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
);
3456 unsigned long long array_blocks
;
3459 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3461 (get_num_migr_units(migr_rec
)-1)) &&
3462 (super
->migr_rec
->rec_status
==
3463 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3466 info
->reshape_progress
= blocks_per_unit
* units
;
3468 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3469 (unsigned long long)units
,
3470 (unsigned long long)blocks_per_unit
,
3471 info
->reshape_progress
);
3473 used_disks
= imsm_num_data_members(prev_map
);
3474 if (used_disks
> 0) {
3475 array_blocks
= per_dev_array_size(map
) *
3477 info
->custom_array_size
=
3478 round_size_to_mb(array_blocks
,
3484 /* we could emulate the checkpointing of
3485 * 'sync_action=check' migrations, but for now
3486 * we just immediately complete them
3489 /* this is handled by container_content_imsm() */
3490 case MIGR_STATE_CHANGE
:
3491 /* FIXME handle other migrations */
3493 /* we are not dirty, so... */
3494 info
->resync_start
= MaxSector
;
3498 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3499 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3501 info
->array
.major_version
= -1;
3502 info
->array
.minor_version
= -2;
3503 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3504 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3505 uuid_from_super_imsm(st
, info
->uuid
);
3509 for (i
=0; i
<map_disks
; i
++) {
3511 if (i
< info
->array
.raid_disks
) {
3512 struct imsm_disk
*dsk
;
3513 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3514 dsk
= get_imsm_disk(super
, j
);
3515 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3522 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3523 int failed
, int look_in_map
);
3525 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3528 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3530 if (is_gen_migration(dev
)) {
3533 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3535 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3536 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3537 if (map2
->map_state
!= map_state
) {
3538 map2
->map_state
= map_state
;
3539 super
->updates_pending
++;
3544 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3548 for (d
= super
->missing
; d
; d
= d
->next
)
3549 if (d
->index
== index
)
3554 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3556 struct intel_super
*super
= st
->sb
;
3557 struct imsm_disk
*disk
;
3558 int map_disks
= info
->array
.raid_disks
;
3559 int max_enough
= -1;
3561 struct imsm_super
*mpb
;
3563 if (super
->current_vol
>= 0) {
3564 getinfo_super_imsm_volume(st
, info
, map
);
3567 memset(info
, 0, sizeof(*info
));
3569 /* Set raid_disks to zero so that Assemble will always pull in valid
3572 info
->array
.raid_disks
= 0;
3573 info
->array
.level
= LEVEL_CONTAINER
;
3574 info
->array
.layout
= 0;
3575 info
->array
.md_minor
= -1;
3576 info
->array
.ctime
= 0; /* N/A for imsm */
3577 info
->array
.utime
= 0;
3578 info
->array
.chunk_size
= 0;
3580 info
->disk
.major
= 0;
3581 info
->disk
.minor
= 0;
3582 info
->disk
.raid_disk
= -1;
3583 info
->reshape_active
= 0;
3584 info
->array
.major_version
= -1;
3585 info
->array
.minor_version
= -2;
3586 strcpy(info
->text_version
, "imsm");
3587 info
->safe_mode_delay
= 0;
3588 info
->disk
.number
= -1;
3589 info
->disk
.state
= 0;
3591 info
->recovery_start
= MaxSector
;
3592 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3593 info
->bb
.supported
= 1;
3595 /* do we have the all the insync disks that we expect? */
3596 mpb
= super
->anchor
;
3597 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3599 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3600 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3601 int failed
, enough
, j
, missing
= 0;
3602 struct imsm_map
*map
;
3605 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3606 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3607 map
= get_imsm_map(dev
, MAP_0
);
3609 /* any newly missing disks?
3610 * (catches single-degraded vs double-degraded)
3612 for (j
= 0; j
< map
->num_members
; j
++) {
3613 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3614 __u32 idx
= ord_to_idx(ord
);
3616 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3617 info
->disk
.raid_disk
= j
;
3619 if (!(ord
& IMSM_ORD_REBUILD
) &&
3620 get_imsm_missing(super
, idx
)) {
3626 if (state
== IMSM_T_STATE_FAILED
)
3628 else if (state
== IMSM_T_STATE_DEGRADED
&&
3629 (state
!= map
->map_state
|| missing
))
3631 else /* we're normal, or already degraded */
3633 if (is_gen_migration(dev
) && missing
) {
3634 /* during general migration we need all disks
3635 * that process is running on.
3636 * No new missing disk is allowed.
3640 /* no more checks necessary
3644 /* in the missing/failed disk case check to see
3645 * if at least one array is runnable
3647 max_enough
= max(max_enough
, enough
);
3649 dprintf("enough: %d\n", max_enough
);
3650 info
->container_enough
= max_enough
;
3653 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3655 disk
= &super
->disks
->disk
;
3656 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3657 info
->component_size
= reserved
;
3658 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3659 /* we don't change info->disk.raid_disk here because
3660 * this state will be finalized in mdmon after we have
3661 * found the 'most fresh' version of the metadata
3663 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3664 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3665 0 : (1 << MD_DISK_SYNC
);
3668 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3669 * ->compare_super may have updated the 'num_raid_devs' field for spares
3671 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3672 uuid_from_super_imsm(st
, info
->uuid
);
3674 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3676 /* I don't know how to compute 'map' on imsm, so use safe default */
3679 for (i
= 0; i
< map_disks
; i
++)
3685 /* allocates memory and fills disk in mdinfo structure
3686 * for each disk in array */
3687 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3689 struct mdinfo
*mddev
;
3690 struct intel_super
*super
= st
->sb
;
3691 struct imsm_disk
*disk
;
3694 if (!super
|| !super
->disks
)
3697 mddev
= xcalloc(1, sizeof(*mddev
));
3701 tmp
= xcalloc(1, sizeof(*tmp
));
3703 tmp
->next
= mddev
->devs
;
3705 tmp
->disk
.number
= count
++;
3706 tmp
->disk
.major
= dl
->major
;
3707 tmp
->disk
.minor
= dl
->minor
;
3708 tmp
->disk
.state
= is_configured(disk
) ?
3709 (1 << MD_DISK_ACTIVE
) : 0;
3710 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3711 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3712 tmp
->disk
.raid_disk
= -1;
3718 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3719 char *update
, char *devname
, int verbose
,
3720 int uuid_set
, char *homehost
)
3722 /* For 'assemble' and 'force' we need to return non-zero if any
3723 * change was made. For others, the return value is ignored.
3724 * Update options are:
3725 * force-one : This device looks a bit old but needs to be included,
3726 * update age info appropriately.
3727 * assemble: clear any 'faulty' flag to allow this device to
3729 * force-array: Array is degraded but being forced, mark it clean
3730 * if that will be needed to assemble it.
3732 * newdev: not used ????
3733 * grow: Array has gained a new device - this is currently for
3735 * resync: mark as dirty so a resync will happen.
3736 * name: update the name - preserving the homehost
3737 * uuid: Change the uuid of the array to match watch is given
3739 * Following are not relevant for this imsm:
3740 * sparc2.2 : update from old dodgey metadata
3741 * super-minor: change the preferred_minor number
3742 * summaries: update redundant counters.
3743 * homehost: update the recorded homehost
3744 * _reshape_progress: record new reshape_progress position.
3747 struct intel_super
*super
= st
->sb
;
3748 struct imsm_super
*mpb
;
3750 /* we can only update container info */
3751 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3754 mpb
= super
->anchor
;
3756 if (strcmp(update
, "uuid") == 0) {
3757 /* We take this to mean that the family_num should be updated.
3758 * However that is much smaller than the uuid so we cannot really
3759 * allow an explicit uuid to be given. And it is hard to reliably
3761 * So if !uuid_set we know the current uuid is random and just used
3762 * the first 'int' and copy it to the other 3 positions.
3763 * Otherwise we require the 4 'int's to be the same as would be the
3764 * case if we are using a random uuid. So an explicit uuid will be
3765 * accepted as long as all for ints are the same... which shouldn't hurt
3768 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3771 if (info
->uuid
[0] != info
->uuid
[1] ||
3772 info
->uuid
[1] != info
->uuid
[2] ||
3773 info
->uuid
[2] != info
->uuid
[3])
3779 mpb
->orig_family_num
= info
->uuid
[0];
3780 } else if (strcmp(update
, "assemble") == 0)
3785 /* successful update? recompute checksum */
3787 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3792 static size_t disks_to_mpb_size(int disks
)
3796 size
= sizeof(struct imsm_super
);
3797 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3798 size
+= 2 * sizeof(struct imsm_dev
);
3799 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3800 size
+= (4 - 2) * sizeof(struct imsm_map
);
3801 /* 4 possible disk_ord_tbl's */
3802 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3803 /* maximum bbm log */
3804 size
+= sizeof(struct bbm_log
);
3809 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3810 unsigned long long data_offset
)
3812 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3815 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3818 static void free_devlist(struct intel_super
*super
)
3820 struct intel_dev
*dv
;
3822 while (super
->devlist
) {
3823 dv
= super
->devlist
->next
;
3824 free(super
->devlist
->dev
);
3825 free(super
->devlist
);
3826 super
->devlist
= dv
;
3830 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3832 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3835 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3839 * 0 same, or first was empty, and second was copied
3840 * 1 second had wrong number
3842 * 3 wrong other info
3844 struct intel_super
*first
= st
->sb
;
3845 struct intel_super
*sec
= tst
->sb
;
3852 /* in platform dependent environment test if the disks
3853 * use the same Intel hba
3854 * If not on Intel hba at all, allow anything.
