2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
47 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
51 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
53 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
54 /* supports RAID CNG */
55 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
56 /* supports expanded stripe sizes of 256K, 512K and 1MB */
57 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
59 /* The OROM Support RST Caching of Volumes */
60 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
61 /* The OROM supports creating disks greater than 2TB */
62 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
63 /* The OROM supports Bad Block Management */
64 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
66 /* THe OROM Supports NVM Caching of Volumes */
67 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
68 /* The OROM supports creating volumes greater than 2TB */
69 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
70 /* originally for PMP, now it's wasted b/c. Never use this bit! */
71 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
72 /* Verify MPB contents against checksum after reading MPB */
73 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
75 /* Define all supported attributes that have to be accepted by mdadm
77 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
79 MPB_ATTRIB_2TB_DISK | \
84 MPB_ATTRIB_EXP_STRIPE_SIZE | \
87 /* Define attributes that are unused but not harmful */
88 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
90 #define MPB_SECTOR_CNT 2210
91 #define IMSM_RESERVED_SECTORS 8192
92 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2048
93 #define SECT_PER_MB_SHIFT 11
94 #define MAX_SECTOR_SIZE 4096
95 #define MULTIPLE_PPL_AREA_SIZE_IMSM (1024 * 1024) /* Size of the whole
100 * This macro let's us ensure that no-one accidentally
101 * changes the size of a struct
103 #define ASSERT_SIZE(_struct, size) \
104 static inline void __assert_size_##_struct(void) \
108 case (sizeof(struct _struct) == size): break; \
112 /* Disk configuration info. */
113 #define IMSM_MAX_DEVICES 255
115 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
116 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
117 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
118 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
119 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
120 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
121 #define JOURNAL_DISK __cpu_to_le32(0x2000000) /* Device marked as Journaling Drive */
122 __u32 status
; /* 0xF0 - 0xF3 */
123 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
124 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
125 #define IMSM_DISK_FILLERS 3
126 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
128 ASSERT_SIZE(imsm_disk
, 48)
130 /* map selector for map managment
136 /* RAID map configuration infos. */
138 __u32 pba_of_lba0_lo
; /* start address of partition */
139 __u32 blocks_per_member_lo
;/* blocks per member */
140 __u32 num_data_stripes_lo
; /* number of data stripes */
141 __u16 blocks_per_strip
;
142 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
143 #define IMSM_T_STATE_NORMAL 0
144 #define IMSM_T_STATE_UNINITIALIZED 1
145 #define IMSM_T_STATE_DEGRADED 2
146 #define IMSM_T_STATE_FAILED 3
148 #define IMSM_T_RAID0 0
149 #define IMSM_T_RAID1 1
150 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
151 __u8 num_members
; /* number of member disks */
152 __u8 num_domains
; /* number of parity domains */
153 __u8 failed_disk_num
; /* valid only when state is degraded */
155 __u32 pba_of_lba0_hi
;
156 __u32 blocks_per_member_hi
;
157 __u32 num_data_stripes_hi
;
158 __u32 filler
[4]; /* expansion area */
159 #define IMSM_ORD_REBUILD (1 << 24)
160 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
161 * top byte contains some flags
164 ASSERT_SIZE(imsm_map
, 52)
167 __u32 curr_migr_unit
;
168 __u32 checkpoint_id
; /* id to access curr_migr_unit */
169 __u8 migr_state
; /* Normal or Migrating */
171 #define MIGR_REBUILD 1
172 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
173 #define MIGR_GEN_MIGR 3
174 #define MIGR_STATE_CHANGE 4
175 #define MIGR_REPAIR 5
176 __u8 migr_type
; /* Initializing, Rebuilding, ... */
177 #define RAIDVOL_CLEAN 0
178 #define RAIDVOL_DIRTY 1
179 #define RAIDVOL_DSRECORD_VALID 2
181 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
182 __u16 verify_errors
; /* number of mismatches */
183 __u16 bad_blocks
; /* number of bad blocks during verify */
185 struct imsm_map map
[1];
186 /* here comes another one if migr_state */
188 ASSERT_SIZE(imsm_vol
, 84)
191 __u8 volume
[MAX_RAID_SERIAL_LEN
];
194 #define DEV_BOOTABLE __cpu_to_le32(0x01)
195 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
196 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
197 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
198 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
199 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
200 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
201 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
202 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
203 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
204 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
205 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
206 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
207 __u32 status
; /* Persistent RaidDev status */
208 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
212 __u8 cng_master_disk
;
216 __u16 my_vol_raid_dev_num
; /* Used in Unique volume Id for this RaidDev */
222 /* Unique Volume Id of the NvCache Volume associated with this volume */
223 __u32 nvc_vol_orig_family_num
;
224 __u16 nvc_vol_raid_dev_num
;
227 #define RWH_DISTRIBUTED 1
228 #define RWH_JOURNALING_DRIVE 2
229 #define RWH_MULTIPLE_DISTRIBUTED 3
230 #define RWH_MULTIPLE_PPLS_JOURNALING_DRIVE 4
231 #define RWH_MULTIPLE_OFF 5
232 __u8 rwh_policy
; /* Raid Write Hole Policy */
233 __u8 jd_serial
[MAX_RAID_SERIAL_LEN
]; /* Journal Drive serial number */
236 #define IMSM_DEV_FILLERS 3
237 __u32 filler
[IMSM_DEV_FILLERS
];
240 ASSERT_SIZE(imsm_dev
, 164)
243 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
244 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
245 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
246 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
247 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
248 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
249 __u32 attributes
; /* 0x34 - 0x37 */
250 __u8 num_disks
; /* 0x38 Number of configured disks */
251 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
252 __u8 error_log_pos
; /* 0x3A */
253 __u8 fill
[1]; /* 0x3B */
254 __u32 cache_size
; /* 0x3c - 0x40 in mb */
255 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
256 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
257 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
258 __u16 num_raid_devs_created
; /* 0x4C - 0x4D Used for generating unique
259 * volume IDs for raid_dev created in this array
262 __u16 filler1
; /* 0x4E - 0x4F */
263 __u64 creation_time
; /* 0x50 - 0x57 Array creation time */
264 #define IMSM_FILLERS 32
265 __u32 filler
[IMSM_FILLERS
]; /* 0x58 - 0xD7 RAID_MPB_FILLERS */
266 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
267 /* here comes imsm_dev[num_raid_devs] */
268 /* here comes BBM logs */
270 ASSERT_SIZE(imsm_super
, 264)
272 #define BBM_LOG_MAX_ENTRIES 254
273 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
274 #define BBM_LOG_SIGNATURE 0xabadb10c
276 struct bbm_log_block_addr
{
279 } __attribute__ ((__packed__
));
281 struct bbm_log_entry
{
282 __u8 marked_count
; /* Number of blocks marked - 1 */
283 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
284 struct bbm_log_block_addr defective_block_start
;
285 } __attribute__ ((__packed__
));
288 __u32 signature
; /* 0xABADB10C */
290 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
292 ASSERT_SIZE(bbm_log
, 2040)
294 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
296 #define BLOCKS_PER_KB (1024/512)
298 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
300 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
302 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
303 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
304 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
307 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
308 * be recovered using srcMap */
309 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
310 * already been migrated and must
311 * be recovered from checkpoint area */
313 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of single PPL, without the header */
316 __u32 rec_status
; /* Status used to determine how to restart
317 * migration in case it aborts
319 __u32 curr_migr_unit_lo
; /* 0..numMigrUnits-1 */
320 __u32 family_num
; /* Family number of MPB
321 * containing the RaidDev
322 * that is migrating */
323 __u32 ascending_migr
; /* True if migrating in increasing
325 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
326 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
328 * advances per unit-of-operation */
329 __u32 ckpt_area_pba_lo
; /* Pba of first block of ckpt copy area */
330 __u32 dest_1st_member_lba_lo
; /* First member lba on first
331 * stripe of destination */
332 __u32 num_migr_units_lo
; /* Total num migration units-of-op */
333 __u32 post_migr_vol_cap
; /* Size of volume after
334 * migration completes */
335 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
336 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
337 * migration ckpt record was read from
338 * (for recovered migrations) */
339 __u32 curr_migr_unit_hi
; /* 0..numMigrUnits-1 high order 32 bits */
340 __u32 ckpt_area_pba_hi
; /* Pba of first block of ckpt copy area
341 * high order 32 bits */
342 __u32 dest_1st_member_lba_hi
; /* First member lba on first stripe of
343 * destination - high order 32 bits */
344 __u32 num_migr_units_hi
; /* Total num migration units-of-op
345 * high order 32 bits */
347 ASSERT_SIZE(migr_record
, 64)
352 * 2: metadata does not match
360 struct md_list
*next
;
363 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
365 static __u8
migr_type(struct imsm_dev
*dev
)
367 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
368 dev
->status
& DEV_VERIFY_AND_FIX
)
371 return dev
->vol
.migr_type
;
374 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
376 /* for compatibility with older oroms convert MIGR_REPAIR, into
377 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
379 if (migr_type
== MIGR_REPAIR
) {
380 dev
->vol
.migr_type
= MIGR_VERIFY
;
381 dev
->status
|= DEV_VERIFY_AND_FIX
;
383 dev
->vol
.migr_type
= migr_type
;
384 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
388 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
390 return ROUND_UP(bytes
, sector_size
) / sector_size
;
393 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
394 unsigned int sector_size
)
396 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
400 struct imsm_dev
*dev
;
401 struct intel_dev
*next
;
406 enum sys_dev_type type
;
409 struct intel_hba
*next
;
416 /* internal representation of IMSM metadata */
419 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
420 struct imsm_super
*anchor
; /* immovable parameters */
423 void *migr_rec_buf
; /* buffer for I/O operations */
424 struct migr_record
*migr_rec
; /* migration record */
426 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
427 array, it indicates that mdmon is allowed to clean migration
429 size_t len
; /* size of the 'buf' allocation */
430 size_t extra_space
; /* extra space in 'buf' that is not used yet */
431 void *next_buf
; /* for realloc'ing buf from the manager */
433 int updates_pending
; /* count of pending updates for mdmon */
434 int current_vol
; /* index of raid device undergoing creation */
435 unsigned long long create_offset
; /* common start for 'current_vol' */
436 __u32 random
; /* random data for seeding new family numbers */
437 struct intel_dev
*devlist
;
438 unsigned int sector_size
; /* sector size of used member drives */
442 __u8 serial
[MAX_RAID_SERIAL_LEN
];
445 struct imsm_disk disk
;
448 struct extent
*e
; /* for determining freespace @ create */
449 int raiddisk
; /* slot to fill in autolayout */
451 } *disks
, *current_disk
;
452 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
454 struct dl
*missing
; /* disks removed while we weren't looking */
455 struct bbm_log
*bbm_log
;
456 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
457 const struct imsm_orom
*orom
; /* platform firmware support */
458 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
459 struct md_bb bb
; /* memory for get_bad_blocks call */
463 struct imsm_disk disk
;
464 #define IMSM_UNKNOWN_OWNER (-1)
466 struct intel_disk
*next
;
470 unsigned long long start
, size
;
473 /* definitions of reshape process types */
474 enum imsm_reshape_type
{
480 /* definition of messages passed to imsm_process_update */
481 enum imsm_update_type
{
482 update_activate_spare
,
486 update_add_remove_disk
,
487 update_reshape_container_disks
,
488 update_reshape_migration
,
490 update_general_migration_checkpoint
,
492 update_prealloc_badblocks_mem
,
496 struct imsm_update_activate_spare
{
497 enum imsm_update_type type
;
501 struct imsm_update_activate_spare
*next
;
507 unsigned long long size
;
514 enum takeover_direction
{
518 struct imsm_update_takeover
{
519 enum imsm_update_type type
;
521 enum takeover_direction direction
;
524 struct imsm_update_reshape
{
525 enum imsm_update_type type
;
529 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
532 struct imsm_update_reshape_migration
{
533 enum imsm_update_type type
;
536 /* fields for array migration changes
543 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
546 struct imsm_update_size_change
{
547 enum imsm_update_type type
;
552 struct imsm_update_general_migration_checkpoint
{
553 enum imsm_update_type type
;
554 __u32 curr_migr_unit
;
558 __u8 serial
[MAX_RAID_SERIAL_LEN
];
561 struct imsm_update_create_array
{
562 enum imsm_update_type type
;
567 struct imsm_update_kill_array
{
568 enum imsm_update_type type
;
572 struct imsm_update_rename_array
{
573 enum imsm_update_type type
;
574 __u8 name
[MAX_RAID_SERIAL_LEN
];
578 struct imsm_update_add_remove_disk
{
579 enum imsm_update_type type
;
582 struct imsm_update_prealloc_bb_mem
{
583 enum imsm_update_type type
;
586 struct imsm_update_rwh_policy
{
587 enum imsm_update_type type
;
592 static const char *_sys_dev_type
[] = {
593 [SYS_DEV_UNKNOWN
] = "Unknown",
594 [SYS_DEV_SAS
] = "SAS",
595 [SYS_DEV_SATA
] = "SATA",
596 [SYS_DEV_NVME
] = "NVMe",
597 [SYS_DEV_VMD
] = "VMD"
600 const char *get_sys_dev_type(enum sys_dev_type type
)
602 if (type
>= SYS_DEV_MAX
)
603 type
= SYS_DEV_UNKNOWN
;
605 return _sys_dev_type
[type
];
608 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
610 struct intel_hba
*result
= xmalloc(sizeof(*result
));
612 result
->type
= device
->type
;
613 result
->path
= xstrdup(device
->path
);
615 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
621 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
623 struct intel_hba
*result
;
625 for (result
= hba
; result
; result
= result
->next
) {
626 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
632 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
634 struct intel_hba
*hba
;
636 /* check if disk attached to Intel HBA */
637 hba
= find_intel_hba(super
->hba
, device
);
640 /* Check if HBA is already attached to super */
641 if (super
->hba
== NULL
) {
642 super
->hba
= alloc_intel_hba(device
);
647 /* Intel metadata allows for all disks attached to the same type HBA.
648 * Do not support HBA types mixing
650 if (device
->type
!= hba
->type
)
653 /* Multiple same type HBAs can be used if they share the same OROM */
654 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
656 if (device_orom
!= super
->orom
)
662 hba
->next
= alloc_intel_hba(device
);
666 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
668 struct sys_dev
*list
, *elem
;
671 if ((list
= find_intel_devices()) == NULL
)
675 disk_path
= (char *) devname
;
677 disk_path
= diskfd_to_devpath(fd
);
682 for (elem
= list
; elem
; elem
= elem
->next
)
683 if (path_attached_to_hba(disk_path
, elem
->path
))
686 if (disk_path
!= devname
)
692 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
695 static struct supertype
*match_metadata_desc_imsm(char *arg
)
697 struct supertype
*st
;
699 if (strcmp(arg
, "imsm") != 0 &&
700 strcmp(arg
, "default") != 0
704 st
= xcalloc(1, sizeof(*st
));
705 st
->ss
= &super_imsm
;
706 st
->max_devs
= IMSM_MAX_DEVICES
;
707 st
->minor_version
= 0;
712 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
714 return &mpb
->sig
[MPB_SIG_LEN
];
717 /* retrieve a disk directly from the anchor when the anchor is known to be
718 * up-to-date, currently only at load time
720 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
722 if (index
>= mpb
->num_disks
)
724 return &mpb
->disk
[index
];
727 /* retrieve the disk description based on a index of the disk
730 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
734 for (d
= super
->disks
; d
; d
= d
->next
)
735 if (d
->index
== index
)
740 /* retrieve a disk from the parsed metadata */
741 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
745 dl
= get_imsm_dl_disk(super
, index
);
752 /* generate a checksum directly from the anchor when the anchor is known to be
753 * up-to-date, currently only at load or write_super after coalescing
755 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
757 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
758 __u32
*p
= (__u32
*) mpb
;
762 sum
+= __le32_to_cpu(*p
);
766 return sum
- __le32_to_cpu(mpb
->check_sum
);
769 static size_t sizeof_imsm_map(struct imsm_map
*map
)
771 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
774 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
776 /* A device can have 2 maps if it is in the middle of a migration.
778 * MAP_0 - we return the first map
779 * MAP_1 - we return the second map if it exists, else NULL
780 * MAP_X - we return the second map if it exists, else the first
782 struct imsm_map
*map
= &dev
->vol
.map
[0];
783 struct imsm_map
*map2
= NULL
;
785 if (dev
->vol
.migr_state
)
786 map2
= (void *)map
+ sizeof_imsm_map(map
);
788 switch (second_map
) {
805 /* return the size of the device.
806 * migr_state increases the returned size if map[0] were to be duplicated
808 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
810 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
811 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
813 /* migrating means an additional map */
814 if (dev
->vol
.migr_state
)
815 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
817 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
822 /* retrieve disk serial number list from a metadata update */
823 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
826 struct disk_info
*inf
;
828 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
829 sizeof_imsm_dev(&update
->dev
, 0);
834 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
840 if (index
>= mpb
->num_raid_devs
)
843 /* devices start after all disks */
844 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
846 for (i
= 0; i
<= index
; i
++)
848 return _mpb
+ offset
;
850 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
855 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
857 struct intel_dev
*dv
;
859 if (index
>= super
->anchor
->num_raid_devs
)
861 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
862 if (dv
->index
== index
)
867 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
870 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
871 __le16_to_cpu(addr
->w1
));
874 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
876 struct bbm_log_block_addr addr
;
878 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
879 addr
.dw1
= __cpu_to_le32((__u32
)(sec
>> 16) & 0xffffffff);
883 /* get size of the bbm log */
884 static __u32
get_imsm_bbm_log_size(struct bbm_log
*log
)
886 if (!log
|| log
->entry_count
== 0)
889 return sizeof(log
->signature
) +
890 sizeof(log
->entry_count
) +
891 log
->entry_count
* sizeof(struct bbm_log_entry
);
894 /* check if bad block is not partially stored in bbm log */
895 static int is_stored_in_bbm(struct bbm_log
*log
, const __u8 idx
, const unsigned
896 long long sector
, const int length
, __u32
*pos
)
900 for (i
= *pos
; i
< log
->entry_count
; i
++) {
901 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
902 unsigned long long bb_start
;
903 unsigned long long bb_end
;
905 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
906 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
908 if ((entry
->disk_ordinal
== idx
) && (bb_start
>= sector
) &&
909 (bb_end
<= sector
+ length
)) {
917 /* record new bad block in bbm log */
918 static int record_new_badblock(struct bbm_log
*log
, const __u8 idx
, unsigned
919 long long sector
, int length
)
923 struct bbm_log_entry
*entry
= NULL
;
925 while (is_stored_in_bbm(log
, idx
, sector
, length
, &pos
)) {
926 struct bbm_log_entry
*e
= &log
->marked_block_entries
[pos
];
928 if ((e
->marked_count
+ 1 == BBM_LOG_MAX_LBA_ENTRY_VAL
) &&
929 (__le48_to_cpu(&e
->defective_block_start
) == sector
)) {
930 sector
+= BBM_LOG_MAX_LBA_ENTRY_VAL
;
931 length
-= BBM_LOG_MAX_LBA_ENTRY_VAL
;
940 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
941 BBM_LOG_MAX_LBA_ENTRY_VAL
;
942 entry
->defective_block_start
= __cpu_to_le48(sector
);
943 entry
->marked_count
= cnt
- 1;
950 new_bb
= ROUND_UP(length
, BBM_LOG_MAX_LBA_ENTRY_VAL
) /
951 BBM_LOG_MAX_LBA_ENTRY_VAL
;
952 if (log
->entry_count
+ new_bb
> BBM_LOG_MAX_ENTRIES
)
956 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
957 BBM_LOG_MAX_LBA_ENTRY_VAL
;
958 struct bbm_log_entry
*entry
=
959 &log
->marked_block_entries
[log
->entry_count
];
961 entry
->defective_block_start
= __cpu_to_le48(sector
);
962 entry
->marked_count
= cnt
- 1;
963 entry
->disk_ordinal
= idx
;
974 /* clear all bad blocks for given disk */
975 static void clear_disk_badblocks(struct bbm_log
*log
, const __u8 idx
)
979 while (i
< log
->entry_count
) {
980 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
982 if (entries
[i
].disk_ordinal
== idx
) {
983 if (i
< log
->entry_count
- 1)
984 entries
[i
] = entries
[log
->entry_count
- 1];
992 /* clear given bad block */
993 static int clear_badblock(struct bbm_log
*log
, const __u8 idx
, const unsigned
994 long long sector
, const int length
) {
997 while (i
< log
->entry_count
) {
998 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
1000 if ((entries
[i
].disk_ordinal
== idx
) &&
1001 (__le48_to_cpu(&entries
[i
].defective_block_start
) ==
1002 sector
) && (entries
[i
].marked_count
+ 1 == length
)) {
1003 if (i
< log
->entry_count
- 1)
1004 entries
[i
] = entries
[log
->entry_count
- 1];
1014 /* allocate and load BBM log from metadata */
1015 static int load_bbm_log(struct intel_super
*super
)
1017 struct imsm_super
*mpb
= super
->anchor
;
1018 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1020 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
1021 if (!super
->bbm_log
)
1025 struct bbm_log
*log
= (void *)mpb
+
1026 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1030 if (bbm_log_size
< sizeof(log
->signature
) +
1031 sizeof(log
->entry_count
))
1034 entry_count
= __le32_to_cpu(log
->entry_count
);
1035 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
1036 (entry_count
> BBM_LOG_MAX_ENTRIES
))
1040 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
1041 entry_count
* sizeof(struct bbm_log_entry
))
1044 memcpy(super
->bbm_log
, log
, bbm_log_size
);
1046 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
1047 super
->bbm_log
->entry_count
= 0;
1053 /* checks if bad block is within volume boundaries */
1054 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
1055 const unsigned long long start_sector
,
1056 const unsigned long long size
)
1058 unsigned long long bb_start
;
1059 unsigned long long bb_end
;
1061 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1062 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1064 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1065 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1071 /* get list of bad blocks on a drive for a volume */
1072 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1073 const unsigned long long start_sector
,
1074 const unsigned long long size
,
1080 for (i
= 0; i
< log
->entry_count
; i
++) {
1081 const struct bbm_log_entry
*ent
=
1082 &log
->marked_block_entries
[i
];
1083 struct md_bb_entry
*bb
;
1085 if ((ent
->disk_ordinal
== idx
) &&
1086 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1088 if (!bbs
->entries
) {
1089 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1095 bb
= &bbs
->entries
[count
++];
1096 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1097 bb
->length
= ent
->marked_count
+ 1;
1105 * == MAP_0 get first map
1106 * == MAP_1 get second map
1107 * == MAP_X than get map according to the current migr_state
1109 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1113 struct imsm_map
*map
;
1115 map
= get_imsm_map(dev
, second_map
);
1117 /* top byte identifies disk under rebuild */
1118 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1121 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1122 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1124 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1126 return ord_to_idx(ord
);
1129 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1131 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1134 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
1139 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1140 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1141 if (ord_to_idx(ord
) == idx
)
1148 static int get_imsm_raid_level(struct imsm_map
*map
)
1150 if (map
->raid_level
== 1) {
1151 if (map
->num_members
== 2)
1157 return map
->raid_level
;
1160 static int cmp_extent(const void *av
, const void *bv
)
1162 const struct extent
*a
= av
;
1163 const struct extent
*b
= bv
;
1164 if (a
->start
< b
->start
)
1166 if (a
->start
> b
->start
)
1171 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1173 int memberships
= 0;
1176 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1177 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1178 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1180 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
1187 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1189 static int split_ull(unsigned long long n
, void *lo
, void *hi
)
1191 if (lo
== 0 || hi
== 0)
1193 __put_unaligned32(__cpu_to_le32((__u32
)n
), lo
);
1194 __put_unaligned32(__cpu_to_le32((n
>> 32)), hi
);
1198 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1200 return (unsigned long long)__le32_to_cpu(lo
) |
1201 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1204 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1208 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1211 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1215 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1218 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1222 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1225 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1229 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1232 static unsigned long long imsm_dev_size(struct imsm_dev
*dev
)
1236 return join_u32(dev
->size_low
, dev
->size_high
);
1239 static unsigned long long migr_chkp_area_pba(struct migr_record
*migr_rec
)
1241 if (migr_rec
== NULL
)
1243 return join_u32(migr_rec
->ckpt_area_pba_lo
,
1244 migr_rec
->ckpt_area_pba_hi
);
1247 static unsigned long long current_migr_unit(struct migr_record
*migr_rec
)
1249 if (migr_rec
== NULL
)
1251 return join_u32(migr_rec
->curr_migr_unit_lo
,
1252 migr_rec
->curr_migr_unit_hi
);
1255 static unsigned long long migr_dest_1st_member_lba(struct migr_record
*migr_rec
)
1257 if (migr_rec
== NULL
)
1259 return join_u32(migr_rec
->dest_1st_member_lba_lo
,
1260 migr_rec
->dest_1st_member_lba_hi
);
1263 static unsigned long long get_num_migr_units(struct migr_record
*migr_rec
)
1265 if (migr_rec
== NULL
)
1267 return join_u32(migr_rec
->num_migr_units_lo
,
1268 migr_rec
->num_migr_units_hi
);
1271 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1273 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1276 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1278 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1281 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1283 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1286 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1288 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1291 static void set_imsm_dev_size(struct imsm_dev
*dev
, unsigned long long n
)
1293 split_ull(n
, &dev
->size_low
, &dev
->size_high
);
1296 static void set_migr_chkp_area_pba(struct migr_record
*migr_rec
,
1297 unsigned long long n
)
1299 split_ull(n
, &migr_rec
->ckpt_area_pba_lo
, &migr_rec
->ckpt_area_pba_hi
);
1302 static void set_current_migr_unit(struct migr_record
*migr_rec
,
1303 unsigned long long n
)
1305 split_ull(n
, &migr_rec
->curr_migr_unit_lo
,
1306 &migr_rec
->curr_migr_unit_hi
);
1309 static void set_migr_dest_1st_member_lba(struct migr_record
*migr_rec
,
1310 unsigned long long n
)
1312 split_ull(n
, &migr_rec
->dest_1st_member_lba_lo
,
1313 &migr_rec
->dest_1st_member_lba_hi
);
1316 static void set_num_migr_units(struct migr_record
*migr_rec
,
1317 unsigned long long n
)
1319 split_ull(n
, &migr_rec
->num_migr_units_lo
,
1320 &migr_rec
->num_migr_units_hi
);
1323 static unsigned long long per_dev_array_size(struct imsm_map
*map
)
1325 unsigned long long array_size
= 0;
1330 array_size
= num_data_stripes(map
) * map
->blocks_per_strip
;
1331 if (get_imsm_raid_level(map
) == 1 || get_imsm_raid_level(map
) == 10)
1337 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
,
1338 int get_minimal_reservation
)
1340 /* find a list of used extents on the given physical device */
1341 struct extent
*rv
, *e
;
1343 int memberships
= count_memberships(dl
, super
);
1346 /* trim the reserved area for spares, so they can join any array
1347 * regardless of whether the OROM has assigned sectors from the
1348 * IMSM_RESERVED_SECTORS region
1350 if (dl
->index
== -1 || get_minimal_reservation
)
1351 reservation
= imsm_min_reserved_sectors(super
);
1353 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1355 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
1358 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1359 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1360 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1362 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1363 e
->start
= pba_of_lba0(map
);
1364 e
->size
= per_dev_array_size(map
);
1368 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1370 /* determine the start of the metadata
1371 * when no raid devices are defined use the default
1372 * ...otherwise allow the metadata to truncate the value
1373 * as is the case with older versions of imsm
1376 struct extent
*last
= &rv
[memberships
- 1];
1377 unsigned long long remainder
;
1379 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1380 /* round down to 1k block to satisfy precision of the kernel
1384 /* make sure remainder is still sane */
1385 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1386 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1387 if (reservation
> remainder
)
1388 reservation
= remainder
;
1390 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1395 /* try to determine how much space is reserved for metadata from
1396 * the last get_extents() entry, otherwise fallback to the
1399 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1405 /* for spares just return a minimal reservation which will grow
1406 * once the spare is picked up by an array
1408 if (dl
->index
== -1)
1409 return MPB_SECTOR_CNT
;
1411 e
= get_extents(super
, dl
, 0);
1413 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1415 /* scroll to last entry */
1416 for (i
= 0; e
[i
].size
; i
++)
1419 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1426 static int is_spare(struct imsm_disk
*disk
)
1428 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1431 static int is_configured(struct imsm_disk
*disk
)
1433 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1436 static int is_failed(struct imsm_disk
*disk
)
1438 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1441 static int is_journal(struct imsm_disk
*disk
)
1443 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1446 /* round array size down to closest MB and ensure it splits evenly
1449 static unsigned long long round_size_to_mb(unsigned long long size
, unsigned int
1453 size
= (size
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
1459 static int able_to_resync(int raid_level
, int missing_disks
)
1461 int max_missing_disks
= 0;
1463 switch (raid_level
) {
1465 max_missing_disks
= 1;
1468 max_missing_disks
= 0;
1470 return missing_disks
<= max_missing_disks
;
1473 /* try to determine how much space is reserved for metadata from
1474 * the last get_extents() entry on the smallest active disk,
1475 * otherwise fallback to the default
1477 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1481 unsigned long long min_active
;
1483 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1484 struct dl
*dl
, *dl_min
= NULL
;
1490 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1493 unsigned long long blocks
= total_blocks(&dl
->disk
);
1494 if (blocks
< min_active
|| min_active
== 0) {
1496 min_active
= blocks
;
1502 /* find last lba used by subarrays on the smallest active disk */
1503 e
= get_extents(super
, dl_min
, 0);
1506 for (i
= 0; e
[i
].size
; i
++)
1509 remainder
= min_active
- e
[i
].start
;
1512 /* to give priority to recovery we should not require full
1513 IMSM_RESERVED_SECTORS from the spare */
1514 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1516 /* if real reservation is smaller use that value */
1517 return (remainder
< rv
) ? remainder
: rv
;
1521 * Return minimum size of a spare and sector size
1522 * that can be used in this array
1524 int get_spare_criteria_imsm(struct supertype
*st
, struct spare_criteria
*c
)
1526 struct intel_super
*super
= st
->sb
;
1530 unsigned long long size
= 0;
1537 /* find first active disk in array */
1539 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1543 /* find last lba used by subarrays */
1544 e
= get_extents(super
, dl
, 0);
1547 for (i
= 0; e
[i
].size
; i
++)
1550 size
= e
[i
-1].start
+ e
[i
-1].size
;
1553 /* add the amount of space needed for metadata */
1554 size
+= imsm_min_reserved_sectors(super
);
1556 c
->min_size
= size
* 512;
1557 c
->sector_size
= super
->sector_size
;
1562 static int is_gen_migration(struct imsm_dev
*dev
);
1564 #define IMSM_4K_DIV 8
1566 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1567 struct imsm_dev
*dev
);
1569 static void print_imsm_dev(struct intel_super
*super
,
1570 struct imsm_dev
*dev
,
1576 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1577 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1581 printf("[%.16s]:\n", dev
->volume
);
1582 printf(" Subarray : %d\n", super
->current_vol
);
1583 printf(" UUID : %s\n", uuid
);
1584 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1586 printf(" <-- %d", get_imsm_raid_level(map2
));
1588 printf(" Members : %d", map
->num_members
);
1590 printf(" <-- %d", map2
->num_members
);
1592 printf(" Slots : [");
1593 for (i
= 0; i
< map
->num_members
; i
++) {
1594 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1595 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1600 for (i
= 0; i
< map2
->num_members
; i
++) {
1601 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1602 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1607 printf(" Failed disk : ");
1608 if (map
->failed_disk_num
== 0xff)
1611 printf("%i", map
->failed_disk_num
);
1613 slot
= get_imsm_disk_slot(map
, disk_idx
);
1615 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1616 printf(" This Slot : %d%s\n", slot
,
1617 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1619 printf(" This Slot : ?\n");
1620 printf(" Sector Size : %u\n", super
->sector_size
);
1621 sz
= imsm_dev_size(dev
);
1622 printf(" Array Size : %llu%s\n",
1623 (unsigned long long)sz
* 512 / super
->sector_size
,
1624 human_size(sz
* 512));
1625 sz
= blocks_per_member(map
);
1626 printf(" Per Dev Size : %llu%s\n",
1627 (unsigned long long)sz
* 512 / super
->sector_size
,
1628 human_size(sz
* 512));
1629 printf(" Sector Offset : %llu\n",
1631 printf(" Num Stripes : %llu\n",
1632 num_data_stripes(map
));
1633 printf(" Chunk Size : %u KiB",
1634 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1636 printf(" <-- %u KiB",
1637 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1639 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1640 printf(" Migrate State : ");
1641 if (dev
->vol
.migr_state
) {
1642 if (migr_type(dev
) == MIGR_INIT
)
1643 printf("initialize\n");
1644 else if (migr_type(dev
) == MIGR_REBUILD
)
1645 printf("rebuild\n");
1646 else if (migr_type(dev
) == MIGR_VERIFY
)
1648 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1649 printf("general migration\n");
1650 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1651 printf("state change\n");
1652 else if (migr_type(dev
) == MIGR_REPAIR
)
1655 printf("<unknown:%d>\n", migr_type(dev
));
1658 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1659 if (dev
->vol
.migr_state
) {
1660 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1662 printf(" <-- %s", map_state_str
[map
->map_state
]);
1663 printf("\n Checkpoint : %u ",
1664 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1665 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1668 printf("(%llu)", (unsigned long long)
1669 blocks_per_migr_unit(super
, dev
));
1672 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1674 printf(" RWH Policy : ");
1675 if (dev
->rwh_policy
== RWH_OFF
|| dev
->rwh_policy
== RWH_MULTIPLE_OFF
)
1677 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1678 printf("PPL distributed\n");
1679 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1680 printf("PPL journaling drive\n");
1681 else if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
1682 printf("Multiple distributed PPLs\n");
1683 else if (dev
->rwh_policy
== RWH_MULTIPLE_PPLS_JOURNALING_DRIVE
)
1684 printf("Multiple PPLs on journaling drive\n");
1686 printf("<unknown:%d>\n", dev
->rwh_policy
);
1688 printf(" Volume ID : %u\n", dev
->my_vol_raid_dev_num
);
1691 static void print_imsm_disk(struct imsm_disk
*disk
,
1694 unsigned int sector_size
) {
1695 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1698 if (index
< -1 || !disk
)
1702 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1704 printf(" Disk%02d Serial : %s\n", index
, str
);
1706 printf(" Disk Serial : %s\n", str
);
1707 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1708 is_configured(disk
) ? " active" : "",
1709 is_failed(disk
) ? " failed" : "",
1710 is_journal(disk
) ? " journal" : "");
1711 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1712 sz
= total_blocks(disk
) - reserved
;
1713 printf(" Usable Size : %llu%s\n",
1714 (unsigned long long)sz
* 512 / sector_size
,
1715 human_size(sz
* 512));
1718 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1720 struct migr_record
*migr_rec
= super
->migr_rec
;
1722 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1723 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1724 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1725 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1726 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1727 set_migr_chkp_area_pba(migr_rec
,
1728 migr_chkp_area_pba(migr_rec
) / IMSM_4K_DIV
);
1729 set_migr_dest_1st_member_lba(migr_rec
,
1730 migr_dest_1st_member_lba(migr_rec
) / IMSM_4K_DIV
);
1733 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1735 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1738 void convert_to_4k(struct intel_super
*super
)
1740 struct imsm_super
*mpb
= super
->anchor
;
1741 struct imsm_disk
*disk
;
1743 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1745 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1746 disk
= __get_imsm_disk(mpb
, i
);
1748 convert_to_4k_imsm_disk(disk
);
1750 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1751 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1752 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1754 set_imsm_dev_size(dev
, imsm_dev_size(dev
)/IMSM_4K_DIV
);
1755 dev
->vol
.curr_migr_unit
/= IMSM_4K_DIV
;
1758 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1759 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1760 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1762 if (dev
->vol
.migr_state
) {
1764 map
= get_imsm_map(dev
, MAP_1
);
1765 set_blocks_per_member(map
,
1766 blocks_per_member(map
)/IMSM_4K_DIV
);
1767 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1768 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1772 struct bbm_log
*log
= (void *)mpb
+
1773 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1776 for (i
= 0; i
< log
->entry_count
; i
++) {
1777 struct bbm_log_entry
*entry
=
1778 &log
->marked_block_entries
[i
];
1780 __u8 count
= entry
->marked_count
+ 1;
1781 unsigned long long sector
=
1782 __le48_to_cpu(&entry
->defective_block_start
);
1784 entry
->defective_block_start
=
1785 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1786 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1790 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1793 void examine_migr_rec_imsm(struct intel_super
*super
)
1795 struct migr_record
*migr_rec
= super
->migr_rec
;
1796 struct imsm_super
*mpb
= super
->anchor
;
1799 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1800 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1801 struct imsm_map
*map
;
1804 if (is_gen_migration(dev
) == 0)
1807 printf("\nMigration Record Information:");
1809 /* first map under migration */
1810 map
= get_imsm_map(dev
, MAP_0
);
1812 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1813 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1814 printf(" Empty\n ");
1815 printf("Examine one of first two disks in array\n");
1818 printf("\n Status : ");
1819 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1822 printf("Contains Data\n");
1823 printf(" Current Unit : %llu\n",
1824 current_migr_unit(migr_rec
));
1825 printf(" Family : %u\n",
1826 __le32_to_cpu(migr_rec
->family_num
));
1827 printf(" Ascending : %u\n",
1828 __le32_to_cpu(migr_rec
->ascending_migr
));
1829 printf(" Blocks Per Unit : %u\n",
1830 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1831 printf(" Dest. Depth Per Unit : %u\n",
1832 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1833 printf(" Checkpoint Area pba : %llu\n",
1834 migr_chkp_area_pba(migr_rec
));
1835 printf(" First member lba : %llu\n",
1836 migr_dest_1st_member_lba(migr_rec
));
1837 printf(" Total Number of Units : %llu\n",
1838 get_num_migr_units(migr_rec
));
1839 printf(" Size of volume : %llu\n",
1840 join_u32(migr_rec
->post_migr_vol_cap
,
1841 migr_rec
->post_migr_vol_cap_hi
));
1842 printf(" Record was read from : %u\n",
1843 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1849 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1851 struct migr_record
*migr_rec
= super
->migr_rec
;
1853 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1854 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1855 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1856 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1857 &migr_rec
->post_migr_vol_cap
,
1858 &migr_rec
->post_migr_vol_cap_hi
);
1859 set_migr_chkp_area_pba(migr_rec
,
1860 migr_chkp_area_pba(migr_rec
) * IMSM_4K_DIV
);
1861 set_migr_dest_1st_member_lba(migr_rec
,
1862 migr_dest_1st_member_lba(migr_rec
) * IMSM_4K_DIV
);
1865 void convert_from_4k(struct intel_super
*super
)
1867 struct imsm_super
*mpb
= super
->anchor
;
1868 struct imsm_disk
*disk
;
1870 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1872 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1873 disk
= __get_imsm_disk(mpb
, i
);
1875 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1878 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1879 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1880 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1882 set_imsm_dev_size(dev
, imsm_dev_size(dev
)*IMSM_4K_DIV
);
1883 dev
->vol
.curr_migr_unit
*= IMSM_4K_DIV
;
1886 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1887 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1888 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1890 if (dev
->vol
.migr_state
) {
1892 map
= get_imsm_map(dev
, MAP_1
);
1893 set_blocks_per_member(map
,
1894 blocks_per_member(map
)*IMSM_4K_DIV
);
1895 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1896 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1900 struct bbm_log
*log
= (void *)mpb
+
1901 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1904 for (i
= 0; i
< log
->entry_count
; i
++) {
1905 struct bbm_log_entry
*entry
=
1906 &log
->marked_block_entries
[i
];
1908 __u8 count
= entry
->marked_count
+ 1;
1909 unsigned long long sector
=
1910 __le48_to_cpu(&entry
->defective_block_start
);
1912 entry
->defective_block_start
=
1913 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1914 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1918 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1921 /*******************************************************************************
1922 * function: imsm_check_attributes
1923 * Description: Function checks if features represented by attributes flags
1924 * are supported by mdadm.
