
10345 /* allocate memory for added disks */
10346 for (i
= 0; i
< num_members
; i
++) {
10347 size
= sizeof(struct dl
);
10348 space
= xmalloc(size
);
10353 /* allocate memory for new device */
10354 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10355 (num_members
* sizeof(__u32
));
10356 space
= xmalloc(size
);
10360 len
= disks_to_mpb_size(num_members
* 2);
10365 case update_reshape_container_disks
: {
10366 /* Every raid device in the container is about to
10367 * gain some more devices, and we will enter a
10369 * So each 'imsm_map' will be bigger, and the imsm_vol
10370 * will now hold 2 of them.
10371 * Thus we need new 'struct imsm_dev' allocations sized
10372 * as sizeof_imsm_dev but with more devices in both maps.
10374 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10375 struct intel_dev
*dl
;
10376 void **space_tail
= (void**)&update
->space_list
;
10378 if (update
->len
< (int)sizeof(*u
))
10381 dprintf("for update_reshape\n");
10383 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10384 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10386 if (u
->new_raid_disks
> u
->old_raid_disks
)
10387 size
+= sizeof(__u32
)*2*
10388 (u
->new_raid_disks
- u
->old_raid_disks
);
10392 *space_tail
= NULL
;
10395 len
= disks_to_mpb_size(u
->new_raid_disks
);
10396 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10399 case update_reshape_migration
: {
10400 /* for migration level 0->5 we need to add disks
10401 * so the same as for container operation we will copy
10402 * device to the bigger location.
10403 * in memory prepared device and new disk area are prepared
10404 * for usage in process update
10406 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10407 struct intel_dev
*id
;
10408 void **space_tail
= (void **)&update
->space_list
;
10411 int current_level
= -1;
10413 if (update
->len
< (int)sizeof(*u
))
10416 dprintf("for update_reshape\n");
10418 /* add space for bigger array in update
10420 for (id
= super
->devlist
; id
; id
= id
->next
) {
10421 if (id
->index
== (unsigned)u
->subdev
) {
10422 size
= sizeof_imsm_dev(id
->dev
, 1);
10423 if (u
->new_raid_disks
> u
->old_raid_disks
)
10424 size
+= sizeof(__u32
)*2*
10425 (u
->new_raid_disks
- u
->old_raid_disks
);
10429 *space_tail
= NULL
;
10433 if (update
->space_list
== NULL
)
10436 /* add space for disk in update
10438 size
= sizeof(struct dl
);
10442 *space_tail
= NULL
;
10444 /* add spare device to update
10446 for (id
= super
->devlist
; id
; id
= id
->next
)
10447 if (id
->index
== (unsigned)u
->subdev
) {
10448 struct imsm_dev
*dev
;
10449 struct imsm_map
*map
;
10451 dev
= get_imsm_dev(super
, u
->subdev
);
10452 map
= get_imsm_map(dev
, MAP_0
);
10453 current_level
= map
->raid_level
;
10456 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10457 struct mdinfo
*spares
;
10459 spares
= get_spares_for_grow(st
);
10462 struct mdinfo
*dev
;
10464 dev
= spares
->devs
;
10467 makedev(dev
->disk
.major
,
10469 dl
= get_disk_super(super
,
10472 dl
->index
= u
->old_raid_disks
;
10475 sysfs_free(spares
);
10478 len
= disks_to_mpb_size(u
->new_raid_disks
);
10479 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10482 case update_size_change
: {
10483 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10487 case update_activate_spare
: {
10488 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10492 case update_create_array
: {
10493 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10494 struct intel_dev
*dv
;
10495 struct imsm_dev
*dev
= &u
->dev
;
10496 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10498 struct disk_info
*inf
;
10502 if (update
->len
< (int)sizeof(*u
))
10505 inf
= get_disk_info(u
);
10506 len
= sizeof_imsm_dev(dev
, 1);
10507 /* allocate a new super->devlist entry */
10508 dv
= xmalloc(sizeof(*dv
));
10509 dv
->dev
= xmalloc(len
);
10510 update
->space
= dv
;
10512 /* count how many spares will be converted to members */
10513 for (i
= 0; i
< map
->num_members
; i
++) {
10514 dl
= serial_to_dl(inf
[i
].serial
, super
);
10516 /* hmm maybe it failed?, nothing we can do about
10521 if (count_memberships(dl
, super
) == 0)
10524 len
+= activate
* sizeof(struct imsm_disk
);
10527 case update_kill_array
: {
10528 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10532 case update_rename_array
: {
10533 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10537 case update_add_remove_disk
:
10538 /* no update->len needed */
10540 case update_prealloc_badblocks_mem
:
10541 super
->extra_space
+= sizeof(struct bbm_log
) -
10542 get_imsm_bbm_log_size(super
->bbm_log
);
10544 case update_rwh_policy
: {
10545 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10553 /* check if we need a larger metadata buffer */
10554 if (super
->next_buf
)
10555 buf_len
= super
->next_len
;
10557 buf_len
= super
->len
;
10559 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10560 /* ok we need a larger buf than what is currently allocated
10561 * if this allocation fails process_update will notice that
10562 * ->next_len is set and ->next_buf is NULL
10564 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10565 super
->extra_space
+ len
, sector_size
);
10566 if (super
->next_buf
)
10567 free(super
->next_buf
);
10569 super
->next_len
= buf_len
;
10570 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10571 memset(super
->next_buf
, 0, buf_len
);
10573 super
->next_buf
= NULL
;
10578 /* must be called while manager is quiesced */
10579 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10581 struct imsm_super
*mpb
= super
->anchor
;
10583 struct imsm_dev
*dev
;
10584 struct imsm_map
*map
;
10585 unsigned int i
, j
, num_members
;
10586 __u32 ord
, ord_map0
;
10587 struct bbm_log
*log
= super
->bbm_log
;
10589 dprintf("deleting device[%d] from imsm_super\n", index
);
10591 /* shift all indexes down one */
10592 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10593 if (iter
->index
> (int)index
)
10595 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10596 if (iter
->index
> (int)index
)
10599 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10600 dev
= get_imsm_dev(super
, i
);
10601 map
= get_imsm_map(dev
, MAP_0
);
10602 num_members
= map
->num_members
;
10603 for (j
= 0; j
< num_members
; j
++) {
10604 /* update ord entries being careful not to propagate
10605 * ord-flags to the first map
10607 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10608 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10610 if (ord_to_idx(ord
) <= index
)
10613 map
= get_imsm_map(dev
, MAP_0
);
10614 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10615 map
= get_imsm_map(dev
, MAP_1
);
10617 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10621 for (i
= 0; i
< log
->entry_count
; i
++) {
10622 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10624 if (entry
->disk_ordinal
<= index
)
10626 entry
->disk_ordinal
--;
10630 super
->updates_pending
++;
10632 struct dl
*dl
= *dlp
;
10634 *dlp
= (*dlp
)->next
;
10635 __free_imsm_disk(dl
, 1);
10639 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10640 struct intel_super
*super
,
10641 struct imsm_dev
*dev
)
10647 struct imsm_map
*map
;
10650 ret_val
= raid_disks
/2;
10651 /* check map if all disks pairs not failed
10654 map
= get_imsm_map(dev
, MAP_0
);
10655 for (i
= 0; i
< ret_val
; i
++) {
10656 int degradation
= 0;
10657 if (get_imsm_disk(super
, i
) == NULL
)
10659 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10661 if (degradation
== 2)
10664 map
= get_imsm_map(dev
, MAP_1
);
10665 /* if there is no second map
10666 * result can be returned
10670 /* check degradation in second map
10672 for (i
= 0; i
< ret_val
; i
++) {
10673 int degradation
= 0;
10674 if (get_imsm_disk(super
, i
) == NULL
)
10676 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10678 if (degradation
== 2)
10692 /*******************************************************************************
10693 * Function: validate_container_imsm
10694 * Description: This routine validates container after assemble,
10695 * eg. if devices in container are under the same controller.