3856 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3857 if (first
->hba
->type
!= sec
->hba
->type
) {
3859 "HBAs of devices do not match %s != %s\n",
3860 get_sys_dev_type(first
->hba
->type
),
3861 get_sys_dev_type(sec
->hba
->type
));
3864 if (first
->orom
!= sec
->orom
) {
3866 "HBAs of devices do not match %s != %s\n",
3867 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3872 /* if an anchor does not have num_raid_devs set then it is a free
3875 if (first
->anchor
->num_raid_devs
> 0 &&
3876 sec
->anchor
->num_raid_devs
> 0) {
3877 /* Determine if these disks might ever have been
3878 * related. Further disambiguation can only take place
3879 * in load_super_imsm_all
3881 __u32 first_family
= first
->anchor
->orig_family_num
;
3882 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3884 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3885 MAX_SIGNATURE_LENGTH
) != 0)
3888 if (first_family
== 0)
3889 first_family
= first
->anchor
->family_num
;
3890 if (sec_family
== 0)
3891 sec_family
= sec
->anchor
->family_num
;
3893 if (first_family
!= sec_family
)
3898 /* if 'first' is a spare promote it to a populated mpb with sec's
3901 if (first
->anchor
->num_raid_devs
== 0 &&
3902 sec
->anchor
->num_raid_devs
> 0) {
3904 struct intel_dev
*dv
;
3905 struct imsm_dev
*dev
;
3907 /* we need to copy raid device info from sec if an allocation
3908 * fails here we don't associate the spare
3910 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3911 dv
= xmalloc(sizeof(*dv
));
3912 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3915 dv
->next
= first
->devlist
;
3916 first
->devlist
= dv
;
3918 if (i
< sec
->anchor
->num_raid_devs
) {
3919 /* allocation failure */
3920 free_devlist(first
);
3921 pr_err("imsm: failed to associate spare\n");
3924 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3925 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3926 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3927 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3928 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3929 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3935 static void fd2devname(int fd
, char *name
)
3939 char dname
[PATH_MAX
];
3944 if (fstat(fd
, &st
) != 0)
3946 sprintf(path
, "/sys/dev/block/%d:%d",
3947 major(st
.st_rdev
), minor(st
.st_rdev
));
3949 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3954 nm
= strrchr(dname
, '/');
3957 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3961 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3964 char *name
= fd2kname(fd
);
3969 if (strncmp(name
, "nvme", 4) != 0)
3972 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3974 return load_sys(path
, buf
, buf_len
);
3977 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3979 static int imsm_read_serial(int fd
, char *devname
,
3980 __u8
*serial
, size_t serial_buf_len
)
3989 memset(buf
, 0, sizeof(buf
));
3991 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3994 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3996 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3997 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3998 fd2devname(fd
, (char *) serial
);
4004 pr_err("Failed to retrieve serial for %s\n",
4009 /* trim all whitespace and non-printable characters and convert
4012 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4015 /* ':' is reserved for use in placeholder serial
4016 * numbers for missing disks
4027 if (len
> serial_buf_len
) {
4028 /* truncate leading characters */
4029 dest
+= len
- serial_buf_len
;
4030 len
= serial_buf_len
;
4033 memset(serial
, 0, serial_buf_len
);
4034 memcpy(serial
, dest
, len
);
4039 static int serialcmp(__u8
*s1
, __u8
*s2
)
4041 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4044 static void serialcpy(__u8
*dest
, __u8
*src
)
4046 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4049 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4053 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4054 if (serialcmp(dl
->serial
, serial
) == 0)
4060 static struct imsm_disk
*
4061 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4065 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4066 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4068 if (serialcmp(disk
->serial
, serial
) == 0) {
4079 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4081 struct imsm_disk
*disk
;
4086 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4088 rv
= imsm_read_serial(fd
, devname
, serial
, MAX_RAID_SERIAL_LEN
);
4093 dl
= xcalloc(1, sizeof(*dl
));
4096 dl
->major
= major(stb
.st_rdev
);
4097 dl
->minor
= minor(stb
.st_rdev
);
4098 dl
->next
= super
->disks
;
4099 dl
->fd
= keep_fd
? fd
: -1;
4100 assert(super
->disks
== NULL
);
4102 serialcpy(dl
->serial
, serial
);
4105 fd2devname(fd
, name
);
4107 dl
->devname
= xstrdup(devname
);
4109 dl
->devname
= xstrdup(name
);
4111 /* look up this disk's index in the current anchor */
4112 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4115 /* only set index on disks that are a member of a
4116 * populated contianer, i.e. one with raid_devs
4118 if (is_failed(&dl
->disk
))
4120 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4127 /* When migrating map0 contains the 'destination' state while map1
4128 * contains the current state. When not migrating map0 contains the
4129 * current state. This routine assumes that map[0].map_state is set to
4130 * the current array state before being called.
4132 * Migration is indicated by one of the following states
4133 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4134 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4135 * map1state=unitialized)
4136 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4138 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4139 * map1state=degraded)
4140 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4143 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4144 __u8 to_state
, int migr_type
)
4146 struct imsm_map
*dest
;
4147 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4149 dev
->vol
.migr_state
= 1;
4150 set_migr_type(dev
, migr_type
);
4151 dev
->vol
.curr_migr_unit
= 0;
4152 dest
= get_imsm_map(dev
, MAP_1
);
4154 /* duplicate and then set the target end state in map[0] */
4155 memcpy(dest
, src
, sizeof_imsm_map(src
));
4156 if (migr_type
== MIGR_GEN_MIGR
) {
4160 for (i
= 0; i
< src
->num_members
; i
++) {
4161 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4162 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4166 if (migr_type
== MIGR_GEN_MIGR
)
4167 /* Clear migration record */
4168 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4170 src
->map_state
= to_state
;
4173 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4176 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4177 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4181 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4182 * completed in the last migration.