1926 * attributes - Attributes read from metadata
1928 * 0 - passed attributes contains unsupported features flags
1929 * 1 - all features are supported
1930 ******************************************************************************/
1931 static int imsm_check_attributes(__u32 attributes
)
1934 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1936 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1938 not_supported
&= attributes
;
1939 if (not_supported
) {
1940 pr_err("(IMSM): Unsupported attributes : %x\n",
1941 (unsigned)__le32_to_cpu(not_supported
));
1942 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1943 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1944 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1946 if (not_supported
& MPB_ATTRIB_2TB
) {
1947 dprintf("\t\tMPB_ATTRIB_2TB\n");
1948 not_supported
^= MPB_ATTRIB_2TB
;
1950 if (not_supported
& MPB_ATTRIB_RAID0
) {
1951 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1952 not_supported
^= MPB_ATTRIB_RAID0
;
1954 if (not_supported
& MPB_ATTRIB_RAID1
) {
1955 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1956 not_supported
^= MPB_ATTRIB_RAID1
;
1958 if (not_supported
& MPB_ATTRIB_RAID10
) {
1959 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1960 not_supported
^= MPB_ATTRIB_RAID10
;
1962 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1963 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1964 not_supported
^= MPB_ATTRIB_RAID1E
;
1966 if (not_supported
& MPB_ATTRIB_RAID5
) {
1967 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1968 not_supported
^= MPB_ATTRIB_RAID5
;
1970 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1971 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1972 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1974 if (not_supported
& MPB_ATTRIB_BBM
) {
1975 dprintf("\t\tMPB_ATTRIB_BBM\n");
1976 not_supported
^= MPB_ATTRIB_BBM
;
1978 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1979 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1980 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1982 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1983 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1984 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1986 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1987 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1988 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1990 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1991 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1992 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1994 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1995 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1996 not_supported
^= MPB_ATTRIB_NEVER_USE
;
2000 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
2008 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
2010 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
2012 struct intel_super
*super
= st
->sb
;
2013 struct imsm_super
*mpb
= super
->anchor
;
2014 char str
[MAX_SIGNATURE_LENGTH
];
2019 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2021 time_t creation_time
;
2023 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
2024 str
[MPB_SIG_LEN
-1] = '\0';
2025 printf(" Magic : %s\n", str
);
2026 printf(" Version : %s\n", get_imsm_version(mpb
));
2027 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
2028 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
2029 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
2030 creation_time
= __le64_to_cpu(mpb
->creation_time
);
2031 printf(" Creation Time : %.24s\n",
2032 creation_time
? ctime(&creation_time
) : "Unknown");
2033 printf(" Attributes : ");
2034 if (imsm_check_attributes(mpb
->attributes
))
2035 printf("All supported\n");
2037 printf("not supported\n");
2038 getinfo_super_imsm(st
, &info
, NULL
);
2039 fname_from_uuid(st
, &info
, nbuf
, ':');
2040 printf(" UUID : %s\n", nbuf
+ 5);
2041 sum
= __le32_to_cpu(mpb
->check_sum
);
2042 printf(" Checksum : %08x %s\n", sum
,
2043 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
2044 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
2045 printf(" Disks : %d\n", mpb
->num_disks
);
2046 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
2047 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
2048 super
->disks
->index
, reserved
, super
->sector_size
);
2049 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
2050 struct bbm_log
*log
= super
->bbm_log
;
2053 printf("Bad Block Management Log:\n");
2054 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
2055 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
2056 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
2058 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2060 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2062 super
->current_vol
= i
;
2063 getinfo_super_imsm(st
, &info
, NULL
);
2064 fname_from_uuid(st
, &info
, nbuf
, ':');
2065 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
2067 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2068 if (i
== super
->disks
->index
)
2070 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
2071 super
->sector_size
);
2074 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2075 if (dl
->index
== -1)
2076 print_imsm_disk(&dl
->disk
, -1, reserved
,
2077 super
->sector_size
);
2079 examine_migr_rec_imsm(super
);
2082 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
2084 /* We just write a generic IMSM ARRAY entry */
2087 struct intel_super
*super
= st
->sb
;
2089 if (!super
->anchor
->num_raid_devs
) {
2090 printf("ARRAY metadata=imsm\n");
2094 getinfo_super_imsm(st
, &info
, NULL
);
2095 fname_from_uuid(st
, &info
, nbuf
, ':');
2096 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
2099 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
2101 /* We just write a generic IMSM ARRAY entry */
2105 struct intel_super
*super
= st
->sb
;
2108 if (!super
->anchor
->num_raid_devs
)
2111 getinfo_super_imsm(st
, &info
, NULL
);
2112 fname_from_uuid(st
, &info
, nbuf
, ':');
2113 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2114 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2116 super
->current_vol
= i
;
2117 getinfo_super_imsm(st
, &info
, NULL
);
2118 fname_from_uuid(st
, &info
, nbuf1
, ':');
2119 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
2120 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
2124 static void export_examine_super_imsm(struct supertype
*st
)
2126 struct intel_super
*super
= st
->sb
;
2127 struct imsm_super
*mpb
= super
->anchor
;
2131 getinfo_super_imsm(st
, &info
, NULL
);
2132 fname_from_uuid(st
, &info
, nbuf
, ':');
2133 printf("MD_METADATA=imsm\n");
2134 printf("MD_LEVEL=container\n");
2135 printf("MD_UUID=%s\n", nbuf
+5);
2136 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
2137 printf("MD_CREATION_TIME=%llu\n", __le64_to_cpu(mpb
->creation_time
));
2140 static void detail_super_imsm(struct supertype
*st
, char *homehost
,
2145 struct intel_super
*super
= st
->sb
;
2146 int temp_vol
= super
->current_vol
;
2149 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2151 getinfo_super_imsm(st
, &info
, NULL
);
2152 fname_from_uuid(st
, &info
, nbuf
, ':');
2153 printf("\n UUID : %s\n", nbuf
+ 5);
2155 super
->current_vol
= temp_vol
;
2158 static void brief_detail_super_imsm(struct supertype
*st
, char *subarray
)
2162 struct intel_super
*super
= st
->sb
;
2163 int temp_vol
= super
->current_vol
;
2166 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2168 getinfo_super_imsm(st
, &info
, NULL
);
2169 fname_from_uuid(st
, &info
, nbuf
, ':');
2170 printf(" UUID=%s", nbuf
+ 5);
2172 super
->current_vol
= temp_vol
;
2175 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
,
2176 size_t serial_buf_len
);
2177 static void fd2devname(int fd
, char *name
);
2179 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2181 /* dump an unsorted list of devices attached to AHCI Intel storage
2182 * controller, as well as non-connected ports
2184 int hba_len
= strlen(hba_path
) + 1;
2189 unsigned long port_mask
= (1 << port_count
) - 1;
2191 if (port_count
> (int)sizeof(port_mask
) * 8) {
2193 pr_err("port_count %d out of range\n", port_count
);
2197 /* scroll through /sys/dev/block looking for devices attached to
2200 dir
= opendir("/sys/dev/block");
2204 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2215 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2217 path
= devt_to_devpath(makedev(major
, minor
));
2220 if (!path_attached_to_hba(path
, hba_path
)) {
2226 /* retrieve the scsi device type */
2227 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2229 pr_err("failed to allocate 'device'\n");
2233 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2234 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2236 pr_err("failed to read device type for %s\n",
2242 type
= strtoul(buf
, NULL
, 10);
2244 /* if it's not a disk print the vendor and model */
2245 if (!(type
== 0 || type
== 7 || type
== 14)) {
2248 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2249 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2250 strncpy(vendor
, buf
, sizeof(vendor
));
2251 vendor
[sizeof(vendor
) - 1] = '\0';
2252 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2253 while (isspace(*c
) || *c
== '\0')
2257 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2258 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2259 strncpy(model
, buf
, sizeof(model
));
2260 model
[sizeof(model
) - 1] = '\0';
2261 c
= (char *) &model
[sizeof(model
) - 1];
2262 while (isspace(*c
) || *c
== '\0')
2266 if (vendor
[0] && model
[0])
2267 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2269 switch (type
) { /* numbers from hald/linux/device.c */
2270 case 1: sprintf(buf
, "tape"); break;
2271 case 2: sprintf(buf
, "printer"); break;
2272 case 3: sprintf(buf
, "processor"); break;
2274 case 5: sprintf(buf
, "cdrom"); break;
2275 case 6: sprintf(buf
, "scanner"); break;
2276 case 8: sprintf(buf
, "media_changer"); break;
2277 case 9: sprintf(buf
, "comm"); break;
2278 case 12: sprintf(buf
, "raid"); break;
2279 default: sprintf(buf
, "unknown");
2285 /* chop device path to 'host%d' and calculate the port number */
2286 c
= strchr(&path
[hba_len
], '/');
2289 pr_err("%s - invalid path name\n", path
+ hba_len
);
2294 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2295 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2299 *c
= '/'; /* repair the full string */
2300 pr_err("failed to determine port number for %s\n",
2307 /* mark this port as used */
2308 port_mask
&= ~(1 << port
);
2310 /* print out the device information */
2312 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2316 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2318 printf(" Port%d : - disk info unavailable -\n", port
);
2320 fd2devname(fd
, buf
);
2321 printf(" Port%d : %s", port
, buf
);
2322 if (imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2324 printf(" (%s)\n", buf
);
2339 for (i
= 0; i
< port_count
; i
++)
2340 if (port_mask
& (1 << i
))
2341 printf(" Port%d : - no device attached -\n", i
);
2347 static int print_nvme_info(struct sys_dev
*hba
)
2355 dir
= opendir("/sys/block/");
2359 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2360 if (strstr(ent
->d_name
, "nvme")) {
2361 sprintf(buf
, "/sys/block/%s", ent
->d_name
);
2362 rp
= realpath(buf
, NULL
);
2365 if (path_attached_to_hba(rp
, hba
->path
)) {
2366 fd
= open_dev(ent
->d_name
);
2372 fd2devname(fd
, buf
);
2373 if (hba
->type
== SYS_DEV_VMD
)
2374 printf(" NVMe under VMD : %s", buf
);
2375 else if (hba
->type
== SYS_DEV_NVME
)
2376 printf(" NVMe Device : %s", buf
);
2377 if (!imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2379 printf(" (%s)\n", buf
);
2392 static void print_found_intel_controllers(struct sys_dev
*elem
)
2394 for (; elem
; elem
= elem
->next
) {
2395 pr_err("found Intel(R) ");
2396 if (elem
->type
== SYS_DEV_SATA
)
2397 fprintf(stderr
, "SATA ");
2398 else if (elem
->type
== SYS_DEV_SAS
)
2399 fprintf(stderr
, "SAS ");
2400 else if (elem
->type
== SYS_DEV_NVME
)
2401 fprintf(stderr
, "NVMe ");
2403 if (elem
->type
== SYS_DEV_VMD
)
2404 fprintf(stderr
, "VMD domain");
2406 fprintf(stderr
, "RAID controller");
2409 fprintf(stderr
, " at %s", elem
->pci_id
);
2410 fprintf(stderr
, ".\n");
2415 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2422 if ((dir
= opendir(hba_path
)) == NULL
)
2425 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2428 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2429 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2431 if (*port_count
== 0)
2433 else if (host
< host_base
)
2436 if (host
+ 1 > *port_count
+ host_base
)
2437 *port_count
= host
+ 1 - host_base
;
2443 static void print_imsm_capability(const struct imsm_orom
*orom
)
2445 printf(" Platform : Intel(R) ");
2446 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2447 printf("Matrix Storage Manager\n");
2448 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2449 printf("Virtual RAID on CPU\n");
2451 printf("Rapid Storage Technology%s\n",
2452 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2453 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2454 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2455 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2456 printf(" RAID Levels :%s%s%s%s%s\n",
2457 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2458 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2459 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2460 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2461 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2462 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2463 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2464 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2465 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2466 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2467 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2468 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2469 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2470 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2471 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2472 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2473 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2474 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2475 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2476 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2477 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2478 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2479 printf(" 2TB volumes :%s supported\n",
2480 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2481 printf(" 2TB disks :%s supported\n",
2482 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2483 printf(" Max Disks : %d\n", orom
->tds
);
2484 printf(" Max Volumes : %d per array, %d per %s\n",
2485 orom
->vpa
, orom
->vphba
,
2486 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2490 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2492 printf("MD_FIRMWARE_TYPE=imsm\n");
2493 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2494 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2495 orom
->hotfix_ver
, orom
->build
);
2496 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2497 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2498 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2499 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2500 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2501 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2502 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2503 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2504 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2505 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2506 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2507 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2508 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2509 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2510 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2511 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2512 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2513 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2514 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2515 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2516 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2517 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2518 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2519 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2520 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2521 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2522 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2523 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2526 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2528 /* There are two components to imsm platform support, the ahci SATA
2529 * controller and the option-rom. To find the SATA controller we
2530 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2531 * controller with the Intel vendor id is present. This approach
2532 * allows mdadm to leverage the kernel's ahci detection logic, with the
2533 * caveat that if ahci.ko is not loaded mdadm will not be able to
2534 * detect platform raid capabilities. The option-rom resides in a
2535 * platform "Adapter ROM". We scan for its signature to retrieve the
2536 * platform capabilities. If raid support is disabled in the BIOS the
2537 * option-rom capability structure will not be available.
2539 struct sys_dev
*list
, *hba
;
2544 if (enumerate_only
) {
2545 if (check_env("IMSM_NO_PLATFORM"))
2547 list
= find_intel_devices();
2550 for (hba
= list
; hba
; hba
= hba
->next
) {
2551 if (find_imsm_capability(hba
)) {
2561 list
= find_intel_devices();
2564 pr_err("no active Intel(R) RAID controller found.\n");
2566 } else if (verbose
> 0)
2567 print_found_intel_controllers(list
);
2569 for (hba
= list
; hba
; hba
= hba
->next
) {
2570 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2572 if (!find_imsm_capability(hba
)) {
2574 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2575 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2576 get_sys_dev_type(hba
->type
));
2582 if (controller_path
&& result
== 1) {
2583 pr_err("no active Intel(R) RAID controller found under %s\n",
2588 const struct orom_entry
*entry
;
2590 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2591 if (entry
->type
== SYS_DEV_VMD
) {
2592 print_imsm_capability(&entry
->orom
);
2593 printf(" 3rd party NVMe :%s supported\n",
2594 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2595 for (hba
= list
; hba
; hba
= hba
->next
) {
2596 if (hba
->type
== SYS_DEV_VMD
) {
2598 printf(" I/O Controller : %s (%s)\n",
2599 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2600 if (print_nvme_info(hba
)) {
2602 pr_err("failed to get devices attached to VMD domain.\n");
2611 print_imsm_capability(&entry
->orom
);
2612 if (entry
->type
== SYS_DEV_NVME
) {
2613 for (hba
= list
; hba
; hba
= hba
->next
) {
2614 if (hba
->type
== SYS_DEV_NVME
)
2615 print_nvme_info(hba
);
2621 struct devid_list
*devid
;
2622 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2623 hba
= device_by_id(devid
->devid
);
2627 printf(" I/O Controller : %s (%s)\n",
2628 hba
->path
, get_sys_dev_type(hba
->type
));
2629 if (hba
->type
== SYS_DEV_SATA
) {
2630 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2631 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2633 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2644 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2646 struct sys_dev
*list
, *hba
;
2649 list
= find_intel_devices();
2652 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2657 for (hba
= list
; hba
; hba
= hba
->next
) {
2658 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2660 if (!find_imsm_capability(hba
) && verbose
> 0) {
2662 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2663 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2669 const struct orom_entry
*entry
;
2671 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2672 if (entry
->type
== SYS_DEV_VMD
) {
2673 for (hba
= list
; hba
; hba
= hba
->next
)
2674 print_imsm_capability_export(&entry
->orom
);
2677 print_imsm_capability_export(&entry
->orom
);
2683 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2685 /* the imsm metadata format does not specify any host
2686 * identification information. We return -1 since we can never
2687 * confirm nor deny whether a given array is "meant" for this
2688 * host. We rely on compare_super and the 'family_num' fields to
2689 * exclude member disks that do not belong, and we rely on
2690 * mdadm.conf to specify the arrays that should be assembled.
2691 * Auto-assembly may still pick up "foreign" arrays.
2697 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2699 /* The uuid returned here is used for:
2700 * uuid to put into bitmap file (Create, Grow)
2701 * uuid for backup header when saving critical section (Grow)
2702 * comparing uuids when re-adding a device into an array
2703 * In these cases the uuid required is that of the data-array,
2704 * not the device-set.
2705 * uuid to recognise same set when adding a missing device back
2706 * to an array. This is a uuid for the device-set.
2708 * For each of these we can make do with a truncated
2709 * or hashed uuid rather than the original, as long as
2711 * In each case the uuid required is that of the data-array,
2712 * not the device-set.
2714 /* imsm does not track uuid's so we synthesis one using sha1 on
2715 * - The signature (Which is constant for all imsm array, but no matter)
2716 * - the orig_family_num of the container
2717 * - the index number of the volume
2718 * - the 'serial' number of the volume.
2719 * Hopefully these are all constant.
2721 struct intel_super
*super
= st
->sb
;
2724 struct sha1_ctx ctx
;
2725 struct imsm_dev
*dev
= NULL
;
2728 /* some mdadm versions failed to set ->orig_family_num, in which
2729 * case fall back to ->family_num. orig_family_num will be
2730 * fixed up with the first metadata update.
2732 family_num
= super
->anchor
->orig_family_num
;
2733 if (family_num
== 0)
2734 family_num
= super
->anchor
->family_num
;
2735 sha1_init_ctx(&ctx
);
2736 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2737 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2738 if (super
->current_vol
>= 0)
2739 dev
= get_imsm_dev(super
, super
->current_vol
);
2741 __u32 vol
= super
->current_vol
;
2742 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2743 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2745 sha1_finish_ctx(&ctx
, buf
);
2746 memcpy(uuid
, buf
, 4*4);
2751 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2753 __u8
*v
= get_imsm_version(mpb
);
2754 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2755 char major
[] = { 0, 0, 0 };
2756 char minor
[] = { 0 ,0, 0 };
2757 char patch
[] = { 0, 0, 0 };
2758 char *ver_parse
[] = { major
, minor
, patch
};
2762 while (*v
!= '\0' && v
< end
) {
2763 if (*v
!= '.' && j
< 2)
2764 ver_parse
[i
][j
++] = *v
;
2772 *m
= strtol(minor
, NULL
, 0);
2773 *p
= strtol(patch
, NULL
, 0);
2777 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2779 /* migr_strip_size when repairing or initializing parity */
2780 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2781 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2783 switch (get_imsm_raid_level(map
)) {
2788 return 128*1024 >> 9;
2792 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2794 /* migr_strip_size when rebuilding a degraded disk, no idea why
2795 * this is different than migr_strip_size_resync(), but it's good
2798 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2799 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2801 switch (get_imsm_raid_level(map
)) {
2804 if (map
->num_members
% map
->num_domains
== 0)
2805 return 128*1024 >> 9;
2809 return max((__u32
) 64*1024 >> 9, chunk
);
2811 return 128*1024 >> 9;
2815 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2817 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2818 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2819 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2820 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2822 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2825 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2827 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2828 int level
= get_imsm_raid_level(lo
);
2830 if (level
== 1 || level
== 10) {
2831 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2833 return hi
->num_domains
;
2835 return num_stripes_per_unit_resync(dev
);
2838 static __u8
imsm_num_data_members(struct imsm_map
*map
)
2840 /* named 'imsm_' because raid0, raid1 and raid10
2841 * counter-intuitively have the same number of data disks
2843 switch (get_imsm_raid_level(map
)) {
2845 return map
->num_members
;
2849 return map
->num_members
/2;
2851 return map
->num_members
- 1;
2853 dprintf("unsupported raid level\n");
2858 static unsigned long long calc_component_size(struct imsm_map
*map
,
2859 struct imsm_dev
*dev
)
2861 unsigned long long component_size
;
2862 unsigned long long dev_size
= imsm_dev_size(dev
);
2863 long long calc_dev_size
= 0;
2864 unsigned int member_disks
= imsm_num_data_members(map
);
2866 if (member_disks
== 0)
2869 component_size
= per_dev_array_size(map
);
2870 calc_dev_size
= component_size
* member_disks
;
2872 /* Component size is rounded to 1MB so difference between size from
2873 * metadata and size calculated from num_data_stripes equals up to
2874 * 2048 blocks per each device. If the difference is higher it means
2875 * that array size was expanded and num_data_stripes was not updated.
2877 if (llabs(calc_dev_size
- (long long)dev_size
) >
2878 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
2879 component_size
= dev_size
/ member_disks
;
2880 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
2881 component_size
/ map
->blocks_per_strip
,
2882 num_data_stripes(map
));
2885 return component_size
;
2888 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2890 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2891 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2893 switch(get_imsm_raid_level(map
)) {
2896 return chunk
* map
->num_domains
;
2898 return chunk
* map
->num_members
;
2904 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2906 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2907 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2908 __u32 strip
= block
/ chunk
;
2910 switch (get_imsm_raid_level(map
)) {
2913 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2914 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2916 return vol_stripe
* chunk
+ block
% chunk
;
2918 __u32 stripe
= strip
/ (map
->num_members
- 1);
2920 return stripe
* chunk
+ block
% chunk
;
2927 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2928 struct imsm_dev
*dev
)
2930 /* calculate the conversion factor between per member 'blocks'
2931 * (md/{resync,rebuild}_start) and imsm migration units, return
2932 * 0 for the 'not migrating' and 'unsupported migration' cases
2934 if (!dev
->vol
.migr_state
)
2937 switch (migr_type(dev
)) {
2938 case MIGR_GEN_MIGR
: {
2939 struct migr_record
*migr_rec
= super
->migr_rec
;
2940 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2945 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2946 __u32 stripes_per_unit
;
2947 __u32 blocks_per_unit
;
2956 /* yes, this is really the translation of migr_units to
2957 * per-member blocks in the 'resync' case
2959 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2960 migr_chunk
= migr_strip_blocks_resync(dev
);
2961 disks
= imsm_num_data_members(map
);
2962 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2963 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2964 segment
= blocks_per_unit
/ stripe
;
2965 block_rel
= blocks_per_unit
- segment
* stripe
;
2966 parity_depth
= parity_segment_depth(dev
);
2967 block_map
= map_migr_block(dev
, block_rel
);
2968 return block_map
+ parity_depth
* segment
;
2970 case MIGR_REBUILD
: {
2971 __u32 stripes_per_unit
;
2974 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2975 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2976 return migr_chunk
* stripes_per_unit
;
2978 case MIGR_STATE_CHANGE
:
2984 static int imsm_level_to_layout(int level
)
2992 return ALGORITHM_LEFT_ASYMMETRIC
;
2999 /*******************************************************************************
3000 * Function: read_imsm_migr_rec
3001 * Description: Function reads imsm migration record from last sector of disk
3003 * fd : disk descriptor
3004 * super : metadata info
3008 ******************************************************************************/
3009 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
3012 unsigned int sector_size
= super
->sector_size
;
3013 unsigned long long dsize
;
3015 get_dev_size(fd
, NULL
, &dsize
);
3016 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
3018 pr_err("Cannot seek to anchor block: %s\n",
3022 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
3023 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3024 MIGR_REC_BUF_SECTORS
*sector_size
) {
3025 pr_err("Cannot read migr record block: %s\n",
3030 if (sector_size
== 4096)
3031 convert_from_4k_imsm_migr_rec(super
);
3037 static struct imsm_dev
*imsm_get_device_during_migration(
3038 struct intel_super
*super
)
3041 struct intel_dev
*dv
;
3043 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
3044 if (is_gen_migration(dv
->dev
))
3050 /*******************************************************************************
3051 * Function: load_imsm_migr_rec
3052 * Description: Function reads imsm migration record (it is stored at the last
3055 * super : imsm internal array info
3056 * info : general array info
3060 * -2 : no migration in progress
3061 ******************************************************************************/
3062 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
3069 struct imsm_dev
*dev
;
3070 struct imsm_map
*map
;
3073 /* find map under migration */
3074 dev
= imsm_get_device_during_migration(super
);
3075 /* nothing to load,no migration in progress?
3081 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
3082 /* read only from one of the first two slots */
3083 if ((sd
->disk
.raid_disk
< 0) ||
3084 (sd
->disk
.raid_disk
> 1))
3087 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3088 fd
= dev_open(nm
, O_RDONLY
);
3094 map
= get_imsm_map(dev
, MAP_0
);
3095 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3096 /* skip spare and failed disks
3100 /* read only from one of the first two slots */
3102 slot
= get_imsm_disk_slot(map
, dl
->index
);
3103 if (map
== NULL
|| slot
> 1 || slot
< 0)
3105 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3106 fd
= dev_open(nm
, O_RDONLY
);
3113 retval
= read_imsm_migr_rec(fd
, super
);
3121 /*******************************************************************************
3122 * function: imsm_create_metadata_checkpoint_update
3123 * Description: It creates update for checkpoint change.