10698 * info : linked list with info about devices used in array
10702 ******************************************************************************/
10703 int validate_container_imsm(struct mdinfo
*info
)
10705 if (check_no_platform())
10708 struct sys_dev
*idev
;
10709 struct sys_dev
*hba
= NULL
;
10710 struct sys_dev
*intel_devices
= find_intel_devices();
10711 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10712 info
->disk
.minor
), 1, NULL
);
10714 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10715 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10724 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10725 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10729 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10730 struct mdinfo
*dev
;
10732 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10733 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
,
10734 dev
->disk
.minor
), 1, NULL
);
10736 struct sys_dev
*hba2
= NULL
;
10737 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10738 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10746 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10747 get_orom_by_device_id(hba2
->dev_id
);
10749 if (hba2
&& hba
->type
!= hba2
->type
) {
10750 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10751 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10755 if (orom
!= orom2
) {
10756 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10757 " This operation is not supported and can lead to data loss.\n");
10762 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10763 " This operation is not supported and can lead to data loss.\n");
10771 /*******************************************************************************
10772 * Function: imsm_record_badblock
10773 * Description: This routine stores new bad block record in BBM log
10776 * a : array containing a bad block
10777 * slot : disk number containing a bad block
10778 * sector : bad block sector
10779 * length : bad block sectors range
10783 ******************************************************************************/
10784 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10785 unsigned long long sector
, int length
)
10787 struct intel_super
*super
= a
->container
->sb
;
10791 ord
= imsm_disk_slot_to_ord(a
, slot
);
10795 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10798 super
->updates_pending
++;
10802 /*******************************************************************************
10803 * Function: imsm_clear_badblock
10804 * Description: This routine clears bad block record from BBM log
10807 * a : array containing a bad block
10808 * slot : disk number containing a bad block
10809 * sector : bad block sector
10810 * length : bad block sectors range
10814 ******************************************************************************/
10815 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10816 unsigned long long sector
, int length
)
10818 struct intel_super
*super
= a
->container
->sb
;
10822 ord
= imsm_disk_slot_to_ord(a
, slot
);
10826 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10828 super
->updates_pending
++;
10832 /*******************************************************************************
10833 * Function: imsm_get_badblocks
10834 * Description: This routine get list of bad blocks for an array
10838 * slot : disk number
10840 * bb : structure containing bad blocks
10842 ******************************************************************************/
10843 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10845 int inst
= a
->info
.container_member
;
10846 struct intel_super
*super
= a
->container
->sb
;
10847 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10848 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10851 ord
= imsm_disk_slot_to_ord(a
, slot
);
10855 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10856 per_dev_array_size(map
), &super
->bb
);
10860 /*******************************************************************************
10861 * Function: examine_badblocks_imsm
10862 * Description: Prints list of bad blocks on a disk to the standard output
10865 * st : metadata handler
10866 * fd : open file descriptor for device
10867 * devname : device name
10871 ******************************************************************************/
10872 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10874 struct intel_super
*super
= st
->sb
;
10875 struct bbm_log
*log
= super
->bbm_log
;
10876 struct dl
*d
= NULL
;
10879 for (d
= super
->disks
; d
; d
= d
->next
) {
10880 if (strcmp(d
->devname
, devname
) == 0)
10884 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10885 pr_err("%s doesn't appear to be part of a raid array\n",
10892 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10894 for (i
= 0; i
< log
->entry_count
; i
++) {
10895 if (entry
[i
].disk_ordinal
== d
->index
) {
10896 unsigned long long sector
= __le48_to_cpu(
10897 &entry
[i
].defective_block_start
);
10898 int cnt
= entry
[i
].marked_count
+ 1;
10901 printf("Bad-blocks on %s:\n", devname
);
10905 printf("%20llu for %d sectors\n", sector
, cnt
);
10911 printf("No bad-blocks list configured on %s\n", devname
);
10915 /*******************************************************************************
10916 * Function: init_migr_record_imsm
10917 * Description: Function inits imsm migration record
10919 * super : imsm internal array info
10920 * dev : device under migration
10921 * info : general array info to find the smallest device
10924 ******************************************************************************/
10925 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10926 struct mdinfo
*info
)
10928 struct intel_super
*super
= st
->sb
;
10929 struct migr_record
*migr_rec
= super
->migr_rec
;
10930 int new_data_disks
;
10931 unsigned long long dsize
, dev_sectors
;
10932 long long unsigned min_dev_sectors
= -1LLU;
10933 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10934 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10935 unsigned long long num_migr_units
;
10936 unsigned long long array_blocks
;
10937 struct dl
*dl_disk
= NULL
;
10939 memset(migr_rec
, 0, sizeof(struct migr_record
));
10940 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10942 /* only ascending reshape supported now */
10943 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10945 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10946 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10947 migr_rec
->dest_depth_per_unit
*=
10948 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10949 new_data_disks
= imsm_num_data_members(map_dest
);
10950 migr_rec
->blocks_per_unit
=
10951 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10952 migr_rec
->dest_depth_per_unit
=
10953 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10954 array_blocks
= info
->component_size
* new_data_disks
;
10956 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10958 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10960 set_num_migr_units(migr_rec
, num_migr_units
);
10962 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10963 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10965 /* Find the smallest dev */
10966 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
10967 /* ignore spares in container */
10968 if (dl_disk
->index
< 0)
10970 get_dev_size(dl_disk
->fd
, NULL
, &dsize
);
10971 dev_sectors
= dsize
/ 512;
10972 if (dev_sectors
< min_dev_sectors
)
10973 min_dev_sectors
= dev_sectors
;
10975 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
10976 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10978 write_imsm_migr_rec(st
);
10983 /*******************************************************************************
10984 * Function: save_backup_imsm
10985 * Description: Function saves critical data stripes to Migration Copy Area
10986 * and updates the current migration unit status.
10987 * Use restore_stripes() to form a destination stripe,
10988 * and to write it to the Copy Area.
10990 * st : supertype information
10991 * dev : imsm device that backup is saved for
10992 * info : general array info
10993 * buf : input buffer
10994 * length : length of data to backup (blocks_per_unit)
10998 ******************************************************************************/
10999 int save_backup_imsm(struct supertype
*st
,
11000 struct imsm_dev
*dev
,
11001 struct mdinfo
*info
,
11006 struct intel_super
*super
= st
->sb
;
11008 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
11009 int new_disks
= map_dest
->num_members
;
11010 int dest_layout
= 0;
11011 int dest_chunk
, targets
[new_disks
];
11012 unsigned long long start
, target_offsets
[new_disks
];
11013 int data_disks
= imsm_num_data_members(map_dest
);
11015 for (i
= 0; i
< new_disks
; i
++) {
11016 struct dl
*dl_disk
= get_imsm_dl_disk(super
, i
);
11017 if (dl_disk
&& is_fd_valid(dl_disk
->fd
))
11018 targets
[i
] = dl_disk
->fd
;
11023 start
= info
->reshape_progress
* 512;
11024 for (i
= 0; i
< new_disks
; i
++) {
11025 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
11026 /* move back copy area adderss, it will be moved forward
11027 * in restore_stripes() using start input variable
11029 target_offsets
[i
] -= start
/data_disks
;
11032 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
11033 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
11035 if (restore_stripes(targets
, /* list of dest devices */
11036 target_offsets
, /* migration record offsets */
11039 map_dest
->raid_level
,
11041 -1, /* source backup file descriptor */
11042 0, /* input buf offset
11043 * always 0 buf is already offseted */
11047 pr_err("Error restoring stripes\n");
11057 /*******************************************************************************
11058 * Function: save_checkpoint_imsm
11059 * Description: Function called for current unit status update
11060 * in the migration record. It writes it to disk.
11062 * super : imsm internal array info
11063 * info : general array info
11067 * 2: failure, means no valid migration record
11068 * / no general migration in progress /
11069 ******************************************************************************/
11070 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
11072 struct intel_super
*super
= st
->sb
;
11073 unsigned long long blocks_per_unit
;
11074 unsigned long long curr_migr_unit
;
11076 if (load_imsm_migr_rec(super
) != 0) {
11077 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
11081 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
11082 if (blocks_per_unit
== 0) {
11083 dprintf("imsm: no migration in progress.\n");
11086 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
11087 /* check if array is alligned to copy area
11088 * if it is not alligned, add one to current migration unit value
11089 * this can happend on array reshape finish only
11091 if (info
->reshape_progress
% blocks_per_unit
)
11094 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
11095 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
11096 set_migr_dest_1st_member_lba(super
->migr_rec
,
11097 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
11099 if (write_imsm_migr_rec(st
) < 0) {
11100 dprintf("imsm: Cannot write migration record outside backup area\n");
11107 /*******************************************************************************
11108 * Function: recover_backup_imsm
11109 * Description: Function recovers critical data from the Migration Copy Area
11110 * while assembling an array.
11112 * super : imsm internal array info
11113 * info : general array info
11115 * 0 : success (or there is no data to recover)
11117 ******************************************************************************/
11118 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
11120 struct intel_super
*super
= st
->sb
;
11121 struct migr_record
*migr_rec
= super
->migr_rec
;
11122 struct imsm_map
*map_dest
;
11123 struct intel_dev
*id
= NULL
;
11124 unsigned long long read_offset
;
11125 unsigned long long write_offset
;
11127 int new_disks
, err
;
11130 unsigned int sector_size
= super
->sector_size
;
11131 unsigned long long curr_migr_unit
= current_migr_unit(migr_rec
);
11132 unsigned long long num_migr_units
= get_num_migr_units(migr_rec
);
11133 char buffer
[SYSFS_MAX_BUF_SIZE
];
11134 int skipped_disks
= 0;
11135 struct dl
*dl_disk
;
11137 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, sizeof(buffer
));
11141 /* recover data only during assemblation */
11142 if (strncmp(buffer
, "inactive", 8) != 0)
11144 /* no data to recover */
11145 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
11147 if (curr_migr_unit
>= num_migr_units
)
11150 /* find device during reshape */
11151 for (id
= super
->devlist
; id
; id
= id
->next
)
11152 if (is_gen_migration(id
->dev
))
11157 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
11158 new_disks
= map_dest
->num_members
;
11160 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
11162 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
11163 pba_of_lba0(map_dest
)) * 512;
11165 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11166 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
11169 for (dl_disk
= super
->disks
; dl_disk
; dl_disk
= dl_disk
->next
) {
11170 if (dl_disk
->index
< 0)
11173 if (!is_fd_valid(dl_disk
->fd
)) {
11177 if (lseek64(dl_disk
->fd
, read_offset
, SEEK_SET
) < 0) {
11178 pr_err("Cannot seek to block: %s\n",
11183 if (read(dl_disk
->fd
, buf
, unit_len
) != (ssize_t
)unit_len
) {
11184 pr_err("Cannot read copy area block: %s\n",
11189 if (lseek64(dl_disk
->fd
, write_offset
, SEEK_SET
) < 0) {
11190 pr_err("Cannot seek to block: %s\n",
11195 if (write(dl_disk
->fd
, buf
, unit_len
) != (ssize_t
)unit_len
) {
11196 pr_err("Cannot restore block: %s\n",
11203 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11207 pr_err("Cannot restore data from backup. Too many failed disks\n");
11211 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11212 /* ignore error == 2, this can mean end of reshape here
11214 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11224 * test_and_add_drive_controller_policy_imsm() - add disk controller to policies list.