4184 * FIXME add support for raid-level-migration
4186 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4187 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4188 /* when final map state is other than expected
4189 * merge maps (not for migration)
4193 for (i
= 0; i
< prev
->num_members
; i
++)
4194 for (j
= 0; j
< map
->num_members
; j
++)
4195 /* during online capacity expansion
4196 * disks position can be changed
4197 * if takeover is used
4199 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4200 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4201 map
->disk_ord_tbl
[j
] |=
4202 prev
->disk_ord_tbl
[i
];
4205 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4206 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4209 dev
->vol
.migr_state
= 0;
4210 set_migr_type(dev
, 0);
4211 dev
->vol
.curr_migr_unit
= 0;
4212 map
->map_state
= map_state
;
4215 static int parse_raid_devices(struct intel_super
*super
)
4218 struct imsm_dev
*dev_new
;
4219 size_t len
, len_migr
;
4221 size_t space_needed
= 0;
4222 struct imsm_super
*mpb
= super
->anchor
;
4224 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4225 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4226 struct intel_dev
*dv
;
4228 len
= sizeof_imsm_dev(dev_iter
, 0);
4229 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4231 space_needed
+= len_migr
- len
;
4233 dv
= xmalloc(sizeof(*dv
));
4234 if (max_len
< len_migr
)
4236 if (max_len
> len_migr
)
4237 space_needed
+= max_len
- len_migr
;
4238 dev_new
= xmalloc(max_len
);
4239 imsm_copy_dev(dev_new
, dev_iter
);
4242 dv
->next
= super
->devlist
;
4243 super
->devlist
= dv
;
4246 /* ensure that super->buf is large enough when all raid devices
4249 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4252 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4253 super
->sector_size
);
4254 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4257 memcpy(buf
, super
->buf
, super
->len
);
4258 memset(buf
+ super
->len
, 0, len
- super
->len
);
4264 super
->extra_space
+= space_needed
;
4269 /*******************************************************************************
4270 * Function: check_mpb_migr_compatibility
4271 * Description: Function checks for unsupported migration features:
4272 * - migration optimization area (pba_of_lba0)
4273 * - descending reshape (ascending_migr)
4275 * super : imsm metadata information
4277 * 0 : migration is compatible
4278 * -1 : migration is not compatible
4279 ******************************************************************************/
4280 int check_mpb_migr_compatibility(struct intel_super
*super
)
4282 struct imsm_map
*map0
, *map1
;
4283 struct migr_record
*migr_rec
= super
->migr_rec
;
4286 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4287 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4290 dev_iter
->vol
.migr_state
== 1 &&
4291 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4292 /* This device is migrating */
4293 map0
= get_imsm_map(dev_iter
, MAP_0
);
4294 map1
= get_imsm_map(dev_iter
, MAP_1
);
4295 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4296 /* migration optimization area was used */
4298 if (migr_rec
->ascending_migr
== 0 &&
4299 migr_rec
->dest_depth_per_unit
> 0)
4300 /* descending reshape not supported yet */
4307 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4309 /* load_imsm_mpb - read matrix metadata
4310 * allocates super->mpb to be freed by free_imsm
4312 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4314 unsigned long long dsize
;
4315 unsigned long long sectors
;
4316 unsigned int sector_size
= super
->sector_size
;
4318 struct imsm_super
*anchor
;
4321 get_dev_size(fd
, NULL
, &dsize
);
4322 if (dsize
< 2*sector_size
) {
4324 pr_err("%s: device to small for imsm\n",
4329 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4331 pr_err("Cannot seek to anchor block on %s: %s\n",
4332 devname
, strerror(errno
));
4336 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4338 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4341 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4343 pr_err("Cannot read anchor block on %s: %s\n",
4344 devname
, strerror(errno
));
4349 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4351 pr_err("no IMSM anchor on %s\n", devname
);
4356 __free_imsm(super
, 0);
4357 /* reload capability and hba */
4359 /* capability and hba must be updated with new super allocation */
4360 find_intel_hba_capability(fd
, super
, devname
);
4361 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4362 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4364 pr_err("unable to allocate %zu byte mpb buffer\n",
4369 memcpy(super
->buf
, anchor
, sector_size
);
4371 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4374 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4375 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4376 pr_err("could not allocate migr_rec buffer\n");
4380 super
->clean_migration_record_by_mdmon
= 0;
4383 check_sum
= __gen_imsm_checksum(super
->anchor
);
4384 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4386 pr_err("IMSM checksum %x != %x on %s\n",
4388 __le32_to_cpu(super
->anchor
->check_sum
),
4396 /* read the extended mpb */
4397 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4399 pr_err("Cannot seek to extended mpb on %s: %s\n",
4400 devname
, strerror(errno
));
4404 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4405 super
->len
- sector_size
) != super
->len
- sector_size
) {
4407 pr_err("Cannot read extended mpb on %s: %s\n",
4408 devname
, strerror(errno
));
4412 check_sum
= __gen_imsm_checksum(super
->anchor
);
4413 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4415 pr_err("IMSM checksum %x != %x on %s\n",
4416 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4424 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4426 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4427 static void clear_hi(struct intel_super
*super
)
4429 struct imsm_super
*mpb
= super
->anchor
;
4431 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4433 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4434 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4435 disk
->total_blocks_hi
= 0;
4437 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4438 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4441 for (n
= 0; n
< 2; ++n
) {
4442 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4445 map
->pba_of_lba0_hi
= 0;
4446 map
->blocks_per_member_hi
= 0;
4447 map
->num_data_stripes_hi
= 0;
4453 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4457 err
= load_imsm_mpb(fd
, super
, devname
);
4460 if (super
->sector_size
== 4096)
4461 convert_from_4k(super
);
4462 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4465 err
= parse_raid_devices(super
);
4468 err
= load_bbm_log(super
);
4473 static void __free_imsm_disk(struct dl
*d
)
4485 static void free_imsm_disks(struct intel_super
*super
)
4489 while (super
->disks
) {
4491 super
->disks
= d
->next
;
4492 __free_imsm_disk(d
);
4494 while (super
->disk_mgmt_list
) {
4495 d
= super
->disk_mgmt_list
;
4496 super
->disk_mgmt_list
= d
->next
;
4497 __free_imsm_disk(d
);
4499 while (super
->missing
) {
4501 super
->missing
= d
->next
;
4502 __free_imsm_disk(d
);
4507 /* free all the pieces hanging off of a super pointer */
4508 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4510 struct intel_hba
*elem
, *next
;
4516 /* unlink capability description */
4518 if (super
->migr_rec_buf
) {
4519 free(super
->migr_rec_buf
);
4520 super
->migr_rec_buf
= NULL
;
4523 free_imsm_disks(super
);
4524 free_devlist(super
);
4528 free((void *)elem
->path
);
4534 free(super
->bbm_log
);
4538 static void free_imsm(struct intel_super
*super
)
4540 __free_imsm(super
, 1);
4541 free(super
->bb
.entries
);
4545 static void free_super_imsm(struct supertype
*st
)
4547 struct intel_super
*super
= st
->sb
;
4556 static struct intel_super
*alloc_super(void)
4558 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4560 super
->current_vol
= -1;
4561 super
->create_offset
= ~((unsigned long long) 0);
4563 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4564 sizeof(struct md_bb_entry
));
4565 if (!super
->bb
.entries
) {
4574 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4576 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4578 struct sys_dev
*hba_name
;
4581 if (fd
>= 0 && test_partition(fd
)) {
4582 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4586 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4591 hba_name
= find_disk_attached_hba(fd
, NULL
);
4594 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4598 rv
= attach_hba_to_super(super
, hba_name
);
4601 struct intel_hba
*hba
= super
->hba
;
4603 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4604 " but the container is assigned to Intel(R) %s %s (",
4606 get_sys_dev_type(hba_name
->type
),
4607 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4608 hba_name
->pci_id
? : "Err!",
4609 get_sys_dev_type(super
->hba
->type
),
4610 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4613 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4615 fprintf(stderr
, ", ");
4618 fprintf(stderr
, ").\n"
4619 " Mixing devices attached to different controllers is not allowed.\n");
4623 super
->orom
= find_imsm_capability(hba_name
);
4630 /* find_missing - helper routine for load_super_imsm_all that identifies
4631 * disks that have disappeared from the system. This routine relies on
4632 * the mpb being uptodate, which it is at load time.
4634 static int find_missing(struct intel_super
*super
)
4637 struct imsm_super
*mpb
= super
->anchor
;
4639 struct imsm_disk
*disk
;
4641 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4642 disk
= __get_imsm_disk(mpb
, i
);
4643 dl
= serial_to_dl(disk
->serial
, super
);
4647 dl
= xmalloc(sizeof(*dl
));
4651 dl
->devname
= xstrdup("missing");
4653 serialcpy(dl
->serial
, disk
->serial
);
4656 dl
->next
= super
->missing
;
4657 super
->missing
= dl
;
4663 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4665 struct intel_disk
*idisk
= disk_list
;
4668 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4670 idisk
= idisk
->next
;
4676 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4677 struct intel_super
*super
,
4678 struct intel_disk
**disk_list
)
4680 struct imsm_disk
*d
= &super
->disks
->disk
;
4681 struct imsm_super
*mpb
= super
->anchor
;
4684 for (i
= 0; i
< tbl_size
; i
++) {
4685 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4686 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4688 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4689 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4690 dprintf("mpb from %d:%d matches %d:%d\n",
4691 super
->disks
->major
,
4692 super
->disks
->minor
,
4693 table
[i
]->disks
->major
,
4694 table
[i
]->disks
->minor
);
4698 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4699 is_configured(d
) == is_configured(tbl_d
)) &&
4700 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4701 /* current version of the mpb is a
4702 * better candidate than the one in
4703 * super_table, but copy over "cross
4704 * generational" status
4706 struct intel_disk
*idisk
;
4708 dprintf("mpb from %d:%d replaces %d:%d\n",
4709 super
->disks
->major
,
4710 super
->disks
->minor
,
4711 table
[i
]->disks
->major
,
4712 table
[i
]->disks
->minor
);
4714 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4715 if (idisk
&& is_failed(&idisk
->disk
))
4716 tbl_d
->status
|= FAILED_DISK
;
4719 struct intel_disk
*idisk
;
4720 struct imsm_disk
*disk
;
4722 /* tbl_mpb is more up to date, but copy
4723 * over cross generational status before
4726 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4727 if (disk
&& is_failed(disk
))
4728 d
->status
|= FAILED_DISK
;
4730 idisk
= disk_list_get(d
->serial
, *disk_list
);
4733 if (disk
&& is_configured(disk
))
4734 idisk
->disk
.status
|= CONFIGURED_DISK
;
4737 dprintf("mpb from %d:%d prefer %d:%d\n",
4738 super
->disks
->major
,
4739 super
->disks
->minor
,
4740 table
[i
]->disks
->major
,
4741 table
[i
]->disks
->minor
);
4749 table
[tbl_size
++] = super
;
4753 /* update/extend the merged list of imsm_disk records */
4754 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4755 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4756 struct intel_disk
*idisk
;
4758 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4760 idisk
->disk
.status
|= disk
->status
;
4761 if (is_configured(&idisk
->disk
) ||
4762 is_failed(&idisk
->disk
))
4763 idisk
->disk
.status
&= ~(SPARE_DISK
);
4765 idisk
= xcalloc(1, sizeof(*idisk
));
4766 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4767 idisk
->disk
= *disk
;
4768 idisk
->next
= *disk_list
;
4772 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4779 static struct intel_super
*
4780 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4783 struct imsm_super
*mpb
= super
->anchor
;
4787 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4788 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4789 struct intel_disk
*idisk
;
4791 idisk
= disk_list_get(disk
->serial
, disk_list
);
4793 if (idisk
->owner
== owner
||
4794 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4797 dprintf("'%.16s' owner %d != %d\n",
4798 disk
->serial
, idisk
->owner
,
4801 dprintf("unknown disk %x [%d]: %.16s\n",
4802 __le32_to_cpu(mpb
->family_num
), i
,
4808 if (ok_count
== mpb
->num_disks
)
4813 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4815 struct intel_super
*s
;
4817 for (s
= super_list
; s
; s
= s
->next
) {
4818 if (family_num
!= s
->anchor
->family_num
)
4820 pr_err("Conflict, offlining family %#x on '%s'\n",
4821 __le32_to_cpu(family_num
), s
->disks
->devname
);
4825 static struct intel_super
*
4826 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4828 struct intel_super
*super_table
[len
];
4829 struct intel_disk
*disk_list
= NULL
;
4830 struct intel_super
*champion
, *spare
;
4831 struct intel_super
*s
, **del
;
4836 memset(super_table
, 0, sizeof(super_table
));
4837 for (s
= *super_list
; s
; s
= s
->next
)
4838 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4840 for (i
= 0; i
< tbl_size
; i
++) {
4841 struct imsm_disk
*d
;
4842 struct intel_disk
*idisk
;
4843 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4846 d
= &s
->disks
->disk
;
4848 /* 'd' must appear in merged disk list for its
4849 * configuration to be valid
4851 idisk
= disk_list_get(d
->serial
, disk_list
);
4852 if (idisk
&& idisk
->owner
== i
)
4853 s
= validate_members(s
, disk_list
, i
);
4858 dprintf("marking family: %#x from %d:%d offline\n",
4860 super_table
[i
]->disks
->major
,
4861 super_table
[i
]->disks
->minor
);
4865 /* This is where the mdadm implementation differs from the Windows
4866 * driver which has no strict concept of a container. We can only
4867 * assemble one family from a container, so when returning a prodigal
4868 * array member to this system the code will not be able to disambiguate
4869 * the container contents that should be assembled ("foreign" versus
4870 * "local"). It requires user intervention to set the orig_family_num
4871 * to a new value to establish a new container. The Windows driver in
4872 * this situation fixes up the volume name in place and manages the
4873 * foreign array as an independent entity.