3125 * super : imsm internal array info
3126 * u : pointer to prepared update
3129 * If length is equal to 0, input pointer u contains no update
3130 ******************************************************************************/
3131 static int imsm_create_metadata_checkpoint_update(
3132 struct intel_super
*super
,
3133 struct imsm_update_general_migration_checkpoint
**u
)
3136 int update_memory_size
= 0;
3138 dprintf("(enter)\n");
3144 /* size of all update data without anchor */
3145 update_memory_size
=
3146 sizeof(struct imsm_update_general_migration_checkpoint
);
3148 *u
= xcalloc(1, update_memory_size
);
3150 dprintf("error: cannot get memory\n");
3153 (*u
)->type
= update_general_migration_checkpoint
;
3154 (*u
)->curr_migr_unit
= current_migr_unit(super
->migr_rec
);
3155 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3157 return update_memory_size
;
3160 static void imsm_update_metadata_locally(struct supertype
*st
,
3161 void *buf
, int len
);
3163 /*******************************************************************************
3164 * Function: write_imsm_migr_rec
3165 * Description: Function writes imsm migration record
3166 * (at the last sector of disk)
3168 * super : imsm internal array info
3172 ******************************************************************************/
3173 static int write_imsm_migr_rec(struct supertype
*st
)
3175 struct intel_super
*super
= st
->sb
;
3176 unsigned int sector_size
= super
->sector_size
;
3177 unsigned long long dsize
;
3183 struct imsm_update_general_migration_checkpoint
*u
;
3184 struct imsm_dev
*dev
;
3185 struct imsm_map
*map
;
3187 /* find map under migration */
3188 dev
= imsm_get_device_during_migration(super
);
3189 /* if no migration, write buffer anyway to clear migr_record
3190 * on disk based on first available device
3193 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3194 super
->current_vol
);
3196 map
= get_imsm_map(dev
, MAP_0
);
3198 if (sector_size
== 4096)
3199 convert_to_4k_imsm_migr_rec(super
);
3200 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3203 /* skip failed and spare devices */
3206 /* write to 2 first slots only */
3208 slot
= get_imsm_disk_slot(map
, sd
->index
);
3209 if (map
== NULL
|| slot
> 1 || slot
< 0)
3212 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3213 fd
= dev_open(nm
, O_RDWR
);
3216 get_dev_size(fd
, NULL
, &dsize
);
3217 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3219 pr_err("Cannot seek to anchor block: %s\n",
3223 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3224 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3225 MIGR_REC_BUF_SECTORS
*sector_size
) {
3226 pr_err("Cannot write migr record block: %s\n",
3233 if (sector_size
== 4096)
3234 convert_from_4k_imsm_migr_rec(super
);
3235 /* update checkpoint information in metadata */
3236 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3238 dprintf("imsm: Cannot prepare update\n");
3241 /* update metadata locally */
3242 imsm_update_metadata_locally(st
, u
, len
);
3243 /* and possibly remotely */
3244 if (st
->update_tail
) {
3245 append_metadata_update(st
, u
, len
);
3246 /* during reshape we do all work inside metadata handler
3247 * manage_reshape(), so metadata update has to be triggered
3250 flush_metadata_updates(st
);
3251 st
->update_tail
= &st
->updates
;
3262 /* spare/missing disks activations are not allowe when
3263 * array/container performs reshape operation, because
3264 * all arrays in container works on the same disks set
3266 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3269 struct intel_dev
*i_dev
;
3270 struct imsm_dev
*dev
;
3272 /* check whole container
3274 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3276 if (is_gen_migration(dev
)) {
3277 /* No repair during any migration in container
3285 static unsigned long long imsm_component_size_alignment_check(int level
,
3287 unsigned int sector_size
,
3288 unsigned long long component_size
)
3290 unsigned int component_size_alignment
;
3292 /* check component size alignment
3294 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3296 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3297 level
, chunk_size
, component_size
,
3298 component_size_alignment
);
3300 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3301 dprintf("imsm: reported component size aligned from %llu ",
3303 component_size
-= component_size_alignment
;
3304 dprintf_cont("to %llu (%i).\n",
3305 component_size
, component_size_alignment
);
3308 return component_size
;
3311 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3313 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3314 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3316 return pba_of_lba0(map
) +
3317 (num_data_stripes(map
) * map
->blocks_per_strip
);
3320 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3322 struct intel_super
*super
= st
->sb
;
3323 struct migr_record
*migr_rec
= super
->migr_rec
;
3324 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3325 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3326 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3327 struct imsm_map
*map_to_analyse
= map
;
3329 int map_disks
= info
->array
.raid_disks
;
3331 memset(info
, 0, sizeof(*info
));
3333 map_to_analyse
= prev_map
;
3335 dl
= super
->current_disk
;
3337 info
->container_member
= super
->current_vol
;
3338 info
->array
.raid_disks
= map
->num_members
;
3339 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3340 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3341 info
->array
.md_minor
= -1;
3342 info
->array
.ctime
= 0;
3343 info
->array
.utime
= 0;
3344 info
->array
.chunk_size
=
3345 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3346 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3347 info
->custom_array_size
= imsm_dev_size(dev
);
3348 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3350 if (is_gen_migration(dev
)) {
3351 info
->reshape_active
= 1;
3352 info
->new_level
= get_imsm_raid_level(map
);
3353 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3354 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3355 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3356 if (info
->delta_disks
) {
3357 /* this needs to be applied to every array
3360 info
->reshape_active
= CONTAINER_RESHAPE
;
3362 /* We shape information that we give to md might have to be
3363 * modify to cope with md's requirement for reshaping arrays.
3364 * For example, when reshaping a RAID0, md requires it to be
3365 * presented as a degraded RAID4.
3366 * Also if a RAID0 is migrating to a RAID5 we need to specify
3367 * the array as already being RAID5, but the 'before' layout
3368 * is a RAID4-like layout.
3370 switch (info
->array
.level
) {
3372 switch(info
->new_level
) {
3374 /* conversion is happening as RAID4 */
3375 info
->array
.level
= 4;
3376 info
->array
.raid_disks
+= 1;
3379 /* conversion is happening as RAID5 */
3380 info
->array
.level
= 5;
3381 info
->array
.layout
= ALGORITHM_PARITY_N
;
3382 info
->delta_disks
-= 1;
3385 /* FIXME error message */
3386 info
->array
.level
= UnSet
;
3392 info
->new_level
= UnSet
;
3393 info
->new_layout
= UnSet
;
3394 info
->new_chunk
= info
->array
.chunk_size
;
3395 info
->delta_disks
= 0;
3399 info
->disk
.major
= dl
->major
;
3400 info
->disk
.minor
= dl
->minor
;
3401 info
->disk
.number
= dl
->index
;
3402 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3406 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3407 info
->component_size
= calc_component_size(map
, dev
);
3408 info
->component_size
= imsm_component_size_alignment_check(
3410 info
->array
.chunk_size
,
3412 info
->component_size
);
3413 info
->bb
.supported
= 1;
3415 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3416 info
->recovery_start
= MaxSector
;
3418 if (info
->array
.level
== 5 &&
3419 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3420 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3421 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3422 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3423 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3424 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3426 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3428 } else if (info
->array
.level
<= 0) {
3429 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3431 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3434 info
->reshape_progress
= 0;
3435 info
->resync_start
= MaxSector
;
3436 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3437 !(info
->array
.state
& 1)) &&
3438 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3439 info
->resync_start
= 0;
3441 if (dev
->vol
.migr_state
) {
3442 switch (migr_type(dev
)) {
3445 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3447 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3449 info
->resync_start
= blocks_per_unit
* units
;
3452 case MIGR_GEN_MIGR
: {
3453 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3455 __u64 units
= current_migr_unit(migr_rec
);
3458 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3460 (get_num_migr_units(migr_rec
)-1)) &&
3461 (super
->migr_rec
->rec_status
==
3462 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3465 info
->reshape_progress
= blocks_per_unit
* units
;
3467 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3468 (unsigned long long)units
,
3469 (unsigned long long)blocks_per_unit
,
3470 info
->reshape_progress
);
3472 used_disks
= imsm_num_data_members(prev_map
);
3473 if (used_disks
> 0) {
3474 info
->custom_array_size
= per_dev_array_size(map
) *
3479 /* we could emulate the checkpointing of
3480 * 'sync_action=check' migrations, but for now
3481 * we just immediately complete them
3484 /* this is handled by container_content_imsm() */
3485 case MIGR_STATE_CHANGE
:
3486 /* FIXME handle other migrations */
3488 /* we are not dirty, so... */
3489 info
->resync_start
= MaxSector
;
3493 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3494 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3496 info
->array
.major_version
= -1;
3497 info
->array
.minor_version
= -2;
3498 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3499 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3500 uuid_from_super_imsm(st
, info
->uuid
);
3504 for (i
=0; i
<map_disks
; i
++) {
3506 if (i
< info
->array
.raid_disks
) {
3507 struct imsm_disk
*dsk
;
3508 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3509 dsk
= get_imsm_disk(super
, j
);
3510 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3517 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3518 int failed
, int look_in_map
);
3520 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3523 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3525 if (is_gen_migration(dev
)) {
3528 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3530 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3531 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3532 if (map2
->map_state
!= map_state
) {
3533 map2
->map_state
= map_state
;
3534 super
->updates_pending
++;
3539 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3543 for (d
= super
->missing
; d
; d
= d
->next
)
3544 if (d
->index
== index
)
3549 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3551 struct intel_super
*super
= st
->sb
;
3552 struct imsm_disk
*disk
;
3553 int map_disks
= info
->array
.raid_disks
;
3554 int max_enough
= -1;
3556 struct imsm_super
*mpb
;
3558 if (super
->current_vol
>= 0) {
3559 getinfo_super_imsm_volume(st
, info
, map
);
3562 memset(info
, 0, sizeof(*info
));
3564 /* Set raid_disks to zero so that Assemble will always pull in valid
3567 info
->array
.raid_disks
= 0;
3568 info
->array
.level
= LEVEL_CONTAINER
;
3569 info
->array
.layout
= 0;
3570 info
->array
.md_minor
= -1;
3571 info
->array
.ctime
= 0; /* N/A for imsm */
3572 info
->array
.utime
= 0;
3573 info
->array
.chunk_size
= 0;
3575 info
->disk
.major
= 0;
3576 info
->disk
.minor
= 0;
3577 info
->disk
.raid_disk
= -1;
3578 info
->reshape_active
= 0;
3579 info
->array
.major_version
= -1;
3580 info
->array
.minor_version
= -2;
3581 strcpy(info
->text_version
, "imsm");
3582 info
->safe_mode_delay
= 0;
3583 info
->disk
.number
= -1;
3584 info
->disk
.state
= 0;
3586 info
->recovery_start
= MaxSector
;
3587 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3588 info
->bb
.supported
= 1;
3590 /* do we have the all the insync disks that we expect? */
3591 mpb
= super
->anchor
;
3592 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3594 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3595 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3596 int failed
, enough
, j
, missing
= 0;
3597 struct imsm_map
*map
;
3600 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3601 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3602 map
= get_imsm_map(dev
, MAP_0
);
3604 /* any newly missing disks?
3605 * (catches single-degraded vs double-degraded)
3607 for (j
= 0; j
< map
->num_members
; j
++) {
3608 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3609 __u32 idx
= ord_to_idx(ord
);
3611 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3612 info
->disk
.raid_disk
= j
;
3614 if (!(ord
& IMSM_ORD_REBUILD
) &&
3615 get_imsm_missing(super
, idx
)) {
3621 if (state
== IMSM_T_STATE_FAILED
)
3623 else if (state
== IMSM_T_STATE_DEGRADED
&&
3624 (state
!= map
->map_state
|| missing
))
3626 else /* we're normal, or already degraded */
3628 if (is_gen_migration(dev
) && missing
) {
3629 /* during general migration we need all disks
3630 * that process is running on.
3631 * No new missing disk is allowed.
3635 /* no more checks necessary
3639 /* in the missing/failed disk case check to see
3640 * if at least one array is runnable
3642 max_enough
= max(max_enough
, enough
);
3644 dprintf("enough: %d\n", max_enough
);
3645 info
->container_enough
= max_enough
;
3648 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3650 disk
= &super
->disks
->disk
;
3651 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3652 info
->component_size
= reserved
;
3653 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3654 /* we don't change info->disk.raid_disk here because
3655 * this state will be finalized in mdmon after we have
3656 * found the 'most fresh' version of the metadata
3658 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3659 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3660 0 : (1 << MD_DISK_SYNC
);
3663 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3664 * ->compare_super may have updated the 'num_raid_devs' field for spares
3666 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3667 uuid_from_super_imsm(st
, info
->uuid
);
3669 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3671 /* I don't know how to compute 'map' on imsm, so use safe default */
3674 for (i
= 0; i
< map_disks
; i
++)
3680 /* allocates memory and fills disk in mdinfo structure
3681 * for each disk in array */
3682 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3684 struct mdinfo
*mddev
;
3685 struct intel_super
*super
= st
->sb
;
3686 struct imsm_disk
*disk
;
3689 if (!super
|| !super
->disks
)
3692 mddev
= xcalloc(1, sizeof(*mddev
));
3696 tmp
= xcalloc(1, sizeof(*tmp
));
3698 tmp
->next
= mddev
->devs
;
3700 tmp
->disk
.number
= count
++;
3701 tmp
->disk
.major
= dl
->major
;
3702 tmp
->disk
.minor
= dl
->minor
;
3703 tmp
->disk
.state
= is_configured(disk
) ?
3704 (1 << MD_DISK_ACTIVE
) : 0;
3705 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3706 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3707 tmp
->disk
.raid_disk
= -1;
3713 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3714 char *update
, char *devname
, int verbose
,
3715 int uuid_set
, char *homehost
)
3717 /* For 'assemble' and 'force' we need to return non-zero if any
3718 * change was made. For others, the return value is ignored.
3719 * Update options are:
3720 * force-one : This device looks a bit old but needs to be included,
3721 * update age info appropriately.
3722 * assemble: clear any 'faulty' flag to allow this device to
3724 * force-array: Array is degraded but being forced, mark it clean
3725 * if that will be needed to assemble it.
3727 * newdev: not used ????
3728 * grow: Array has gained a new device - this is currently for
3730 * resync: mark as dirty so a resync will happen.
3731 * name: update the name - preserving the homehost
3732 * uuid: Change the uuid of the array to match watch is given
3734 * Following are not relevant for this imsm:
3735 * sparc2.2 : update from old dodgey metadata
3736 * super-minor: change the preferred_minor number
3737 * summaries: update redundant counters.
3738 * homehost: update the recorded homehost
3739 * _reshape_progress: record new reshape_progress position.
3742 struct intel_super
*super
= st
->sb
;
3743 struct imsm_super
*mpb
;
3745 /* we can only update container info */
3746 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3749 mpb
= super
->anchor
;
3751 if (strcmp(update
, "uuid") == 0) {
3752 /* We take this to mean that the family_num should be updated.
3753 * However that is much smaller than the uuid so we cannot really
3754 * allow an explicit uuid to be given. And it is hard to reliably
3756 * So if !uuid_set we know the current uuid is random and just used
3757 * the first 'int' and copy it to the other 3 positions.
3758 * Otherwise we require the 4 'int's to be the same as would be the
3759 * case if we are using a random uuid. So an explicit uuid will be
3760 * accepted as long as all for ints are the same... which shouldn't hurt
3763 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3766 if (info
->uuid
[0] != info
->uuid
[1] ||
3767 info
->uuid
[1] != info
->uuid
[2] ||
3768 info
->uuid
[2] != info
->uuid
[3])
3774 mpb
->orig_family_num
= info
->uuid
[0];
3775 } else if (strcmp(update
, "assemble") == 0)
3780 /* successful update? recompute checksum */
3782 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3787 static size_t disks_to_mpb_size(int disks
)
3791 size
= sizeof(struct imsm_super
);
3792 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3793 size
+= 2 * sizeof(struct imsm_dev
);
3794 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3795 size
+= (4 - 2) * sizeof(struct imsm_map
);
3796 /* 4 possible disk_ord_tbl's */
3797 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3798 /* maximum bbm log */
3799 size
+= sizeof(struct bbm_log
);
3804 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3805 unsigned long long data_offset
)
3807 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3810 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3813 static void free_devlist(struct intel_super
*super
)
3815 struct intel_dev
*dv
;
3817 while (super
->devlist
) {
3818 dv
= super
->devlist
->next
;
3819 free(super
->devlist
->dev
);
3820 free(super
->devlist
);
3821 super
->devlist
= dv
;
3825 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3827 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3830 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3834 * 0 same, or first was empty, and second was copied
3835 * 1 second had wrong number
3837 * 3 wrong other info
3839 struct intel_super
*first
= st
->sb
;
3840 struct intel_super
*sec
= tst
->sb
;
3847 /* in platform dependent environment test if the disks
3848 * use the same Intel hba
3849 * If not on Intel hba at all, allow anything.
3851 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3852 if (first
->hba
->type
!= sec
->hba
->type
) {
3854 "HBAs of devices do not match %s != %s\n",
3855 get_sys_dev_type(first
->hba
->type
),
3856 get_sys_dev_type(sec
->hba
->type
));
3859 if (first
->orom
!= sec
->orom
) {
3861 "HBAs of devices do not match %s != %s\n",
3862 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3867 /* if an anchor does not have num_raid_devs set then it is a free
3870 if (first
->anchor
->num_raid_devs
> 0 &&
3871 sec
->anchor
->num_raid_devs
> 0) {
3872 /* Determine if these disks might ever have been
3873 * related. Further disambiguation can only take place
3874 * in load_super_imsm_all
3876 __u32 first_family
= first
->anchor
->orig_family_num
;
3877 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3879 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3880 MAX_SIGNATURE_LENGTH
) != 0)
3883 if (first_family
== 0)
3884 first_family
= first
->anchor
->family_num
;
3885 if (sec_family
== 0)
3886 sec_family
= sec
->anchor
->family_num
;
3888 if (first_family
!= sec_family
)
3893 /* if 'first' is a spare promote it to a populated mpb with sec's
3896 if (first
->anchor
->num_raid_devs
== 0 &&
3897 sec
->anchor
->num_raid_devs
> 0) {
3899 struct intel_dev
*dv
;
3900 struct imsm_dev
*dev
;
3902 /* we need to copy raid device info from sec if an allocation
3903 * fails here we don't associate the spare
3905 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3906 dv
= xmalloc(sizeof(*dv
));
3907 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3910 dv
->next
= first
->devlist
;
3911 first
->devlist
= dv
;
3913 if (i
< sec
->anchor
->num_raid_devs
) {
3914 /* allocation failure */
3915 free_devlist(first
);
3916 pr_err("imsm: failed to associate spare\n");
3919 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3920 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3921 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3922 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3923 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3924 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3930 static void fd2devname(int fd
, char *name
)
3934 char dname
[PATH_MAX
];
3939 if (fstat(fd
, &st
) != 0)
3941 sprintf(path
, "/sys/dev/block/%d:%d",
3942 major(st
.st_rdev
), minor(st
.st_rdev
));
3944 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3949 nm
= strrchr(dname
, '/');
3952 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3956 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3959 char *name
= fd2kname(fd
);
3964 if (strncmp(name
, "nvme", 4) != 0)
3967 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3969 return load_sys(path
, buf
, buf_len
);
3972 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3974 static int imsm_read_serial(int fd
, char *devname
,
3975 __u8
*serial
, size_t serial_buf_len
)
3984 memset(buf
, 0, sizeof(buf
));
3986 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3989 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3991 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3992 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3993 fd2devname(fd
, (char *) serial
);
3999 pr_err("Failed to retrieve serial for %s\n",
4004 /* trim all whitespace and non-printable characters and convert
4007 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4010 /* ':' is reserved for use in placeholder serial
4011 * numbers for missing disks
4022 if (len
> serial_buf_len
) {
4023 /* truncate leading characters */
4024 dest
+= len
- serial_buf_len
;
4025 len
= serial_buf_len
;
4028 memset(serial
, 0, serial_buf_len
);
4029 memcpy(serial
, dest
, len
);
4034 static int serialcmp(__u8
*s1
, __u8
*s2
)
4036 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4039 static void serialcpy(__u8
*dest
, __u8
*src
)
4041 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4044 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4048 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4049 if (serialcmp(dl
->serial
, serial
) == 0)
4055 static struct imsm_disk
*
4056 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4060 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4061 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4063 if (serialcmp(disk
->serial
, serial
) == 0) {
4074 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4076 struct imsm_disk
*disk
;
4081 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4083 rv
= imsm_read_serial(fd
, devname
, serial
, MAX_RAID_SERIAL_LEN
);
4088 dl
= xcalloc(1, sizeof(*dl
));
4091 dl
->major
= major(stb
.st_rdev
);
4092 dl
->minor
= minor(stb
.st_rdev
);
4093 dl
->next
= super
->disks
;
4094 dl
->fd
= keep_fd
? fd
: -1;
4095 assert(super
->disks
== NULL
);
4097 serialcpy(dl
->serial
, serial
);
4100 fd2devname(fd
, name
);
4102 dl
->devname
= xstrdup(devname
);
4104 dl
->devname
= xstrdup(name
);
4106 /* look up this disk's index in the current anchor */
4107 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4110 /* only set index on disks that are a member of a
4111 * populated contianer, i.e. one with raid_devs
4113 if (is_failed(&dl
->disk
))
4115 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4122 /* When migrating map0 contains the 'destination' state while map1
4123 * contains the current state. When not migrating map0 contains the
4124 * current state. This routine assumes that map[0].map_state is set to
4125 * the current array state before being called.
4127 * Migration is indicated by one of the following states
4128 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4129 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4130 * map1state=unitialized)
4131 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4133 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4134 * map1state=degraded)
4135 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4138 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4139 __u8 to_state
, int migr_type
)
4141 struct imsm_map
*dest
;
4142 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4144 dev
->vol
.migr_state
= 1;
4145 set_migr_type(dev
, migr_type
);
4146 dev
->vol
.curr_migr_unit
= 0;
4147 dest
= get_imsm_map(dev
, MAP_1
);
4149 /* duplicate and then set the target end state in map[0] */
4150 memcpy(dest
, src
, sizeof_imsm_map(src
));
4151 if (migr_type
== MIGR_GEN_MIGR
) {
4155 for (i
= 0; i
< src
->num_members
; i
++) {
4156 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4157 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4161 if (migr_type
== MIGR_GEN_MIGR
)
4162 /* Clear migration record */
4163 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4165 src
->map_state
= to_state
;
4168 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4171 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4172 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4176 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4177 * completed in the last migration.
4179 * FIXME add support for raid-level-migration
4181 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4182 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4183 /* when final map state is other than expected
4184 * merge maps (not for migration)
4188 for (i
= 0; i
< prev
->num_members
; i
++)
4189 for (j
= 0; j
< map
->num_members
; j
++)
4190 /* during online capacity expansion
4191 * disks position can be changed
4192 * if takeover is used
4194 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4195 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4196 map
->disk_ord_tbl
[j
] |=
4197 prev
->disk_ord_tbl
[i
];
4200 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4201 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4204 dev
->vol
.migr_state
= 0;
4205 set_migr_type(dev
, 0);
4206 dev
->vol
.curr_migr_unit
= 0;
4207 map
->map_state
= map_state
;
4210 static int parse_raid_devices(struct intel_super
*super
)
4213 struct imsm_dev
*dev_new
;
4214 size_t len
, len_migr
;
4216 size_t space_needed
= 0;
4217 struct imsm_super
*mpb
= super
->anchor
;
4219 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4220 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4221 struct intel_dev
*dv
;
4223 len
= sizeof_imsm_dev(dev_iter
, 0);
4224 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4226 space_needed
+= len_migr
- len
;
4228 dv
= xmalloc(sizeof(*dv
));
4229 if (max_len
< len_migr
)
4231 if (max_len
> len_migr
)
4232 space_needed
+= max_len
- len_migr
;
4233 dev_new
= xmalloc(max_len
);
4234 imsm_copy_dev(dev_new
, dev_iter
);
4237 dv
->next
= super
->devlist
;
4238 super
->devlist
= dv
;
4241 /* ensure that super->buf is large enough when all raid devices
4244 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4247 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4248 super
->sector_size
);
4249 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4252 memcpy(buf
, super
->buf
, super
->len
);
4253 memset(buf
+ super
->len
, 0, len
- super
->len
);
4259 super
->extra_space
+= space_needed
;
4264 /*******************************************************************************
4265 * Function: check_mpb_migr_compatibility
4266 * Description: Function checks for unsupported migration features:
4267 * - migration optimization area (pba_of_lba0)
4268 * - descending reshape (ascending_migr)
4270 * super : imsm metadata information
4272 * 0 : migration is compatible
4273 * -1 : migration is not compatible
4274 ******************************************************************************/
4275 int check_mpb_migr_compatibility(struct intel_super
*super
)
4277 struct imsm_map
*map0
, *map1
;
4278 struct migr_record
*migr_rec
= super
->migr_rec
;
4281 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4282 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4285 dev_iter
->vol
.migr_state
== 1 &&
4286 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4287 /* This device is migrating */
4288 map0
= get_imsm_map(dev_iter
, MAP_0
);
4289 map1
= get_imsm_map(dev_iter
, MAP_1
);
4290 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4291 /* migration optimization area was used */
4293 if (migr_rec
->ascending_migr
== 0 &&
4294 migr_rec
->dest_depth_per_unit
> 0)
4295 /* descending reshape not supported yet */
4302 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4304 /* load_imsm_mpb - read matrix metadata
4305 * allocates super->mpb to be freed by free_imsm
4307 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4309 unsigned long long dsize
;
4310 unsigned long long sectors
;
4311 unsigned int sector_size
= super
->sector_size
;
4313 struct imsm_super
*anchor
;
4316 get_dev_size(fd
, NULL
, &dsize
);
4317 if (dsize
< 2*sector_size
) {
4319 pr_err("%s: device to small for imsm\n",
4324 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4326 pr_err("Cannot seek to anchor block on %s: %s\n",
4327 devname
, strerror(errno
));
4331 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4333 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4336 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4338 pr_err("Cannot read anchor block on %s: %s\n",
4339 devname
, strerror(errno
));
4344 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4346 pr_err("no IMSM anchor on %s\n", devname
);
4351 __free_imsm(super
, 0);
4352 /* reload capability and hba */
4354 /* capability and hba must be updated with new super allocation */
4355 find_intel_hba_capability(fd
, super
, devname
);
4356 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4357 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4359 pr_err("unable to allocate %zu byte mpb buffer\n",
4364 memcpy(super
->buf
, anchor
, sector_size
);
4366 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4369 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4370 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4371 pr_err("could not allocate migr_rec buffer\n");
4375 super
->clean_migration_record_by_mdmon
= 0;
4378 check_sum
= __gen_imsm_checksum(super
->anchor
);
4379 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4381 pr_err("IMSM checksum %x != %x on %s\n",
4383 __le32_to_cpu(super
->anchor
->check_sum
),
4391 /* read the extended mpb */
4392 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4394 pr_err("Cannot seek to extended mpb on %s: %s\n",
4395 devname
, strerror(errno
));
4399 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4400 super
->len
- sector_size
) != super
->len
- sector_size
) {
4402 pr_err("Cannot read extended mpb on %s: %s\n",
4403 devname
, strerror(errno
));
4407 check_sum
= __gen_imsm_checksum(super
->anchor
);
4408 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4410 pr_err("IMSM checksum %x != %x on %s\n",
4411 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4419 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4421 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4422 static void clear_hi(struct intel_super
*super
)
4424 struct imsm_super
*mpb
= super
->anchor
;
4426 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4428 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4429 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4430 disk
->total_blocks_hi
= 0;
4432 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4433 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4436 for (n
= 0; n
< 2; ++n
) {
4437 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4440 map
->pba_of_lba0_hi
= 0;
4441 map
->blocks_per_member_hi
= 0;
4442 map
->num_data_stripes_hi
= 0;
4448 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4452 err
= load_imsm_mpb(fd
, super
, devname
);
4455 if (super
->sector_size
== 4096)
4456 convert_from_4k(super
);
4457 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4460 err
= parse_raid_devices(super
);
4463 err
= load_bbm_log(super
);
4468 static void __free_imsm_disk(struct dl
*d
)
4480 static void free_imsm_disks(struct intel_super
*super
)
4484 while (super
->disks
) {
4486 super
->disks
= d
->next
;
4487 __free_imsm_disk(d
);
4489 while (super
->disk_mgmt_list
) {
4490 d
= super
->disk_mgmt_list
;
4491 super
->disk_mgmt_list
= d
->next
;
4492 __free_imsm_disk(d
);
4494 while (super
->missing
) {
4496 super
->missing
= d
->next
;
4497 __free_imsm_disk(d
);
4502 /* free all the pieces hanging off of a super pointer */
4503 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4505 struct intel_hba
*elem
, *next
;
4511 /* unlink capability description */
4513 if (super
->migr_rec_buf
) {
4514 free(super
->migr_rec_buf
);
4515 super
->migr_rec_buf
= NULL
;
4518 free_imsm_disks(super
);
4519 free_devlist(super
);
4523 free((void *)elem
->path
);
4529 free(super
->bbm_log
);
4533 static void free_imsm(struct intel_super
*super
)
4535 __free_imsm(super
, 1);
4536 free(super
->bb
.entries
);
4540 static void free_super_imsm(struct supertype
*st
)
4542 struct intel_super
*super
= st
->sb
;
4551 static struct intel_super
*alloc_super(void)
4553 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4555 super
->current_vol
= -1;
4556 super
->create_offset
= ~((unsigned long long) 0);
4558 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4559 sizeof(struct md_bb_entry
));
4560 if (!super
->bb
.entries
) {
4569 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4571 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4573 struct sys_dev
*hba_name
;
4576 if (fd
>= 0 && test_partition(fd
)) {
4577 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4581 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4586 hba_name
= find_disk_attached_hba(fd
, NULL
);
4589 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4593 rv
= attach_hba_to_super(super
, hba_name
);
4596 struct intel_hba
*hba
= super
->hba
;
4598 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4599 " but the container is assigned to Intel(R) %s %s (",
4601 get_sys_dev_type(hba_name
->type
),
4602 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4603 hba_name
->pci_id
? : "Err!",
4604 get_sys_dev_type(super
->hba
->type
),
4605 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4608 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4610 fprintf(stderr
, ", ");
4613 fprintf(stderr
, ").\n"
4614 " Mixing devices attached to different controllers is not allowed.\n");
4618 super
->orom
= find_imsm_capability(hba_name
);
4625 /* find_missing - helper routine for load_super_imsm_all that identifies
4626 * disks that have disappeared from the system. This routine relies on
4627 * the mpb being uptodate, which it is at load time.