11225 * @type: Policy type to search on list.
11226 * @pols: List of currently recorded policies.
11227 * @disk_fd: File descriptor of the device to check.
11228 * @hba: The hba disk is attached, could be NULL if verification is disabled.
11229 * @verbose: verbose flag.
11231 * IMSM cares about drive physical placement. If @hba is not set, it adds unknown policy.
11232 * If there is no controller policy on pols we are free to add first one. If there is a policy then,
11233 * new must be the same - no controller mixing allowed.
11235 static mdadm_status_t
11236 test_and_add_drive_controller_policy_imsm(const char * const type
, dev_policy_t
**pols
, int disk_fd
,
11237 struct sys_dev
*hba
, const int verbose
)
11239 const char *controller_policy
= get_sys_dev_type(SYS_DEV_UNKNOWN
);
11240 struct dev_policy
*pol
= pol_find(*pols
, (char *)type
);
11241 char devname
[MAX_RAID_SERIAL_LEN
];
11244 controller_policy
= get_sys_dev_type(hba
->type
);
11247 pol_add(pols
, (char *)type
, (char *)controller_policy
, "imsm");
11248 return MDADM_STATUS_SUCCESS
;
11251 if (strcmp(pol
->value
, controller_policy
) == 0)
11252 return MDADM_STATUS_SUCCESS
;
11254 fd2devname(disk_fd
, devname
);
11255 pr_vrb("Intel(R) raid controller \"%s\" found for %s, but \"%s\" was detected earlier\n",
11256 controller_policy
, devname
, pol
->value
);
11257 pr_vrb("Disks under different controllers cannot be used, aborting\n");
11259 return MDADM_STATUS_ERROR
;
11262 struct imsm_drive_policy
{
11264 mdadm_status_t (*test_and_add_drive_policy
)(const char * const type
,
11265 struct dev_policy
**pols
, int disk_fd
,
11266 struct sys_dev
*hba
, const int verbose
);
11269 struct imsm_drive_policy imsm_policies
[] = {
11270 {"controller", test_and_add_drive_controller_policy_imsm
},
11273 mdadm_status_t
test_and_add_drive_policies_imsm(struct dev_policy
**pols
, int disk_fd
,
11276 struct imsm_drive_policy
*imsm_pol
;
11277 struct sys_dev
*hba
= NULL
;
11278 char path
[PATH_MAX
];
11279 mdadm_status_t ret
;
11282 /* If imsm platform verification is disabled, do not search for hba. */
11283 if (check_no_platform() != 1) {
11284 if (!diskfd_to_devpath(disk_fd
, 1, path
)) {
11285 pr_vrb("IMSM: Failed to retrieve device path by file descriptor.\n");
11286 return MDADM_STATUS_ERROR
;
11289 hba
= find_disk_attached_hba(disk_fd
, path
);
11291 pr_vrb("IMSM: Failed to find hba for %s\n", path
);
11292 return MDADM_STATUS_ERROR
;
11296 for (i
= 0; i
< ARRAY_SIZE(imsm_policies
); i
++) {
11297 imsm_pol
= &imsm_policies
[i
];
11299 ret
= imsm_pol
->test_and_add_drive_policy(imsm_pol
->type
, pols
, disk_fd
, hba
,
11301 if (ret
!= MDADM_STATUS_SUCCESS
)
11302 /* Inherit error code */
11306 return MDADM_STATUS_SUCCESS
;
11310 * get_spare_criteria_imsm() - set spare criteria.
11312 * @mddev_path: path to md device devnode, it must be container.
11313 * @c: spare_criteria struct to fill, not NULL.
11315 * If superblock is not loaded, use mddev_path to load_container. It must be given in this case.
11316 * Filles size and sector size accordingly to superblock.
11318 mdadm_status_t
get_spare_criteria_imsm(struct supertype
*st
, char *mddev_path
,
11319 struct spare_criteria
*c
)
11321 mdadm_status_t ret
= MDADM_STATUS_ERROR
;
11322 bool free_superblock
= false;
11323 unsigned long long size
= 0;
11324 struct intel_super
*super
;
11329 /* If no superblock and no mddev_path, we cannot load superblock. */
11330 assert(st
->sb
|| mddev_path
);
11333 int fd
= open(mddev_path
, O_RDONLY
);
11336 if (!is_fd_valid(fd
))
11337 return MDADM_STATUS_ERROR
;
11340 if (load_container_imsm(st
, fd
, st
->devnm
)) {
11342 return MDADM_STATUS_ERROR
;
11344 free_superblock
= true;
11347 rv
= mddev_test_and_add_drive_policies(st
, &c
->pols
, fd
, 0);
11350 if (rv
!= MDADM_STATUS_SUCCESS
)
11356 /* find first active disk in array */
11358 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
11364 /* find last lba used by subarrays */
11365 e
= get_extents(super
, dl
, 0);
11369 for (i
= 0; e
[i
].size
; i
++)
11372 size
= e
[i
- 1].start
+ e
[i
- 1].size
;
11375 /* add the amount of space needed for metadata */
11376 size
+= imsm_min_reserved_sectors(super
);
11378 c
->min_size
= size
* 512;
11379 c
->sector_size
= super
->sector_size
;
11380 c
->criteria_set
= true;
11381 ret
= MDADM_STATUS_SUCCESS
;
11384 if (free_superblock
)
11385 free_super_imsm(st
);
11387 if (ret
!= MDADM_STATUS_SUCCESS
)
11388 c
->criteria_set
= false;
11393 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11395 static char devnm
[32];
11396 char subdev_name
[20];
11397 struct mdstat_ent
*mdstat
;
11399 sprintf(subdev_name
, "%d", subdev
);
11400 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11404 strcpy(devnm
, mdstat
->devnm
);
11405 free_mdstat(mdstat
);
11409 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11410 struct geo_params
*geo
,
11411 int *old_raid_disks
,
11414 /* currently we only support increasing the number of devices
11415 * for a container. This increases the number of device for each
11416 * member array. They must all be RAID0 or RAID5.
11419 struct mdinfo
*info
, *member
;
11420 int devices_that_can_grow
= 0;
11422 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11424 if (geo
->size
> 0 ||
11425 geo
->level
!= UnSet
||
11426 geo
->layout
!= UnSet
||
11427 geo
->chunksize
!= 0 ||
11428 geo
->raid_disks
== UnSet
) {
11429 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11433 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11434 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11438 info
= container_content_imsm(st
, NULL
);
11439 for (member
= info
; member
; member
= member
->next
) {
11442 dprintf("imsm: checking device_num: %i\n",
11443 member
->container_member
);
11445 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11446 /* we work on container for Online Capacity Expansion
11447 * only so raid_disks has to grow
11449 dprintf("imsm: for container operation raid disks increase is required\n");
11453 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11454 /* we cannot use this container with other raid level
11456 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11457 info
->array
.level
);
11460 /* check for platform support
11461 * for this raid level configuration
11463 struct intel_super
*super
= st
->sb
;
11464 if (!is_raid_level_supported(super
->orom
,
11465 member
->array
.level
,
11466 geo
->raid_disks
)) {
11467 dprintf("platform does not support raid%d with %d disk%s\n",
11470 geo
->raid_disks
> 1 ? "s" : "");
11473 /* check if component size is aligned to chunk size
11475 if (info
->component_size
%
11476 (info
->array
.chunk_size
/512)) {
11477 dprintf("Component size is not aligned to chunk size\n");
11482 if (*old_raid_disks
&&
11483 info
->array
.raid_disks
!= *old_raid_disks
)
11485 *old_raid_disks
= info
->array
.raid_disks
;
11487 /* All raid5 and raid0 volumes in container
11488 * have to be ready for Online Capacity Expansion
11489 * so they need to be assembled. We have already
11490 * checked that no recovery etc is happening.
11492 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11493 st
->container_devnm
);
11494 if (result
== NULL
) {
11495 dprintf("imsm: cannot find array\n");
11498 devices_that_can_grow
++;
11501 if (!member
&& devices_that_can_grow
)
11505 dprintf("Container operation allowed\n");
11507 dprintf("Error: %i\n", ret_val
);
11512 /* Function: get_spares_for_grow
11513 * Description: Allocates memory and creates list of spare devices
11514 * avaliable in container. Checks if spare drive size is acceptable.