4878 for (i
= 0; i
< tbl_size
; i
++) {
4879 struct intel_super
*tbl_ent
= super_table
[i
];
4885 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4890 if (s
&& !is_spare
) {
4891 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4893 } else if (!s
&& !is_spare
)
4906 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4907 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4909 /* collect all dl's onto 'champion', and update them to
4910 * champion's version of the status
4912 for (s
= *super_list
; s
; s
= s
->next
) {
4913 struct imsm_super
*mpb
= champion
->anchor
;
4914 struct dl
*dl
= s
->disks
;
4919 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4921 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4922 struct imsm_disk
*disk
;
4924 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4927 /* only set index on disks that are a member of
4928 * a populated contianer, i.e. one with
4931 if (is_failed(&dl
->disk
))
4933 else if (is_spare(&dl
->disk
))
4939 if (i
>= mpb
->num_disks
) {
4940 struct intel_disk
*idisk
;
4942 idisk
= disk_list_get(dl
->serial
, disk_list
);
4943 if (idisk
&& is_spare(&idisk
->disk
) &&
4944 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4952 dl
->next
= champion
->disks
;
4953 champion
->disks
= dl
;
4957 /* delete 'champion' from super_list */
4958 for (del
= super_list
; *del
; ) {
4959 if (*del
== champion
) {
4960 *del
= (*del
)->next
;
4963 del
= &(*del
)->next
;
4965 champion
->next
= NULL
;
4969 struct intel_disk
*idisk
= disk_list
;
4971 disk_list
= disk_list
->next
;
4979 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4980 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4981 int major
, int minor
, int keep_fd
);
4983 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4984 int *max
, int keep_fd
);
4986 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4987 char *devname
, struct md_list
*devlist
,
4990 struct intel_super
*super_list
= NULL
;
4991 struct intel_super
*super
= NULL
;
4996 /* 'fd' is an opened container */
4997 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4999 /* get super block from devlist devices */
5000 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
5003 /* all mpbs enter, maybe one leaves */
5004 super
= imsm_thunderdome(&super_list
, i
);
5010 if (find_missing(super
) != 0) {
5016 /* load migration record */
5017 err
= load_imsm_migr_rec(super
, NULL
);
5019 /* migration is in progress,
5020 * but migr_rec cannot be loaded,
5026 /* Check migration compatibility */
5027 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5028 pr_err("Unsupported migration detected");
5030 fprintf(stderr
, " on %s\n", devname
);
5032 fprintf(stderr
, " (IMSM).\n");
5041 while (super_list
) {
5042 struct intel_super
*s
= super_list
;
5044 super_list
= super_list
->next
;
5053 strcpy(st
->container_devnm
, fd2devnm(fd
));
5055 st
->container_devnm
[0] = 0;
5056 if (err
== 0 && st
->ss
== NULL
) {
5057 st
->ss
= &super_imsm
;
5058 st
->minor_version
= 0;
5059 st
->max_devs
= IMSM_MAX_DEVICES
;
5065 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5066 int *max
, int keep_fd
)
5068 struct md_list
*tmpdev
;
5072 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5073 if (tmpdev
->used
!= 1)
5075 if (tmpdev
->container
== 1) {
5077 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5079 pr_err("cannot open device %s: %s\n",
5080 tmpdev
->devname
, strerror(errno
));
5084 err
= get_sra_super_block(fd
, super_list
,
5085 tmpdev
->devname
, &lmax
,
5094 int major
= major(tmpdev
->st_rdev
);
5095 int minor
= minor(tmpdev
->st_rdev
);
5096 err
= get_super_block(super_list
,
5113 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5114 int major
, int minor
, int keep_fd
)
5116 struct intel_super
*s
;
5128 sprintf(nm
, "%d:%d", major
, minor
);
5129 dfd
= dev_open(nm
, O_RDWR
);
5135 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5136 find_intel_hba_capability(dfd
, s
, devname
);
5137 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5139 /* retry the load if we might have raced against mdmon */
5140 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5141 for (retry
= 0; retry
< 3; retry
++) {
5143 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5149 s
->next
= *super_list
;
5157 if (dfd
>= 0 && !keep_fd
)
5164 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5171 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5175 if (sra
->array
.major_version
!= -1 ||
5176 sra
->array
.minor_version
!= -2 ||
5177 strcmp(sra
->text_version
, "imsm") != 0) {
5182 devnm
= fd2devnm(fd
);
5183 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5184 if (get_super_block(super_list
, devnm
, devname
,
5185 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5196 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5198 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5201 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5203 struct intel_super
*super
;
5207 if (test_partition(fd
))
5208 /* IMSM not allowed on partitions */
5211 free_super_imsm(st
);
5213 super
= alloc_super();
5214 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5217 /* Load hba and capabilities if they exist.
5218 * But do not preclude loading metadata in case capabilities or hba are
5219 * non-compliant and ignore_hw_compat is set.