4629 static int find_missing(struct intel_super
*super
)
4632 struct imsm_super
*mpb
= super
->anchor
;
4634 struct imsm_disk
*disk
;
4636 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4637 disk
= __get_imsm_disk(mpb
, i
);
4638 dl
= serial_to_dl(disk
->serial
, super
);
4642 dl
= xmalloc(sizeof(*dl
));
4646 dl
->devname
= xstrdup("missing");
4648 serialcpy(dl
->serial
, disk
->serial
);
4651 dl
->next
= super
->missing
;
4652 super
->missing
= dl
;
4658 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4660 struct intel_disk
*idisk
= disk_list
;
4663 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4665 idisk
= idisk
->next
;
4671 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4672 struct intel_super
*super
,
4673 struct intel_disk
**disk_list
)
4675 struct imsm_disk
*d
= &super
->disks
->disk
;
4676 struct imsm_super
*mpb
= super
->anchor
;
4679 for (i
= 0; i
< tbl_size
; i
++) {
4680 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4681 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4683 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4684 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4685 dprintf("mpb from %d:%d matches %d:%d\n",
4686 super
->disks
->major
,
4687 super
->disks
->minor
,
4688 table
[i
]->disks
->major
,
4689 table
[i
]->disks
->minor
);
4693 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4694 is_configured(d
) == is_configured(tbl_d
)) &&
4695 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4696 /* current version of the mpb is a
4697 * better candidate than the one in
4698 * super_table, but copy over "cross
4699 * generational" status
4701 struct intel_disk
*idisk
;
4703 dprintf("mpb from %d:%d replaces %d:%d\n",
4704 super
->disks
->major
,
4705 super
->disks
->minor
,
4706 table
[i
]->disks
->major
,
4707 table
[i
]->disks
->minor
);
4709 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4710 if (idisk
&& is_failed(&idisk
->disk
))
4711 tbl_d
->status
|= FAILED_DISK
;
4714 struct intel_disk
*idisk
;
4715 struct imsm_disk
*disk
;
4717 /* tbl_mpb is more up to date, but copy
4718 * over cross generational status before
4721 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4722 if (disk
&& is_failed(disk
))
4723 d
->status
|= FAILED_DISK
;
4725 idisk
= disk_list_get(d
->serial
, *disk_list
);
4728 if (disk
&& is_configured(disk
))
4729 idisk
->disk
.status
|= CONFIGURED_DISK
;
4732 dprintf("mpb from %d:%d prefer %d:%d\n",
4733 super
->disks
->major
,
4734 super
->disks
->minor
,
4735 table
[i
]->disks
->major
,
4736 table
[i
]->disks
->minor
);
4744 table
[tbl_size
++] = super
;
4748 /* update/extend the merged list of imsm_disk records */
4749 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4750 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4751 struct intel_disk
*idisk
;
4753 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4755 idisk
->disk
.status
|= disk
->status
;
4756 if (is_configured(&idisk
->disk
) ||
4757 is_failed(&idisk
->disk
))
4758 idisk
->disk
.status
&= ~(SPARE_DISK
);
4760 idisk
= xcalloc(1, sizeof(*idisk
));
4761 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4762 idisk
->disk
= *disk
;
4763 idisk
->next
= *disk_list
;
4767 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4774 static struct intel_super
*
4775 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4778 struct imsm_super
*mpb
= super
->anchor
;
4782 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4783 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4784 struct intel_disk
*idisk
;
4786 idisk
= disk_list_get(disk
->serial
, disk_list
);
4788 if (idisk
->owner
== owner
||
4789 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4792 dprintf("'%.16s' owner %d != %d\n",
4793 disk
->serial
, idisk
->owner
,
4796 dprintf("unknown disk %x [%d]: %.16s\n",
4797 __le32_to_cpu(mpb
->family_num
), i
,
4803 if (ok_count
== mpb
->num_disks
)
4808 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4810 struct intel_super
*s
;
4812 for (s
= super_list
; s
; s
= s
->next
) {
4813 if (family_num
!= s
->anchor
->family_num
)
4815 pr_err("Conflict, offlining family %#x on '%s'\n",
4816 __le32_to_cpu(family_num
), s
->disks
->devname
);
4820 static struct intel_super
*
4821 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4823 struct intel_super
*super_table
[len
];
4824 struct intel_disk
*disk_list
= NULL
;
4825 struct intel_super
*champion
, *spare
;
4826 struct intel_super
*s
, **del
;
4831 memset(super_table
, 0, sizeof(super_table
));
4832 for (s
= *super_list
; s
; s
= s
->next
)
4833 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4835 for (i
= 0; i
< tbl_size
; i
++) {
4836 struct imsm_disk
*d
;
4837 struct intel_disk
*idisk
;
4838 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4841 d
= &s
->disks
->disk
;
4843 /* 'd' must appear in merged disk list for its
4844 * configuration to be valid
4846 idisk
= disk_list_get(d
->serial
, disk_list
);
4847 if (idisk
&& idisk
->owner
== i
)
4848 s
= validate_members(s
, disk_list
, i
);
4853 dprintf("marking family: %#x from %d:%d offline\n",
4855 super_table
[i
]->disks
->major
,
4856 super_table
[i
]->disks
->minor
);
4860 /* This is where the mdadm implementation differs from the Windows
4861 * driver which has no strict concept of a container. We can only
4862 * assemble one family from a container, so when returning a prodigal
4863 * array member to this system the code will not be able to disambiguate
4864 * the container contents that should be assembled ("foreign" versus
4865 * "local"). It requires user intervention to set the orig_family_num
4866 * to a new value to establish a new container. The Windows driver in
4867 * this situation fixes up the volume name in place and manages the
4868 * foreign array as an independent entity.
4873 for (i
= 0; i
< tbl_size
; i
++) {
4874 struct intel_super
*tbl_ent
= super_table
[i
];
4880 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4885 if (s
&& !is_spare
) {
4886 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4888 } else if (!s
&& !is_spare
)
4901 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4902 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4904 /* collect all dl's onto 'champion', and update them to
4905 * champion's version of the status
4907 for (s
= *super_list
; s
; s
= s
->next
) {
4908 struct imsm_super
*mpb
= champion
->anchor
;
4909 struct dl
*dl
= s
->disks
;
4914 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4916 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4917 struct imsm_disk
*disk
;
4919 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4922 /* only set index on disks that are a member of
4923 * a populated contianer, i.e. one with
4926 if (is_failed(&dl
->disk
))
4928 else if (is_spare(&dl
->disk
))
4934 if (i
>= mpb
->num_disks
) {
4935 struct intel_disk
*idisk
;
4937 idisk
= disk_list_get(dl
->serial
, disk_list
);
4938 if (idisk
&& is_spare(&idisk
->disk
) &&
4939 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4947 dl
->next
= champion
->disks
;
4948 champion
->disks
= dl
;
4952 /* delete 'champion' from super_list */
4953 for (del
= super_list
; *del
; ) {
4954 if (*del
== champion
) {
4955 *del
= (*del
)->next
;
4958 del
= &(*del
)->next
;
4960 champion
->next
= NULL
;
4964 struct intel_disk
*idisk
= disk_list
;
4966 disk_list
= disk_list
->next
;
4974 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4975 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4976 int major
, int minor
, int keep_fd
);
4978 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4979 int *max
, int keep_fd
);
4981 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4982 char *devname
, struct md_list
*devlist
,
4985 struct intel_super
*super_list
= NULL
;
4986 struct intel_super
*super
= NULL
;
4991 /* 'fd' is an opened container */
4992 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4994 /* get super block from devlist devices */
4995 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4998 /* all mpbs enter, maybe one leaves */
4999 super
= imsm_thunderdome(&super_list
, i
);
5005 if (find_missing(super
) != 0) {
5011 /* load migration record */
5012 err
= load_imsm_migr_rec(super
, NULL
);
5014 /* migration is in progress,
5015 * but migr_rec cannot be loaded,
5021 /* Check migration compatibility */
5022 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5023 pr_err("Unsupported migration detected");
5025 fprintf(stderr
, " on %s\n", devname
);
5027 fprintf(stderr
, " (IMSM).\n");
5036 while (super_list
) {
5037 struct intel_super
*s
= super_list
;
5039 super_list
= super_list
->next
;
5048 strcpy(st
->container_devnm
, fd2devnm(fd
));
5050 st
->container_devnm
[0] = 0;
5051 if (err
== 0 && st
->ss
== NULL
) {
5052 st
->ss
= &super_imsm
;
5053 st
->minor_version
= 0;
5054 st
->max_devs
= IMSM_MAX_DEVICES
;
5060 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5061 int *max
, int keep_fd
)
5063 struct md_list
*tmpdev
;
5067 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5068 if (tmpdev
->used
!= 1)
5070 if (tmpdev
->container
== 1) {
5072 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5074 pr_err("cannot open device %s: %s\n",
5075 tmpdev
->devname
, strerror(errno
));
5079 err
= get_sra_super_block(fd
, super_list
,
5080 tmpdev
->devname
, &lmax
,
5089 int major
= major(tmpdev
->st_rdev
);
5090 int minor
= minor(tmpdev
->st_rdev
);
5091 err
= get_super_block(super_list
,
5108 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5109 int major
, int minor
, int keep_fd
)
5111 struct intel_super
*s
;
5123 sprintf(nm
, "%d:%d", major
, minor
);
5124 dfd
= dev_open(nm
, O_RDWR
);
5130 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5131 find_intel_hba_capability(dfd
, s
, devname
);
5132 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5134 /* retry the load if we might have raced against mdmon */
5135 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5136 for (retry
= 0; retry
< 3; retry
++) {
5138 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5144 s
->next
= *super_list
;
5152 if (dfd
>= 0 && !keep_fd
)
5159 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5166 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5170 if (sra
->array
.major_version
!= -1 ||
5171 sra
->array
.minor_version
!= -2 ||
5172 strcmp(sra
->text_version
, "imsm") != 0) {
5177 devnm
= fd2devnm(fd
);
5178 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5179 if (get_super_block(super_list
, devnm
, devname
,
5180 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5191 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5193 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5196 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5198 struct intel_super
*super
;
5202 if (test_partition(fd
))
5203 /* IMSM not allowed on partitions */
5206 free_super_imsm(st
);
5208 super
= alloc_super();
5209 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5212 /* Load hba and capabilities if they exist.
5213 * But do not preclude loading metadata in case capabilities or hba are
5214 * non-compliant and ignore_hw_compat is set.
5216 rv
= find_intel_hba_capability(fd
, super
, devname
);
5217 /* no orom/efi or non-intel hba of the disk */
5218 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5220 pr_err("No OROM/EFI properties for %s\n", devname
);
5224 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5226 /* retry the load if we might have raced against mdmon */
5228 struct mdstat_ent
*mdstat
= NULL
;
5229 char *name
= fd2kname(fd
);
5232 mdstat
= mdstat_by_component(name
);
5234 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5235 for (retry
= 0; retry
< 3; retry
++) {
5237 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5243 free_mdstat(mdstat
);
5248 pr_err("Failed to load all information sections on %s\n", devname
);
5254 if (st
->ss
== NULL
) {
5255 st
->ss
= &super_imsm
;
5256 st
->minor_version
= 0;
5257 st
->max_devs
= IMSM_MAX_DEVICES
;
5260 /* load migration record */
5261 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5262 /* Check for unsupported migration features */
5263 if (check_mpb_migr_compatibility(super
) != 0) {
5264 pr_err("Unsupported migration detected");
5266 fprintf(stderr
, " on %s\n", devname
);
5268 fprintf(stderr
, " (IMSM).\n");
5276 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5278 if (info
->level
== 1)
5280 return info
->chunk_size
>> 9;
5283 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5284 unsigned long long size
)
5286 if (info
->level
== 1)
5289 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5292 static void imsm_update_version_info(struct intel_super
*super
)
5294 /* update the version and attributes */
5295 struct imsm_super
*mpb
= super
->anchor
;
5297 struct imsm_dev
*dev
;
5298 struct imsm_map
*map
;
5301 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5302 dev
= get_imsm_dev(super
, i
);
5303 map
= get_imsm_map(dev
, MAP_0
);
5304 if (__le32_to_cpu(dev
->size_high
) > 0)
5305 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5307 /* FIXME detect when an array spans a port multiplier */
5309 mpb
->attributes
|= MPB_ATTRIB_PM
;
5312 if (mpb
->num_raid_devs
> 1 ||
5313 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5314 version
= MPB_VERSION_ATTRIBS
;
5315 switch (get_imsm_raid_level(map
)) {
5316 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5317 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5318 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5319 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5322 if (map
->num_members
>= 5)
5323 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5324 else if (dev
->status
== DEV_CLONE_N_GO
)
5325 version
= MPB_VERSION_CNG
;
5326 else if (get_imsm_raid_level(map
) == 5)
5327 version
= MPB_VERSION_RAID5
;
5328 else if (map
->num_members
>= 3)
5329 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5330 else if (get_imsm_raid_level(map
) == 1)
5331 version
= MPB_VERSION_RAID1
;
5333 version
= MPB_VERSION_RAID0
;
5335 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5339 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5341 struct imsm_super
*mpb
= super
->anchor
;
5342 char *reason
= NULL
;
5344 size_t len
= strlen(name
);
5348 while (isspace(start
[len
- 1]))
5350 while (*start
&& isspace(*start
))
5352 memmove(name
, start
, len
+ 1);
5355 if (len
> MAX_RAID_SERIAL_LEN
)
5356 reason
= "must be 16 characters or less";
5358 reason
= "must be a non-empty string";
5360 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5361 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5363 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5364 reason
= "already exists";
5369 if (reason
&& !quiet
)
5370 pr_err("imsm volume name %s\n", reason
);
5375 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5376 struct shape
*s
, char *name
,
5377 char *homehost
, int *uuid
,
5378 long long data_offset
)
5380 /* We are creating a volume inside a pre-existing container.
5381 * so st->sb is already set.
5383 struct intel_super
*super
= st
->sb
;
5384 unsigned int sector_size
= super
->sector_size
;
5385 struct imsm_super
*mpb
= super
->anchor
;
5386 struct intel_dev
*dv
;
5387 struct imsm_dev
*dev
;
5388 struct imsm_vol
*vol
;
5389 struct imsm_map
*map
;
5390 int idx
= mpb
->num_raid_devs
;
5393 unsigned long long array_blocks
;
5394 size_t size_old
, size_new
;
5395 unsigned long long num_data_stripes
;
5396 unsigned int data_disks
;
5397 unsigned long long size_per_member
;
5399 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5400 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5404 /* ensure the mpb is large enough for the new data */
5405 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5406 size_new
= disks_to_mpb_size(info
->nr_disks
);
5407 if (size_new
> size_old
) {
5409 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5411 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5412 pr_err("could not allocate new mpb\n");
5415 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5416 MIGR_REC_BUF_SECTORS
*
5417 MAX_SECTOR_SIZE
) != 0) {
5418 pr_err("could not allocate migr_rec buffer\n");
5424 memcpy(mpb_new
, mpb
, size_old
);
5427 super
->anchor
= mpb_new
;
5428 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5429 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5430 super
->len
= size_round
;
5432 super
->current_vol
= idx
;
5434 /* handle 'failed_disks' by either:
5435 * a) create dummy disk entries in the table if this the first
5436 * volume in the array. We add them here as this is the only
5437 * opportunity to add them. add_to_super_imsm_volume()
5438 * handles the non-failed disks and continues incrementing
5440 * b) validate that 'failed_disks' matches the current number
5441 * of missing disks if the container is populated
5443 if (super
->current_vol
== 0) {
5445 for (i
= 0; i
< info
->failed_disks
; i
++) {
5446 struct imsm_disk
*disk
;
5449 disk
= __get_imsm_disk(mpb
, i
);
5450 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5451 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5452 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5453 "missing:%d", (__u8
)i
);
5455 find_missing(super
);
5460 for (d
= super
->missing
; d
; d
= d
->next
)
5462 if (info
->failed_disks
> missing
) {
5463 pr_err("unable to add 'missing' disk to container\n");
5468 if (!check_name(super
, name
, 0))
5470 dv
= xmalloc(sizeof(*dv
));
5471 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5473 * Explicitly allow truncating to not confuse gcc's
5474 * -Werror=stringop-truncation
5476 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5477 memcpy(dev
->volume
, name
, namelen
);
5478 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5479 info
->layout
, info
->chunk_size
,
5480 s
->size
* BLOCKS_PER_KB
);
5481 data_disks
= get_data_disks(info
->level
, info
->layout
,
5483 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5484 size_per_member
= array_blocks
/ data_disks
;
5486 set_imsm_dev_size(dev
, array_blocks
);
5487 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5489 vol
->migr_state
= 0;
5490 set_migr_type(dev
, MIGR_INIT
);
5491 vol
->dirty
= !info
->state
;
5492 vol
->curr_migr_unit
= 0;
5493 map
= get_imsm_map(dev
, MAP_0
);
5494 set_pba_of_lba0(map
, super
->create_offset
);
5495 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5496 map
->failed_disk_num
= ~0;
5497 if (info
->level
> 0)
5498 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5499 : IMSM_T_STATE_UNINITIALIZED
);
5501 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5502 IMSM_T_STATE_NORMAL
;
5505 if (info
->level
== 1 && info
->raid_disks
> 2) {
5508 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5512 map
->raid_level
= info
->level
;
5513 if (info
->level
== 10) {
5514 map
->raid_level
= 1;
5515 map
->num_domains
= info
->raid_disks
/ 2;
5516 } else if (info
->level
== 1)
5517 map
->num_domains
= info
->raid_disks
;
5519 map
->num_domains
= 1;
5521 /* info->size is only int so use the 'size' parameter instead */
5522 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5523 num_data_stripes
/= map
->num_domains
;
5524 set_num_data_stripes(map
, num_data_stripes
);
5526 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5527 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5531 map
->num_members
= info
->raid_disks
;
5532 for (i
= 0; i
< map
->num_members
; i
++) {
5533 /* initialized in add_to_super */
5534 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5536 mpb
->num_raid_devs
++;
5537 mpb
->num_raid_devs_created
++;
5538 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5540 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5541 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5542 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5543 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5547 pr_err("imsm does not support consistency policy %s\n",
5548 map_num(consistency_policies
, s
->consistency_policy
));
5553 dv
->index
= super
->current_vol
;
5554 dv
->next
= super
->devlist
;
5555 super
->devlist
= dv
;
5557 imsm_update_version_info(super
);
5562 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5563 struct shape
*s
, char *name
,
5564 char *homehost
, int *uuid
,
5565 unsigned long long data_offset
)
5567 /* This is primarily called by Create when creating a new array.
5568 * We will then get add_to_super called for each component, and then
5569 * write_init_super called to write it out to each device.
5570 * For IMSM, Create can create on fresh devices or on a pre-existing
5572 * To create on a pre-existing array a different method will be called.
5573 * This one is just for fresh drives.
5575 struct intel_super
*super
;
5576 struct imsm_super
*mpb
;
5580 if (data_offset
!= INVALID_SECTORS
) {
5581 pr_err("data-offset not supported by imsm\n");
5586 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5590 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5592 mpb_size
= MAX_SECTOR_SIZE
;
5594 super
= alloc_super();
5596 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5601 pr_err("could not allocate superblock\n");
5604 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5605 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5606 pr_err("could not allocate migr_rec buffer\n");
5611 memset(super
->buf
, 0, mpb_size
);
5613 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5617 /* zeroing superblock */
5621 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5623 version
= (char *) mpb
->sig
;
5624 strcpy(version
, MPB_SIGNATURE
);
5625 version
+= strlen(MPB_SIGNATURE
);
5626 strcpy(version
, MPB_VERSION_RAID0
);
5631 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5633 unsigned int member_sector_size
;
5636 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5640 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5642 if (member_sector_size
!= super
->sector_size
)
5647 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5648 int fd
, char *devname
)
5650 struct intel_super
*super
= st
->sb
;
5651 struct imsm_super
*mpb
= super
->anchor
;
5652 struct imsm_disk
*_disk
;
5653 struct imsm_dev
*dev
;
5654 struct imsm_map
*map
;
5658 dev
= get_imsm_dev(super
, super
->current_vol
);
5659 map
= get_imsm_map(dev
, MAP_0
);
5661 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5662 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5668 /* we're doing autolayout so grab the pre-marked (in
5669 * validate_geometry) raid_disk
5671 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5672 if (dl
->raiddisk
== dk
->raid_disk
)
5675 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5676 if (dl
->major
== dk
->major
&&
5677 dl
->minor
== dk
->minor
)
5682 pr_err("%s is not a member of the same container\n", devname
);
5686 if (mpb
->num_disks
== 0)
5687 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5688 &super
->sector_size
))
5691 if (!drive_validate_sector_size(super
, dl
)) {
5692 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5696 /* add a pristine spare to the metadata */
5697 if (dl
->index
< 0) {
5698 dl
->index
= super
->anchor
->num_disks
;
5699 super
->anchor
->num_disks
++;
5701 /* Check the device has not already been added */
5702 slot
= get_imsm_disk_slot(map
, dl
->index
);
5704 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5705 pr_err("%s has been included in this array twice\n",
5709 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5710 dl
->disk
.status
= CONFIGURED_DISK
;
5712 /* update size of 'missing' disks to be at least as large as the
5713 * largest acitve member (we only have dummy missing disks when
5714 * creating the first volume)
5716 if (super
->current_vol
== 0) {
5717 for (df
= super
->missing
; df
; df
= df
->next
) {
5718 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5719 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5720 _disk
= __get_imsm_disk(mpb
, df
->index
);
5725 /* refresh unset/failed slots to point to valid 'missing' entries */
5726 for (df
= super
->missing
; df
; df
= df
->next
)
5727 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5728 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5730 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5732 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5733 if (is_gen_migration(dev
)) {
5734 struct imsm_map
*map2
= get_imsm_map(dev
,
5736 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5737 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5738 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5741 if ((unsigned)df
->index
==
5743 set_imsm_ord_tbl_ent(map2
,
5749 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5753 /* if we are creating the first raid device update the family number */
5754 if (super
->current_vol
== 0) {
5756 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5758 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5759 if (!_dev
|| !_disk
) {
5760 pr_err("BUG mpb setup error\n");
5766 sum
+= __gen_imsm_checksum(mpb
);
5767 mpb
->family_num
= __cpu_to_le32(sum
);
5768 mpb
->orig_family_num
= mpb
->family_num
;
5769 mpb
->creation_time
= __cpu_to_le64((__u64
)time(NULL
));
5771 super
->current_disk
= dl
;
5776 * Function marks disk as spare and restores disk serial
5777 * in case it was previously marked as failed by takeover operation
5779 * -1 : critical error
5780 * 0 : disk is marked as spare but serial is not set
5783 int mark_spare(struct dl
*disk
)
5785 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5792 if (!imsm_read_serial(disk
->fd
, NULL
, serial
, MAX_RAID_SERIAL_LEN
)) {
5793 /* Restore disk serial number, because takeover marks disk
5794 * as failed and adds to serial ':0' before it becomes
5797 serialcpy(disk
->serial
, serial
);
5798 serialcpy(disk
->disk
.serial
, serial
);
5801 disk
->disk
.status
= SPARE_DISK
;
5808 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
);
5810 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5811 int fd
, char *devname
,
5812 unsigned long long data_offset
)
5814 struct intel_super
*super
= st
->sb
;
5816 unsigned long long size
;
5817 unsigned int member_sector_size
;
5822 /* If we are on an RAID enabled platform check that the disk is
5823 * attached to the raid controller.
5824 * We do not need to test disks attachment for container based additions,
5825 * they shall be already tested when container was created/assembled.
5827 rv
= find_intel_hba_capability(fd
, super
, devname
);
5828 /* no orom/efi or non-intel hba of the disk */
5830 dprintf("capability: %p fd: %d ret: %d\n",
5831 super
->orom
, fd
, rv
);
5835 if (super
->current_vol
>= 0)
5836 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5839 dd
= xcalloc(sizeof(*dd
), 1);
5840 dd
->major
= major(stb
.st_rdev
);
5841 dd
->minor
= minor(stb
.st_rdev
);
5842 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5845 dd
->action
= DISK_ADD
;
5846 rv
= imsm_read_serial(fd
, devname
, dd
->serial
, MAX_RAID_SERIAL_LEN
);
5848 pr_err("failed to retrieve scsi serial, aborting\n");
5854 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5855 (super
->hba
->type
== SYS_DEV_VMD
))) {
5857 char *devpath
= diskfd_to_devpath(fd
);
5858 char controller_path
[PATH_MAX
];
5861 pr_err("failed to get devpath, aborting\n");
5868 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5871 if (devpath_to_vendor(controller_path
) == 0x8086) {
5873 * If Intel's NVMe drive has serial ended with
5874 * "-A","-B","-1" or "-2" it means that this is "x8"
5875 * device (double drive on single PCIe card).
5876 * User should be warned about potential data loss.
5878 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5879 /* Skip empty character at the end */
5880 if (dd
->serial
[i
] == 0)
5883 if (((dd
->serial
[i
] == 'A') ||
5884 (dd
->serial
[i
] == 'B') ||
5885 (dd
->serial
[i
] == '1') ||
5886 (dd
->serial
[i
] == '2')) &&
5887 (dd
->serial
[i
-1] == '-'))
5888 pr_err("\tThe action you are about to take may put your data at risk.\n"
5889 "\tPlease note that x8 devices may consist of two separate x4 devices "
5890 "located on a single PCIe port.\n"
5891 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5894 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5895 !imsm_orom_has_tpv_support(super
->orom
)) {
5896 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5897 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
5904 get_dev_size(fd
, NULL
, &size
);
5905 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5907 if (super
->sector_size
== 0) {
5908 /* this a first device, so sector_size is not set yet */
5909 super
->sector_size
= member_sector_size
;
5912 /* clear migr_rec when adding disk to container */
5913 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5914 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5916 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5917 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5918 MIGR_REC_BUF_SECTORS
*member_sector_size
)
5919 perror("Write migr_rec failed");
5923 serialcpy(dd
->disk
.serial
, dd
->serial
);
5924 set_total_blocks(&dd
->disk
, size
);
5925 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5926 struct imsm_super
*mpb
= super
->anchor
;
5927 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5930 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5931 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5933 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5935 if (st
->update_tail
) {
5936 dd
->next
= super
->disk_mgmt_list
;
5937 super
->disk_mgmt_list
= dd
;
5939 /* this is called outside of mdmon
5940 * write initial spare metadata
5941 * mdmon will overwrite it.
5943 dd
->next
= super
->disks
;
5945 write_super_imsm_spare(super
, dd
);
5951 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5953 struct intel_super
*super
= st
->sb
;
5956 /* remove from super works only in mdmon - for communication
5957 * manager - monitor. Check if communication memory buffer
5960 if (!st
->update_tail
) {
5961 pr_err("shall be used in mdmon context only\n");
5964 dd
= xcalloc(1, sizeof(*dd
));
5965 dd
->major
= dk
->major
;
5966 dd
->minor
= dk
->minor
;
5969 dd
->action
= DISK_REMOVE
;
5971 dd
->next
= super
->disk_mgmt_list
;
5972 super
->disk_mgmt_list
= dd
;
5977 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5980 char buf
[MAX_SECTOR_SIZE
];
5981 struct imsm_super anchor
;
5982 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5985 static int write_super_imsm_spare(struct intel_super
*super
, struct dl
*d
)
5987 struct imsm_super
*mpb
= super
->anchor
;
5988 struct imsm_super
*spare
= &spare_record
.anchor
;
5994 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5995 spare
->generation_num
= __cpu_to_le32(1UL);
5996 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5997 spare
->num_disks
= 1;
5998 spare
->num_raid_devs
= 0;
5999 spare
->cache_size
= mpb
->cache_size
;
6000 spare
->pwr_cycle_count
= __cpu_to_le32(1);
6002 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
6003 MPB_SIGNATURE MPB_VERSION_RAID0
);
6005 spare
->disk
[0] = d
->disk
;
6006 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
6007 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6009 if (super
->sector_size
== 4096)
6010 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6012 sum
= __gen_imsm_checksum(spare
);
6013 spare
->family_num
= __cpu_to_le32(sum
);
6014 spare
->orig_family_num
= 0;
6015 sum
= __gen_imsm_checksum(spare
);
6016 spare
->check_sum
= __cpu_to_le32(sum
);
6018 if (store_imsm_mpb(d
->fd
, spare
)) {
6019 pr_err("failed for device %d:%d %s\n",
6020 d
->major
, d
->minor
, strerror(errno
));
6026 /* spare records have their own family number and do not have any defined raid
6029 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
6033 for (d
= super
->disks
; d
; d
= d
->next
) {
6037 if (write_super_imsm_spare(super
, d
))
6049 static int write_super_imsm(struct supertype
*st
, int doclose
)
6051 struct intel_super
*super
= st
->sb
;
6052 unsigned int sector_size
= super
->sector_size
;
6053 struct imsm_super
*mpb
= super
->anchor
;
6059 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6061 int clear_migration_record
= 1;
6064 /* 'generation' is incremented everytime the metadata is written */
6065 generation
= __le32_to_cpu(mpb
->generation_num
);
6067 mpb
->generation_num
= __cpu_to_le32(generation
);
6069 /* fix up cases where previous mdadm releases failed to set
6072 if (mpb
->orig_family_num
== 0)
6073 mpb
->orig_family_num
= mpb
->family_num
;
6075 for (d
= super
->disks
; d
; d
= d
->next
) {
6079 mpb
->disk
[d
->index
] = d
->disk
;
6083 for (d
= super
->missing
; d
; d
= d
->next
) {
6084 mpb
->disk
[d
->index
] = d
->disk
;
6087 mpb
->num_disks
= num_disks
;
6088 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6090 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6091 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6092 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6094 imsm_copy_dev(dev
, dev2
);
6095 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6097 if (is_gen_migration(dev2
))
6098 clear_migration_record
= 0;
6101 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6104 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6105 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6107 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6109 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6110 mpb_size
+= bbm_log_size
;
6111 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6114 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6117 /* recalculate checksum */
6118 sum
= __gen_imsm_checksum(mpb
);
6119 mpb
->check_sum
= __cpu_to_le32(sum
);
6121 if (super
->clean_migration_record_by_mdmon
) {
6122 clear_migration_record
= 1;
6123 super
->clean_migration_record_by_mdmon
= 0;
6125 if (clear_migration_record
)
6126 memset(super
->migr_rec_buf
, 0,
6127 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6129 if (sector_size
== 4096)
6130 convert_to_4k(super
);
6132 /* write the mpb for disks that compose raid devices */
6133 for (d
= super
->disks
; d
; d
= d
->next
) {
6134 if (d
->index
< 0 || is_failed(&d
->disk
))
6137 if (clear_migration_record
) {
6138 unsigned long long dsize
;
6140 get_dev_size(d
->fd
, NULL
, &dsize
);
6141 if (lseek64(d
->fd
, dsize
- sector_size
,
6143 if ((unsigned int)write(d
->fd
,
6144 super
->migr_rec_buf
,
6145 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6146 MIGR_REC_BUF_SECTORS
*sector_size
)
6147 perror("Write migr_rec failed");
6151 if (store_imsm_mpb(d
->fd
, mpb
))
6153 "failed for device %d:%d (fd: %d)%s\n",
6155 d
->fd
, strerror(errno
));
6164 return write_super_imsm_spares(super
, doclose
);
6169 static int create_array(struct supertype
*st
, int dev_idx
)
6172 struct imsm_update_create_array
*u
;
6173 struct intel_super
*super
= st
->sb
;
6174 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6175 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6176 struct disk_info
*inf
;
6177 struct imsm_disk
*disk
;
6180 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6181 sizeof(*inf
) * map
->num_members
;
6183 u
->type
= update_create_array
;
6184 u
->dev_idx
= dev_idx
;
6185 imsm_copy_dev(&u
->dev
, dev
);
6186 inf
= get_disk_info(u
);
6187 for (i
= 0; i
< map
->num_members
; i
++) {
6188 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6190 disk
= get_imsm_disk(super
, idx
);
6192 disk
= get_imsm_missing(super
, idx
);
6193 serialcpy(inf
[i
].serial
, disk
->serial
);
6195 append_metadata_update(st
, u
, len
);
6200 static int mgmt_disk(struct supertype
*st
)
6202 struct intel_super
*super
= st
->sb
;
6204 struct imsm_update_add_remove_disk
*u
;
6206 if (!super
->disk_mgmt_list
)
6211 u
->type
= update_add_remove_disk
;
6212 append_metadata_update(st
, u
, len
);
6217 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6219 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6221 struct ppl_header
*ppl_hdr
= buf
;
6224 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6226 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6228 perror("Failed to seek to PPL header location");
6232 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6234 perror("Write PPL header failed");
6243 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6245 struct intel_super
*super
= st
->sb
;
6247 struct ppl_header
*ppl_hdr
;
6250 /* first clear entire ppl space */
6251 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6255 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6257 pr_err("Failed to allocate PPL header buffer\n");
6261 memset(buf
, 0, PPL_HEADER_SIZE
);
6263 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6264 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6266 if (info
->mismatch_cnt
) {
6268 * We are overwriting an invalid ppl. Make one entry with wrong
6269 * checksum to prevent the kernel from skipping resync.
6271 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6272 ppl_hdr
->entries
[0].checksum
= ~0;
6275 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6281 static int is_rebuilding(struct imsm_dev
*dev
);
6283 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6284 struct mdinfo
*disk
)
6286 struct intel_super
*super
= st
->sb
;
6288 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6290 struct ppl_header
*ppl_hdr
= NULL
;
6292 struct imsm_dev
*dev
;
6295 unsigned long long ppl_offset
= 0;
6296 unsigned long long prev_gen_num
= 0;
6298 if (disk
->disk
.raid_disk
< 0)
6301 dev
= get_imsm_dev(super
, info
->container_member
);
6302 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6303 d
= get_imsm_dl_disk(super
, idx
);
6305 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6308 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6309 pr_err("Failed to allocate PPL header buffer\n");
6315 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6318 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6320 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6322 perror("Failed to seek to PPL header location");
6327 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6328 perror("Read PPL header failed");
6335 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6336 ppl_hdr
->checksum
= 0;
6338 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6339 dprintf("Wrong PPL header checksum on %s\n",
6344 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6345 /* previous was newest, it was already checked */
6349 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6350 super
->anchor
->orig_family_num
)) {
6351 dprintf("Wrong PPL header signature on %s\n",
6358 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6360 ppl_offset
+= PPL_HEADER_SIZE
;
6361 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6363 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6366 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6378 * Update metadata to use mutliple PPLs area (1MB).