11515 * Parameters: Pointer to the supertype structure
11516 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11519 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11521 struct spare_criteria sc
= {0};
11522 struct mdinfo
*spares
;
11524 get_spare_criteria_imsm(st
, NULL
, &sc
);
11525 spares
= container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11527 dev_policy_free(sc
.pols
);
11532 /******************************************************************************
11533 * function: imsm_create_metadata_update_for_reshape
11534 * Function creates update for whole IMSM container.
11536 ******************************************************************************/
11537 static int imsm_create_metadata_update_for_reshape(
11538 struct supertype
*st
,
11539 struct geo_params
*geo
,
11540 int old_raid_disks
,
11541 struct imsm_update_reshape
**updatep
)
11543 struct intel_super
*super
= st
->sb
;
11544 struct imsm_super
*mpb
= super
->anchor
;
11545 int update_memory_size
;
11546 struct imsm_update_reshape
*u
;
11547 struct mdinfo
*spares
;
11550 struct mdinfo
*dev
;
11552 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11554 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11556 /* size of all update data without anchor */
11557 update_memory_size
= sizeof(struct imsm_update_reshape
);
11559 /* now add space for spare disks that we need to add. */
11560 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11562 u
= xcalloc(1, update_memory_size
);
11563 u
->type
= update_reshape_container_disks
;
11564 u
->old_raid_disks
= old_raid_disks
;
11565 u
->new_raid_disks
= geo
->raid_disks
;
11567 /* now get spare disks list
11569 spares
= get_spares_for_grow(st
);
11571 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11572 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11577 /* we have got spares
11578 * update disk list in imsm_disk list table in anchor
11580 dprintf("imsm: %i spares are available.\n\n",
11581 spares
->array
.spare_disks
);
11583 dev
= spares
->devs
;
11584 for (i
= 0; i
< delta_disks
; i
++) {
11589 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11591 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11592 dl
->index
= mpb
->num_disks
;
11600 sysfs_free(spares
);
11602 dprintf("imsm: reshape update preparation :");
11603 if (i
== delta_disks
) {
11604 dprintf_cont(" OK\n");
11606 return update_memory_size
;
11609 dprintf_cont(" Error\n");
11614 /******************************************************************************
11615 * function: imsm_create_metadata_update_for_size_change()
11616 * Creates update for IMSM array for array size change.
11618 ******************************************************************************/
11619 static int imsm_create_metadata_update_for_size_change(
11620 struct supertype
*st
,
11621 struct geo_params
*geo
,
11622 struct imsm_update_size_change
**updatep
)
11624 struct intel_super
*super
= st
->sb
;
11625 int update_memory_size
;
11626 struct imsm_update_size_change
*u
;
11628 dprintf("(enter) New size = %llu\n", geo
->size
);
11630 /* size of all update data without anchor */
11631 update_memory_size
= sizeof(struct imsm_update_size_change
);
11633 u
= xcalloc(1, update_memory_size
);
11634 u
->type
= update_size_change
;
11635 u
->subdev
= super
->current_vol
;
11636 u
->new_size
= geo
->size
;
11638 dprintf("imsm: reshape update preparation : OK\n");
11641 return update_memory_size
;
11644 /******************************************************************************
11645 * function: imsm_create_metadata_update_for_migration()
11646 * Creates update for IMSM array.
11648 ******************************************************************************/
11649 static int imsm_create_metadata_update_for_migration(
11650 struct supertype
*st
,
11651 struct geo_params
*geo
,
11652 struct imsm_update_reshape_migration
**updatep
)
11654 struct intel_super
*super
= st
->sb
;
11655 int update_memory_size
;
11656 int current_chunk_size
;
11657 struct imsm_update_reshape_migration
*u
;
11658 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11659 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11660 int previous_level
= -1;
11662 dprintf("(enter) New Level = %i\n", geo
->level
);
11664 /* size of all update data without anchor */
11665 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11667 u
= xcalloc(1, update_memory_size
);
11668 u
->type
= update_reshape_migration
;
11669 u
->subdev
= super
->current_vol
;
11670 u
->new_level
= geo
->level
;
11671 u
->new_layout
= geo
->layout
;
11672 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11673 u
->new_disks
[0] = -1;
11674 u
->new_chunksize
= -1;
11676 current_chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) / 2;
11678 if (geo
->chunksize
!= current_chunk_size
) {
11679 u
->new_chunksize
= geo
->chunksize
/ 1024;
11680 dprintf("imsm: chunk size change from %i to %i\n",
11681 current_chunk_size
, u
->new_chunksize
);
11683 previous_level
= map
->raid_level
;
11685 if (geo
->level
== 5 && previous_level
== 0) {
11686 struct mdinfo
*spares
= NULL
;
11688 u
->new_raid_disks
++;
11689 spares
= get_spares_for_grow(st
);
11690 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11692 sysfs_free(spares
);
11693 update_memory_size
= 0;
11694 pr_err("cannot get spare device for requested migration\n");
11697 sysfs_free(spares
);
11699 dprintf("imsm: reshape update preparation : OK\n");
11702 return update_memory_size
;
11705 static void imsm_update_metadata_locally(struct supertype
*st
,
11706 void *buf
, int len
)
11708 struct metadata_update mu
;
11713 mu
.space_list
= NULL
;
11715 if (imsm_prepare_update(st
, &mu
))
11716 imsm_process_update(st
, &mu
);
11718 while (mu
.space_list
) {
11719 void **space
= mu
.space_list
;
11720 mu
.space_list
= *space
;
11726 * imsm_analyze_expand() - check expand properties and calculate new size.
11727 * @st: imsm supertype.
11728 * @geo: new geometry params.
11729 * @array: array info.
11730 * @direction: reshape direction.
11732 * Obtain free space after the &array and verify if expand to requested size is
11733 * possible. If geo->size is set to %MAX_SIZE, assume that max free size is
11737 * On success %IMSM_STATUS_OK is returned, geo->size and geo->raid_disks are
11739 * On error, %IMSM_STATUS_ERROR is returned.
11741 static imsm_status_t
imsm_analyze_expand(struct supertype
*st
,
11742 struct geo_params
*geo
,
11743 struct mdinfo
*array
,
11746 struct intel_super
*super
= st
->sb
;
11747 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11748 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11749 int data_disks
= imsm_num_data_members(map
);
11751 unsigned long long current_size
;
11752 unsigned long long free_size
;
11753 unsigned long long new_size
;
11754 unsigned long long max_size
;
11756 const int chunk_kib
= geo
->chunksize
/ 1024;
11759 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11761 * Accept size for rollback only.