5221 rv
= find_intel_hba_capability(fd
, super
, devname
);
5222 /* no orom/efi or non-intel hba of the disk */
5223 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5225 pr_err("No OROM/EFI properties for %s\n", devname
);
5229 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5231 /* retry the load if we might have raced against mdmon */
5233 struct mdstat_ent
*mdstat
= NULL
;
5234 char *name
= fd2kname(fd
);
5237 mdstat
= mdstat_by_component(name
);
5239 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5240 for (retry
= 0; retry
< 3; retry
++) {
5242 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5248 free_mdstat(mdstat
);
5253 pr_err("Failed to load all information sections on %s\n", devname
);
5259 if (st
->ss
== NULL
) {
5260 st
->ss
= &super_imsm
;
5261 st
->minor_version
= 0;
5262 st
->max_devs
= IMSM_MAX_DEVICES
;
5265 /* load migration record */
5266 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5267 /* Check for unsupported migration features */
5268 if (check_mpb_migr_compatibility(super
) != 0) {
5269 pr_err("Unsupported migration detected");
5271 fprintf(stderr
, " on %s\n", devname
);
5273 fprintf(stderr
, " (IMSM).\n");
5281 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5283 if (info
->level
== 1)
5285 return info
->chunk_size
>> 9;
5288 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5289 unsigned long long size
)
5291 if (info
->level
== 1)
5294 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5297 static void imsm_update_version_info(struct intel_super
*super
)
5299 /* update the version and attributes */
5300 struct imsm_super
*mpb
= super
->anchor
;
5302 struct imsm_dev
*dev
;
5303 struct imsm_map
*map
;
5306 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5307 dev
= get_imsm_dev(super
, i
);
5308 map
= get_imsm_map(dev
, MAP_0
);
5309 if (__le32_to_cpu(dev
->size_high
) > 0)
5310 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5312 /* FIXME detect when an array spans a port multiplier */
5314 mpb
->attributes
|= MPB_ATTRIB_PM
;
5317 if (mpb
->num_raid_devs
> 1 ||
5318 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5319 version
= MPB_VERSION_ATTRIBS
;
5320 switch (get_imsm_raid_level(map
)) {
5321 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5322 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5323 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5324 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5327 if (map
->num_members
>= 5)
5328 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5329 else if (dev
->status
== DEV_CLONE_N_GO
)
5330 version
= MPB_VERSION_CNG
;
5331 else if (get_imsm_raid_level(map
) == 5)
5332 version
= MPB_VERSION_RAID5
;
5333 else if (map
->num_members
>= 3)
5334 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5335 else if (get_imsm_raid_level(map
) == 1)
5336 version
= MPB_VERSION_RAID1
;
5338 version
= MPB_VERSION_RAID0
;
5340 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5344 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5346 struct imsm_super
*mpb
= super
->anchor
;
5347 char *reason
= NULL
;
5349 size_t len
= strlen(name
);
5353 while (isspace(start
[len
- 1]))
5355 while (*start
&& isspace(*start
))
5357 memmove(name
, start
, len
+ 1);
5360 if (len
> MAX_RAID_SERIAL_LEN
)
5361 reason
= "must be 16 characters or less";
5363 reason
= "must be a non-empty string";
5365 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5366 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5368 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5369 reason
= "already exists";
5374 if (reason
&& !quiet
)
5375 pr_err("imsm volume name %s\n", reason
);
5380 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5381 struct shape
*s
, char *name
,
5382 char *homehost
, int *uuid
,
5383 long long data_offset
)
5385 /* We are creating a volume inside a pre-existing container.
5386 * so st->sb is already set.
5388 struct intel_super
*super
= st
->sb
;
5389 unsigned int sector_size
= super
->sector_size
;
5390 struct imsm_super
*mpb
= super
->anchor
;
5391 struct intel_dev
*dv
;
5392 struct imsm_dev
*dev
;
5393 struct imsm_vol
*vol
;
5394 struct imsm_map
*map
;
5395 int idx
= mpb
->num_raid_devs
;
5398 unsigned long long array_blocks
;
5399 size_t size_old
, size_new
;
5400 unsigned long long num_data_stripes
;
5401 unsigned int data_disks
;
5402 unsigned long long size_per_member
;
5404 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5405 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5409 /* ensure the mpb is large enough for the new data */
5410 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5411 size_new
= disks_to_mpb_size(info
->nr_disks
);
5412 if (size_new
> size_old
) {
5414 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5416 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5417 pr_err("could not allocate new mpb\n");
5420 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5421 MIGR_REC_BUF_SECTORS
*
5422 MAX_SECTOR_SIZE
) != 0) {
5423 pr_err("could not allocate migr_rec buffer\n");
5429 memcpy(mpb_new
, mpb
, size_old
);
5432 super
->anchor
= mpb_new
;
5433 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5434 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5435 super
->len
= size_round
;
5437 super
->current_vol
= idx
;
5439 /* handle 'failed_disks' by either:
5440 * a) create dummy disk entries in the table if this the first
5441 * volume in the array. We add them here as this is the only
5442 * opportunity to add them. add_to_super_imsm_volume()
5443 * handles the non-failed disks and continues incrementing
5445 * b) validate that 'failed_disks' matches the current number
5446 * of missing disks if the container is populated
5448 if (super
->current_vol
== 0) {
5450 for (i
= 0; i
< info
->failed_disks
; i
++) {
5451 struct imsm_disk
*disk
;
5454 disk
= __get_imsm_disk(mpb
, i
);
5455 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5456 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5457 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5458 "missing:%d", (__u8
)i
);
5460 find_missing(super
);
5465 for (d
= super
->missing
; d
; d
= d
->next
)
5467 if (info
->failed_disks
> missing
) {
5468 pr_err("unable to add 'missing' disk to container\n");
5473 if (!check_name(super
, name
, 0))
5475 dv
= xmalloc(sizeof(*dv
));
5476 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5478 * Explicitly allow truncating to not confuse gcc's
5479 * -Werror=stringop-truncation
5481 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5482 memcpy(dev
->volume
, name
, namelen
);
5483 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5484 info
->layout
, info
->chunk_size
,
5485 s
->size
* BLOCKS_PER_KB
);
5486 data_disks
= get_data_disks(info
->level
, info
->layout
,
5488 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5489 size_per_member
= array_blocks
/ data_disks
;
5491 set_imsm_dev_size(dev
, array_blocks
);
5492 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5494 vol
->migr_state
= 0;
5495 set_migr_type(dev
, MIGR_INIT
);
5496 vol
->dirty
= !info
->state
;
5497 vol
->curr_migr_unit
= 0;
5498 map
= get_imsm_map(dev
, MAP_0
);
5499 set_pba_of_lba0(map
, super
->create_offset
);
5500 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5501 map
->failed_disk_num
= ~0;
5502 if (info
->level
> 0)
5503 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5504 : IMSM_T_STATE_UNINITIALIZED
);
5506 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5507 IMSM_T_STATE_NORMAL
;
5510 if (info
->level
== 1 && info
->raid_disks
> 2) {
5513 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5517 map
->raid_level
= info
->level
;
5518 if (info
->level
== 10) {
5519 map
->raid_level
= 1;
5520 map
->num_domains
= info
->raid_disks
/ 2;
5521 } else if (info
->level
== 1)
5522 map
->num_domains
= info
->raid_disks
;
5524 map
->num_domains
= 1;
5526 /* info->size is only int so use the 'size' parameter instead */
5527 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5528 num_data_stripes
/= map
->num_domains
;
5529 set_num_data_stripes(map
, num_data_stripes
);
5531 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5532 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5536 map
->num_members
= info
->raid_disks
;
5537 for (i
= 0; i
< map
->num_members
; i
++) {
5538 /* initialized in add_to_super */
5539 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5541 mpb
->num_raid_devs
++;
5542 mpb
->num_raid_devs_created
++;
5543 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5545 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5546 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5547 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5548 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5552 pr_err("imsm does not support consistency policy %s\n",
5553 map_num(consistency_policies
, s
->consistency_policy
));
5558 dv
->index
= super
->current_vol
;
5559 dv
->next
= super
->devlist
;
5560 super
->devlist
= dv
;
5562 imsm_update_version_info(super
);
5567 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5568 struct shape
*s
, char *name
,
5569 char *homehost
, int *uuid
,
5570 unsigned long long data_offset
)
5572 /* This is primarily called by Create when creating a new array.
5573 * We will then get add_to_super called for each component, and then
5574 * write_init_super called to write it out to each device.
5575 * For IMSM, Create can create on fresh devices or on a pre-existing
5577 * To create on a pre-existing array a different method will be called.
5578 * This one is just for fresh drives.