6379 * This is done once for all RAID members
6381 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6382 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6384 struct mdinfo
*member_dev
;
6386 sprintf(subarray
, "%d", info
->container_member
);
6388 if (mdmon_running(st
->container_devnm
))
6389 st
->update_tail
= &st
->updates
;
6391 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6392 pr_err("Failed to update subarray %s\n",
6395 if (st
->update_tail
)
6396 flush_metadata_updates(st
);
6398 st
->ss
->sync_metadata(st
);
6399 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6400 for (member_dev
= info
->devs
; member_dev
;
6401 member_dev
= member_dev
->next
)
6402 member_dev
->ppl_size
=
6403 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6408 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6410 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6411 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6412 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6413 (is_rebuilding(dev
) &&
6414 dev
->vol
.curr_migr_unit
== 0 &&
6415 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6416 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6418 info
->mismatch_cnt
++;
6419 } else if (ret
== 0 &&
6420 ppl_hdr
->entries_count
== 0 &&
6421 is_rebuilding(dev
) &&
6422 info
->resync_start
== 0) {
6424 * The header has no entries - add a single empty entry and
6425 * rewrite the header to prevent the kernel from going into
6426 * resync after an interrupted rebuild.
6428 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6429 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6437 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6439 struct intel_super
*super
= st
->sb
;
6443 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6444 info
->array
.level
!= 5)
6447 for (d
= super
->disks
; d
; d
= d
->next
) {
6448 if (d
->index
< 0 || is_failed(&d
->disk
))
6451 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6459 static int write_init_super_imsm(struct supertype
*st
)
6461 struct intel_super
*super
= st
->sb
;
6462 int current_vol
= super
->current_vol
;
6466 getinfo_super_imsm(st
, &info
, NULL
);
6468 /* we are done with current_vol reset it to point st at the container */
6469 super
->current_vol
= -1;
6471 if (st
->update_tail
) {
6472 /* queue the recently created array / added disk
6473 * as a metadata update */
6475 /* determine if we are creating a volume or adding a disk */
6476 if (current_vol
< 0) {
6477 /* in the mgmt (add/remove) disk case we are running
6478 * in mdmon context, so don't close fd's
6482 rv
= write_init_ppl_imsm_all(st
, &info
);
6484 rv
= create_array(st
, current_vol
);
6488 for (d
= super
->disks
; d
; d
= d
->next
)
6489 Kill(d
->devname
, NULL
, 0, -1, 1);
6490 if (current_vol
>= 0)
6491 rv
= write_init_ppl_imsm_all(st
, &info
);
6493 rv
= write_super_imsm(st
, 1);
6499 static int store_super_imsm(struct supertype
*st
, int fd
)
6501 struct intel_super
*super
= st
->sb
;
6502 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6507 if (super
->sector_size
== 4096)
6508 convert_to_4k(super
);
6509 return store_imsm_mpb(fd
, mpb
);
6512 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6513 int layout
, int raiddisks
, int chunk
,
6514 unsigned long long size
,
6515 unsigned long long data_offset
,
6517 unsigned long long *freesize
,
6521 unsigned long long ldsize
;
6522 struct intel_super
*super
;
6525 if (level
!= LEVEL_CONTAINER
)
6530 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6533 pr_err("imsm: Cannot open %s: %s\n",
6534 dev
, strerror(errno
));
6537 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6542 /* capabilities retrieve could be possible
6543 * note that there is no fd for the disks in array.
6545 super
= alloc_super();
6550 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6556 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6560 fd2devname(fd
, str
);
6561 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6562 fd
, str
, super
->orom
, rv
, raiddisks
);
6564 /* no orom/efi or non-intel hba of the disk */
6571 if (raiddisks
> super
->orom
->tds
) {
6573 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6574 raiddisks
, super
->orom
->tds
);
6578 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6579 (ldsize
>> 9) >> 32 > 0) {
6581 pr_err("%s exceeds maximum platform supported size\n", dev
);
6587 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6593 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6595 const unsigned long long base_start
= e
[*idx
].start
;
6596 unsigned long long end
= base_start
+ e
[*idx
].size
;
6599 if (base_start
== end
)
6603 for (i
= *idx
; i
< num_extents
; i
++) {
6604 /* extend overlapping extents */
6605 if (e
[i
].start
>= base_start
&&
6606 e
[i
].start
<= end
) {
6609 if (e
[i
].start
+ e
[i
].size
> end
)
6610 end
= e
[i
].start
+ e
[i
].size
;
6611 } else if (e
[i
].start
> end
) {
6617 return end
- base_start
;
6620 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6622 /* build a composite disk with all known extents and generate a new
6623 * 'maxsize' given the "all disks in an array must share a common start
6624 * offset" constraint
6626 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6630 unsigned long long pos
;
6631 unsigned long long start
= 0;
6632 unsigned long long maxsize
;
6633 unsigned long reserve
;
6635 /* coalesce and sort all extents. also, check to see if we need to
6636 * reserve space between member arrays
6639 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6642 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6645 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6650 while (i
< sum_extents
) {
6651 e
[j
].start
= e
[i
].start
;
6652 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6654 if (e
[j
-1].size
== 0)
6663 unsigned long long esize
;
6665 esize
= e
[i
].start
- pos
;
6666 if (esize
>= maxsize
) {
6671 pos
= e
[i
].start
+ e
[i
].size
;
6673 } while (e
[i
-1].size
);
6679 /* FIXME assumes volume at offset 0 is the first volume in a
6682 if (start_extent
> 0)
6683 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6687 if (maxsize
< reserve
)
6690 super
->create_offset
= ~((unsigned long long) 0);
6691 if (start
+ reserve
> super
->create_offset
)
6692 return 0; /* start overflows create_offset */
6693 super
->create_offset
= start
+ reserve
;
6695 return maxsize
- reserve
;
6698 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6700 if (level
< 0 || level
== 6 || level
== 4)
6703 /* if we have an orom prevent invalid raid levels */
6706 case 0: return imsm_orom_has_raid0(orom
);
6709 return imsm_orom_has_raid1e(orom
);
6710 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6711 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6712 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6715 return 1; /* not on an Intel RAID platform so anything goes */
6721 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6722 int dpa
, int verbose
)
6724 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6725 struct mdstat_ent
*memb
;
6731 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6732 if (memb
->metadata_version
&&
6733 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6734 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6735 !is_subarray(memb
->metadata_version
+9) &&
6737 struct dev_member
*dev
= memb
->members
;
6739 while(dev
&& (fd
< 0)) {
6740 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6741 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6743 fd
= open(path
, O_RDONLY
, 0);
6744 if (num
<= 0 || fd
< 0) {
6745 pr_vrb("Cannot open %s: %s\n",
6746 dev
->name
, strerror(errno
));
6752 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6753 struct mdstat_ent
*vol
;
6754 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6755 if (vol
->active
> 0 &&
6756 vol
->metadata_version
&&
6757 is_container_member(vol
, memb
->devnm
)) {
6762 if (*devlist
&& (found
< dpa
)) {
6763 dv
= xcalloc(1, sizeof(*dv
));
6764 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6765 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6768 dv
->next
= *devlist
;
6776 free_mdstat(mdstat
);
6781 static struct md_list
*
6782 get_loop_devices(void)
6785 struct md_list
*devlist
= NULL
;
6788 for(i
= 0; i
< 12; i
++) {
6789 dv
= xcalloc(1, sizeof(*dv
));
6790 dv
->devname
= xmalloc(40);
6791 sprintf(dv
->devname
, "/dev/loop%d", i
);
6799 static struct md_list
*
6800 get_devices(const char *hba_path
)
6802 struct md_list
*devlist
= NULL
;
6809 devlist
= get_loop_devices();
6812 /* scroll through /sys/dev/block looking for devices attached to
6815 dir
= opendir("/sys/dev/block");
6816 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6821 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6823 path
= devt_to_devpath(makedev(major
, minor
));
6826 if (!path_attached_to_hba(path
, hba_path
)) {
6833 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6835 fd2devname(fd
, buf
);
6838 pr_err("cannot open device: %s\n",
6843 dv
= xcalloc(1, sizeof(*dv
));
6844 dv
->devname
= xstrdup(buf
);
6851 devlist
= devlist
->next
;
6861 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6862 int verbose
, int *found
)
6864 struct md_list
*tmpdev
;
6866 struct supertype
*st
;
6868 /* first walk the list of devices to find a consistent set
6869 * that match the criterea, if that is possible.
6870 * We flag the ones we like with 'used'.
6873 st
= match_metadata_desc_imsm("imsm");
6875 pr_vrb("cannot allocate memory for imsm supertype\n");
6879 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6880 char *devname
= tmpdev
->devname
;
6882 struct supertype
*tst
;
6884 if (tmpdev
->used
> 1)
6886 tst
= dup_super(st
);
6888 pr_vrb("cannot allocate memory for imsm supertype\n");
6891 tmpdev
->container
= 0;
6892 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6894 dprintf("cannot open device %s: %s\n",
6895 devname
, strerror(errno
));
6897 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
6899 } else if (must_be_container(dfd
)) {
6900 struct supertype
*cst
;
6901 cst
= super_by_fd(dfd
, NULL
);
6903 dprintf("cannot recognize container type %s\n",
6906 } else if (tst
->ss
!= st
->ss
) {
6907 dprintf("non-imsm container - ignore it: %s\n",
6910 } else if (!tst
->ss
->load_container
||
6911 tst
->ss
->load_container(tst
, dfd
, NULL
))
6914 tmpdev
->container
= 1;
6917 cst
->ss
->free_super(cst
);
6919 tmpdev
->st_rdev
= rdev
;
6920 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6921 dprintf("no RAID superblock on %s\n",
6924 } else if (tst
->ss
->compare_super
== NULL
) {
6925 dprintf("Cannot assemble %s metadata on %s\n",
6926 tst
->ss
->name
, devname
);
6932 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6933 /* Ignore unrecognised devices during auto-assembly */
6938 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6940 if (st
->minor_version
== -1)
6941 st
->minor_version
= tst
->minor_version
;
6943 if (memcmp(info
.uuid
, uuid_zero
,
6944 sizeof(int[4])) == 0) {
6945 /* this is a floating spare. It cannot define
6946 * an array unless there are no more arrays of
6947 * this type to be found. It can be included
6948 * in an array of this type though.
6954 if (st
->ss
!= tst
->ss
||
6955 st
->minor_version
!= tst
->minor_version
||
6956 st
->ss
->compare_super(st
, tst
) != 0) {
6957 /* Some mismatch. If exactly one array matches this host,
6958 * we can resolve on that one.
6959 * Or, if we are auto assembling, we just ignore the second
6962 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6968 dprintf("found: devname: %s\n", devname
);
6972 tst
->ss
->free_super(tst
);
6976 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6977 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6978 for (iter
= head
; iter
; iter
= iter
->next
) {
6979 dprintf("content->text_version: %s vol\n",
6980 iter
->text_version
);
6981 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6982 /* do not assemble arrays with unsupported
6984 dprintf("Cannot activate member %s.\n",
6985 iter
->text_version
);
6992 dprintf("No valid super block on device list: err: %d %p\n",
6996 dprintf("no more devices to examine\n");
6999 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7000 if (tmpdev
->used
== 1 && tmpdev
->found
) {
7002 if (count
< tmpdev
->found
)
7005 count
-= tmpdev
->found
;
7008 if (tmpdev
->used
== 1)
7013 st
->ss
->free_super(st
);
7017 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
7020 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
7022 const struct orom_entry
*entry
;
7023 struct devid_list
*dv
, *devid_list
;
7028 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7029 if (strstr(idev
->path
, hba_path
))
7033 if (!idev
|| !idev
->dev_id
)
7036 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7038 if (!entry
|| !entry
->devid_list
)
7041 devid_list
= entry
->devid_list
;
7042 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7043 struct md_list
*devlist
;
7044 struct sys_dev
*device
= NULL
;
7049 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7051 device
= device_by_id(dv
->devid
);
7054 hpath
= device
->path
;
7058 devlist
= get_devices(hpath
);
7059 /* if no intel devices return zero volumes */
7060 if (devlist
== NULL
)
7063 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7065 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7066 if (devlist
== NULL
)
7070 count
+= count_volumes_list(devlist
,
7074 dprintf("found %d count: %d\n", found
, count
);
7077 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7080 struct md_list
*dv
= devlist
;
7081 devlist
= devlist
->next
;
7089 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7093 if (hba
->type
== SYS_DEV_VMD
) {
7094 struct sys_dev
*dev
;
7097 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7098 if (dev
->type
== SYS_DEV_VMD
)
7099 count
+= __count_volumes(dev
->path
, dpa
,
7104 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7107 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7109 /* up to 512 if the plaform supports it, otherwise the platform max.
7110 * 128 if no platform detected
7112 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7114 return min(512, (1 << fs
));
7118 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7119 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7121 /* check/set platform and metadata limits/defaults */
7122 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7123 pr_vrb("platform supports a maximum of %d disks per array\n",
7128 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7129 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7130 pr_vrb("platform does not support raid%d with %d disk%s\n",
7131 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7135 if (*chunk
== 0 || *chunk
== UnSet
)
7136 *chunk
= imsm_default_chunk(super
->orom
);
7138 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7139 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7143 if (layout
!= imsm_level_to_layout(level
)) {
7145 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7146 else if (level
== 10)
7147 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7149 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7154 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7155 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7156 pr_vrb("platform does not support a volume size over 2TB\n");
7163 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7164 * FIX ME add ahci details
7166 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7167 int layout
, int raiddisks
, int *chunk
,
7168 unsigned long long size
,
7169 unsigned long long data_offset
,
7171 unsigned long long *freesize
,
7175 struct intel_super
*super
= st
->sb
;
7176 struct imsm_super
*mpb
;
7178 unsigned long long pos
= 0;
7179 unsigned long long maxsize
;
7183 /* We must have the container info already read in. */
7187 mpb
= super
->anchor
;
7189 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7190 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7194 /* General test: make sure there is space for
7195 * 'raiddisks' device extents of size 'size' at a given
7198 unsigned long long minsize
= size
;
7199 unsigned long long start_offset
= MaxSector
;
7202 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7203 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7208 e
= get_extents(super
, dl
, 0);
7211 unsigned long long esize
;
7212 esize
= e
[i
].start
- pos
;
7213 if (esize
>= minsize
)
7215 if (found
&& start_offset
== MaxSector
) {
7218 } else if (found
&& pos
!= start_offset
) {
7222 pos
= e
[i
].start
+ e
[i
].size
;
7224 } while (e
[i
-1].size
);
7229 if (dcnt
< raiddisks
) {
7231 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7238 /* This device must be a member of the set */
7239 if (!stat_is_blkdev(dev
, &rdev
))
7241 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7242 if (dl
->major
== (int)major(rdev
) &&
7243 dl
->minor
== (int)minor(rdev
))
7248 pr_err("%s is not in the same imsm set\n", dev
);
7250 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7251 /* If a volume is present then the current creation attempt
7252 * cannot incorporate new spares because the orom may not
7253 * understand this configuration (all member disks must be
7254 * members of each array in the container).
7256 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7257 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7259 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7260 mpb
->num_disks
!= raiddisks
) {
7261 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7265 /* retrieve the largest free space block */
7266 e
= get_extents(super
, dl
, 0);
7271 unsigned long long esize
;
7273 esize
= e
[i
].start
- pos
;
7274 if (esize
>= maxsize
)
7276 pos
= e
[i
].start
+ e
[i
].size
;
7278 } while (e
[i
-1].size
);
7283 pr_err("unable to determine free space for: %s\n",
7287 if (maxsize
< size
) {
7289 pr_err("%s not enough space (%llu < %llu)\n",
7290 dev
, maxsize
, size
);
7294 /* count total number of extents for merge */
7296 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7298 i
+= dl
->extent_cnt
;
7300 maxsize
= merge_extents(super
, i
);
7302 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7303 pr_err("attempting to create a second volume with size less then remaining space.\n");
7305 if (maxsize
< size
|| maxsize
== 0) {
7308 pr_err("no free space left on device. Aborting...\n");
7310 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7316 *freesize
= maxsize
;
7319 int count
= count_volumes(super
->hba
,
7320 super
->orom
->dpa
, verbose
);
7321 if (super
->orom
->vphba
<= count
) {
7322 pr_vrb("platform does not support more than %d raid volumes.\n",
7323 super
->orom
->vphba
);
7330 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7331 unsigned long long size
, int chunk
,
7332 unsigned long long *freesize
)
7334 struct intel_super
*super
= st
->sb
;
7335 struct imsm_super
*mpb
= super
->anchor
;
7340 unsigned long long maxsize
;
7341 unsigned long long minsize
;
7345 /* find the largest common start free region of the possible disks */
7349 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7355 /* don't activate new spares if we are orom constrained
7356 * and there is already a volume active in the container
7358 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7361 e
= get_extents(super
, dl
, 0);
7364 for (i
= 1; e
[i
-1].size
; i
++)
7372 maxsize
= merge_extents(super
, extent_cnt
);
7376 minsize
= chunk
* 2;
7378 if (cnt
< raiddisks
||
7379 (super
->orom
&& used
&& used
!= raiddisks
) ||
7380 maxsize
< minsize
||
7382 pr_err("not enough devices with space to create array.\n");
7383 return 0; /* No enough free spaces large enough */
7394 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7395 pr_err("attempting to create a second volume with size less then remaining space.\n");
7397 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7399 dl
->raiddisk
= cnt
++;
7403 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7408 static int reserve_space(struct supertype
*st
, int raiddisks
,
7409 unsigned long long size
, int chunk
,
7410 unsigned long long *freesize
)
7412 struct intel_super
*super
= st
->sb
;
7417 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7420 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7422 dl
->raiddisk
= cnt
++;
7429 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7430 int raiddisks
, int *chunk
, unsigned long long size
,
7431 unsigned long long data_offset
,
7432 char *dev
, unsigned long long *freesize
,
7433 int consistency_policy
, int verbose
)
7440 * if given unused devices create a container
7441 * if given given devices in a container create a member volume
7443 if (level
== LEVEL_CONTAINER
) {
7444 /* Must be a fresh device to add to a container */
7445 return validate_geometry_imsm_container(st
, level
, layout
,
7454 * Size is given in sectors.
7456 if (size
&& (size
< 2048)) {
7457 pr_err("Given size must be greater than 1M.\n");
7458 /* Depends on algorithm in Create.c :
7459 * if container was given (dev == NULL) return -1,
7460 * if block device was given ( dev != NULL) return 0.
7462 return dev
? -1 : 0;
7467 struct intel_super
*super
= st
->sb
;
7468 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7469 raiddisks
, chunk
, size
,
7472 /* we are being asked to automatically layout a
7473 * new volume based on the current contents of
7474 * the container. If the the parameters can be
7475 * satisfied reserve_space will record the disks,
7476 * start offset, and size of the volume to be
7477 * created. add_to_super and getinfo_super
7478 * detect when autolayout is in progress.
7480 /* assuming that freesize is always given when array is
7482 if (super
->orom
&& freesize
) {
7484 count
= count_volumes(super
->hba
,
7485 super
->orom
->dpa
, verbose
);
7486 if (super
->orom
->vphba
<= count
) {
7487 pr_vrb("platform does not support more than %d raid volumes.\n",
7488 super
->orom
->vphba
);
7493 return reserve_space(st
, raiddisks
, size
,
7499 /* creating in a given container */
7500 return validate_geometry_imsm_volume(st
, level
, layout
,
7501 raiddisks
, chunk
, size
,
7503 dev
, freesize
, verbose
);
7506 /* This device needs to be a device in an 'imsm' container */
7507 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7510 pr_err("Cannot create this array on device %s\n",
7515 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7517 pr_err("Cannot open %s: %s\n",
7518 dev
, strerror(errno
));
7521 /* Well, it is in use by someone, maybe an 'imsm' container. */
7522 cfd
= open_container(fd
);
7526 pr_err("Cannot use %s: It is busy\n",
7530 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7531 if (sra
&& sra
->array
.major_version
== -1 &&
7532 strcmp(sra
->text_version
, "imsm") == 0)
7536 /* This is a member of a imsm container. Load the container
7537 * and try to create a volume
7539 struct intel_super
*super
;
7541 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7543 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7545 return validate_geometry_imsm_volume(st
, level
, layout
,
7547 size
, data_offset
, dev
,
7554 pr_err("failed container membership check\n");
7560 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7562 struct intel_super
*super
= st
->sb
;
7564 if (level
&& *level
== UnSet
)
7565 *level
= LEVEL_CONTAINER
;
7567 if (level
&& layout
&& *layout
== UnSet
)
7568 *layout
= imsm_level_to_layout(*level
);
7570 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7571 *chunk
= imsm_default_chunk(super
->orom
);
7574 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7576 static int kill_subarray_imsm(struct supertype
*st
, char *subarray_id
)
7578 /* remove the subarray currently referenced by subarray_id */
7580 struct intel_dev
**dp
;
7581 struct intel_super
*super
= st
->sb
;
7582 __u8 current_vol
= strtoul(subarray_id
, NULL
, 10);
7583 struct imsm_super
*mpb
= super
->anchor
;
7585 if (mpb
->num_raid_devs
== 0)
7588 /* block deletions that would change the uuid of active subarrays
7590 * FIXME when immutable ids are available, but note that we'll
7591 * also need to fixup the invalidated/active subarray indexes in
7594 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7597 if (i
< current_vol
)
7599 sprintf(subarray
, "%u", i
);
7600 if (is_subarray_active(subarray
, st
->devnm
)) {
7601 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7608 if (st
->update_tail
) {
7609 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7611 u
->type
= update_kill_array
;
7612 u
->dev_idx
= current_vol
;
7613 append_metadata_update(st
, u
, sizeof(*u
));
7618 for (dp
= &super
->devlist
; *dp
;)
7619 if ((*dp
)->index
== current_vol
) {
7622 handle_missing(super
, (*dp
)->dev
);
7623 if ((*dp
)->index
> current_vol
)
7628 /* no more raid devices, all active components are now spares,
7629 * but of course failed are still failed
7631 if (--mpb
->num_raid_devs
== 0) {
7634 for (d
= super
->disks
; d
; d
= d
->next
)
7639 super
->updates_pending
++;
7644 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7645 char *update
, struct mddev_ident
*ident
)
7647 /* update the subarray currently referenced by ->current_vol */
7648 struct intel_super
*super
= st
->sb
;
7649 struct imsm_super
*mpb
= super
->anchor
;
7651 if (strcmp(update
, "name") == 0) {
7652 char *name
= ident
->name
;
7656 if (is_subarray_active(subarray
, st
->devnm
)) {
7657 pr_err("Unable to update name of active subarray\n");
7661 if (!check_name(super
, name
, 0))
7664 vol
= strtoul(subarray
, &ep
, 10);
7665 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7668 if (st
->update_tail
) {
7669 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7671 u
->type
= update_rename_array
;
7673 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7674 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7675 append_metadata_update(st
, u
, sizeof(*u
));
7677 struct imsm_dev
*dev
;
7680 dev
= get_imsm_dev(super
, vol
);
7681 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7682 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7683 memcpy(dev
->volume
, name
, namelen
);
7684 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7685 dev
= get_imsm_dev(super
, i
);
7686 handle_missing(super
, dev
);
7688 super
->updates_pending
++;
7690 } else if (strcmp(update
, "ppl") == 0 ||
7691 strcmp(update
, "no-ppl") == 0) {
7694 int vol
= strtoul(subarray
, &ep
, 10);
7696 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7699 if (strcmp(update
, "ppl") == 0)
7700 new_policy
= RWH_MULTIPLE_DISTRIBUTED
;
7702 new_policy
= RWH_MULTIPLE_OFF
;
7704 if (st
->update_tail
) {
7705 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7707 u
->type
= update_rwh_policy
;
7709 u
->new_policy
= new_policy
;
7710 append_metadata_update(st
, u
, sizeof(*u
));
7712 struct imsm_dev
*dev
;
7714 dev
= get_imsm_dev(super
, vol
);
7715 dev
->rwh_policy
= new_policy
;
7716 super
->updates_pending
++;
7724 static int is_gen_migration(struct imsm_dev
*dev
)
7729 if (!dev
->vol
.migr_state
)
7732 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7738 static int is_rebuilding(struct imsm_dev
*dev
)
7740 struct imsm_map
*migr_map
;
7742 if (!dev
->vol
.migr_state
)
7745 if (migr_type(dev
) != MIGR_REBUILD
)
7748 migr_map
= get_imsm_map(dev
, MAP_1
);
7750 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7756 static int is_initializing(struct imsm_dev
*dev
)
7758 struct imsm_map
*migr_map
;
7760 if (!dev
->vol
.migr_state
)
7763 if (migr_type(dev
) != MIGR_INIT
)
7766 migr_map
= get_imsm_map(dev
, MAP_1
);
7768 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7774 static void update_recovery_start(struct intel_super
*super
,
7775 struct imsm_dev
*dev
,
7776 struct mdinfo
*array
)
7778 struct mdinfo
*rebuild
= NULL
;
7782 if (!is_rebuilding(dev
))
7785 /* Find the rebuild target, but punt on the dual rebuild case */
7786 for (d
= array
->devs
; d
; d
= d
->next
)
7787 if (d
->recovery_start
== 0) {
7794 /* (?) none of the disks are marked with
7795 * IMSM_ORD_REBUILD, so assume they are missing and the
7796 * disk_ord_tbl was not correctly updated
7798 dprintf("failed to locate out-of-sync disk\n");
7802 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7803 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7806 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7808 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7810 /* Given a container loaded by load_super_imsm_all,
7811 * extract information about all the arrays into
7813 * If 'subarray' is given, just extract info about that array.
7815 * For each imsm_dev create an mdinfo, fill it in,
7816 * then look for matching devices in super->disks
7817 * and create appropriate device mdinfo.
7819 struct intel_super
*super
= st
->sb
;
7820 struct imsm_super
*mpb
= super
->anchor
;
7821 struct mdinfo
*rest
= NULL
;
7825 int spare_disks
= 0;
7826 int current_vol
= super
->current_vol
;
7828 /* do not assemble arrays when not all attributes are supported */
7829 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7831 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7834 /* count spare devices, not used in maps
7836 for (d
= super
->disks
; d
; d
= d
->next
)
7840 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7841 struct imsm_dev
*dev
;
7842 struct imsm_map
*map
;
7843 struct imsm_map
*map2
;
7844 struct mdinfo
*this;
7851 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7854 dev
= get_imsm_dev(super
, i
);
7855 map
= get_imsm_map(dev
, MAP_0
);
7856 map2
= get_imsm_map(dev
, MAP_1
);
7857 level
= get_imsm_raid_level(map
);
7859 /* do not publish arrays that are in the middle of an
7860 * unsupported migration
7862 if (dev
->vol
.migr_state
&&
7863 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7864 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7868 /* do not publish arrays that are not support by controller's
7872 this = xmalloc(sizeof(*this));
7874 super
->current_vol
= i
;
7875 getinfo_super_imsm_volume(st
, this, NULL
);
7877 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7878 /* mdadm does not support all metadata features- set the bit in all arrays state */
7879 if (!validate_geometry_imsm_orom(super
,
7880 level
, /* RAID level */
7881 imsm_level_to_layout(level
),
7882 map
->num_members
, /* raid disks */
7883 &chunk
, imsm_dev_size(dev
),
7885 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7887 this->array
.state
|=
7888 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7889 (1<<MD_SB_BLOCK_VOLUME
);
7892 /* if array has bad blocks, set suitable bit in all arrays state */
7894 this->array
.state
|=
7895 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7896 (1<<MD_SB_BLOCK_VOLUME
);
7898 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7899 unsigned long long recovery_start
;
7900 struct mdinfo
*info_d
;
7908 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7909 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7910 for (d
= super
->disks
; d
; d
= d
->next
)
7911 if (d
->index
== idx
)
7914 recovery_start
= MaxSector
;
7917 if (d
&& is_failed(&d
->disk
))
7919 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
7921 if (!(ord
& IMSM_ORD_REBUILD
))
7922 this->array
.working_disks
++;
7924 * if we skip some disks the array will be assmebled degraded;
7925 * reset resync start to avoid a dirty-degraded
7926 * situation when performing the intial sync
7931 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
7932 if ((!able_to_resync(level
, missing
) ||
7933 recovery_start
== 0))
7934 this->resync_start
= MaxSector
;
7937 * FIXME handle dirty degraded
7944 info_d
= xcalloc(1, sizeof(*info_d
));
7945 info_d
->next
= this->devs
;
7946 this->devs
= info_d
;
7948 info_d
->disk
.number
= d
->index
;
7949 info_d
->disk
.major
= d
->major
;
7950 info_d
->disk
.minor
= d
->minor
;
7951 info_d
->disk
.raid_disk
= slot
;
7952 info_d
->recovery_start
= recovery_start
;
7954 if (slot
< map2
->num_members
)
7955 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7957 this->array
.spare_disks
++;
7959 if (slot
< map
->num_members
)
7960 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7962 this->array
.spare_disks
++;
7965 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7966 info_d
->data_offset
= pba_of_lba0(map
);
7967 info_d
->component_size
= calc_component_size(map
, dev
);
7969 if (map
->raid_level
== 5) {
7970 info_d
->ppl_sector
= this->ppl_sector
;
7971 info_d
->ppl_size
= this->ppl_size
;
7972 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
7973 recovery_start
== 0)
7974 this->resync_start
= 0;
7977 info_d
->bb
.supported
= 1;
7978 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7979 info_d
->data_offset
,
7980 info_d
->component_size
,
7983 /* now that the disk list is up-to-date fixup recovery_start */
7984 update_recovery_start(super
, dev
, this);
7985 this->array
.spare_disks
+= spare_disks
;
7987 /* check for reshape */
7988 if (this->reshape_active
== 1)
7989 recover_backup_imsm(st
, this);
7993 super
->current_vol
= current_vol
;
7997 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7998 int failed
, int look_in_map
)
8000 struct imsm_map
*map
;
8002 map
= get_imsm_map(dev
, look_in_map
);
8005 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
8006 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
8008 switch (get_imsm_raid_level(map
)) {
8010 return IMSM_T_STATE_FAILED
;
8013 if (failed
< map
->num_members
)
8014 return IMSM_T_STATE_DEGRADED
;
8016 return IMSM_T_STATE_FAILED
;
8021 * check to see if any mirrors have failed, otherwise we
8022 * are degraded. Even numbered slots are mirrored on
8026 /* gcc -Os complains that this is unused */
8027 int insync
= insync
;
8029 for (i
= 0; i
< map
->num_members
; i
++) {
8030 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8031 int idx
= ord_to_idx(ord
);
8032 struct imsm_disk
*disk
;
8034 /* reset the potential in-sync count on even-numbered
8035 * slots. num_copies is always 2 for imsm raid10
8040 disk
= get_imsm_disk(super
, idx
);
8041 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8044 /* no in-sync disks left in this mirror the
8048 return IMSM_T_STATE_FAILED
;
8051 return IMSM_T_STATE_DEGRADED
;
8055 return IMSM_T_STATE_DEGRADED
;
8057 return IMSM_T_STATE_FAILED
;
8063 return map
->map_state
;
8066 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8071 struct imsm_disk
*disk
;
8072 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8073 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8074 struct imsm_map
*map_for_loop
;
8079 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8080 * disks that are being rebuilt. New failures are recorded to
8081 * map[0]. So we look through all the disks we started with and
8082 * see if any failures are still present, or if any new ones
8086 if (prev
&& (map
->num_members
< prev
->num_members
))
8087 map_for_loop
= prev
;
8089 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8091 /* when MAP_X is passed both maps failures are counted
8094 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8095 i
< prev
->num_members
) {
8096 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8097 idx_1
= ord_to_idx(ord
);
8099 disk
= get_imsm_disk(super
, idx_1
);
8100 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8103 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8104 i
< map
->num_members
) {
8105 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8106 idx
= ord_to_idx(ord
);
8109 disk
= get_imsm_disk(super
, idx
);
8110 if (!disk
|| is_failed(disk
) ||
8111 ord
& IMSM_ORD_REBUILD
)
8120 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8123 struct intel_super
*super
= c
->sb
;
8124 struct imsm_super
*mpb
= super
->anchor
;
8125 struct imsm_update_prealloc_bb_mem u
;
8127 if (atoi(inst
) >= mpb
->num_raid_devs
) {
8128 pr_err("subarry index %d, out of range\n", atoi(inst
));
8132 dprintf("imsm: open_new %s\n", inst
);
8133 a
->info
.container_member
= atoi(inst
);
8135 u
.type
= update_prealloc_badblocks_mem
;
8136 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8141 static int is_resyncing(struct imsm_dev
*dev
)
8143 struct imsm_map
*migr_map
;
8145 if (!dev
->vol
.migr_state
)
8148 if (migr_type(dev
) == MIGR_INIT
||
8149 migr_type(dev
) == MIGR_REPAIR
)
8152 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8155 migr_map
= get_imsm_map(dev
, MAP_1
);
8157 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8158 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8164 /* return true if we recorded new information */
8165 static int mark_failure(struct intel_super
*super
,
8166 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8170 struct imsm_map
*map
;
8171 char buf
[MAX_RAID_SERIAL_LEN
+3];
8172 unsigned int len
, shift
= 0;
8174 /* new failures are always set in map[0] */
8175 map
= get_imsm_map(dev
, MAP_0
);
8177 slot
= get_imsm_disk_slot(map
, idx
);
8181 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8182 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8185 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8186 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8188 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8189 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8190 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8192 disk
->status
|= FAILED_DISK
;
8193 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8194 /* mark failures in second map if second map exists and this disk
8196 * This is valid for migration, initialization and rebuild
8198 if (dev
->vol
.migr_state
) {
8199 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8200 int slot2
= get_imsm_disk_slot(map2
, idx
);
8202 if (slot2
< map2
->num_members
&& slot2
>= 0)
8203 set_imsm_ord_tbl_ent(map2
, slot2
,
8204 idx
| IMSM_ORD_REBUILD
);
8206 if (map
->failed_disk_num
== 0xff ||
8207 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8208 map
->failed_disk_num
= slot
;
8210 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8215 static void mark_missing(struct intel_super
*super
,
8216 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8218 mark_failure(super
, dev
, disk
, idx
);
8220 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8223 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8224 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8227 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8231 if (!super
->missing
)
8234 /* When orom adds replacement for missing disk it does
8235 * not remove entry of missing disk, but just updates map with
8236 * new added disk. So it is not enough just to test if there is
8237 * any missing disk, we have to look if there are any failed disks
8238 * in map to stop migration */
8240 dprintf("imsm: mark missing\n");
8241 /* end process for initialization and rebuild only
8243 if (is_gen_migration(dev
) == 0) {
8244 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8248 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8249 struct imsm_map
*map1
;
8250 int i
, ord
, ord_map1
;
8253 for (i
= 0; i
< map
->num_members
; i
++) {
8254 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8255 if (!(ord
& IMSM_ORD_REBUILD
))
8258 map1
= get_imsm_map(dev
, MAP_1
);
8262 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8263 if (ord_map1
& IMSM_ORD_REBUILD
)
8268 map_state
= imsm_check_degraded(super
, dev
,
8270 end_migration(dev
, super
, map_state
);
8274 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8275 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8276 super
->updates_pending
++;
8279 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8282 unsigned long long array_blocks
;
8283 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8284 int used_disks
= imsm_num_data_members(map
);
8286 if (used_disks
== 0) {
8287 /* when problems occures
8288 * return current array_blocks value
8290 array_blocks
= imsm_dev_size(dev
);
8292 return array_blocks
;
8295 /* set array size in metadata
8298 /* OLCE size change is caused by added disks
8300 array_blocks
= per_dev_array_size(map
) * used_disks
;
8302 /* Online Volume Size Change
8303 * Using available free space
8305 array_blocks
= new_size
;
8307 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8308 set_imsm_dev_size(dev
, array_blocks
);
8310 return array_blocks
;
8313 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8315 static void imsm_progress_container_reshape(struct intel_super
*super
)
8317 /* if no device has a migr_state, but some device has a
8318 * different number of members than the previous device, start
8319 * changing the number of devices in this device to match
8322 struct imsm_super
*mpb
= super
->anchor
;
8323 int prev_disks
= -1;
8327 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8328 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8329 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8330 struct imsm_map
*map2
;
8331 int prev_num_members
;
8333 if (dev
->vol
.migr_state
)
8336 if (prev_disks
== -1)
8337 prev_disks
= map
->num_members
;
8338 if (prev_disks
== map
->num_members
)
8341 /* OK, this array needs to enter reshape mode.