11763 new_size
= geo
->size
* 2;
11767 if (data_disks
== 0) {
11768 pr_err("imsm: Cannot retrieve data disks.\n");
11769 return IMSM_STATUS_ERROR
;
11771 current_size
= array
->custom_array_size
/ data_disks
;
11773 rv
= imsm_get_free_size(super
, dev
->vol
.map
->num_members
, 0, chunk_kib
, &free_size
, true);
11774 if (rv
!= IMSM_STATUS_OK
) {
11775 pr_err("imsm: Cannot find free space for expand.\n");
11776 return IMSM_STATUS_ERROR
;
11778 max_size
= round_member_size_to_mb(free_size
+ current_size
);
11780 if (geo
->size
== MAX_SIZE
)
11781 new_size
= max_size
;
11783 new_size
= round_member_size_to_mb(geo
->size
* 2);
11785 if (new_size
== 0) {
11786 pr_err("imsm: Rounded requested size is 0.\n");
11787 return IMSM_STATUS_ERROR
;
11790 if (new_size
> max_size
) {
11791 pr_err("imsm: Rounded requested size (%llu) is larger than free space available (%llu).\n",
11792 new_size
, max_size
);
11793 return IMSM_STATUS_ERROR
;
11796 if (new_size
== current_size
) {
11797 pr_err("imsm: Rounded requested size (%llu) is same as current size (%llu).\n",
11798 new_size
, current_size
);
11799 return IMSM_STATUS_ERROR
;
11802 if (new_size
< current_size
) {
11803 pr_err("imsm: Size reduction is not supported, rounded requested size (%llu) is smaller than current (%llu).\n",
11804 new_size
, current_size
);
11805 return IMSM_STATUS_ERROR
;
11809 dprintf("imsm: New size per member is %llu.\n", new_size
);
11810 geo
->size
= data_disks
* new_size
;
11811 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11812 return IMSM_STATUS_OK
;
11815 /***************************************************************************
11816 * Function: imsm_analyze_change
11817 * Description: Function analyze change for single volume
11818 * and validate if transition is supported
11819 * Parameters: Geometry parameters, supertype structure,
11820 * metadata change direction (apply/rollback)
11821 * Returns: Operation type code on success, -1 if fail
11822 ****************************************************************************/
11823 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11824 struct geo_params
*geo
,
11827 struct mdinfo info
;
11829 int check_devs
= 0;
11831 /* number of added/removed disks in operation result */
11832 int devNumChange
= 0;
11833 /* imsm compatible layout value for array geometry verification */
11834 int imsm_layout
= -1;
11837 getinfo_super_imsm_volume(st
, &info
, NULL
);
11838 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11839 geo
->level
!= UnSet
) {
11840 switch (info
.array
.level
) {
11842 if (geo
->level
== 5) {
11843 change
= CH_MIGRATION
;
11844 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11845 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11847 goto analyse_change_exit
;
11849 imsm_layout
= geo
->layout
;
11851 devNumChange
= 1; /* parity disk added */
11852 } else if (geo
->level
== 10) {
11853 change
= CH_TAKEOVER
;
11855 devNumChange
= 2; /* two mirrors added */
11856 imsm_layout
= 0x102; /* imsm supported layout */
11861 if (geo
->level
== 0) {
11862 change
= CH_TAKEOVER
;
11864 devNumChange
= -(geo
->raid_disks
/2);
11865 imsm_layout
= 0; /* imsm raid0 layout */
11869 if (change
== -1) {
11870 pr_err("Error. Level Migration from %d to %d not supported!\n",
11871 info
.array
.level
, geo
->level
);
11872 goto analyse_change_exit
;
11875 geo
->level
= info
.array
.level
;
11877 if (geo
->layout
!= info
.array
.layout
&&
11878 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11879 change
= CH_MIGRATION
;
11880 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11881 geo
->layout
== 5) {
11882 /* reshape 5 -> 4 */
11883 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11884 geo
->layout
== 0) {
11885 /* reshape 4 -> 5 */
11889 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11890 info
.array
.layout
, geo
->layout
);
11892 goto analyse_change_exit
;
11895 geo
->layout
= info
.array
.layout
;
11896 if (imsm_layout
== -1)
11897 imsm_layout
= info
.array
.layout
;
11900 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11901 geo
->chunksize
!= info
.array
.chunk_size
) {
11902 if (info
.array
.level
== 10) {
11903 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11905 goto analyse_change_exit
;
11906 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11907 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11908 geo
->chunksize
/1024, info
.component_size
/2);
11910 goto analyse_change_exit
;
11912 change
= CH_MIGRATION
;
11914 geo
->chunksize
= info
.array
.chunk_size
;
11917 if (geo
->size
> 0) {
11918 if (change
!= -1) {
11919 pr_err("Error. Size change should be the only one at a time.\n");
11921 goto analyse_change_exit
;
11924 rv
= imsm_analyze_expand(st
, geo
, &info
, direction
);
11925 if (rv
!= IMSM_STATUS_OK
)
11926 goto analyse_change_exit
;
11927 change
= CH_ARRAY_SIZE
;
11930 chunk
= geo
->chunksize
/ 1024;
11931 if (!validate_geometry_imsm(st
,
11934 geo
->raid_disks
+ devNumChange
,
11936 geo
->size
, INVALID_SECTORS
,
11937 0, 0, info
.consistency_policy
, 1))
11941 struct intel_super
*super
= st
->sb
;
11942 struct imsm_super
*mpb
= super
->anchor
;
11944 if (mpb
->num_raid_devs
> 1) {
11945 pr_err("Error. Cannot perform operation on %s- for this operation "
11946 "it MUST be single array in container\n", geo
->dev_name
);
11951 analyse_change_exit
:
11952 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11953 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11954 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11960 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11962 struct intel_super
*super
= st
->sb
;
11963 struct imsm_update_takeover
*u
;
11965 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11967 u
->type
= update_takeover
;
11968 u
->subarray
= super
->current_vol
;
11970 /* 10->0 transition */
11971 if (geo
->level
== 0)
11972 u
->direction
= R10_TO_R0
;
11974 /* 0->10 transition */
11975 if (geo
->level
== 10)
11976 u
->direction
= R0_TO_R10
;
11978 /* update metadata locally */
11979 imsm_update_metadata_locally(st
, u
,
11980 sizeof(struct imsm_update_takeover
));
11981 /* and possibly remotely */
11982 if (st
->update_tail
)
11983 append_metadata_update(st
, u
,
11984 sizeof(struct imsm_update_takeover
));
11991 /* Flush size update if size calculated by num_data_stripes is higher than
11992 * imsm_dev_size to eliminate differences during reshape.
11993 * Mdmon will recalculate them correctly.
11994 * If subarray index is not set then check whole container.
11996 * 0 - no error occurred
11997 * 1 - error detected
11999 static int imsm_fix_size_mismatch(struct supertype
*st
, int subarray_index
)
12001 struct intel_super
*super
= st
->sb
;
12002 int tmp
= super
->current_vol
;
12006 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
12007 if (subarray_index
>= 0 && i
!= subarray_index
)
12009 super
->current_vol
= i
;
12010 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
12011 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
12012 unsigned int disc_count
= imsm_num_data_members(map
);
12013 struct geo_params geo
;
12014 struct imsm_update_size_change
*update
;
12015 unsigned long long calc_size
= per_dev_array_size(map
) * disc_count
;
12016 unsigned long long d_size
= imsm_dev_size(dev
);
12019 if (calc_size
== d_size
)
12022 /* There is a difference, confirm that imsm_dev_size is
12023 * smaller and push update.
12025 if (d_size
> calc_size
) {
12026 pr_err("imsm: dev size of subarray %d is incorrect\n",
12030 memset(&geo
, 0, sizeof(struct geo_params
));
12032 u_size
= imsm_create_metadata_update_for_size_change(st
, &geo
,
12034 imsm_update_metadata_locally(st
, update
, u_size
);
12035 if (st
->update_tail
) {
12036 append_metadata_update(st
, update
, u_size
);
12037 flush_metadata_updates(st
);
12038 st
->update_tail
= &st
->updates
;
12040 imsm_sync_metadata(st
);
12046 super
->current_vol
= tmp
;
12050 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
12052 int layout
, int chunksize
, int raid_disks
,
12053 int delta_disks
, char *backup
, char *dev
,
12054 int direction
, int verbose
)
12057 struct geo_params geo
;
12059 dprintf("(enter)\n");
12061 memset(&geo
, 0, sizeof(struct geo_params
));
12063 geo
.dev_name
= dev
;
12064 strcpy(geo
.devnm
, st
->devnm
);
12067 geo
.layout
= layout
;
12068 geo
.chunksize
= chunksize
;
12069 geo
.raid_disks
= raid_disks
;
12070 if (delta_disks
!= UnSet
)
12071 geo
.raid_disks
+= delta_disks
;
12073 dprintf("for level : %i\n", geo
.level
);
12074 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
12076 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
12077 /* On container level we can only increase number of devices. */
12078 dprintf("imsm: info: Container operation\n");
12079 int old_raid_disks
= 0;
12081 if (imsm_reshape_is_allowed_on_container(
12082 st
, &geo
, &old_raid_disks
, direction
)) {
12083 struct imsm_update_reshape
*u
= NULL
;
12086 if (imsm_fix_size_mismatch(st
, -1)) {
12087 dprintf("imsm: Cannot fix size mismatch\n");
12088 goto exit_imsm_reshape_super
;
12091 len
= imsm_create_metadata_update_for_reshape(
12092 st
, &geo
, old_raid_disks
, &u
);
12095 dprintf("imsm: Cannot prepare update\n");
12096 goto exit_imsm_reshape_super
;
12100 /* update metadata locally */
12101 imsm_update_metadata_locally(st
, u
, len
);
12102 /* and possibly remotely */
12103 if (st
->update_tail
)
12104 append_metadata_update(st
, u
, len
);
12109 pr_err("(imsm) Operation is not allowed on this container\n");
12112 /* On volume level we support following operations
12113 * - takeover: raid10 -> raid0; raid0 -> raid10
12114 * - chunk size migration
12115 * - migration: raid5 -> raid0; raid0 -> raid5
12117 struct intel_super
*super
= st
->sb
;
12118 struct intel_dev
*dev
= super
->devlist
;
12120 dprintf("imsm: info: Volume operation\n");
12121 /* find requested device */
12124 imsm_find_array_devnm_by_subdev(
12125 dev
->index
, st
->container_devnm
);
12126 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
12131 pr_err("Cannot find %s (%s) subarray\n",
12132 geo
.dev_name
, geo
.devnm
);
12133 goto exit_imsm_reshape_super
;
12135 super
->current_vol
= dev
->index
;
12136 change
= imsm_analyze_change(st
, &geo
, direction
);
12139 ret_val
= imsm_takeover(st
, &geo
);
12141 case CH_MIGRATION
: {
12142 struct imsm_update_reshape_migration
*u
= NULL
;
12144 imsm_create_metadata_update_for_migration(
12147 dprintf("imsm: Cannot prepare update\n");
12151 /* update metadata locally */
12152 imsm_update_metadata_locally(st
, u
, len
);
12153 /* and possibly remotely */
12154 if (st
->update_tail
)
12155 append_metadata_update(st
, u
, len
);
12160 case CH_ARRAY_SIZE
: {
12161 struct imsm_update_size_change
*u
= NULL
;
12163 imsm_create_metadata_update_for_size_change(
12166 dprintf("imsm: Cannot prepare update\n");
12170 /* update metadata locally */
12171 imsm_update_metadata_locally(st
, u
, len
);
12172 /* and possibly remotely */
12173 if (st
->update_tail
)
12174 append_metadata_update(st
, u
, len
);
12184 exit_imsm_reshape_super
:
12185 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
12189 #define COMPLETED_OK 0
12190 #define COMPLETED_NONE 1
12191 #define COMPLETED_DELAYED 2
12193 static int read_completed(int fd
, unsigned long long *val
)
12196 char buf
[SYSFS_MAX_BUF_SIZE
];
12198 ret
= sysfs_fd_get_str(fd
, buf
, sizeof(buf
));
12202 ret
= COMPLETED_OK
;
12203 if (str_is_none(buf
) == true) {
12204 ret
= COMPLETED_NONE
;
12205 } else if (strncmp(buf
, "delayed", 7) == 0) {
12206 ret
= COMPLETED_DELAYED
;
12209 *val
= strtoull(buf
, &ep
, 0);
12210 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
12216 /*******************************************************************************
12217 * Function: wait_for_reshape_imsm
12218 * Description: Function writes new sync_max value and waits until
12219 * reshape process reach new position
12221 * sra : general array info
12222 * ndata : number of disks in new array's layout
12225 * 1 : there is no reshape in progress,
12227 ******************************************************************************/
12228 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
12230 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
12232 unsigned long long completed
;
12233 /* to_complete : new sync_max position */
12234 unsigned long long to_complete
= sra
->reshape_progress
;
12235 unsigned long long position_to_set
= to_complete
/ ndata
;
12237 if (!is_fd_valid(fd
)) {
12238 dprintf("cannot open reshape_position\n");
12243 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
12245 dprintf("cannot read reshape_position (no reshape in progres)\n");
12249 sleep_for(0, MSEC_TO_NSEC(30), true);
12254 if (completed
> position_to_set
) {
12255 dprintf("wrong next position to set %llu (%llu)\n",
12256 to_complete
, position_to_set
);
12260 dprintf("Position set: %llu\n", position_to_set
);
12261 if (sysfs_set_num(sra
, NULL
, "sync_max",
12262 position_to_set
) != 0) {
12263 dprintf("cannot set reshape position to %llu\n",
12271 char action
[SYSFS_MAX_BUF_SIZE
];
12272 int timeout
= 3000;
12274 sysfs_wait(fd
, &timeout
);
12275 if (sysfs_get_str(sra
, NULL
, "sync_action",
12276 action
, sizeof(action
)) > 0 &&
12277 strncmp(action
, "reshape", 7) != 0) {
12278 if (strncmp(action
, "idle", 4) == 0)
12284 rc
= read_completed(fd
, &completed
);
12286 dprintf("cannot read reshape_position (in loop)\n");
12289 } else if (rc
== COMPLETED_NONE
)
12291 } while (completed
< position_to_set
);
12297 /*******************************************************************************
12298 * Function: check_degradation_change
12299 * Description: Check that array hasn't become failed.