5580 struct intel_super
*super
;
5581 struct imsm_super
*mpb
;
5585 if (data_offset
!= INVALID_SECTORS
) {
5586 pr_err("data-offset not supported by imsm\n");
5591 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5595 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5597 mpb_size
= MAX_SECTOR_SIZE
;
5599 super
= alloc_super();
5601 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5606 pr_err("could not allocate superblock\n");
5609 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5610 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5611 pr_err("could not allocate migr_rec buffer\n");
5616 memset(super
->buf
, 0, mpb_size
);
5618 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5622 /* zeroing superblock */
5626 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5628 version
= (char *) mpb
->sig
;
5629 strcpy(version
, MPB_SIGNATURE
);
5630 version
+= strlen(MPB_SIGNATURE
);
5631 strcpy(version
, MPB_VERSION_RAID0
);
5636 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5638 unsigned int member_sector_size
;
5641 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5645 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5647 if (member_sector_size
!= super
->sector_size
)
5652 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5653 int fd
, char *devname
)
5655 struct intel_super
*super
= st
->sb
;
5656 struct imsm_super
*mpb
= super
->anchor
;
5657 struct imsm_disk
*_disk
;
5658 struct imsm_dev
*dev
;
5659 struct imsm_map
*map
;
5663 dev
= get_imsm_dev(super
, super
->current_vol
);
5664 map
= get_imsm_map(dev
, MAP_0
);
5666 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5667 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5673 /* we're doing autolayout so grab the pre-marked (in
5674 * validate_geometry) raid_disk
5676 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5677 if (dl
->raiddisk
== dk
->raid_disk
)
5680 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5681 if (dl
->major
== dk
->major
&&
5682 dl
->minor
== dk
->minor
)
5687 pr_err("%s is not a member of the same container\n", devname
);
5691 if (mpb
->num_disks
== 0)
5692 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5693 &super
->sector_size
))
5696 if (!drive_validate_sector_size(super
, dl
)) {
5697 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5701 /* add a pristine spare to the metadata */
5702 if (dl
->index
< 0) {
5703 dl
->index
= super
->anchor
->num_disks
;
5704 super
->anchor
->num_disks
++;
5706 /* Check the device has not already been added */
5707 slot
= get_imsm_disk_slot(map
, dl
->index
);
5709 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5710 pr_err("%s has been included in this array twice\n",
5714 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5715 dl
->disk
.status
= CONFIGURED_DISK
;
5717 /* update size of 'missing' disks to be at least as large as the
5718 * largest acitve member (we only have dummy missing disks when
5719 * creating the first volume)
5721 if (super
->current_vol
== 0) {
5722 for (df
= super
->missing
; df
; df
= df
->next
) {
5723 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5724 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5725 _disk
= __get_imsm_disk(mpb
, df
->index
);
5730 /* refresh unset/failed slots to point to valid 'missing' entries */
5731 for (df
= super
->missing
; df
; df
= df
->next
)
5732 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5733 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5735 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5737 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5738 if (is_gen_migration(dev
)) {
5739 struct imsm_map
*map2
= get_imsm_map(dev
,
5741 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5742 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5743 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5746 if ((unsigned)df
->index
==
5748 set_imsm_ord_tbl_ent(map2
,
5754 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5758 /* if we are creating the first raid device update the family number */
5759 if (super
->current_vol
== 0) {
5761 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5763 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5764 if (!_dev
|| !_disk
) {
5765 pr_err("BUG mpb setup error\n");
5771 sum
+= __gen_imsm_checksum(mpb
);
5772 mpb
->family_num
= __cpu_to_le32(sum
);
5773 mpb
->orig_family_num
= mpb
->family_num
;
5774 mpb
->creation_time
= __cpu_to_le64((__u64
)time(NULL
));
5776 super
->current_disk
= dl
;
5781 * Function marks disk as spare and restores disk serial
5782 * in case it was previously marked as failed by takeover operation
5784 * -1 : critical error
5785 * 0 : disk is marked as spare but serial is not set
5788 int mark_spare(struct dl
*disk
)
5790 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5797 if (!imsm_read_serial(disk
->fd
, NULL
, serial
, MAX_RAID_SERIAL_LEN
)) {
5798 /* Restore disk serial number, because takeover marks disk
5799 * as failed and adds to serial ':0' before it becomes
5802 serialcpy(disk
->serial
, serial
);
5803 serialcpy(disk
->disk
.serial
, serial
);
5806 disk
->disk
.status
= SPARE_DISK
;
5812 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5813 int fd
, char *devname
,
5814 unsigned long long data_offset
)
5816 struct intel_super
*super
= st
->sb
;
5818 unsigned long long size
;
5819 unsigned int member_sector_size
;
5824 /* If we are on an RAID enabled platform check that the disk is
5825 * attached to the raid controller.
5826 * We do not need to test disks attachment for container based additions,
5827 * they shall be already tested when container was created/assembled.
5829 rv
= find_intel_hba_capability(fd
, super
, devname
);
5830 /* no orom/efi or non-intel hba of the disk */
5832 dprintf("capability: %p fd: %d ret: %d\n",
5833 super
->orom
, fd
, rv
);
5837 if (super
->current_vol
>= 0)
5838 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5841 dd
= xcalloc(sizeof(*dd
), 1);
5842 dd
->major
= major(stb
.st_rdev
);
5843 dd
->minor
= minor(stb
.st_rdev
);
5844 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5847 dd
->action
= DISK_ADD
;
5848 rv
= imsm_read_serial(fd
, devname
, dd
->serial
, MAX_RAID_SERIAL_LEN
);
5850 pr_err("failed to retrieve scsi serial, aborting\n");
5856 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5857 (super
->hba
->type
== SYS_DEV_VMD
))) {
5859 char *devpath
= diskfd_to_devpath(fd
);
5860 char controller_path
[PATH_MAX
];
5863 pr_err("failed to get devpath, aborting\n");
5870 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5873 if (devpath_to_vendor(controller_path
) == 0x8086) {
5875 * If Intel's NVMe drive has serial ended with
5876 * "-A","-B","-1" or "-2" it means that this is "x8"
5877 * device (double drive on single PCIe card).
5878 * User should be warned about potential data loss.
5880 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5881 /* Skip empty character at the end */
5882 if (dd
->serial
[i
] == 0)
5885 if (((dd
->serial
[i
] == 'A') ||
5886 (dd
->serial
[i
] == 'B') ||
5887 (dd
->serial
[i
] == '1') ||
5888 (dd
->serial
[i
] == '2')) &&
5889 (dd
->serial
[i
-1] == '-'))
5890 pr_err("\tThe action you are about to take may put your data at risk.\n"
5891 "\tPlease note that x8 devices may consist of two separate x4 devices "
5892 "located on a single PCIe port.\n"
5893 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5896 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5897 !imsm_orom_has_tpv_support(super
->orom
)) {
5898 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5899 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
5906 get_dev_size(fd
, NULL
, &size
);
5907 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5909 if (super
->sector_size
== 0) {
5910 /* this a first device, so sector_size is not set yet */
5911 super
->sector_size
= member_sector_size
;
5914 /* clear migr_rec when adding disk to container */
5915 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5916 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5918 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5919 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5920 MIGR_REC_BUF_SECTORS
*member_sector_size
)
5921 perror("Write migr_rec failed");
5925 serialcpy(dd
->disk
.serial
, dd
->serial
);
5926 set_total_blocks(&dd
->disk
, size
);
5927 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5928 struct imsm_super
*mpb
= super
->anchor
;
5929 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5932 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5933 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5935 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5937 if (st
->update_tail
) {
5938 dd
->next
= super
->disk_mgmt_list
;
5939 super
->disk_mgmt_list
= dd
;
5941 dd
->next
= super
->disks
;
5943 super
->updates_pending
++;
5949 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5951 struct intel_super
*super
= st
->sb
;
5954 /* remove from super works only in mdmon - for communication
5955 * manager - monitor. Check if communication memory buffer
5958 if (!st
->update_tail
) {
5959 pr_err("shall be used in mdmon context only\n");
5962 dd
= xcalloc(1, sizeof(*dd
));
5963 dd
->major
= dk
->major
;
5964 dd
->minor
= dk
->minor
;
5967 dd
->action
= DISK_REMOVE
;
5969 dd
->next
= super
->disk_mgmt_list
;
5970 super
->disk_mgmt_list
= dd
;
5975 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5978 char buf
[MAX_SECTOR_SIZE
];
5979 struct imsm_super anchor
;
5980 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5982 /* spare records have their own family number and do not have any defined raid
5985 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5987 struct imsm_super
*mpb
= super
->anchor
;
5988 struct imsm_super
*spare
= &spare_record
.anchor
;
5992 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5993 spare
->generation_num
= __cpu_to_le32(1UL);
5994 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5995 spare
->num_disks
= 1;
5996 spare
->num_raid_devs
= 0;
5997 spare
->cache_size
= mpb
->cache_size
;
5998 spare
->pwr_cycle_count
= __cpu_to_le32(1);
6000 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
6001 MPB_SIGNATURE MPB_VERSION_RAID0
);
6003 for (d
= super
->disks
; d
; d
= d
->next
) {
6007 spare
->disk
[0] = d
->disk
;
6008 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
6009 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6011 if (super
->sector_size
== 4096)
6012 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6014 sum
= __gen_imsm_checksum(spare
);
6015 spare
->family_num
= __cpu_to_le32(sum
);
6016 spare
->orig_family_num
= 0;
6017 sum
= __gen_imsm_checksum(spare
);
6018 spare
->check_sum
= __cpu_to_le32(sum
);
6020 if (store_imsm_mpb(d
->fd
, spare
)) {
6021 pr_err("failed for device %d:%d %s\n",
6022 d
->major
, d
->minor