8342 * i.e it needs a migr_state
8345 copy_map_size
= sizeof_imsm_map(map
);
8346 prev_num_members
= map
->num_members
;
8347 map
->num_members
= prev_disks
;
8348 dev
->vol
.migr_state
= 1;
8349 dev
->vol
.curr_migr_unit
= 0;
8350 set_migr_type(dev
, MIGR_GEN_MIGR
);
8351 for (i
= prev_num_members
;
8352 i
< map
->num_members
; i
++)
8353 set_imsm_ord_tbl_ent(map
, i
, i
);
8354 map2
= get_imsm_map(dev
, MAP_1
);
8355 /* Copy the current map */
8356 memcpy(map2
, map
, copy_map_size
);
8357 map2
->num_members
= prev_num_members
;
8359 imsm_set_array_size(dev
, -1);
8360 super
->clean_migration_record_by_mdmon
= 1;
8361 super
->updates_pending
++;
8365 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8366 * states are handled in imsm_set_disk() with one exception, when a
8367 * resync is stopped due to a new failure this routine will set the
8368 * 'degraded' state for the array.
8370 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8372 int inst
= a
->info
.container_member
;
8373 struct intel_super
*super
= a
->container
->sb
;
8374 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8375 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8376 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8377 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8378 __u32 blocks_per_unit
;
8380 if (dev
->vol
.migr_state
&&
8381 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8382 /* array state change is blocked due to reshape action
8384 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8385 * - finish the reshape (if last_checkpoint is big and action != reshape)
8386 * - update curr_migr_unit
8388 if (a
->curr_action
== reshape
) {
8389 /* still reshaping, maybe update curr_migr_unit */
8390 goto mark_checkpoint
;
8392 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8393 /* for some reason we aborted the reshape.
8395 * disable automatic metadata rollback
8396 * user action is required to recover process
8399 struct imsm_map
*map2
=
8400 get_imsm_map(dev
, MAP_1
);
8401 dev
->vol
.migr_state
= 0;
8402 set_migr_type(dev
, 0);
8403 dev
->vol
.curr_migr_unit
= 0;
8405 sizeof_imsm_map(map2
));
8406 super
->updates_pending
++;
8409 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8410 unsigned long long array_blocks
;
8414 used_disks
= imsm_num_data_members(map
);
8415 if (used_disks
> 0) {
8417 per_dev_array_size(map
) *
8420 round_size_to_mb(array_blocks
,
8422 a
->info
.custom_array_size
= array_blocks
;
8423 /* encourage manager to update array
8427 a
->check_reshape
= 1;
8429 /* finalize online capacity expansion/reshape */
8430 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8432 mdi
->disk
.raid_disk
,
8435 imsm_progress_container_reshape(super
);
8440 /* before we activate this array handle any missing disks */
8441 if (consistent
== 2)
8442 handle_missing(super
, dev
);
8444 if (consistent
== 2 &&
8445 (!is_resync_complete(&a
->info
) ||
8446 map_state
!= IMSM_T_STATE_NORMAL
||
8447 dev
->vol
.migr_state
))
8450 if (is_resync_complete(&a
->info
)) {
8451 /* complete intialization / resync,
8452 * recovery and interrupted recovery is completed in
8455 if (is_resyncing(dev
)) {
8456 dprintf("imsm: mark resync done\n");
8457 end_migration(dev
, super
, map_state
);
8458 super
->updates_pending
++;
8459 a
->last_checkpoint
= 0;
8461 } else if ((!is_resyncing(dev
) && !failed
) &&
8462 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8463 /* mark the start of the init process if nothing is failed */
8464 dprintf("imsm: mark resync start\n");
8465 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8466 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8468 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8469 super
->updates_pending
++;
8473 /* skip checkpointing for general migration,
8474 * it is controlled in mdadm
8476 if (is_gen_migration(dev
))
8477 goto skip_mark_checkpoint
;
8479 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8480 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8481 if (blocks_per_unit
) {
8485 units
= a
->last_checkpoint
/ blocks_per_unit
;
8488 /* check that we did not overflow 32-bits, and that
8489 * curr_migr_unit needs updating
8491 if (units32
== units
&&
8493 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8494 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8495 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8496 super
->updates_pending
++;
8500 skip_mark_checkpoint
:
8501 /* mark dirty / clean */
8502 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8503 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8504 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8506 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8508 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8509 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8510 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8511 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8513 super
->updates_pending
++;
8519 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8521 int inst
= a
->info
.container_member
;
8522 struct intel_super
*super
= a
->container
->sb
;
8523 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8524 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8526 if (slot
> map
->num_members
) {
8527 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8528 slot
, map
->num_members
- 1);
8535 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8538 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8540 int inst
= a
->info
.container_member
;
8541 struct intel_super
*super
= a
->container
->sb
;
8542 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8543 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8544 struct imsm_disk
*disk
;
8546 int recovery_not_finished
= 0;
8550 int rebuild_done
= 0;
8553 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8557 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8558 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8560 /* check for new failures */
8561 if (disk
&& (state
& DS_FAULTY
)) {
8562 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8563 super
->updates_pending
++;
8566 /* check if in_sync */
8567 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8568 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8570 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8572 super
->updates_pending
++;
8575 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8576 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8578 /* check if recovery complete, newly degraded, or failed */
8579 dprintf("imsm: Detected transition to state ");
8580 switch (map_state
) {
8581 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8582 dprintf("normal: ");
8583 if (is_rebuilding(dev
)) {
8584 dprintf_cont("while rebuilding");
8585 /* check if recovery is really finished */
8586 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8587 if (mdi
->recovery_start
!= MaxSector
) {
8588 recovery_not_finished
= 1;
8591 if (recovery_not_finished
) {
8593 dprintf("Rebuild has not finished yet, state not changed");
8594 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8595 a
->last_checkpoint
= mdi
->recovery_start
;
8596 super
->updates_pending
++;
8600 end_migration(dev
, super
, map_state
);
8601 map
= get_imsm_map(dev
, MAP_0
);
8602 map
->failed_disk_num
= ~0;
8603 super
->updates_pending
++;
8604 a
->last_checkpoint
= 0;
8607 if (is_gen_migration(dev
)) {
8608 dprintf_cont("while general migration");
8609 if (a
->last_checkpoint
>= a
->info
.component_size
)
8610 end_migration(dev
, super
, map_state
);
8612 map
->map_state
= map_state
;
8613 map
= get_imsm_map(dev
, MAP_0
);
8614 map
->failed_disk_num
= ~0;
8615 super
->updates_pending
++;
8619 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8620 dprintf_cont("degraded: ");
8621 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8622 dprintf_cont("mark degraded");
8623 map
->map_state
= map_state
;
8624 super
->updates_pending
++;
8625 a
->last_checkpoint
= 0;
8628 if (is_rebuilding(dev
)) {
8629 dprintf_cont("while rebuilding ");
8630 if (state
& DS_FAULTY
) {
8631 dprintf_cont("removing failed drive ");
8632 if (n
== map
->failed_disk_num
) {
8633 dprintf_cont("end migration");
8634 end_migration(dev
, super
, map_state
);
8635 a
->last_checkpoint
= 0;
8637 dprintf_cont("fail detected during rebuild, changing map state");
8638 map
->map_state
= map_state
;
8640 super
->updates_pending
++;
8646 /* check if recovery is really finished */
8647 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8648 if (mdi
->recovery_start
!= MaxSector
) {
8649 recovery_not_finished
= 1;
8652 if (recovery_not_finished
) {
8654 dprintf_cont("Rebuild has not finished yet");
8655 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8656 a
->last_checkpoint
=
8657 mdi
->recovery_start
;
8658 super
->updates_pending
++;
8663 dprintf_cont(" Rebuild done, still degraded");
8664 end_migration(dev
, super
, map_state
);
8665 a
->last_checkpoint
= 0;
8666 super
->updates_pending
++;
8668 for (i
= 0; i
< map
->num_members
; i
++) {
8669 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8671 if (idx
& IMSM_ORD_REBUILD
)
8672 map
->failed_disk_num
= i
;
8674 super
->updates_pending
++;
8677 if (is_gen_migration(dev
)) {
8678 dprintf_cont("while general migration");
8679 if (a
->last_checkpoint
>= a
->info
.component_size
)
8680 end_migration(dev
, super
, map_state
);
8682 map
->map_state
= map_state
;
8683 manage_second_map(super
, dev
);
8685 super
->updates_pending
++;
8688 if (is_initializing(dev
)) {
8689 dprintf_cont("while initialization.");
8690 map
->map_state
= map_state
;
8691 super
->updates_pending
++;
8695 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8696 dprintf_cont("failed: ");
8697 if (is_gen_migration(dev
)) {
8698 dprintf_cont("while general migration");
8699 map
->map_state
= map_state
;
8700 super
->updates_pending
++;
8703 if (map
->map_state
!= map_state
) {
8704 dprintf_cont("mark failed");
8705 end_migration(dev
, super
, map_state
);
8706 super
->updates_pending
++;
8707 a
->last_checkpoint
= 0;
8712 dprintf_cont("state %i\n", map_state
);
8717 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8720 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8721 unsigned long long dsize
;
8722 unsigned long long sectors
;
8723 unsigned int sector_size
;
8725 get_dev_sector_size(fd
, NULL
, §or_size
);
8726 get_dev_size(fd
, NULL
, &dsize
);
8728 if (mpb_size
> sector_size
) {
8729 /* -1 to account for anchor */
8730 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8732 /* write the extended mpb to the sectors preceeding the anchor */
8733 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8737 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8738 sector_size
* sectors
) != sector_size
* sectors
)
8742 /* first block is stored on second to last sector of the disk */
8743 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8746 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8752 static void imsm_sync_metadata(struct supertype
*container
)
8754 struct intel_super
*super
= container
->sb
;
8756 dprintf("sync metadata: %d\n", super
->updates_pending
);
8757 if (!super
->updates_pending
)
8760 write_super_imsm(container
, 0);
8762 super
->updates_pending
= 0;
8765 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8767 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8768 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8771 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8775 if (dl
&& is_failed(&dl
->disk
))
8779 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8784 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8785 struct active_array
*a
, int activate_new
,
8786 struct mdinfo
*additional_test_list
)
8788 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8789 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8790 struct imsm_super
*mpb
= super
->anchor
;
8791 struct imsm_map
*map
;
8792 unsigned long long pos
;
8797 __u32 array_start
= 0;
8798 __u32 array_end
= 0;
8800 struct mdinfo
*test_list
;
8802 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8803 /* If in this array, skip */
8804 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8805 if (d
->state_fd
>= 0 &&
8806 d
->disk
.major
== dl
->major
&&
8807 d
->disk
.minor
== dl
->minor
) {
8808 dprintf("%x:%x already in array\n",
8809 dl
->major
, dl
->minor
);
8814 test_list
= additional_test_list
;
8816 if (test_list
->disk
.major
== dl
->major
&&
8817 test_list
->disk
.minor
== dl
->minor
) {
8818 dprintf("%x:%x already in additional test list\n",
8819 dl
->major
, dl
->minor
);
8822 test_list
= test_list
->next
;
8827 /* skip in use or failed drives */
8828 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8830 dprintf("%x:%x status (failed: %d index: %d)\n",
8831 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8835 /* skip pure spares when we are looking for partially
8836 * assimilated drives
8838 if (dl
->index
== -1 && !activate_new
)
8841 if (!drive_validate_sector_size(super
, dl
))
8844 /* Does this unused device have the requisite free space?
8845 * It needs to be able to cover all member volumes
8847 ex
= get_extents(super
, dl
, 1);
8849 dprintf("cannot get extents\n");
8852 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8853 dev
= get_imsm_dev(super
, i
);
8854 map
= get_imsm_map(dev
, MAP_0
);
8856 /* check if this disk is already a member of
8859 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8865 array_start
= pba_of_lba0(map
);
8866 array_end
= array_start
+
8867 per_dev_array_size(map
) - 1;
8870 /* check that we can start at pba_of_lba0 with
8871 * num_data_stripes*blocks_per_stripe of space
8873 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8877 pos
= ex
[j
].start
+ ex
[j
].size
;
8879 } while (ex
[j
-1].size
);
8886 if (i
< mpb
->num_raid_devs
) {
8887 dprintf("%x:%x does not have %u to %u available\n",
8888 dl
->major
, dl
->minor
, array_start
, array_end
);
8898 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8900 struct imsm_dev
*dev2
;
8901 struct imsm_map
*map
;
8907 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8909 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8910 if (state
== IMSM_T_STATE_FAILED
) {
8911 map
= get_imsm_map(dev2
, MAP_0
);
8914 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8916 * Check if failed disks are deleted from intel
8917 * disk list or are marked to be deleted
8919 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8920 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8922 * Do not rebuild the array if failed disks
8923 * from failed sub-array are not removed from
8927 is_failed(&idisk
->disk
) &&
8928 (idisk
->action
!= DISK_REMOVE
))
8936 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8937 struct metadata_update
**updates
)
8940 * Find a device with unused free space and use it to replace a
8941 * failed/vacant region in an array. We replace failed regions one a
8942 * array at a time. The result is that a new spare disk will be added
8943 * to the first failed array and after the monitor has finished
8944 * propagating failures the remainder will be consumed.
8946 * FIXME add a capability for mdmon to request spares from another
8950 struct intel_super
*super
= a
->container
->sb
;
8951 int inst
= a
->info
.container_member
;
8952 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8953 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8954 int failed
= a
->info
.array
.raid_disks
;
8955 struct mdinfo
*rv
= NULL
;
8958 struct metadata_update
*mu
;
8960 struct imsm_update_activate_spare
*u
;
8965 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8966 if ((d
->curr_state
& DS_FAULTY
) &&
8968 /* wait for Removal to happen */
8970 if (d
->state_fd
>= 0)
8974 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8975 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8977 if (imsm_reshape_blocks_arrays_changes(super
))
8980 /* Cannot activate another spare if rebuild is in progress already
8982 if (is_rebuilding(dev
)) {
8983 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8987 if (a
->info
.array
.level
== 4)
8988 /* No repair for takeovered array
8989 * imsm doesn't support raid4
8993 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8994 IMSM_T_STATE_DEGRADED
)
8997 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8998 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
9003 * If there are any failed disks check state of the other volume.
9004 * Block rebuild if the another one is failed until failed disks
9005 * are removed from container.
9008 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
9009 MAX_RAID_SERIAL_LEN
, dev
->volume
);
9010 /* check if states of the other volumes allow for rebuild */
9011 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
9013 allowed
= imsm_rebuild_allowed(a
->container
,
9021 /* For each slot, if it is not working, find a spare */
9022 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
9023 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9024 if (d
->disk
.raid_disk
== i
)
9026 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
9027 if (d
&& (d
->state_fd
>= 0))
9031 * OK, this device needs recovery. Try to re-add the
9032 * previous occupant of this slot, if this fails see if
9033 * we can continue the assimilation of a spare that was
9034 * partially assimilated, finally try to activate a new
9037 dl
= imsm_readd(super
, i
, a
);
9039 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9041 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9045 /* found a usable disk with enough space */
9046 di
= xcalloc(1, sizeof(*di
));
9048 /* dl->index will be -1 in the case we are activating a
9049 * pristine spare. imsm_process_update() will create a
9050 * new index in this case. Once a disk is found to be
9051 * failed in all member arrays it is kicked from the
9054 di
->disk
.number
= dl
->index
;
9056 /* (ab)use di->devs to store a pointer to the device
9059 di
->devs
= (struct mdinfo
*) dl
;
9061 di
->disk
.raid_disk
= i
;
9062 di
->disk
.major
= dl
->major
;
9063 di
->disk
.minor
= dl
->minor
;
9065 di
->recovery_start
= 0;
9066 di
->data_offset
= pba_of_lba0(map
);
9067 di
->component_size
= a
->info
.component_size
;
9068 di
->container_member
= inst
;
9069 di
->bb
.supported
= 1;
9070 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9071 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9072 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9074 super
->random
= random32();
9078 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9079 i
, di
->data_offset
);
9083 /* No spares found */
9085 /* Now 'rv' has a list of devices to return.
9086 * Create a metadata_update record to update the
9087 * disk_ord_tbl for the array
9089 mu
= xmalloc(sizeof(*mu
));
9090 mu
->buf
= xcalloc(num_spares
,
9091 sizeof(struct imsm_update_activate_spare
));
9093 mu
->space_list
= NULL
;
9094 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9095 mu
->next
= *updates
;
9096 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9098 for (di
= rv
; di
; di
= di
->next
) {
9099 u
->type
= update_activate_spare
;
9100 u
->dl
= (struct dl
*) di
->devs
;
9102 u
->slot
= di
->disk
.raid_disk
;
9113 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9115 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9116 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9117 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9118 struct disk_info
*inf
= get_disk_info(u
);
9119 struct imsm_disk
*disk
;
9123 for (i
= 0; i
< map
->num_members
; i
++) {
9124 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9125 for (j
= 0; j
< new_map
->num_members
; j
++)
9126 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9133 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9137 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9138 if (dl
->major
== major
&& dl
->minor
== minor
)
9143 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9149 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9150 if (dl
->major
== major
&& dl
->minor
== minor
) {
9153 prev
->next
= dl
->next
;
9155 super
->disks
= dl
->next
;
9157 __free_imsm_disk(dl
);
9158 dprintf("removed %x:%x\n", major
, minor
);
9166 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9168 static int add_remove_disk_update(struct intel_super
*super
)
9170 int check_degraded
= 0;
9173 /* add/remove some spares to/from the metadata/contrainer */
9174 while (super
->disk_mgmt_list
) {
9175 struct dl
*disk_cfg
;
9177 disk_cfg
= super
->disk_mgmt_list
;
9178 super
->disk_mgmt_list
= disk_cfg
->next
;
9179 disk_cfg
->next
= NULL
;
9181 if (disk_cfg
->action
== DISK_ADD
) {
9182 disk_cfg
->next
= super
->disks
;
9183 super
->disks
= disk_cfg
;
9185 dprintf("added %x:%x\n",
9186 disk_cfg
->major
, disk_cfg
->minor
);
9187 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9188 dprintf("Disk remove action processed: %x.%x\n",
9189 disk_cfg
->major
, disk_cfg
->minor
);
9190 disk
= get_disk_super(super
,
9194 /* store action status */
9195 disk
->action
= DISK_REMOVE
;
9196 /* remove spare disks only */
9197 if (disk
->index
== -1) {
9198 remove_disk_super(super
,
9202 disk_cfg
->fd
= disk
->fd
;
9206 /* release allocate disk structure */
9207 __free_imsm_disk(disk_cfg
);
9210 return check_degraded
;
9213 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9214 struct intel_super
*super
,
9217 struct intel_dev
*id
;
9218 void **tofree
= NULL
;
9221 dprintf("(enter)\n");
9222 if (u
->subdev
< 0 || u
->subdev
> 1) {
9223 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9226 if (space_list
== NULL
|| *space_list
== NULL
) {
9227 dprintf("imsm: Error: Memory is not allocated\n");
9231 for (id
= super
->devlist
; id
; id
= id
->next
) {
9232 if (id
->index
== (unsigned)u
->subdev
) {
9233 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9234 struct imsm_map
*map
;
9235 struct imsm_dev
*new_dev
=
9236 (struct imsm_dev
*)*space_list
;
9237 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9239 struct dl
*new_disk
;
9241 if (new_dev
== NULL
)
9243 *space_list
= **space_list
;
9244 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9245 map
= get_imsm_map(new_dev
, MAP_0
);
9247 dprintf("imsm: Error: migration in progress");
9251 to_state
= map
->map_state
;
9252 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9254 /* this should not happen */
9255 if (u
->new_disks
[0] < 0) {
9256 map
->failed_disk_num
=
9257 map
->num_members
- 1;
9258 to_state
= IMSM_T_STATE_DEGRADED
;
9260 to_state
= IMSM_T_STATE_NORMAL
;
9262 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9263 if (u
->new_level
> -1)
9264 map
->raid_level
= u
->new_level
;
9265 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9266 if ((u
->new_level
== 5) &&
9267 (migr_map
->raid_level
== 0)) {
9268 int ord
= map
->num_members
- 1;
9269 migr_map
->num_members
--;
9270 if (u
->new_disks
[0] < 0)
9271 ord
|= IMSM_ORD_REBUILD
;
9272 set_imsm_ord_tbl_ent(map
,
9273 map
->num_members
- 1,
9277 tofree
= (void **)dev
;
9279 /* update chunk size
9281 if (u
->new_chunksize
> 0) {
9282 unsigned long long num_data_stripes
;
9283 struct imsm_map
*dest_map
=
9284 get_imsm_map(dev
, MAP_0
);
9286 imsm_num_data_members(dest_map
);
9288 if (used_disks
== 0)
9291 map
->blocks_per_strip
=
9292 __cpu_to_le16(u
->new_chunksize
* 2);
9294 imsm_dev_size(dev
) / used_disks
;
9295 num_data_stripes
/= map
->blocks_per_strip
;
9296 num_data_stripes
/= map
->num_domains
;
9297 set_num_data_stripes(map
, num_data_stripes
);
9300 /* ensure blocks_per_member has valid value
9302 set_blocks_per_member(map
,
9303 per_dev_array_size(map
) +
9304 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9308 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9309 migr_map
->raid_level
== map
->raid_level
)
9312 if (u
->new_disks
[0] >= 0) {
9315 new_disk
= get_disk_super(super
,
9316 major(u
->new_disks
[0]),
9317 minor(u
->new_disks
[0]));
9318 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9319 major(u
->new_disks
[0]),
9320 minor(u
->new_disks
[0]),
9321 new_disk
, new_disk
->index
);
9322 if (new_disk
== NULL
)
9323 goto error_disk_add
;
9325 new_disk
->index
= map
->num_members
- 1;
9326 /* slot to fill in autolayout
9328 new_disk
->raiddisk
= new_disk
->index
;
9329 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9330 new_disk
->disk
.status
&= ~SPARE_DISK
;
9332 goto error_disk_add
;
9335 *tofree
= *space_list
;
9336 /* calculate new size
9338 imsm_set_array_size(new_dev
, -1);
9345 *space_list
= tofree
;
9349 dprintf("Error: imsm: Cannot find disk.\n");
9353 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9354 struct intel_super
*super
)
9356 struct intel_dev
*id
;
9359 dprintf("(enter)\n");
9360 if (u
->subdev
< 0 || u
->subdev
> 1) {
9361 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9365 for (id
= super
->devlist
; id
; id
= id
->next
) {
9366 if (id
->index
== (unsigned)u
->subdev
) {
9367 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9368 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9369 int used_disks
= imsm_num_data_members(map
);
9370 unsigned long long blocks_per_member
;
9371 unsigned long long num_data_stripes
;
9372 unsigned long long new_size_per_disk
;
9374 if (used_disks
== 0)
9377 /* calculate new size
9379 new_size_per_disk
= u
->new_size
/ used_disks
;
9380 blocks_per_member
= new_size_per_disk
+
9381 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9382 num_data_stripes
= new_size_per_disk
/
9383 map
->blocks_per_strip
;
9384 num_data_stripes
/= map
->num_domains
;
9385 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9386 u
->new_size
, new_size_per_disk
,
9388 set_blocks_per_member(map
, blocks_per_member
);
9389 set_num_data_stripes(map
, num_data_stripes
);
9390 imsm_set_array_size(dev
, u
->new_size
);
9400 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9401 struct intel_super
*super
,
9402 struct active_array
*active_array
)
9404 struct imsm_super
*mpb
= super
->anchor
;
9405 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9406 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9407 struct imsm_map
*migr_map
;
9408 struct active_array
*a
;
9409 struct imsm_disk
*disk
;
9416 int second_map_created
= 0;
9418 for (; u
; u
= u
->next
) {
9419 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9424 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9429 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9434 /* count failures (excluding rebuilds and the victim)
9435 * to determine map[0] state
9438 for (i
= 0; i
< map
->num_members
; i
++) {
9441 disk
= get_imsm_disk(super
,
9442 get_imsm_disk_idx(dev
, i
, MAP_X
));
9443 if (!disk
|| is_failed(disk
))
9447 /* adding a pristine spare, assign a new index */
9448 if (dl
->index
< 0) {
9449 dl
->index
= super
->anchor
->num_disks
;
9450 super
->anchor
->num_disks
++;
9453 disk
->status
|= CONFIGURED_DISK
;
9454 disk
->status
&= ~SPARE_DISK
;
9457 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9458 if (!second_map_created
) {
9459 second_map_created
= 1;
9460 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9461 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9463 map
->map_state
= to_state
;
9464 migr_map
= get_imsm_map(dev
, MAP_1
);
9465 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9466 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9467 dl
->index
| IMSM_ORD_REBUILD
);
9469 /* update the family_num to mark a new container
9470 * generation, being careful to record the existing
9471 * family_num in orig_family_num to clean up after
9472 * earlier mdadm versions that neglected to set it.