12301 * info : for sysfs access
12302 * sources : source disks descriptors
12303 * degraded: previous degradation level
12305 * degradation level
12306 ******************************************************************************/
12307 int check_degradation_change(struct mdinfo
*info
,
12311 unsigned long long new_degraded
;
12314 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
12315 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
12316 /* check each device to ensure it is still working */
12319 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
12320 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
12322 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
12323 char sbuf
[SYSFS_MAX_BUF_SIZE
];
12324 int raid_disk
= sd
->disk
.raid_disk
;
12326 if (sysfs_get_str(info
,
12327 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
12328 strstr(sbuf
, "faulty") ||
12329 strstr(sbuf
, "in_sync") == NULL
) {
12330 /* this device is dead */
12331 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
12332 if (raid_disk
>= 0)
12333 close_fd(&sources
[raid_disk
]);
12340 return new_degraded
;
12343 /*******************************************************************************
12344 * Function: imsm_manage_reshape
12345 * Description: Function finds array under reshape and it manages reshape
12346 * process. It creates stripes backups (if required) and sets
12349 * afd : Backup handle (nattive) - not used
12350 * sra : general array info
12351 * reshape : reshape parameters - not used
12352 * st : supertype structure
12353 * blocks : size of critical section [blocks]
12354 * fds : table of source device descriptor
12355 * offsets : start of array (offest per devices)
12357 * destfd : table of destination device descriptor
12358 * destoffsets : table of destination offsets (per device)
12360 * 1 : success, reshape is done
12362 ******************************************************************************/
12363 static int imsm_manage_reshape(
12364 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
12365 struct supertype
*st
, unsigned long backup_blocks
,
12366 int *fds
, unsigned long long *offsets
,
12367 int dests
, int *destfd
, unsigned long long *destoffsets
)
12370 struct intel_super
*super
= st
->sb
;
12371 struct intel_dev
*dv
;
12372 unsigned int sector_size
= super
->sector_size
;
12373 struct imsm_dev
*dev
= NULL
;
12374 struct imsm_map
*map_src
, *map_dest
;
12375 int migr_vol_qan
= 0;
12376 int ndata
, odata
; /* [bytes] */
12377 int chunk
; /* [bytes] */
12378 struct migr_record
*migr_rec
;
12380 unsigned int buf_size
; /* [bytes] */
12381 unsigned long long max_position
; /* array size [bytes] */
12382 unsigned long long next_step
; /* [blocks]/[bytes] */
12383 unsigned long long old_data_stripe_length
;
12384 unsigned long long start_src
; /* [bytes] */
12385 unsigned long long start
; /* [bytes] */
12386 unsigned long long start_buf_shift
; /* [bytes] */
12388 int source_layout
= 0;
12389 int subarray_index
= -1;
12394 if (!fds
|| !offsets
)
12397 /* Find volume during the reshape */
12398 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
12399 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12400 dv
->dev
->vol
.migr_state
== 1) {
12403 subarray_index
= dv
->index
;
12406 /* Only one volume can migrate at the same time */
12407 if (migr_vol_qan
!= 1) {
12408 pr_err("%s", migr_vol_qan
?
12409 "Number of migrating volumes greater than 1\n" :
12410 "There is no volume during migrationg\n");
12414 map_dest
= get_imsm_map(dev
, MAP_0
);
12415 map_src
= get_imsm_map(dev
, MAP_1
);
12416 if (map_src
== NULL
)
12419 ndata
= imsm_num_data_members(map_dest
);
12420 odata
= imsm_num_data_members(map_src
);
12422 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12423 old_data_stripe_length
= odata
* chunk
;
12425 migr_rec
= super
->migr_rec
;
12427 /* initialize migration record for start condition */
12428 if (sra
->reshape_progress
== 0)
12429 init_migr_record_imsm(st
, dev
, sra
);
12431 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12432 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12435 /* Save checkpoint to update migration record for current
12436 * reshape position (in md). It can be farther than current
12437 * reshape position in metadata.
12439 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12440 /* ignore error == 2, this can mean end of reshape here
12442 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12447 /* size for data */
12448 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12449 /* extend buffer size for parity disk */
12450 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12451 /* add space for stripe alignment */
12452 buf_size
+= old_data_stripe_length
;
12453 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12454 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12458 max_position
= sra
->component_size
* ndata
;
12459 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12461 while (current_migr_unit(migr_rec
) <
12462 get_num_migr_units(migr_rec
)) {
12463 /* current reshape position [blocks] */
12464 unsigned long long current_position
=
12465 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12466 * current_migr_unit(migr_rec
);
12467 unsigned long long border
;
12469 /* Check that array hasn't become failed.
12471 degraded
= check_degradation_change(sra
, fds
, degraded
);
12472 if (degraded
> 1) {
12473 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12477 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12479 if ((current_position
+ next_step
) > max_position
)
12480 next_step
= max_position
- current_position
;
12482 start
= current_position
* 512;
12484 /* align reading start to old geometry */
12485 start_buf_shift
= start
% old_data_stripe_length
;
12486 start_src
= start
- start_buf_shift
;
12488 border
= (start_src
/ odata
) - (start
/ ndata
);
12490 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12491 /* save critical stripes to buf
12492 * start - start address of current unit
12493 * to backup [bytes]
12494 * start_src - start address of current unit
12495 * to backup alligned to source array
12498 unsigned long long next_step_filler
;
12499 unsigned long long copy_length
= next_step
* 512;
12501 /* allign copy area length to stripe in old geometry */
12502 next_step_filler
= ((copy_length
+ start_buf_shift
)
12503 % old_data_stripe_length
);
12504 if (next_step_filler
)
12505 next_step_filler
= (old_data_stripe_length
12506 - next_step_filler
);
12507 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12508 start
, start_src
, copy_length
,
12509 start_buf_shift
, next_step_filler
);
12511 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12512 chunk
, map_src
->raid_level
,
12513 source_layout
, 0, NULL
, start_src
,
12515 next_step_filler
+ start_buf_shift
,
12517 dprintf("imsm: Cannot save stripes to buffer\n");
12520 /* Convert data to destination format and store it
12521 * in backup general migration area
12523 if (save_backup_imsm(st
, dev
, sra
,
12524 buf
+ start_buf_shift
, copy_length
)) {
12525 dprintf("imsm: Cannot save stripes to target devices\n");
12528 if (save_checkpoint_imsm(st
, sra
,
12529 UNIT_SRC_IN_CP_AREA
)) {
12530 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12534 /* set next step to use whole border area */
12535 border
/= next_step
;
12537 next_step
*= border
;
12539 /* When data backed up, checkpoint stored,
12540 * kick the kernel to reshape unit of data
12542 next_step
= next_step
+ sra
->reshape_progress
;
12543 /* limit next step to array max position */
12544 if (next_step
> max_position
)
12545 next_step
= max_position
;
12546 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12547 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12548 sra
->reshape_progress
= next_step
;
12550 /* wait until reshape finish */
12551 if (wait_for_reshape_imsm(sra
, ndata
)) {
12552 dprintf("wait_for_reshape_imsm returned error!\n");
12558 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12559 /* ignore error == 2, this can mean end of reshape here
12561 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12567 /* clear migr_rec on disks after successful migration */
12570 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12571 for (d
= super
->disks
; d
; d
= d
->next
) {
12572 if (d
->index
< 0 || is_failed(&d
->disk
))
12574 unsigned long long dsize
;
12576 get_dev_size(d
->fd
, NULL
, &dsize
);
12577 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12579 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12580 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12581 MIGR_REC_BUF_SECTORS
*sector_size
)
12582 perror("Write migr_rec failed");
12586 /* return '1' if done */
12589 /* After the reshape eliminate size mismatch in metadata.