, strerror(errno
));
6034 static int write_super_imsm(struct supertype
*st
, int doclose
)
6036 struct intel_super
*super
= st
->sb
;
6037 unsigned int sector_size
= super
->sector_size
;
6038 struct imsm_super
*mpb
= super
->anchor
;
6044 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6046 int clear_migration_record
= 1;
6049 /* 'generation' is incremented everytime the metadata is written */
6050 generation
= __le32_to_cpu(mpb
->generation_num
);
6052 mpb
->generation_num
= __cpu_to_le32(generation
);
6054 /* fix up cases where previous mdadm releases failed to set
6057 if (mpb
->orig_family_num
== 0)
6058 mpb
->orig_family_num
= mpb
->family_num
;
6060 for (d
= super
->disks
; d
; d
= d
->next
) {
6064 mpb
->disk
[d
->index
] = d
->disk
;
6068 for (d
= super
->missing
; d
; d
= d
->next
) {
6069 mpb
->disk
[d
->index
] = d
->disk
;
6072 mpb
->num_disks
= num_disks
;
6073 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6075 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6076 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6077 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6079 imsm_copy_dev(dev
, dev2
);
6080 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6082 if (is_gen_migration(dev2
))
6083 clear_migration_record
= 0;
6086 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6089 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6090 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6092 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6094 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6095 mpb_size
+= bbm_log_size
;
6096 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6099 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6102 /* recalculate checksum */
6103 sum
= __gen_imsm_checksum(mpb
);
6104 mpb
->check_sum
= __cpu_to_le32(sum
);
6106 if (super
->clean_migration_record_by_mdmon
) {
6107 clear_migration_record
= 1;
6108 super
->clean_migration_record_by_mdmon
= 0;
6110 if (clear_migration_record
)
6111 memset(super
->migr_rec_buf
, 0,
6112 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6114 if (sector_size
== 4096)
6115 convert_to_4k(super
);
6117 /* write the mpb for disks that compose raid devices */
6118 for (d
= super
->disks
; d
; d
= d
->next
) {
6119 if (d
->index
< 0 || is_failed(&d
->disk
))
6122 if (clear_migration_record
) {
6123 unsigned long long dsize
;
6125 get_dev_size(d
->fd
, NULL
, &dsize
);
6126 if (lseek64(d
->fd
, dsize
- sector_size
,
6128 if ((unsigned int)write(d
->fd
,
6129 super
->migr_rec_buf
,
6130 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6131 MIGR_REC_BUF_SECTORS
*sector_size
)
6132 perror("Write migr_rec failed");
6136 if (store_imsm_mpb(d
->fd
, mpb
))
6138 "failed for device %d:%d (fd: %d)%s\n",
6140 d
->fd
, strerror(errno
));
6149 return write_super_imsm_spares(super
, doclose
);
6154 static int create_array(struct supertype
*st
, int dev_idx
)
6157 struct imsm_update_create_array
*u
;
6158 struct intel_super
*super
= st
->sb
;
6159 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6160 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6161 struct disk_info
*inf
;
6162 struct imsm_disk
*disk
;
6165 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6166 sizeof(*inf
) * map
->num_members
;
6168 u
->type
= update_create_array
;
6169 u
->dev_idx
= dev_idx
;
6170 imsm_copy_dev(&u
->dev
, dev
);
6171 inf
= get_disk_info(u
);
6172 for (i
= 0; i
< map
->num_members
; i
++) {
6173 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6175 disk
= get_imsm_disk(super
, idx
);
6177 disk
= get_imsm_missing(super
, idx
);
6178 serialcpy(inf
[i
].serial
, disk
->serial
);
6180 append_metadata_update(st
, u
, len
);
6185 static int mgmt_disk(struct supertype
*st
)
6187 struct intel_super
*super
= st
->sb
;
6189 struct imsm_update_add_remove_disk
*u
;
6191 if (!super
->disk_mgmt_list
)
6196 u
->type
= update_add_remove_disk
;
6197 append_metadata_update(st
, u
, len
);
6202 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6204 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6206 struct ppl_header
*ppl_hdr
= buf
;
6209 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6211 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6213 perror("Failed to seek to PPL header location");
6217 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6219 perror("Write PPL header failed");
6228 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6230 struct intel_super
*super
= st
->sb
;
6232 struct ppl_header
*ppl_hdr
;
6235 /* first clear entire ppl space */
6236 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6240 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6242 pr_err("Failed to allocate PPL header buffer\n");
6246 memset(buf
, 0, PPL_HEADER_SIZE
);
6248 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6249 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6251 if (info
->mismatch_cnt
) {
6253 * We are overwriting an invalid ppl. Make one entry with wrong
6254 * checksum to prevent the kernel from skipping resync.
6256 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6257 ppl_hdr
->entries
[0].checksum
= ~0;
6260 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6266 static int is_rebuilding(struct imsm_dev
*dev
);
6268 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6269 struct mdinfo
*disk
)
6271 struct intel_super
*super
= st
->sb
;
6273 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6275 struct ppl_header
*ppl_hdr
= NULL
;
6277 struct imsm_dev
*dev
;
6280 unsigned long long ppl_offset
= 0;
6281 unsigned long long prev_gen_num
= 0;
6283 if (disk
->disk
.raid_disk
< 0)
6286 dev
= get_imsm_dev(super
, info
->container_member
);
6287 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6288 d
= get_imsm_dl_disk(super
, idx
);
6290 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6293 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6294 pr_err("Failed to allocate PPL header buffer\n");
6300 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6303 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6305 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6307 perror("Failed to seek to PPL header location");
6312 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6313 perror("Read PPL header failed");
6320 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6321 ppl_hdr
->checksum
= 0;
6323 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6324 dprintf("Wrong PPL header checksum on %s\n",
6329 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6330 /* previous was newest, it was already checked */
6334 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6335 super
->anchor
->orig_family_num
)) {
6336 dprintf("Wrong PPL header signature on %s\n",
6343 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6345 ppl_offset
+= PPL_HEADER_SIZE
;
6346 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6348 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6351 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6363 * Update metadata to use mutliple PPLs area (1MB).
6364 * This is done once for all RAID members
6366 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6367 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6369 struct mdinfo
*member_dev
;
6371 sprintf(subarray
, "%d", info
->container_member
);
6373 if (mdmon_running(st
->container_devnm
))
6374 st
->update_tail
= &st
->updates
;
6376 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6377 pr_err("Failed to update subarray %s\n",
6380 if (st
->update_tail
)
6381 flush_metadata_updates(st
);
6383 st
->ss
->sync_metadata(st
);
6384 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6385 for (member_dev
= info
->devs
; member_dev
;
6386 member_dev
= member_dev
->next
)
6387 member_dev
->ppl_size
=
6388 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6393 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6395 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6396 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6397 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6398 (is_rebuilding(dev
) &&
6399 dev
->vol
.curr_migr_unit
== 0 &&
6400 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6401 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6403 info
->mismatch_cnt
++;
6404 } else if (ret
== 0 &&
6405 ppl_hdr
->entries_count
== 0 &&
6406 is_rebuilding(dev
) &&
6407 info
->resync_start
== 0) {
6409 * The header has no entries - add a single empty entry and
6410 * rewrite the header to prevent the kernel from going into
6411 * resync after an interrupted rebuild.
6413 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6414 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6422 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6424 struct intel_super
*super
= st
->sb
;
6428 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6429 info
->array
.level
!= 5)
6432 for (d
= super
->disks
; d
; d
= d
->next
) {
6433 if (d
->index
< 0 || is_failed(&d
->disk
))
6436 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6444 static int write_init_super_imsm(struct supertype
*st
)
6446 struct intel_super
*super
= st
->sb
;
6447 int current_vol
= super
->current_vol
;
6451 getinfo_super_imsm(st
, &info
, NULL
);
6453 /* we are done with current_vol reset it to point st at the container */
6454 super
->current_vol
= -1;
6456 if (st
->update_tail
) {
6457 /* queue the recently created array / added disk
6458 * as a metadata update */
6460 /* determine if we are creating a volume or adding a disk */
6461 if (current_vol
< 0) {
6462 /* in the mgmt (add/remove) disk case we are running
6463 * in mdmon context, so don't close fd's
6467 rv
= write_init_ppl_imsm_all(st
, &info
);
6469 rv
= create_array(st
, current_vol
);
6473 for (d
= super
->disks
; d
; d
= d
->next
)
6474 Kill(d
->devname
, NULL
, 0, -1, 1);
6475 if (current_vol
>= 0)
6476 rv
= write_init_ppl_imsm_all(st
, &info
);
6478 rv
= write_super_imsm(st
, 1);
6484 static int store_super_imsm(struct supertype
*st
, int fd
)
6486 struct intel_super
*super
= st
->sb
;
6487 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6492 if (super
->sector_size
== 4096)
6493 convert_to_4k(super
);
6494 return store_imsm_mpb(fd
, mpb
);
6497 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6498 int layout
, int raiddisks
, int chunk
,
6499 unsigned long long size
,
6500 unsigned long long data_offset
,
6502 unsigned long long *freesize
,
6506 unsigned long long ldsize
;
6507 struct intel_super
*super
;
6510 if (level
!= LEVEL_CONTAINER
)
6515 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6518 pr_err("imsm: Cannot open %s: %s\n",
6519 dev
, strerror(errno
));
6522 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6527 /* capabilities retrieve could be possible
6528 * note that there is no fd for the disks in array.