9474 if (mpb
->orig_family_num
== 0)
9475 mpb
->orig_family_num
= mpb
->family_num
;
9476 mpb
->family_num
+= super
->random
;
9478 /* count arrays using the victim in the metadata */
9480 for (a
= active_array
; a
; a
= a
->next
) {
9481 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9482 map
= get_imsm_map(dev
, MAP_0
);
9484 if (get_imsm_disk_slot(map
, victim
) >= 0)
9488 /* delete the victim if it is no longer being
9494 /* We know that 'manager' isn't touching anything,
9495 * so it is safe to delete
9497 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9498 if ((*dlp
)->index
== victim
)
9501 /* victim may be on the missing list */
9503 for (dlp
= &super
->missing
; *dlp
;
9504 dlp
= &(*dlp
)->next
)
9505 if ((*dlp
)->index
== victim
)
9507 imsm_delete(super
, dlp
, victim
);
9514 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9515 struct intel_super
*super
,
9518 struct dl
*new_disk
;
9519 struct intel_dev
*id
;
9521 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9522 int disk_count
= u
->old_raid_disks
;
9523 void **tofree
= NULL
;
9524 int devices_to_reshape
= 1;
9525 struct imsm_super
*mpb
= super
->anchor
;
9527 unsigned int dev_id
;
9529 dprintf("(enter)\n");
9531 /* enable spares to use in array */
9532 for (i
= 0; i
< delta_disks
; i
++) {
9533 new_disk
= get_disk_super(super
,
9534 major(u
->new_disks
[i
]),
9535 minor(u
->new_disks
[i
]));
9536 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9537 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9538 new_disk
, new_disk
->index
);
9539 if (new_disk
== NULL
||
9540 (new_disk
->index
>= 0 &&
9541 new_disk
->index
< u
->old_raid_disks
))
9542 goto update_reshape_exit
;
9543 new_disk
->index
= disk_count
++;
9544 /* slot to fill in autolayout
9546 new_disk
->raiddisk
= new_disk
->index
;
9547 new_disk
->disk
.status
|=
9549 new_disk
->disk
.status
&= ~SPARE_DISK
;
9552 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9553 mpb
->num_raid_devs
);
9554 /* manage changes in volume
9556 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9557 void **sp
= *space_list
;
9558 struct imsm_dev
*newdev
;
9559 struct imsm_map
*newmap
, *oldmap
;
9561 for (id
= super
->devlist
; id
; id
= id
->next
) {
9562 if (id
->index
== dev_id
)
9571 /* Copy the dev, but not (all of) the map */
9572 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9573 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9574 newmap
= get_imsm_map(newdev
, MAP_0
);
9575 /* Copy the current map */
9576 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9577 /* update one device only
9579 if (devices_to_reshape
) {
9580 dprintf("imsm: modifying subdev: %i\n",
9582 devices_to_reshape
--;
9583 newdev
->vol
.migr_state
= 1;
9584 newdev
->vol
.curr_migr_unit
= 0;
9585 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9586 newmap
->num_members
= u
->new_raid_disks
;
9587 for (i
= 0; i
< delta_disks
; i
++) {
9588 set_imsm_ord_tbl_ent(newmap
,
9589 u
->old_raid_disks
+ i
,
9590 u
->old_raid_disks
+ i
);
9592 /* New map is correct, now need to save old map
9594 newmap
= get_imsm_map(newdev
, MAP_1
);
9595 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9597 imsm_set_array_size(newdev
, -1);
9600 sp
= (void **)id
->dev
;
9605 /* Clear migration record */
9606 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9609 *space_list
= tofree
;
9612 update_reshape_exit
:
9617 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9618 struct intel_super
*super
,
9621 struct imsm_dev
*dev
= NULL
;
9622 struct intel_dev
*dv
;
9623 struct imsm_dev
*dev_new
;
9624 struct imsm_map
*map
;
9628 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9629 if (dv
->index
== (unsigned int)u
->subarray
) {
9637 map
= get_imsm_map(dev
, MAP_0
);
9639 if (u
->direction
== R10_TO_R0
) {
9640 unsigned long long num_data_stripes
;
9642 /* Number of failed disks must be half of initial disk number */
9643 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9644 (map
->num_members
/ 2))
9647 /* iterate through devices to mark removed disks as spare */
9648 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9649 if (dm
->disk
.status
& FAILED_DISK
) {
9650 int idx
= dm
->index
;
9651 /* update indexes on the disk list */
9652 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9653 the index values will end up being correct.... NB */
9654 for (du
= super
->disks
; du
; du
= du
->next
)
9655 if (du
->index
> idx
)
9657 /* mark as spare disk */
9662 map
->num_members
= map
->num_members
/ 2;
9663 map
->map_state
= IMSM_T_STATE_NORMAL
;
9664 map
->num_domains
= 1;
9665 map
->raid_level
= 0;
9666 map
->failed_disk_num
= -1;
9667 num_data_stripes
= imsm_dev_size(dev
) / 2;
9668 num_data_stripes
/= map
->blocks_per_strip
;
9669 set_num_data_stripes(map
, num_data_stripes
);
9672 if (u
->direction
== R0_TO_R10
) {
9674 unsigned long long num_data_stripes
;
9676 /* update slots in current disk list */
9677 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9681 /* create new *missing* disks */
9682 for (i
= 0; i
< map
->num_members
; i
++) {
9683 space
= *space_list
;
9686 *space_list
= *space
;
9688 memcpy(du
, super
->disks
, sizeof(*du
));
9692 du
->index
= (i
* 2) + 1;
9693 sprintf((char *)du
->disk
.serial
,
9694 " MISSING_%d", du
->index
);
9695 sprintf((char *)du
->serial
,
9696 "MISSING_%d", du
->index
);
9697 du
->next
= super
->missing
;
9698 super
->missing
= du
;
9700 /* create new dev and map */
9701 space
= *space_list
;
9704 *space_list
= *space
;
9705 dev_new
= (void *)space
;
9706 memcpy(dev_new
, dev
, sizeof(*dev
));
9707 /* update new map */
9708 map
= get_imsm_map(dev_new
, MAP_0
);
9709 map
->num_members
= map
->num_members
* 2;
9710 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9711 map
->num_domains
= 2;
9712 map
->raid_level
= 1;
9713 num_data_stripes
= imsm_dev_size(dev
) / 2;
9714 num_data_stripes
/= map
->blocks_per_strip
;
9715 num_data_stripes
/= map
->num_domains
;
9716 set_num_data_stripes(map
, num_data_stripes
);
9718 /* replace dev<->dev_new */
9721 /* update disk order table */
9722 for (du
= super
->disks
; du
; du
= du
->next
)
9724 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9725 for (du
= super
->missing
; du
; du
= du
->next
)
9726 if (du
->index
>= 0) {
9727 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9728 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9734 static void imsm_process_update(struct supertype
*st
,
9735 struct metadata_update
*update
)
9738 * crack open the metadata_update envelope to find the update record
9739 * update can be one of:
9740 * update_reshape_container_disks - all the arrays in the container
9741 * are being reshaped to have more devices. We need to mark
9742 * the arrays for general migration and convert selected spares
9743 * into active devices.
9744 * update_activate_spare - a spare device has replaced a failed
9745 * device in an array, update the disk_ord_tbl. If this disk is
9746 * present in all member arrays then also clear the SPARE_DISK
9748 * update_create_array
9750 * update_rename_array
9751 * update_add_remove_disk
9753 struct intel_super
*super
= st
->sb
;
9754 struct imsm_super
*mpb
;
9755 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9757 /* update requires a larger buf but the allocation failed */
9758 if (super
->next_len
&& !super
->next_buf
) {
9759 super
->next_len
= 0;
9763 if (super
->next_buf
) {
9764 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9766 super
->len
= super
->next_len
;
9767 super
->buf
= super
->next_buf
;
9769 super
->next_len
= 0;
9770 super
->next_buf
= NULL
;
9773 mpb
= super
->anchor
;
9776 case update_general_migration_checkpoint
: {
9777 struct intel_dev
*id
;
9778 struct imsm_update_general_migration_checkpoint
*u
=
9779 (void *)update
->buf
;
9781 dprintf("called for update_general_migration_checkpoint\n");
9783 /* find device under general migration */
9784 for (id
= super
->devlist
; id
; id
= id
->next
) {
9785 if (is_gen_migration(id
->dev
)) {
9786 id
->dev
->vol
.curr_migr_unit
=
9787 __cpu_to_le32(u
->curr_migr_unit
);
9788 super
->updates_pending
++;
9793 case update_takeover
: {
9794 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9795 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9796 imsm_update_version_info(super
);
9797 super
->updates_pending
++;
9802 case update_reshape_container_disks
: {
9803 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9804 if (apply_reshape_container_disks_update(
9805 u
, super
, &update
->space_list
))
9806 super
->updates_pending
++;
9809 case update_reshape_migration
: {
9810 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9811 if (apply_reshape_migration_update(
9812 u
, super
, &update
->space_list
))
9813 super
->updates_pending
++;
9816 case update_size_change
: {
9817 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9818 if (apply_size_change_update(u
, super
))
9819 super
->updates_pending
++;
9822 case update_activate_spare
: {
9823 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9824 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9825 super
->updates_pending
++;
9828 case update_create_array
: {
9829 /* someone wants to create a new array, we need to be aware of
9830 * a few races/collisions:
9831 * 1/ 'Create' called by two separate instances of mdadm
9832 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9833 * devices that have since been assimilated via
9835 * In the event this update can not be carried out mdadm will
9836 * (FIX ME) notice that its update did not take hold.
9838 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9839 struct intel_dev
*dv
;
9840 struct imsm_dev
*dev
;
9841 struct imsm_map
*map
, *new_map
;
9842 unsigned long long start
, end
;
9843 unsigned long long new_start
, new_end
;
9845 struct disk_info
*inf
;
9848 /* handle racing creates: first come first serve */
9849 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9850 dprintf("subarray %d already defined\n", u
->dev_idx
);
9854 /* check update is next in sequence */
9855 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9856 dprintf("can not create array %d expected index %d\n",
9857 u
->dev_idx
, mpb
->num_raid_devs
);
9861 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9862 new_start
= pba_of_lba0(new_map
);
9863 new_end
= new_start
+ per_dev_array_size(new_map
);
9864 inf
= get_disk_info(u
);
9866 /* handle activate_spare versus create race:
9867 * check to make sure that overlapping arrays do not include
9870 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9871 dev
= get_imsm_dev(super
, i
);
9872 map
= get_imsm_map(dev
, MAP_0
);
9873 start
= pba_of_lba0(map
);
9874 end
= start
+ per_dev_array_size(map
);
9875 if ((new_start
>= start
&& new_start
<= end
) ||
9876 (start
>= new_start
&& start
<= new_end
))
9881 if (disks_overlap(super
, i
, u
)) {
9882 dprintf("arrays overlap\n");
9887 /* check that prepare update was successful */
9888 if (!update
->space
) {
9889 dprintf("prepare update failed\n");
9893 /* check that all disks are still active before committing
9894 * changes. FIXME: could we instead handle this by creating a
9895 * degraded array? That's probably not what the user expects,
9896 * so better to drop this update on the floor.
9898 for (i
= 0; i
< new_map
->num_members
; i
++) {
9899 dl
= serial_to_dl(inf
[i
].serial
, super
);
9901 dprintf("disk disappeared\n");
9906 super
->updates_pending
++;
9908 /* convert spares to members and fixup ord_tbl */
9909 for (i
= 0; i
< new_map
->num_members
; i
++) {
9910 dl
= serial_to_dl(inf
[i
].serial
, super
);
9911 if (dl
->index
== -1) {
9912 dl
->index
= mpb
->num_disks
;
9914 dl
->disk
.status
|= CONFIGURED_DISK
;
9915 dl
->disk
.status
&= ~SPARE_DISK
;
9917 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9922 update
->space
= NULL
;
9923 imsm_copy_dev(dev
, &u
->dev
);
9924 dv
->index
= u
->dev_idx
;
9925 dv
->next
= super
->devlist
;
9926 super
->devlist
= dv
;
9927 mpb
->num_raid_devs
++;
9929 imsm_update_version_info(super
);
9932 /* mdmon knows how to release update->space, but not
9933 * ((struct intel_dev *) update->space)->dev
9935 if (update
->space
) {
9941 case update_kill_array
: {
9942 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9943 int victim
= u
->dev_idx
;
9944 struct active_array
*a
;
9945 struct intel_dev
**dp
;
9946 struct imsm_dev
*dev
;
9948 /* sanity check that we are not affecting the uuid of
9949 * active arrays, or deleting an active array
9951 * FIXME when immutable ids are available, but note that
9952 * we'll also need to fixup the invalidated/active
9953 * subarray indexes in mdstat
9955 for (a
= st
->arrays
; a
; a
= a
->next
)
9956 if (a
->info
.container_member
>= victim
)
9958 /* by definition if mdmon is running at least one array
9959 * is active in the container, so checking
9960 * mpb->num_raid_devs is just extra paranoia
9962 dev
= get_imsm_dev(super
, victim
);
9963 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9964 dprintf("failed to delete subarray-%d\n", victim
);
9968 for (dp
= &super
->devlist
; *dp
;)
9969 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9972 if ((*dp
)->index
> (unsigned)victim
)
9976 mpb
->num_raid_devs
--;
9977 super
->updates_pending
++;
9980 case update_rename_array
: {
9981 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9982 char name
[MAX_RAID_SERIAL_LEN
+1];
9983 int target
= u
->dev_idx
;
9984 struct active_array
*a
;
9985 struct imsm_dev
*dev
;
9987 /* sanity check that we are not affecting the uuid of
9990 memset(name
, 0, sizeof(name
));
9991 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9992 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9993 for (a
= st
->arrays
; a
; a
= a
->next
)
9994 if (a
->info
.container_member
== target
)
9996 dev
= get_imsm_dev(super
, u
->dev_idx
);
9997 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9998 dprintf("failed to rename subarray-%d\n", target
);
10002 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
10003 super
->updates_pending
++;
10006 case update_add_remove_disk
: {
10007 /* we may be able to repair some arrays if disks are
10008 * being added, check the status of add_remove_disk
10009 * if discs has been added.
10011 if (add_remove_disk_update(super
)) {
10012 struct active_array
*a
;
10014 super
->updates_pending
++;
10015 for (a
= st
->arrays
; a
; a
= a
->next
)
10016 a
->check_degraded
= 1;
10020 case update_prealloc_badblocks_mem
:
10022 case update_rwh_policy
: {
10023 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
10024 int target
= u
->dev_idx
;
10025 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
10027 dprintf("could not find subarray-%d\n", target
);
10031 if (dev
->rwh_policy
!= u
->new_policy
) {
10032 dev
->rwh_policy
= u
->new_policy
;
10033 super
->updates_pending
++;
10038 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10042 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10044 static int imsm_prepare_update(struct supertype
*st
,
10045 struct metadata_update
*update
)
10048 * Allocate space to hold new disk entries, raid-device entries or a new
10049 * mpb if necessary. The manager synchronously waits for updates to
10050 * complete in the monitor, so new mpb buffers allocated here can be
10051 * integrated by the monitor thread without worrying about live pointers
10052 * in the manager thread.
10054 enum imsm_update_type type
;
10055 struct intel_super
*super
= st
->sb
;
10056 unsigned int sector_size
= super
->sector_size
;
10057 struct imsm_super
*mpb
= super
->anchor
;
10061 if (update
->len
< (int)sizeof(type
))
10064 type
= *(enum imsm_update_type
*) update
->buf
;
10067 case update_general_migration_checkpoint
:
10068 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10070 dprintf("called for update_general_migration_checkpoint\n");
10072 case update_takeover
: {
10073 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10074 if (update
->len
< (int)sizeof(*u
))
10076 if (u
->direction
== R0_TO_R10
) {
10077 void **tail
= (void **)&update
->space_list
;
10078 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10079 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10080 int num_members
= map
->num_members
;
10083 /* allocate memory for added disks */
10084 for (i
= 0; i
< num_members
; i
++) {
10085 size
= sizeof(struct dl
);
10086 space
= xmalloc(size
);
10091 /* allocate memory for new device */
10092 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10093 (num_members
* sizeof(__u32
));
10094 space
= xmalloc(size
);
10098 len
= disks_to_mpb_size(num_members
* 2);
10103 case update_reshape_container_disks
: {
10104 /* Every raid device in the container is about to
10105 * gain some more devices, and we will enter a
10107 * So each 'imsm_map' will be bigger, and the imsm_vol
10108 * will now hold 2 of them.
10109 * Thus we need new 'struct imsm_dev' allocations sized
10110 * as sizeof_imsm_dev but with more devices in both maps.
10112 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10113 struct intel_dev
*dl
;
10114 void **space_tail
= (void**)&update
->space_list
;
10116 if (update
->len
< (int)sizeof(*u
))
10119 dprintf("for update_reshape\n");
10121 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10122 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10124 if (u
->new_raid_disks
> u
->old_raid_disks
)
10125 size
+= sizeof(__u32
)*2*
10126 (u
->new_raid_disks
- u
->old_raid_disks
);
10130 *space_tail
= NULL
;
10133 len
= disks_to_mpb_size(u
->new_raid_disks
);
10134 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10137 case update_reshape_migration
: {
10138 /* for migration level 0->5 we need to add disks
10139 * so the same as for container operation we will copy
10140 * device to the bigger location.
10141 * in memory prepared device and new disk area are prepared
10142 * for usage in process update
10144 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10145 struct intel_dev
*id
;
10146 void **space_tail
= (void **)&update
->space_list
;
10149 int current_level
= -1;
10151 if (update
->len
< (int)sizeof(*u
))
10154 dprintf("for update_reshape\n");
10156 /* add space for bigger array in update
10158 for (id
= super
->devlist
; id
; id
= id
->next
) {
10159 if (id
->index
== (unsigned)u
->subdev
) {
10160 size
= sizeof_imsm_dev(id
->dev
, 1);
10161 if (u
->new_raid_disks
> u
->old_raid_disks
)
10162 size
+= sizeof(__u32
)*2*
10163 (u
->new_raid_disks
- u
->old_raid_disks
);
10167 *space_tail
= NULL
;
10171 if (update
->space_list
== NULL
)
10174 /* add space for disk in update
10176 size
= sizeof(struct dl
);
10180 *space_tail
= NULL
;
10182 /* add spare device to update
10184 for (id
= super
->devlist
; id
; id
= id
->next
)
10185 if (id
->index
== (unsigned)u
->subdev
) {
10186 struct imsm_dev
*dev
;
10187 struct imsm_map
*map
;
10189 dev
= get_imsm_dev(super
, u
->subdev
);
10190 map
= get_imsm_map(dev
, MAP_0
);
10191 current_level
= map
->raid_level
;
10194 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10195 struct mdinfo
*spares
;
10197 spares
= get_spares_for_grow(st
);
10200 struct mdinfo
*dev
;
10202 dev
= spares
->devs
;
10205 makedev(dev
->disk
.major
,
10207 dl
= get_disk_super(super
,
10210 dl
->index
= u
->old_raid_disks
;
10213 sysfs_free(spares
);
10216 len
= disks_to_mpb_size(u
->new_raid_disks
);
10217 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10220 case update_size_change
: {
10221 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10225 case update_activate_spare
: {
10226 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10230 case update_create_array
: {
10231 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10232 struct intel_dev
*dv
;
10233 struct imsm_dev
*dev
= &u
->dev
;
10234 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10236 struct disk_info
*inf
;
10240 if (update
->len
< (int)sizeof(*u
))
10243 inf
= get_disk_info(u
);
10244 len
= sizeof_imsm_dev(dev
, 1);
10245 /* allocate a new super->devlist entry */
10246 dv
= xmalloc(sizeof(*dv
));
10247 dv
->dev
= xmalloc(len
);
10248 update
->space
= dv
;
10250 /* count how many spares will be converted to members */
10251 for (i
= 0; i
< map
->num_members
; i
++) {
10252 dl
= serial_to_dl(inf
[i
].serial
, super
);
10254 /* hmm maybe it failed?, nothing we can do about
10259 if (count_memberships(dl
, super
) == 0)
10262 len
+= activate
* sizeof(struct imsm_disk
);
10265 case update_kill_array
: {
10266 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10270 case update_rename_array
: {
10271 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10275 case update_add_remove_disk
:
10276 /* no update->len needed */
10278 case update_prealloc_badblocks_mem
:
10279 super
->extra_space
+= sizeof(struct bbm_log
) -
10280 get_imsm_bbm_log_size(super
->bbm_log
);
10282 case update_rwh_policy
: {
10283 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10291 /* check if we need a larger metadata buffer */
10292 if (super
->next_buf
)
10293 buf_len
= super
->next_len
;
10295 buf_len
= super
->len
;
10297 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10298 /* ok we need a larger buf than what is currently allocated
10299 * if this allocation fails process_update will notice that
10300 * ->next_len is set and ->next_buf is NULL
10302 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10303 super
->extra_space
+ len
, sector_size
);
10304 if (super
->next_buf
)
10305 free(super
->next_buf
);
10307 super
->next_len
= buf_len
;
10308 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10309 memset(super
->next_buf
, 0, buf_len
);
10311 super
->next_buf
= NULL
;
10316 /* must be called while manager is quiesced */
10317 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10319 struct imsm_super
*mpb
= super
->anchor
;
10321 struct imsm_dev
*dev
;
10322 struct imsm_map
*map
;
10323 unsigned int i
, j
, num_members
;
10324 __u32 ord
, ord_map0
;
10325 struct bbm_log
*log
= super
->bbm_log
;
10327 dprintf("deleting device[%d] from imsm_super\n", index
);
10329 /* shift all indexes down one */
10330 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10331 if (iter
->index
> (int)index
)
10333 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10334 if (iter
->index
> (int)index
)
10337 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10338 dev
= get_imsm_dev(super
, i
);
10339 map
= get_imsm_map(dev
, MAP_0
);
10340 num_members
= map
->num_members
;
10341 for (j
= 0; j
< num_members
; j
++) {
10342 /* update ord entries being careful not to propagate
10343 * ord-flags to the first map
10345 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10346 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10348 if (ord_to_idx(ord
) <= index
)
10351 map
= get_imsm_map(dev
, MAP_0
);
10352 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10353 map
= get_imsm_map(dev
, MAP_1
);
10355 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10359 for (i
= 0; i
< log
->entry_count
; i
++) {
10360 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10362 if (entry
->disk_ordinal
<= index
)
10364 entry
->disk_ordinal
--;
10368 super
->updates_pending
++;
10370 struct dl
*dl
= *dlp
;
10372 *dlp
= (*dlp
)->next
;
10373 __free_imsm_disk(dl
);
10377 static void close_targets(int *targets
, int new_disks
)
10384 for (i
= 0; i
< new_disks
; i
++) {
10385 if (targets
[i
] >= 0) {
10392 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10393 struct intel_super
*super
,
10394 struct imsm_dev
*dev
)
10400 struct imsm_map
*map
;
10403 ret_val
= raid_disks
/2;
10404 /* check map if all disks pairs not failed
10407 map
= get_imsm_map(dev
, MAP_0
);
10408 for (i
= 0; i
< ret_val
; i
++) {
10409 int degradation
= 0;
10410 if (get_imsm_disk(super
, i
) == NULL
)
10412 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10414 if (degradation
== 2)
10417 map
= get_imsm_map(dev
, MAP_1
);
10418 /* if there is no second map
10419 * result can be returned
10423 /* check degradation in second map
10425 for (i
= 0; i
< ret_val
; i
++) {
10426 int degradation
= 0;
10427 if (get_imsm_disk(super
, i
) == NULL
)
10429 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10431 if (degradation
== 2)
10445 /*******************************************************************************
10446 * Function: open_backup_targets
10447 * Description: Function opens file descriptors for all devices given in
10450 * info : general array info
10451 * raid_disks : number of disks
10452 * raid_fds : table of device's file descriptors
10453 * super : intel super for raid10 degradation check
10454 * dev : intel device for raid10 degradation check
10458 ******************************************************************************/
10459 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10460 struct intel_super
*super
, struct imsm_dev
*dev
)
10466 for (i
= 0; i
< raid_disks
; i
++)
10469 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10472 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10473 dprintf("disk is faulty!!\n");
10477 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10480 dn
= map_dev(sd
->disk
.major
,
10481 sd
->disk
.minor
, 1);
10482 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10483 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10484 pr_err("cannot open component\n");
10489 /* check if maximum array degradation level is not exceeded
10491 if ((raid_disks
- opened
) >
10492 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10494 pr_err("Not enough disks can be opened.\n");
10495 close_targets(raid_fds
, raid_disks
);
10501 /*******************************************************************************
10502 * Function: validate_container_imsm
10503 * Description: This routine validates container after assemble,
10504 * eg. if devices in container are under the same controller.
10507 * info : linked list with info about devices used in array
10511 ******************************************************************************/
10512 int validate_container_imsm(struct mdinfo
*info
)
10514 if (check_env("IMSM_NO_PLATFORM"))
10517 struct sys_dev
*idev
;
10518 struct sys_dev
*hba
= NULL
;
10519 struct sys_dev
*intel_devices
= find_intel_devices();
10520 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10521 info
->disk
.minor
));
10523 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10524 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10533 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10534 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10538 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10539 struct mdinfo
*dev
;
10541 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10542 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10544 struct sys_dev
*hba2
= NULL
;
10545 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10546 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10554 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10555 get_orom_by_device_id(hba2
->dev_id
);
10557 if (hba2
&& hba
->type
!= hba2
->type
) {
10558 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10559 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10563 if (orom
!= orom2
) {
10564 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10565 " This operation is not supported and can lead to data loss.\n");
10570 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10571 " This operation is not supported and can lead to data loss.\n");
10579 /*******************************************************************************
10580 * Function: imsm_record_badblock
10581 * Description: This routine stores new bad block record in BBM log
10584 * a : array containing a bad block
10585 * slot : disk number containing a bad block
10586 * sector : bad block sector
10587 * length : bad block sectors range
10591 ******************************************************************************/
10592 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10593 unsigned long long sector
, int length
)
10595 struct intel_super
*super
= a
->container
->sb
;
10599 ord
= imsm_disk_slot_to_ord(a
, slot
);
10603 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10606 super
->updates_pending
++;
10610 /*******************************************************************************
10611 * Function: imsm_clear_badblock
10612 * Description: This routine clears bad block record from BBM log
10615 * a : array containing a bad block
10616 * slot : disk number containing a bad block
10617 * sector : bad block sector
10618 * length : bad block sectors range
10622 ******************************************************************************/
10623 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10624 unsigned long long sector
, int length
)
10626 struct intel_super
*super
= a
->container
->sb
;
10630 ord
= imsm_disk_slot_to_ord(a
, slot
);
10634 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10636 super
->updates_pending
++;
10640 /*******************************************************************************
10641 * Function: imsm_get_badblocks
10642 * Description: This routine get list of bad blocks for an array
10646 * slot : disk number
10648 * bb : structure containing bad blocks
10650 ******************************************************************************/
10651 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10653 int inst
= a
->info
.container_member
;
10654 struct intel_super
*super
= a
->container
->sb
;
10655 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10656 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10659 ord
= imsm_disk_slot_to_ord(a
, slot
);
10663 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10664 per_dev_array_size(map
), &super
->bb
);
10668 /*******************************************************************************
10669 * Function: examine_badblocks_imsm
10670 * Description: Prints list of bad blocks on a disk to the standard output
10673 * st : metadata handler
10674 * fd : open file descriptor for device
10675 * devname : device name
10679 ******************************************************************************/
10680 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10682 struct intel_super
*super
= st
->sb
;
10683 struct bbm_log
*log
= super
->bbm_log
;
10684 struct dl
*d
= NULL
;
10687 for (d
= super
->disks
; d
; d
= d
->next
) {
10688 if (strcmp(d
->devname
, devname
) == 0)
10692 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10693 pr_err("%s doesn't appear to be part of a raid array\n",
10700 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10702 for (i
= 0; i
< log
->entry_count
; i
++) {
10703 if (entry
[i
].disk_ordinal
== d
->index
) {
10704 unsigned long long sector
= __le48_to_cpu(
10705 &entry
[i
].defective_block_start
);
10706 int cnt
= entry
[i
].marked_count
+ 1;
10709 printf("Bad-blocks on %s:\n", devname
);
10713 printf("%20llu for %d sectors\n", sector
, cnt
);
10719 printf("No bad-blocks list configured on %s\n", devname
);
10723 /*******************************************************************************
10724 * Function: init_migr_record_imsm
10725 * Description: Function inits imsm migration record
10727 * super : imsm internal array info
10728 * dev : device under migration
10729 * info : general array info to find the smallest device
10732 ******************************************************************************/
10733 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10734 struct mdinfo
*info
)
10736 struct intel_super
*super
= st
->sb
;
10737 struct migr_record
*migr_rec
= super
->migr_rec
;
10738 int new_data_disks
;
10739 unsigned long long dsize
, dev_sectors
;
10740 long long unsigned min_dev_sectors
= -1LLU;
10744 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10745 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10746 unsigned long long num_migr_units
;
10747 unsigned long long array_blocks
;
10749 memset(migr_rec
, 0, sizeof(struct migr_record
));
10750 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10752 /* only ascending reshape supported now */
10753 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10755 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10756 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10757 migr_rec
->dest_depth_per_unit
*=
10758 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10759 new_data_disks
= imsm_num_data_members(map_dest
);
10760 migr_rec
->blocks_per_unit
=
10761 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10762 migr_rec
->dest_depth_per_unit
=
10763 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10764 array_blocks
= info
->component_size
* new_data_disks
;
10766 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10768 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10770 set_num_migr_units(migr_rec
, num_migr_units
);
10772 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10773 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10775 /* Find the smallest dev */
10776 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10777 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10778 fd
= dev_open(nm
, O_RDONLY
);
10781 get_dev_size(fd
, NULL
, &dsize
);
10782 dev_sectors
= dsize
/ 512;
10783 if (dev_sectors
< min_dev_sectors
)
10784 min_dev_sectors
= dev_sectors
;
10787 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
10788 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10790 write_imsm_migr_rec(st
);
10795 /*******************************************************************************
10796 * Function: save_backup_imsm
10797 * Description: Function saves critical data stripes to Migration Copy Area
10798 * and updates the current migration unit status.
10799 * Use restore_stripes() to form a destination stripe,
10800 * and to write it to the Copy Area.
10802 * st : supertype information
10803 * dev : imsm device that backup is saved for
10804 * info : general array info
10805 * buf : input buffer
10806 * length : length of data to backup (blocks_per_unit)
10810 ******************************************************************************/
10811 int save_backup_imsm(struct supertype
*st
,
10812 struct imsm_dev
*dev
,
10813 struct mdinfo
*info
,
10818 struct intel_super
*super
= st
->sb
;
10819 unsigned long long *target_offsets
;
10822 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10823 int new_disks
= map_dest
->num_members
;
10824 int dest_layout
= 0;
10826 unsigned long long start
;
10827 int data_disks
= imsm_num_data_members(map_dest
);
10829 targets
= xmalloc(new_disks
* sizeof(int));
10831 for (i
= 0; i
< new_disks
; i
++)
10834 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10836 start
= info
->reshape_progress
* 512;
10837 for (i
= 0; i
< new_disks
; i
++) {
10838 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
10839 /* move back copy area adderss, it will be moved forward
10840 * in restore_stripes() using start input variable
10842 target_offsets
[i
] -= start
/data_disks
;
10845 if (open_backup_targets(info
, new_disks
, targets
,
10849 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10850 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10852 if (restore_stripes(targets
, /* list of dest devices */
10853 target_offsets
, /* migration record offsets */
10856 map_dest
->raid_level
,
10858 -1, /* source backup file descriptor */
10859 0, /* input buf offset
10860 * always 0 buf is already offseted */
10864 pr_err("Error restoring stripes\n");
10872 close_targets(targets
, new_disks
);
10875 free(target_offsets
);
10880 /*******************************************************************************
10881 * Function: save_checkpoint_imsm
10882 * Description: Function called for current unit status update
10883 * in the migration record. It writes it to disk.
10885 * super : imsm internal array info
10886 * info : general array info
10890 * 2: failure, means no valid migration record
10891 * / no general migration in progress /
10892 ******************************************************************************/
10893 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10895 struct intel_super
*super
= st
->sb
;
10896 unsigned long long blocks_per_unit
;
10897 unsigned long long curr_migr_unit
;
10899 if (load_imsm_migr_rec(super
, info
) != 0) {
10900 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10904 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10905 if (blocks_per_unit
== 0) {
10906 dprintf("imsm: no migration in progress.\n");
10909 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10910 /* check if array is alligned to copy area
10911 * if it is not alligned, add one to current migration unit value
10912 * this can happend on array reshape finish only
10914 if (info
->reshape_progress
% blocks_per_unit
)
10917 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
10918 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10919 set_migr_dest_1st_member_lba(super
->migr_rec
,
10920 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
10922 if (write_imsm_migr_rec(st
) < 0) {
10923 dprintf("imsm: Cannot write migration record outside backup area\n");
10930 /*******************************************************************************
10931 * Function: recover_backup_imsm
10932 * Description: Function recovers critical data from the Migration Copy Area
10933 * while assembling an array.
10935 * super : imsm internal array info
10936 * info : general array info
10938 * 0 : success (or there is no data to recover)
10940 ******************************************************************************/
10941 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10943 struct intel_super
*super
= st
->sb
;
10944 struct migr_record
*migr_rec
= super
->migr_rec
;
10945 struct imsm_map
*map_dest
;
10946 struct intel_dev
*id
= NULL
;
10947 unsigned long long read_offset
;
10948 unsigned long long write_offset
;
10950 int *targets
= NULL
;
10951 int new_disks
, i
, err
;
10954 unsigned int sector_size
= super
->sector_size
;
10955 unsigned long curr_migr_unit
= current_migr_unit(migr_rec
);
10956 unsigned long num_migr_units
= get_num_migr_units(migr_rec
);
10958 int skipped_disks
= 0;
10960 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10964 /* recover data only during assemblation */
10965 if (strncmp(buffer
, "inactive", 8) != 0)
10967 /* no data to recover */
10968 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10970 if (curr_migr_unit
>= num_migr_units
)
10973 /* find device during reshape */
10974 for (id
= super
->devlist
; id
; id
= id
->next
)
10975 if (is_gen_migration(id
->dev
))
10980 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10981 new_disks
= map_dest
->num_members
;
10983 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
10985 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
10986 pba_of_lba0(map_dest
)) * 512;
10988 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10989 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10991 targets
= xcalloc(new_disks
, sizeof(int));
10993 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10994 pr_err("Cannot open some devices belonging to array.\n");
10998 for (i
= 0; i
< new_disks
; i
++) {
10999 if (targets
[i
] < 0) {
11003 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
11004 pr_err("Cannot seek to block: %s\n",
11009 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
11010 pr_err("Cannot read copy area block: %s\n",
11015 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
11016 pr_err("Cannot seek to block: %s\n",
11021 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
11022 pr_err("Cannot restore block: %s\n",
11029 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11033 pr_err("Cannot restore data from backup. Too many failed disks\n");
11037 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11038 /* ignore error == 2, this can mean end of reshape here
11040 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11046 for (i
= 0; i
< new_disks
; i
++)
11055 static char disk_by_path
[] = "/dev/disk/by-path/";
11057 static const char *imsm_get_disk_controller_domain(const char *path
)
11059 char disk_path
[PATH_MAX
];
11063 strcpy(disk_path
, disk_by_path
);
11064 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
11065 if (stat(disk_path
, &st
) == 0) {
11066 struct sys_dev
* hba
;
11069 path
= devt_to_devpath(st
.st_rdev
);
11072 hba
= find_disk_attached_hba(-1, path
);
11073 if (hba
&& hba
->type
== SYS_DEV_SAS
)
11075 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
11077 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
11079 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
11083 dprintf("path: %s hba: %s attached: %s\n",
11084 path
, (hba
) ? hba
->path
: "NULL", drv
);
11090 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11092 static char devnm
[32];
11093 char subdev_name
[20];
11094 struct mdstat_ent
*mdstat
;
11096 sprintf(subdev_name
, "%d", subdev
);
11097 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11101 strcpy(devnm
, mdstat
->devnm
);
11102 free_mdstat(mdstat
);
11106 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11107 struct geo_params
*geo
,
11108 int *old_raid_disks
,
11111 /* currently we only support increasing the number of devices
11112 * for a container. This increases the number of device for each
11113 * member array. They must all be RAID0 or RAID5.