12590 * Don't update md/component_size here, volume hasn't
12591 * to take whole space. It is allowed by kernel.
12592 * md/component_size will be set propoperly after next assembly.
12594 imsm_fix_size_mismatch(st
, subarray_index
);
12598 /* See Grow.c: abort_reshape() for further explanation */
12599 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12600 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12601 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12606 /*******************************************************************************
12607 * Function: calculate_bitmap_min_chunksize
12608 * Description: Calculates the minimal valid bitmap chunk size
12610 * max_bits : indicate how many bits can be used for the bitmap
12611 * data_area_size : the size of the data area covered by the bitmap
12614 * The bitmap chunk size
12615 ******************************************************************************/
12616 static unsigned long long
12617 calculate_bitmap_min_chunksize(unsigned long long max_bits
,
12618 unsigned long long data_area_size
)
12620 unsigned long long min_chunk
=
12621 4096; /* sub-page chunks don't work yet.. */
12622 unsigned long long bits
= data_area_size
/ min_chunk
+ 1;
12624 while (bits
> max_bits
) {
12626 bits
= (bits
+ 1) / 2;
12631 /*******************************************************************************
12632 * Function: calculate_bitmap_chunksize
12633 * Description: Calculates the bitmap chunk size for the given device
12635 * st : supertype information
12636 * dev : device for the bitmap
12639 * The bitmap chunk size
12640 ******************************************************************************/
12641 static unsigned long long calculate_bitmap_chunksize(struct supertype
*st
,
12642 struct imsm_dev
*dev
)
12644 struct intel_super
*super
= st
->sb
;
12645 unsigned long long min_chunksize
;
12646 unsigned long long result
= IMSM_DEFAULT_BITMAP_CHUNKSIZE
;
12647 size_t dev_size
= imsm_dev_size(dev
);
12649 min_chunksize
= calculate_bitmap_min_chunksize(
12650 IMSM_BITMAP_AREA_SIZE
* super
->sector_size
, dev_size
);
12652 if (result
< min_chunksize
)
12653 result
= min_chunksize
;
12658 /*******************************************************************************
12659 * Function: init_bitmap_header
12660 * Description: Initialize the bitmap header structure
12662 * st : supertype information
12663 * bms : bitmap header struct to initialize
12664 * dev : device for the bitmap
12669 ******************************************************************************/
12670 static int init_bitmap_header(struct supertype
*st
, struct bitmap_super_s
*bms
,
12671 struct imsm_dev
*dev
)
12678 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
12679 bms
->version
= __cpu_to_le32(BITMAP_MAJOR_HI
);
12680 bms
->daemon_sleep
= __cpu_to_le32(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
);
12681 bms
->sync_size
= __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
);
12682 bms
->write_behind
= __cpu_to_le32(0);
12684 uuid_from_super_imsm(st
, vol_uuid
);
12685 memcpy(bms
->uuid
, vol_uuid
, 16);
12687 bms
->chunksize
= calculate_bitmap_chunksize(st
, dev
);
12692 /*******************************************************************************
12693 * Function: validate_internal_bitmap_for_drive
12694 * Description: Verify if the bitmap header for a given drive.
12696 * st : supertype information
12697 * offset : The offset from the beginning of the drive where to look for
12698 * the bitmap header.
12699 * d : the drive info
12704 ******************************************************************************/
12705 static int validate_internal_bitmap_for_drive(struct supertype
*st
,
12706 unsigned long long offset
,
12709 struct intel_super
*super
= st
->sb
;
12712 bitmap_super_t
*bms
;
12720 if (posix_memalign(&read_buf
, MAX_SECTOR_SIZE
, IMSM_BITMAP_HEADER_SIZE
))
12724 if (!is_fd_valid(fd
)) {
12725 fd
= open(d
->devname
, O_RDONLY
, 0);
12727 if (!is_fd_valid(fd
)) {
12728 dprintf("cannot open the device %s\n", d
->devname
);
12733 if (lseek64(fd
, offset
* super
->sector_size
, SEEK_SET
) < 0)
12735 if (read(fd
, read_buf
, IMSM_BITMAP_HEADER_SIZE
) !=
12736 IMSM_BITMAP_HEADER_SIZE
)
12739 uuid_from_super_imsm(st
, vol_uuid
);
12742 if ((bms
->magic
!= __cpu_to_le32(BITMAP_MAGIC
)) ||
12743 (bms
->version
!= __cpu_to_le32(BITMAP_MAJOR_HI
)) ||
12744 (!same_uuid((int *)bms
->uuid
, vol_uuid
, st
->ss
->swapuuid
))) {
12745 dprintf("wrong bitmap header detected\n");
12751 if (!is_fd_valid(d
->fd
))
12760 /*******************************************************************************
12761 * Function: validate_internal_bitmap_imsm
12762 * Description: Verify if the bitmap header is in place and with proper data.
12764 * st : supertype information
12767 * 0 : success or device w/o RWH_BITMAP
12769 ******************************************************************************/
12770 static int validate_internal_bitmap_imsm(struct supertype
*st
)
12772 struct intel_super
*super
= st
->sb
;
12773 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
12774 unsigned long long offset
;
12777 if (dev
->rwh_policy
!= RWH_BITMAP
)
12780 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12781 for (d
= super
->disks
; d
; d
= d
->next
) {
12782 if (d
->index
< 0 || is_failed(&d
->disk
))
12785 if (validate_internal_bitmap_for_drive(st
, offset
, d
)) {
12786 pr_err("imsm: bitmap validation failed\n");
12793 /*******************************************************************************
12794 * Function: add_internal_bitmap_imsm
12795 * Description: Mark the volume to use the bitmap and updates the chunk size value.
12797 * st : supertype information
12798 * chunkp : bitmap chunk size
12799 * delay : not used for imsm
12800 * write_behind : not used for imsm
12801 * size : not used for imsm
12802 * may_change : not used for imsm
12803 * amajor : not used for imsm
12808 ******************************************************************************/
12809 static int add_internal_bitmap_imsm(struct supertype
*st
, int *chunkp
,
12810 int delay
, int write_behind
,
12811 unsigned long long size
, int may_change
,
12814 struct intel_super
*super
= st
->sb
;
12815 int vol_idx
= super
->current_vol
;
12816 struct imsm_dev
*dev
;
12818 if (!super
->devlist
|| vol_idx
== -1 || !chunkp
)
12821 dev
= get_imsm_dev(super
, vol_idx
);
12822 dev
->rwh_policy
= RWH_BITMAP
;
12823 *chunkp
= calculate_bitmap_chunksize(st
, dev
);
12827 /*******************************************************************************
12828 * Function: locate_bitmap_imsm
12829 * Description: Seek 'fd' to start of write-intent-bitmap.
12831 * st : supertype information
12832 * fd : file descriptor for the device
12833 * node_num : not used for imsm
12838 ******************************************************************************/
12839 static int locate_bitmap_imsm(struct supertype
*st
, int fd
, int node_num
)
12841 struct intel_super
*super
= st
->sb
;
12842 unsigned long long offset
;
12843 int vol_idx
= super
->current_vol
;
12845 if (!super
->devlist
|| vol_idx
== -1)
12848 offset
= get_bitmap_header_sector(super
, super
->current_vol
);
12849 dprintf("bitmap header offset is %llu\n", offset
);
12851 lseek64(fd
, offset
<< 9, 0);
12856 /*******************************************************************************
12857 * Function: write_init_bitmap_imsm
12858 * Description: Write a bitmap header and prepares the area for the bitmap.