11116 struct mdinfo
*info
, *member
;
11117 int devices_that_can_grow
= 0;
11119 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11121 if (geo
->size
> 0 ||
11122 geo
->level
!= UnSet
||
11123 geo
->layout
!= UnSet
||
11124 geo
->chunksize
!= 0 ||
11125 geo
->raid_disks
== UnSet
) {
11126 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11130 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11131 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11135 info
= container_content_imsm(st
, NULL
);
11136 for (member
= info
; member
; member
= member
->next
) {
11139 dprintf("imsm: checking device_num: %i\n",
11140 member
->container_member
);
11142 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11143 /* we work on container for Online Capacity Expansion
11144 * only so raid_disks has to grow
11146 dprintf("imsm: for container operation raid disks increase is required\n");
11150 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11151 /* we cannot use this container with other raid level
11153 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11154 info
->array
.level
);
11157 /* check for platform support
11158 * for this raid level configuration
11160 struct intel_super
*super
= st
->sb
;
11161 if (!is_raid_level_supported(super
->orom
,
11162 member
->array
.level
,
11163 geo
->raid_disks
)) {
11164 dprintf("platform does not support raid%d with %d disk%s\n",
11167 geo
->raid_disks
> 1 ? "s" : "");
11170 /* check if component size is aligned to chunk size
11172 if (info
->component_size
%
11173 (info
->array
.chunk_size
/512)) {
11174 dprintf("Component size is not aligned to chunk size\n");
11179 if (*old_raid_disks
&&
11180 info
->array
.raid_disks
!= *old_raid_disks
)
11182 *old_raid_disks
= info
->array
.raid_disks
;
11184 /* All raid5 and raid0 volumes in container
11185 * have to be ready for Online Capacity Expansion
11186 * so they need to be assembled. We have already
11187 * checked that no recovery etc is happening.
11189 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11190 st
->container_devnm
);
11191 if (result
== NULL
) {
11192 dprintf("imsm: cannot find array\n");
11195 devices_that_can_grow
++;
11198 if (!member
&& devices_that_can_grow
)
11202 dprintf("Container operation allowed\n");
11204 dprintf("Error: %i\n", ret_val
);
11209 /* Function: get_spares_for_grow
11210 * Description: Allocates memory and creates list of spare devices
11211 * avaliable in container. Checks if spare drive size is acceptable.
11212 * Parameters: Pointer to the supertype structure
11213 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11216 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11218 struct spare_criteria sc
;
11220 get_spare_criteria_imsm(st
, &sc
);
11221 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11224 /******************************************************************************
11225 * function: imsm_create_metadata_update_for_reshape
11226 * Function creates update for whole IMSM container.
11228 ******************************************************************************/
11229 static int imsm_create_metadata_update_for_reshape(
11230 struct supertype
*st
,
11231 struct geo_params
*geo
,
11232 int old_raid_disks
,
11233 struct imsm_update_reshape
**updatep
)
11235 struct intel_super
*super
= st
->sb
;
11236 struct imsm_super
*mpb
= super
->anchor
;
11237 int update_memory_size
;
11238 struct imsm_update_reshape
*u
;
11239 struct mdinfo
*spares
;
11242 struct mdinfo
*dev
;
11244 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11246 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11248 /* size of all update data without anchor */
11249 update_memory_size
= sizeof(struct imsm_update_reshape
);
11251 /* now add space for spare disks that we need to add. */
11252 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11254 u
= xcalloc(1, update_memory_size
);
11255 u
->type
= update_reshape_container_disks
;
11256 u
->old_raid_disks
= old_raid_disks
;
11257 u
->new_raid_disks
= geo
->raid_disks
;
11259 /* now get spare disks list
11261 spares
= get_spares_for_grow(st
);
11263 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11264 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11269 /* we have got spares
11270 * update disk list in imsm_disk list table in anchor
11272 dprintf("imsm: %i spares are available.\n\n",
11273 spares
->array
.spare_disks
);
11275 dev
= spares
->devs
;
11276 for (i
= 0; i
< delta_disks
; i
++) {
11281 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11283 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11284 dl
->index
= mpb
->num_disks
;
11292 sysfs_free(spares
);
11294 dprintf("imsm: reshape update preparation :");
11295 if (i
== delta_disks
) {
11296 dprintf_cont(" OK\n");
11298 return update_memory_size
;
11301 dprintf_cont(" Error\n");
11306 /******************************************************************************
11307 * function: imsm_create_metadata_update_for_size_change()
11308 * Creates update for IMSM array for array size change.
11310 ******************************************************************************/
11311 static int imsm_create_metadata_update_for_size_change(
11312 struct supertype
*st
,
11313 struct geo_params
*geo
,
11314 struct imsm_update_size_change
**updatep
)
11316 struct intel_super
*super
= st
->sb
;
11317 int update_memory_size
;
11318 struct imsm_update_size_change
*u
;
11320 dprintf("(enter) New size = %llu\n", geo
->size
);
11322 /* size of all update data without anchor */
11323 update_memory_size
= sizeof(struct imsm_update_size_change
);
11325 u
= xcalloc(1, update_memory_size
);
11326 u
->type
= update_size_change
;
11327 u
->subdev
= super
->current_vol
;
11328 u
->new_size
= geo
->size
;
11330 dprintf("imsm: reshape update preparation : OK\n");
11333 return update_memory_size
;
11336 /******************************************************************************
11337 * function: imsm_create_metadata_update_for_migration()
11338 * Creates update for IMSM array.
11340 ******************************************************************************/
11341 static int imsm_create_metadata_update_for_migration(
11342 struct supertype
*st
,
11343 struct geo_params
*geo
,
11344 struct imsm_update_reshape_migration
**updatep
)
11346 struct intel_super
*super
= st
->sb
;
11347 int update_memory_size
;
11348 struct imsm_update_reshape_migration
*u
;
11349 struct imsm_dev
*dev
;
11350 int previous_level
= -1;
11352 dprintf("(enter) New Level = %i\n", geo
->level
);
11354 /* size of all update data without anchor */
11355 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11357 u
= xcalloc(1, update_memory_size
);
11358 u
->type
= update_reshape_migration
;
11359 u
->subdev
= super
->current_vol
;
11360 u
->new_level
= geo
->level
;
11361 u
->new_layout
= geo
->layout
;
11362 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11363 u
->new_disks
[0] = -1;
11364 u
->new_chunksize
= -1;
11366 dev
= get_imsm_dev(super
, u
->subdev
);
11368 struct imsm_map
*map
;
11370 map
= get_imsm_map(dev
, MAP_0
);
11372 int current_chunk_size
=
11373 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11375 if (geo
->chunksize
!= current_chunk_size
) {
11376 u
->new_chunksize
= geo
->chunksize
/ 1024;
11377 dprintf("imsm: chunk size change from %i to %i\n",
11378 current_chunk_size
, u
->new_chunksize
);
11380 previous_level
= map
->raid_level
;
11383 if (geo
->level
== 5 && previous_level
== 0) {
11384 struct mdinfo
*spares
= NULL
;
11386 u
->new_raid_disks
++;
11387 spares
= get_spares_for_grow(st
);
11388 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11390 sysfs_free(spares
);
11391 update_memory_size
= 0;
11392 pr_err("cannot get spare device for requested migration\n");
11395 sysfs_free(spares
);
11397 dprintf("imsm: reshape update preparation : OK\n");
11400 return update_memory_size
;
11403 static void imsm_update_metadata_locally(struct supertype
*st
,
11404 void *buf
, int len
)
11406 struct metadata_update mu
;
11411 mu
.space_list
= NULL
;
11413 if (imsm_prepare_update(st
, &mu
))
11414 imsm_process_update(st
, &mu
);
11416 while (mu
.space_list
) {
11417 void **space
= mu
.space_list
;
11418 mu
.space_list
= *space
;
11423 /***************************************************************************
11424 * Function: imsm_analyze_change
11425 * Description: Function analyze change for single volume
11426 * and validate if transition is supported
11427 * Parameters: Geometry parameters, supertype structure,
11428 * metadata change direction (apply/rollback)
11429 * Returns: Operation type code on success, -1 if fail
11430 ****************************************************************************/
11431 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11432 struct geo_params
*geo
,
11435 struct mdinfo info
;
11437 int check_devs
= 0;
11439 /* number of added/removed disks in operation result */
11440 int devNumChange
= 0;
11441 /* imsm compatible layout value for array geometry verification */
11442 int imsm_layout
= -1;
11444 struct imsm_dev
*dev
;
11445 struct imsm_map
*map
;
11446 struct intel_super
*super
;
11447 unsigned long long current_size
;
11448 unsigned long long free_size
;
11449 unsigned long long max_size
;
11452 getinfo_super_imsm_volume(st
, &info
, NULL
);
11453 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11454 geo
->level
!= UnSet
) {
11455 switch (info
.array
.level
) {
11457 if (geo
->level
== 5) {
11458 change
= CH_MIGRATION
;
11459 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11460 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11462 goto analyse_change_exit
;
11464 imsm_layout
= geo
->layout
;
11466 devNumChange
= 1; /* parity disk added */
11467 } else if (geo
->level
== 10) {
11468 change
= CH_TAKEOVER
;
11470 devNumChange
= 2; /* two mirrors added */
11471 imsm_layout
= 0x102; /* imsm supported layout */
11476 if (geo
->level
== 0) {
11477 change
= CH_TAKEOVER
;
11479 devNumChange
= -(geo
->raid_disks
/2);
11480 imsm_layout
= 0; /* imsm raid0 layout */
11484 if (change
== -1) {
11485 pr_err("Error. Level Migration from %d to %d not supported!\n",
11486 info
.array
.level
, geo
->level
);
11487 goto analyse_change_exit
;
11490 geo
->level
= info
.array
.level
;
11492 if (geo
->layout
!= info
.array
.layout
&&
11493 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11494 change
= CH_MIGRATION
;
11495 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11496 geo
->layout
== 5) {
11497 /* reshape 5 -> 4 */
11498 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11499 geo
->layout
== 0) {
11500 /* reshape 4 -> 5 */
11504 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11505 info
.array
.layout
, geo
->layout
);
11507 goto analyse_change_exit
;
11510 geo
->layout
= info
.array
.layout
;
11511 if (imsm_layout
== -1)
11512 imsm_layout
= info
.array
.layout
;
11515 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11516 geo
->chunksize
!= info
.array
.chunk_size
) {
11517 if (info
.array
.level
== 10) {
11518 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11520 goto analyse_change_exit
;
11521 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11522 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11523 geo
->chunksize
/1024, info
.component_size
/2);
11525 goto analyse_change_exit
;
11527 change
= CH_MIGRATION
;
11529 geo
->chunksize
= info
.array
.chunk_size
;
11532 chunk
= geo
->chunksize
/ 1024;
11535 dev
= get_imsm_dev(super
, super
->current_vol
);
11536 map
= get_imsm_map(dev
, MAP_0
);
11537 data_disks
= imsm_num_data_members(map
);
11538 /* compute current size per disk member
11540 current_size
= info
.custom_array_size
/ data_disks
;
11542 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11543 /* align component size
11545 geo
->size
= imsm_component_size_alignment_check(
11546 get_imsm_raid_level(dev
->vol
.map
),
11547 chunk
* 1024, super
->sector_size
,
11549 if (geo
->size
== 0) {
11550 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11552 goto analyse_change_exit
;
11556 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11557 if (change
!= -1) {
11558 pr_err("Error. Size change should be the only one at a time.\n");
11560 goto analyse_change_exit
;
11562 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11563 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11564 super
->current_vol
, st
->devnm
);
11565 goto analyse_change_exit
;
11567 /* check the maximum available size
11569 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11570 0, chunk
, &free_size
);
11572 /* Cannot find maximum available space
11576 max_size
= free_size
+ current_size
;
11577 /* align component size
11579 max_size
= imsm_component_size_alignment_check(
11580 get_imsm_raid_level(dev
->vol
.map
),
11581 chunk
* 1024, super
->sector_size
,
11584 if (geo
->size
== MAX_SIZE
) {
11585 /* requested size change to the maximum available size
11587 if (max_size
== 0) {
11588 pr_err("Error. Cannot find maximum available space.\n");
11590 goto analyse_change_exit
;
11592 geo
->size
= max_size
;
11595 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11596 /* accept size for rollback only
11599 /* round size due to metadata compatibility
11601 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11602 << SECT_PER_MB_SHIFT
;
11603 dprintf("Prepare update for size change to %llu\n",
11605 if (current_size
>= geo
->size
) {
11606 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11607 current_size
, geo
->size
);
11608 goto analyse_change_exit
;
11610 if (max_size
&& geo
->size
> max_size
) {
11611 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11612 max_size
, geo
->size
);
11613 goto analyse_change_exit
;
11616 geo
->size
*= data_disks
;
11617 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11618 change
= CH_ARRAY_SIZE
;
11620 if (!validate_geometry_imsm(st
,
11623 geo
->raid_disks
+ devNumChange
,
11625 geo
->size
, INVALID_SECTORS
,
11626 0, 0, info
.consistency_policy
, 1))
11630 struct intel_super
*super
= st
->sb
;
11631 struct imsm_super
*mpb
= super
->anchor
;
11633 if (mpb
->num_raid_devs
> 1) {
11634 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11640 analyse_change_exit
:
11641 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11642 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11643 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11649 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11651 struct intel_super
*super
= st
->sb
;
11652 struct imsm_update_takeover
*u
;
11654 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11656 u
->type
= update_takeover
;
11657 u
->subarray
= super
->current_vol
;
11659 /* 10->0 transition */
11660 if (geo
->level
== 0)
11661 u
->direction
= R10_TO_R0
;
11663 /* 0->10 transition */
11664 if (geo
->level
== 10)
11665 u
->direction
= R0_TO_R10
;
11667 /* update metadata locally */
11668 imsm_update_metadata_locally(st
, u
,
11669 sizeof(struct imsm_update_takeover
));
11670 /* and possibly remotely */
11671 if (st
->update_tail
)
11672 append_metadata_update(st
, u
,
11673 sizeof(struct imsm_update_takeover
));
11680 /* Flush size update if size calculated by num_data_stripes is higher than
11681 * imsm_dev_size to eliminate differences during reshape.
11682 * Mdmon will recalculate them correctly.
11683 * If subarray index is not set then check whole container.
11685 * 0 - no error occurred
11686 * 1 - error detected
11688 static int imsm_fix_size_mismatch(struct supertype
*st
, int subarray_index
)
11690 struct intel_super
*super
= st
->sb
;
11691 int tmp
= super
->current_vol
;
11695 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
11696 if (subarray_index
>= 0 && i
!= subarray_index
)
11698 super
->current_vol
= i
;
11699 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
11700 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
11701 unsigned int disc_count
= imsm_num_data_members(map
);
11702 struct geo_params geo
;
11703 struct imsm_update_size_change
*update
;
11704 unsigned long long calc_size
= per_dev_array_size(map
) * disc_count
;
11705 unsigned long long d_size
= imsm_dev_size(dev
);
11708 if (calc_size
== d_size
|| dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
11711 /* There is a difference, verify that imsm_dev_size is
11712 * rounded correctly and push update.
11714 if (d_size
!= round_size_to_mb(d_size
, disc_count
)) {
11715 dprintf("imsm: Size of volume %d is not rounded correctly\n",
11719 memset(&geo
, 0, sizeof(struct geo_params
));
11721 u_size
= imsm_create_metadata_update_for_size_change(st
, &geo
,
11724 dprintf("imsm: Cannot prepare size change update\n");
11727 imsm_update_metadata_locally(st
, update
, u_size
);
11728 if (st
->update_tail
) {
11729 append_metadata_update(st
, update
, u_size
);
11730 flush_metadata_updates(st
);
11731 st
->update_tail
= &st
->updates
;
11733 imsm_sync_metadata(st
);
11738 super
->current_vol
= tmp
;
11742 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11744 int layout
, int chunksize
, int raid_disks
,
11745 int delta_disks
, char *backup
, char *dev
,
11746 int direction
, int verbose
)
11749 struct geo_params geo
;
11751 dprintf("(enter)\n");
11753 memset(&geo
, 0, sizeof(struct geo_params
));
11755 geo
.dev_name
= dev
;
11756 strcpy(geo
.devnm
, st
->devnm
);
11759 geo
.layout
= layout
;
11760 geo
.chunksize
= chunksize
;
11761 geo
.raid_disks
= raid_disks
;
11762 if (delta_disks
!= UnSet
)
11763 geo
.raid_disks
+= delta_disks
;
11765 dprintf("for level : %i\n", geo
.level
);
11766 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11768 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11769 /* On container level we can only increase number of devices. */
11770 dprintf("imsm: info: Container operation\n");
11771 int old_raid_disks
= 0;
11773 if (imsm_reshape_is_allowed_on_container(
11774 st
, &geo
, &old_raid_disks
, direction
)) {
11775 struct imsm_update_reshape
*u
= NULL
;
11778 if (imsm_fix_size_mismatch(st
, -1)) {
11779 dprintf("imsm: Cannot fix size mismatch\n");
11780 goto exit_imsm_reshape_super
;
11783 len
= imsm_create_metadata_update_for_reshape(
11784 st
, &geo
, old_raid_disks
, &u
);
11787 dprintf("imsm: Cannot prepare update\n");
11788 goto exit_imsm_reshape_super
;
11792 /* update metadata locally */
11793 imsm_update_metadata_locally(st
, u
, len
);
11794 /* and possibly remotely */
11795 if (st
->update_tail
)
11796 append_metadata_update(st
, u
, len
);
11801 pr_err("(imsm) Operation is not allowed on this container\n");
11804 /* On volume level we support following operations
11805 * - takeover: raid10 -> raid0; raid0 -> raid10
11806 * - chunk size migration
11807 * - migration: raid5 -> raid0; raid0 -> raid5
11809 struct intel_super
*super
= st
->sb
;
11810 struct intel_dev
*dev
= super
->devlist
;
11812 dprintf("imsm: info: Volume operation\n");
11813 /* find requested device */
11816 imsm_find_array_devnm_by_subdev(
11817 dev
->index
, st
->container_devnm
);
11818 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11823 pr_err("Cannot find %s (%s) subarray\n",
11824 geo
.dev_name
, geo
.devnm
);
11825 goto exit_imsm_reshape_super
;
11827 super
->current_vol
= dev
->index
;
11828 change
= imsm_analyze_change(st
, &geo
, direction
);
11831 ret_val
= imsm_takeover(st
, &geo
);
11833 case CH_MIGRATION
: {
11834 struct imsm_update_reshape_migration
*u
= NULL
;
11836 imsm_create_metadata_update_for_migration(
11839 dprintf("imsm: Cannot prepare update\n");
11843 /* update metadata locally */
11844 imsm_update_metadata_locally(st
, u
, len
);
11845 /* and possibly remotely */
11846 if (st
->update_tail
)
11847 append_metadata_update(st
, u
, len
);
11852 case CH_ARRAY_SIZE
: {
11853 struct imsm_update_size_change
*u
= NULL
;
11855 imsm_create_metadata_update_for_size_change(
11858 dprintf("imsm: Cannot prepare update\n");
11862 /* update metadata locally */
11863 imsm_update_metadata_locally(st
, u
, len
);
11864 /* and possibly remotely */
11865 if (st
->update_tail
)
11866 append_metadata_update(st
, u
, len
);
11876 exit_imsm_reshape_super
:
11877 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11881 #define COMPLETED_OK 0
11882 #define COMPLETED_NONE 1
11883 #define COMPLETED_DELAYED 2
11885 static int read_completed(int fd
, unsigned long long *val
)
11890 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11894 ret
= COMPLETED_OK
;
11895 if (strncmp(buf
, "none", 4) == 0) {
11896 ret
= COMPLETED_NONE
;
11897 } else if (strncmp(buf
, "delayed", 7) == 0) {
11898 ret
= COMPLETED_DELAYED
;
11901 *val
= strtoull(buf
, &ep
, 0);
11902 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11908 /*******************************************************************************
11909 * Function: wait_for_reshape_imsm
11910 * Description: Function writes new sync_max value and waits until
11911 * reshape process reach new position
11913 * sra : general array info
11914 * ndata : number of disks in new array's layout
11917 * 1 : there is no reshape in progress,
11919 ******************************************************************************/
11920 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11922 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11924 unsigned long long completed
;
11925 /* to_complete : new sync_max position */
11926 unsigned long long to_complete
= sra
->reshape_progress
;
11927 unsigned long long position_to_set
= to_complete
/ ndata
;
11930 dprintf("cannot open reshape_position\n");
11935 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11937 dprintf("cannot read reshape_position (no reshape in progres)\n");
11946 if (completed
> position_to_set
) {
11947 dprintf("wrong next position to set %llu (%llu)\n",
11948 to_complete
, position_to_set
);
11952 dprintf("Position set: %llu\n", position_to_set
);
11953 if (sysfs_set_num(sra
, NULL
, "sync_max",
11954 position_to_set
) != 0) {
11955 dprintf("cannot set reshape position to %llu\n",
11964 int timeout
= 3000;
11966 sysfs_wait(fd
, &timeout
);
11967 if (sysfs_get_str(sra
, NULL
, "sync_action",
11969 strncmp(action
, "reshape", 7) != 0) {
11970 if (strncmp(action
, "idle", 4) == 0)
11976 rc
= read_completed(fd
, &completed
);
11978 dprintf("cannot read reshape_position (in loop)\n");
11981 } else if (rc
== COMPLETED_NONE
)
11983 } while (completed
< position_to_set
);
11989 /*******************************************************************************
11990 * Function: check_degradation_change
11991 * Description: Check that array hasn't become failed.
11993 * info : for sysfs access
11994 * sources : source disks descriptors
11995 * degraded: previous degradation level
11997 * degradation level
11998 ******************************************************************************/
11999 int check_degradation_change(struct mdinfo
*info
,
12003 unsigned long long new_degraded
;
12006 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
12007 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
12008 /* check each device to ensure it is still working */
12011 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
12012 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
12014 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
12017 if (sysfs_get_str(info
,
12018 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
12019 strstr(sbuf
, "faulty") ||
12020 strstr(sbuf
, "in_sync") == NULL
) {
12021 /* this device is dead */
12022 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
12023 if (sd
->disk
.raid_disk
>= 0 &&
12024 sources
[sd
->disk
.raid_disk
] >= 0) {
12026 sd
->disk
.raid_disk
]);
12027 sources
[sd
->disk
.raid_disk
] =
12036 return new_degraded
;
12039 /*******************************************************************************
12040 * Function: imsm_manage_reshape
12041 * Description: Function finds array under reshape and it manages reshape
12042 * process. It creates stripes backups (if required) and sets
12045 * afd : Backup handle (nattive) - not used
12046 * sra : general array info
12047 * reshape : reshape parameters - not used
12048 * st : supertype structure
12049 * blocks : size of critical section [blocks]
12050 * fds : table of source device descriptor
12051 * offsets : start of array (offest per devices)
12053 * destfd : table of destination device descriptor
12054 * destoffsets : table of destination offsets (per device)
12056 * 1 : success, reshape is done
12058 ******************************************************************************/
12059 static int imsm_manage_reshape(
12060 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
12061 struct supertype
*st
, unsigned long backup_blocks
,
12062 int *fds
, unsigned long long *offsets
,
12063 int dests
, int *destfd
, unsigned long long *destoffsets
)
12066 struct intel_super
*super
= st
->sb
;
12067 struct intel_dev
*dv
;
12068 unsigned int sector_size
= super
->sector_size
;
12069 struct imsm_dev
*dev
= NULL
;
12070 struct imsm_map
*map_src
, *map_dest
;
12071 int migr_vol_qan
= 0;
12072 int ndata
, odata
; /* [bytes] */
12073 int chunk
; /* [bytes] */
12074 struct migr_record
*migr_rec
;
12076 unsigned int buf_size
; /* [bytes] */
12077 unsigned long long max_position
; /* array size [bytes] */
12078 unsigned long long next_step
; /* [blocks]/[bytes] */
12079 unsigned long long old_data_stripe_length
;
12080 unsigned long long start_src
; /* [bytes] */
12081 unsigned long long start
; /* [bytes] */
12082 unsigned long long start_buf_shift
; /* [bytes] */
12084 int source_layout
= 0;
12085 int subarray_index
= -1;
12090 if (!fds
|| !offsets
)
12093 /* Find volume during the reshape */
12094 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
12095 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12096 dv
->dev
->vol
.migr_state
== 1) {
12099 subarray_index
= dv
->index
;
12102 /* Only one volume can migrate at the same time */
12103 if (migr_vol_qan
!= 1) {
12104 pr_err("%s", migr_vol_qan
?
12105 "Number of migrating volumes greater than 1\n" :
12106 "There is no volume during migrationg\n");
12110 map_dest
= get_imsm_map(dev
, MAP_0
);
12111 map_src
= get_imsm_map(dev
, MAP_1
);
12112 if (map_src
== NULL
)
12115 ndata
= imsm_num_data_members(map_dest
);
12116 odata
= imsm_num_data_members(map_src
);
12118 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12119 old_data_stripe_length
= odata
* chunk
;
12121 migr_rec
= super
->migr_rec
;
12123 /* initialize migration record for start condition */
12124 if (sra
->reshape_progress
== 0)
12125 init_migr_record_imsm(st
, dev
, sra
);
12127 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12128 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12131 /* Save checkpoint to update migration record for current
12132 * reshape position (in md). It can be farther than current
12133 * reshape position in metadata.
12135 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12136 /* ignore error == 2, this can mean end of reshape here
12138 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12143 /* size for data */
12144 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12145 /* extend buffer size for parity disk */
12146 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12147 /* add space for stripe alignment */
12148 buf_size
+= old_data_stripe_length
;
12149 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12150 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12154 max_position
= sra
->component_size
* ndata
;
12155 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12157 while (current_migr_unit(migr_rec
) <
12158 get_num_migr_units(migr_rec
)) {
12159 /* current reshape position [blocks] */
12160 unsigned long long current_position
=
12161 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12162 * current_migr_unit(migr_rec
);
12163 unsigned long long border
;
12165 /* Check that array hasn't become failed.
12167 degraded
= check_degradation_change(sra
, fds
, degraded
);
12168 if (degraded
> 1) {
12169 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12173 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12175 if ((current_position
+ next_step
) > max_position
)
12176 next_step
= max_position
- current_position
;
12178 start
= current_position
* 512;
12180 /* align reading start to old geometry */
12181 start_buf_shift
= start
% old_data_stripe_length
;
12182 start_src
= start
- start_buf_shift
;
12184 border
= (start_src
/ odata
) - (start
/ ndata
);
12186 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12187 /* save critical stripes to buf
12188 * start - start address of current unit
12189 * to backup [bytes]
12190 * start_src - start address of current unit
12191 * to backup alligned to source array
12194 unsigned long long next_step_filler
;
12195 unsigned long long copy_length
= next_step
* 512;
12197 /* allign copy area length to stripe in old geometry */
12198 next_step_filler
= ((copy_length
+ start_buf_shift
)
12199 % old_data_stripe_length
);
12200 if (next_step_filler
)
12201 next_step_filler
= (old_data_stripe_length
12202 - next_step_filler
);
12203 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12204 start
, start_src
, copy_length
,
12205 start_buf_shift
, next_step_filler
);
12207 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12208 chunk
, map_src
->raid_level
,
12209 source_layout
, 0, NULL
, start_src
,
12211 next_step_filler
+ start_buf_shift
,
12213 dprintf("imsm: Cannot save stripes to buffer\n");
12216 /* Convert data to destination format and store it
12217 * in backup general migration area
12219 if (save_backup_imsm(st
, dev
, sra
,
12220 buf
+ start_buf_shift
, copy_length
)) {
12221 dprintf("imsm: Cannot save stripes to target devices\n");
12224 if (save_checkpoint_imsm(st
, sra
,
12225 UNIT_SRC_IN_CP_AREA
)) {
12226 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12230 /* set next step to use whole border area */
12231 border
/= next_step
;
12233 next_step
*= border
;
12235 /* When data backed up, checkpoint stored,
12236 * kick the kernel to reshape unit of data
12238 next_step
= next_step
+ sra
->reshape_progress
;
12239 /* limit next step to array max position */
12240 if (next_step
> max_position
)
12241 next_step
= max_position
;
12242 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12243 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12244 sra
->reshape_progress
= next_step
;
12246 /* wait until reshape finish */
12247 if (wait_for_reshape_imsm(sra
, ndata
)) {
12248 dprintf("wait_for_reshape_imsm returned error!\n");
12254 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12255 /* ignore error == 2, this can mean end of reshape here
12257 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12263 /* clear migr_rec on disks after successful migration */
12266 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12267 for (d
= super
->disks
; d
; d
= d
->next
) {
12268 if (d
->index
< 0 || is_failed(&d
->disk
))
12270 unsigned long long dsize
;
12272 get_dev_size(d
->fd
, NULL
, &dsize
);
12273 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12275 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12276 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12277 MIGR_REC_BUF_SECTORS
*sector_size
)
12278 perror("Write migr_rec failed");
12282 /* return '1' if done */
12285 /* After the reshape eliminate size mismatch in metadata.
12286 * Don't update md/component_size here, volume hasn't
12287 * to take whole space. It is allowed by kernel.
12288 * md/component_size will be set propoperly after next assembly.
12290 imsm_fix_size_mismatch(st
, subarray_index
);
12294 /* See Grow.c: abort_reshape() for further explanation */
12295 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12296 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12297 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12302 struct superswitch super_imsm
= {
12303 .examine_super
= examine_super_imsm
,
12304 .brief_examine_super
= brief_examine_super_imsm
,
12305 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12306 .export_examine_super
= export_examine_super_imsm
,
12307 .detail_super
= detail_super_imsm
,
12308 .brief_detail_super
= brief_detail_super_imsm
,
12309 .write_init_super
= write_init_super_imsm
,
12310 .validate_geometry
= validate_geometry_imsm
,
12311 .add_to_super
= add_to_super_imsm
,
12312 .remove_from_super
= remove_from_super_imsm
,
12313 .detail_platform
= detail_platform_imsm
,
12314 .export_detail_platform
= export_detail_platform_imsm
,
12315 .kill_subarray
= kill_subarray_imsm
,
12316 .update_subarray
= update_subarray_imsm
,
12317 .load_container
= load_container_imsm
,
12318 .default_geometry
= default_geometry_imsm
,
12319 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12320 .reshape_super
= imsm_reshape_super
,
12321 .manage_reshape
= imsm_manage_reshape
,
12322 .recover_backup
= recover_backup_imsm
,
12323 .examine_badblocks
= examine_badblocks_imsm
,
12324 .match_home
= match_home_imsm
,
12325 .uuid_from_super
= uuid_from_super_imsm
,
12326 .getinfo_super
= getinfo_super_imsm
,
12327 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12328 .update_super
= update_super_imsm
,
12330 .avail_size
= avail_size_imsm
,
12331 .get_spare_criteria
= get_spare_criteria_imsm
,
12333 .compare_super
= compare_super_imsm
,
12335 .load_super
= load_super_imsm
,
12336 .init_super
= init_super_imsm
,
12337 .store_super
= store_super_imsm
,
12338 .free_super
= free_super_imsm
,
12339 .match_metadata_desc
= match_metadata_desc_imsm
,
12340 .container_content
= container_content_imsm
,
12341 .validate_container
= validate_container_imsm
,
12343 .write_init_ppl
= write_init_ppl_imsm
,
12344 .validate_ppl
= validate_ppl_imsm
,
12350 .open_new
= imsm_open_new
,
12351 .set_array_state
= imsm_set_array_state
,
12352 .set_disk
= imsm_set_disk
,
12353 .sync_metadata
= imsm_sync_metadata
,
12354 .activate_spare
= imsm_activate_spare
,
12355 .process_update
= imsm_process_update
,
12356 .prepare_update
= imsm_prepare_update
,
12357 .record_bad_block
= imsm_record_badblock
,
12358 .clear_bad_block
= imsm_clear_badblock
,
12359 .get_bad_blocks
= imsm_get_badblocks
,