12860 * st : supertype information
12861 * fd : file descriptor for the device
12862 * update : not used for imsm
12867 ******************************************************************************/
12868 static int write_init_bitmap_imsm(struct supertype
*st
, int fd
,
12869 enum bitmap_update update
)
12871 struct intel_super
*super
= st
->sb
;
12872 int vol_idx
= super
->current_vol
;
12874 unsigned long long offset
;
12875 bitmap_super_t bms
= { 0 };
12876 size_t written
= 0;
12881 if (!super
->devlist
|| !super
->sector_size
|| vol_idx
== -1)
12884 struct imsm_dev
*dev
= get_imsm_dev(super
, vol_idx
);
12886 /* first clear the space for bitmap header */
12887 unsigned long long bitmap_area_start
=
12888 get_bitmap_header_sector(super
, vol_idx
);
12890 dprintf("zeroing area start (%llu) and size (%u)\n", bitmap_area_start
,
12891 IMSM_BITMAP_AND_HEADER_SIZE
/ super
->sector_size
);
12892 if (zero_disk_range(fd
, bitmap_area_start
,
12893 IMSM_BITMAP_HEADER_SIZE
/ super
->sector_size
)) {
12894 pr_err("imsm: cannot zeroing the space for the bitmap\n");
12898 /* The bitmap area should be filled with "1"s to perform initial
12901 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
))
12903 memset(buf
, 0xFF, MAX_SECTOR_SIZE
);
12904 offset
= get_bitmap_sector(super
, vol_idx
);
12905 lseek64(fd
, offset
<< 9, 0);
12906 while (written
< IMSM_BITMAP_AREA_SIZE
) {
12907 to_write
= IMSM_BITMAP_AREA_SIZE
- written
;
12908 if (to_write
> MAX_SECTOR_SIZE
)
12909 to_write
= MAX_SECTOR_SIZE
;
12910 rv_num
= write(fd
, buf
, MAX_SECTOR_SIZE
);
12911 if (rv_num
!= MAX_SECTOR_SIZE
) {
12913 dprintf("cannot initialize bitmap area\n");
12919 /* write a bitmap header */
12920 init_bitmap_header(st
, &bms
, dev
);
12921 memset(buf
, 0, MAX_SECTOR_SIZE
);
12922 memcpy(buf
, &bms
, sizeof(bitmap_super_t
));
12923 if (locate_bitmap_imsm(st
, fd
, 0)) {
12925 dprintf("cannot locate the bitmap\n");
12928 if (write(fd
, buf
, MAX_SECTOR_SIZE
) != MAX_SECTOR_SIZE
) {
12930 dprintf("cannot write the bitmap header\n");
12941 /*******************************************************************************
12942 * Function: is_vol_to_setup_bitmap
12943 * Description: Checks if a bitmap should be activated on the dev.
12945 * info : info about the volume to setup the bitmap
12946 * dev : the device to check against bitmap creation
12949 * 0 : bitmap should be set up on the device
12951 ******************************************************************************/
12952 static int is_vol_to_setup_bitmap(struct mdinfo
*info
, struct imsm_dev
*dev
)
12957 if ((strcmp((char *)dev
->volume
, info
->name
) == 0) &&
12958 (dev
->rwh_policy
== RWH_BITMAP
))
12964 /*******************************************************************************
12965 * Function: set_bitmap_sysfs
12966 * Description: Set the sysfs atributes of a given volume to activate the bitmap.
12968 * info : info about the volume where the bitmap should be setup
12969 * chunksize : bitmap chunk size
12970 * location : location of the bitmap
12975 ******************************************************************************/
12976 static int set_bitmap_sysfs(struct mdinfo
*info
, unsigned long long chunksize
,
12979 /* The bitmap/metadata is set to external to allow changing of value for
12980 * bitmap/location. When external is used, the kernel will treat an offset
12981 * related to the device's first lba (in opposition to the "internal" case
12982 * when this value is related to the beginning of the superblock).
12984 if (sysfs_set_str(info
, NULL
, "bitmap/metadata", "external")) {
12985 dprintf("failed to set bitmap/metadata\n");
12989 /* It can only be changed when no bitmap is active.
12990 * Should be bigger than 512 and must be power of 2.
12991 * It is expecting the value in bytes.
12993 if (sysfs_set_num(info
, NULL
, "bitmap/chunksize",
12994 __cpu_to_le32(chunksize
))) {
12995 dprintf("failed to set bitmap/chunksize\n");
12999 /* It is expecting the value in sectors. */
13000 if (sysfs_set_num(info
, NULL
, "bitmap/space",
13001 __cpu_to_le64(IMSM_BITMAP_AREA_SIZE
))) {
13002 dprintf("failed to set bitmap/space\n");
13006 /* Determines the delay between the bitmap updates.
13007 * It is expecting the value in seconds.
13009 if (sysfs_set_num(info
, NULL
, "bitmap/time_base",
13010 __cpu_to_le64(IMSM_DEFAULT_BITMAP_DAEMON_SLEEP
))) {
13011 dprintf("failed to set bitmap/time_base\n");
13015 /* It is expecting the value in sectors with a sign at the beginning. */
13016 if (sysfs_set_str(info
, NULL
, "bitmap/location", location
)) {
13017 dprintf("failed to set bitmap/location\n");
13024 /*******************************************************************************
13025 * Function: set_bitmap_imsm
13026 * Description: Setup the bitmap for the given volume
13028 * st : supertype information
13029 * info : info about the volume where the bitmap should be setup
13034 ******************************************************************************/
13035 static int set_bitmap_imsm(struct supertype
*st
, struct mdinfo
*info
)
13037 struct intel_super
*super
= st
->sb
;
13038 int prev_current_vol
= super
->current_vol
;
13039 struct imsm_dev
*dev
;
13041 char location
[16] = "";
13042 unsigned long long chunksize
;
13043 struct intel_dev
*dev_it
;
13045 for (dev_it
= super
->devlist
; dev_it
; dev_it
= dev_it
->next
) {
13046 super
->current_vol
= dev_it
->index
;
13047 dev
= get_imsm_dev(super
, super
->current_vol
);
13049 if (is_vol_to_setup_bitmap(info
, dev
)) {
13050 if (validate_internal_bitmap_imsm(st
)) {
13051 dprintf("bitmap header validation failed\n");
13055 chunksize
= calculate_bitmap_chunksize(st
, dev
);
13056 dprintf("chunk size is %llu\n", chunksize
);
13058 snprintf(location
, sizeof(location
), "+%llu",
13059 get_bitmap_sector(super
, super
->current_vol
));
13060 dprintf("bitmap offset is %s\n", location
);
13062 if (set_bitmap_sysfs(info
, chunksize
, location
)) {
13063 dprintf("cannot setup the bitmap\n");
13070 super
->current_vol
= prev_current_vol
;
13074 struct superswitch super_imsm
= {
13075 .examine_super
= examine_super_imsm
,
13076 .brief_examine_super
= brief_examine_super_imsm
,
13077 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
13078 .export_examine_super
= export_examine_super_imsm
,
13079 .detail_super
= detail_super_imsm
,
13080 .brief_detail_super
= brief_detail_super_imsm
,
13081 .write_init_super
= write_init_super_imsm
,
13082 .validate_geometry
= validate_geometry_imsm
,
13083 .add_to_super
= add_to_super_imsm
,
13084 .remove_from_super
= remove_from_super_imsm
,
13085 .detail_platform
= detail_platform_imsm
,
13086 .export_detail_platform
= export_detail_platform_imsm
,
13087 .kill_subarray
= kill_subarray_imsm
,
13088 .update_subarray
= update_subarray_imsm
,
13089 .load_container
= load_container_imsm
,
13090 .default_geometry
= default_geometry_imsm
,
13091 .test_and_add_drive_policies
= test_and_add_drive_policies_imsm
,
13092 .reshape_super
= imsm_reshape_super
,
13093 .manage_reshape
= imsm_manage_reshape
,
13094 .recover_backup
= recover_backup_imsm
,
13095 .examine_badblocks
= examine_badblocks_imsm
,
13096 .match_home
= match_home_imsm
,
13097 .uuid_from_super
= uuid_from_super_imsm
,
13098 .getinfo_super
= getinfo_super_imsm
,
13099 .getinfo_super_disks
= getinfo_super_disks_imsm
,
13100 .update_super
= update_super_imsm
,
13102 .avail_size
= avail_size_imsm
,
13103 .get_spare_criteria
= get_spare_criteria_imsm
,
13105 .compare_super
= compare_super_imsm
,
13107 .load_super
= load_super_imsm
,
13108 .init_super
= init_super_imsm
,
13109 .store_super
= store_super_imsm
,
13110 .free_super
= free_super_imsm
,
13111 .match_metadata_desc
= match_metadata_desc_imsm
,
13112 .container_content
= container_content_imsm
,
13113 .validate_container
= validate_container_imsm
,
13115 .add_internal_bitmap
= add_internal_bitmap_imsm
,
13116 .locate_bitmap
= locate_bitmap_imsm
,
13117 .write_bitmap
= write_init_bitmap_imsm
,
13118 .set_bitmap
= set_bitmap_imsm
,
13120 .write_init_ppl
= write_init_ppl_imsm
,
13121 .validate_ppl
= validate_ppl_imsm
,
13128 .open_new
= imsm_open_new
,
13129 .set_array_state
= imsm_set_array_state
,
13130 .set_disk
= imsm_set_disk
,
13131 .sync_metadata
= imsm_sync_metadata
,
13132 .activate_spare
= imsm_activate_spare
,
13133 .process_update
= imsm_process_update
,
13134 .prepare_update
= imsm_prepare_update
,
13135 .record_bad_block
= imsm_record_badblock
,
13136 .clear_bad_block
= imsm_clear_badblock
,
13137 .get_bad_blocks
= imsm_get_badblocks
,