6530 super
= alloc_super();
6535 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6541 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6545 fd2devname(fd
, str
);
6546 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6547 fd
, str
, super
->orom
, rv
, raiddisks
);
6549 /* no orom/efi or non-intel hba of the disk */
6556 if (raiddisks
> super
->orom
->tds
) {
6558 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6559 raiddisks
, super
->orom
->tds
);
6563 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6564 (ldsize
>> 9) >> 32 > 0) {
6566 pr_err("%s exceeds maximum platform supported size\n", dev
);
6572 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6578 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6580 const unsigned long long base_start
= e
[*idx
].start
;
6581 unsigned long long end
= base_start
+ e
[*idx
].size
;
6584 if (base_start
== end
)
6588 for (i
= *idx
; i
< num_extents
; i
++) {
6589 /* extend overlapping extents */
6590 if (e
[i
].start
>= base_start
&&
6591 e
[i
].start
<= end
) {
6594 if (e
[i
].start
+ e
[i
].size
> end
)
6595 end
= e
[i
].start
+ e
[i
].size
;
6596 } else if (e
[i
].start
> end
) {
6602 return end
- base_start
;
6605 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6607 /* build a composite disk with all known extents and generate a new
6608 * 'maxsize' given the "all disks in an array must share a common start
6609 * offset" constraint
6611 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6615 unsigned long long pos
;
6616 unsigned long long start
= 0;
6617 unsigned long long maxsize
;
6618 unsigned long reserve
;
6620 /* coalesce and sort all extents. also, check to see if we need to
6621 * reserve space between member arrays
6624 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6627 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6630 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6635 while (i
< sum_extents
) {
6636 e
[j
].start
= e
[i
].start
;
6637 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6639 if (e
[j
-1].size
== 0)
6648 unsigned long long esize
;
6650 esize
= e
[i
].start
- pos
;
6651 if (esize
>= maxsize
) {
6656 pos
= e
[i
].start
+ e
[i
].size
;
6658 } while (e
[i
-1].size
);
6664 /* FIXME assumes volume at offset 0 is the first volume in a
6667 if (start_extent
> 0)
6668 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6672 if (maxsize
< reserve
)
6675 super
->create_offset
= ~((unsigned long long) 0);
6676 if (start
+ reserve
> super
->create_offset
)
6677 return 0; /* start overflows create_offset */
6678 super
->create_offset
= start
+ reserve
;
6680 return maxsize
- reserve
;
6683 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6685 if (level
< 0 || level
== 6 || level
== 4)
6688 /* if we have an orom prevent invalid raid levels */
6691 case 0: return imsm_orom_has_raid0(orom
);
6694 return imsm_orom_has_raid1e(orom
);
6695 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6696 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6697 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6700 return 1; /* not on an Intel RAID platform so anything goes */
6706 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6707 int dpa
, int verbose
)
6709 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6710 struct mdstat_ent
*memb
;
6716 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6717 if (memb
->metadata_version
&&
6718 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6719 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6720 !is_subarray(memb
->metadata_version
+9) &&
6722 struct dev_member
*dev
= memb
->members
;
6724 while(dev
&& (fd
< 0)) {
6725 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6726 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6728 fd
= open(path
, O_RDONLY
, 0);
6729 if (num
<= 0 || fd
< 0) {
6730 pr_vrb("Cannot open %s: %s\n",
6731 dev
->name
, strerror(errno
));
6737 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6738 struct mdstat_ent
*vol
;
6739 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6740 if (vol
->active
> 0 &&
6741 vol
->metadata_version
&&
6742 is_container_member(vol
, memb
->devnm
)) {
6747 if (*devlist
&& (found
< dpa
)) {
6748 dv
= xcalloc(1, sizeof(*dv
));
6749 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6750 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6753 dv
->next
= *devlist
;
6761 free_mdstat(mdstat
);
6766 static struct md_list
*
6767 get_loop_devices(void)
6770 struct md_list
*devlist
= NULL
;
6773 for(i
= 0; i
< 12; i
++) {
6774 dv
= xcalloc(1, sizeof(*dv
));
6775 dv
->devname
= xmalloc(40);
6776 sprintf(dv
->devname
, "/dev/loop%d", i
);
6784 static struct md_list
*
6785 get_devices(const char *hba_path
)
6787 struct md_list
*devlist
= NULL
;
6794 devlist
= get_loop_devices();
6797 /* scroll through /sys/dev/block looking for devices attached to
6800 dir
= opendir("/sys/dev/block");
6801 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6806 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6808 path
= devt_to_devpath(makedev(major
, minor
));
6811 if (!path_attached_to_hba(path
, hba_path
)) {
6818 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6820 fd2devname(fd
, buf
);
6823 pr_err("cannot open device: %s\n",
6828 dv
= xcalloc(1, sizeof(*dv
));
6829 dv
->devname
= xstrdup(buf
);
6836 devlist
= devlist
->next
;
6846 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6847 int verbose
, int *found
)
6849 struct md_list
*tmpdev
;
6851 struct supertype
*st
;
6853 /* first walk the list of devices to find a consistent set
6854 * that match the criterea, if that is possible.
6855 * We flag the ones we like with 'used'.
6858 st
= match_metadata_desc_imsm("imsm");
6860 pr_vrb("cannot allocate memory for imsm supertype\n");
6864 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6865 char *devname
= tmpdev
->devname
;
6867 struct supertype
*tst
;
6869 if (tmpdev
->used
> 1)
6871 tst
= dup_super(st
);
6873 pr_vrb("cannot allocate memory for imsm supertype\n");
6876 tmpdev
->container
= 0;
6877 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6879 dprintf("cannot open device %s: %s\n",
6880 devname
, strerror(errno
));
6882 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
6884 } else if (must_be_container(dfd
)) {
6885 struct supertype
*cst
;
6886 cst
= super_by_fd(dfd
, NULL
);
6888 dprintf("cannot recognize container type %s\n",
6891 } else if (tst
->ss
!= st
->ss
) {
6892 dprintf("non-imsm container - ignore it: %s\n",
6895 } else if (!tst
->ss
->load_container
||
6896 tst
->ss
->load_container(tst
, dfd
, NULL
))
6899 tmpdev
->container
= 1;
6902 cst
->ss
->free_super(cst
);
6904 tmpdev
->st_rdev
= rdev
;
6905 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6906 dprintf("no RAID superblock on %s\n",
6909 } else if (tst
->ss
->compare_super
== NULL
) {
6910 dprintf("Cannot assemble %s metadata on %s\n",
6911 tst
->ss
->name
, devname
);
6917 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6918 /* Ignore unrecognised devices during auto-assembly */
6923 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6925 if (st
->minor_version
== -1)
6926 st
->minor_version
= tst
->minor_version
;
6928 if (memcmp(info
.uuid
, uuid_zero
,
6929 sizeof(int[4])) == 0) {
6930 /* this is a floating spare. It cannot define
6931 * an array unless there are no more arrays of
6932 * this type to be found. It can be included
6933 * in an array of this type though.
6939 if (st
->ss
!= tst
->ss
||
6940 st
->minor_version
!= tst
->minor_version
||
6941 st
->ss
->compare_super(st
, tst
) != 0) {
6942 /* Some mismatch. If exactly one array matches this host,
6943 * we can resolve on that one.
6944 * Or, if we are auto assembling, we just ignore the second
6947 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6953 dprintf("found: devname: %s\n", devname
);
6957 tst
->ss
->free_super(tst
);
6961 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6962 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6963 for (iter
= head
; iter
; iter
= iter
->next
) {
6964 dprintf("content->text_version: %s vol\n",
6965 iter
->text_version
);
6966 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6967 /* do not assemble arrays with unsupported
6969 dprintf("Cannot activate member %s.\n",
6970 iter
->text_version
);
6977 dprintf("No valid super block on device list: err: %d %p\n",
6981 dprintf("no more devices to examine\n");
6984 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6985 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6987 if (count
< tmpdev
->found
)
6990 count
-= tmpdev
->found
;
6993 if (tmpdev
->used
== 1)
6998 st
->ss
->free_super(st
);
7002 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
7005 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
7007 const struct orom_entry
*entry
;
7008 struct devid_list
*dv
, *devid_list
;
7013 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7014 if (strstr(idev
->path
, hba_path
))
7018 if (!idev
|| !idev
->dev_id
)
7021 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7023 if (!entry
|| !entry
->devid_list
)
7026 devid_list
= entry
->devid_list
;
7027 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7028 struct md_list
*devlist
;
7029 struct sys_dev
*device
= NULL
;
7034 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7036 device
= device_by_id(dv
->devid
);
7039 hpath
= device
->path
;
7043 devlist
= get_devices(hpath
);
7044 /* if no intel devices return zero volumes */
7045 if (devlist
== NULL
)
7048 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7050 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7051 if (devlist
== NULL
)
7055 count
+= count_volumes_list(devlist
,
7059 dprintf("found %d count: %d\n", found
, count
);
7062 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7065 struct md_list
*dv
= devlist
;
7066 devlist
= devlist
->next
;
7074 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7078 if (hba
->type
== SYS_DEV_VMD
) {
7079 struct sys_dev
*dev
;
7082 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7083 if (dev
->type
== SYS_DEV_VMD
)
7084 count
+= __count_volumes(dev
->path
, dpa
,
7089 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7092 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7094 /* up to 512 if the plaform supports it, otherwise the platform max.
7095 * 128 if no platform detected
7097 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7099 return min(512, (1 << fs
));
7103 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7104 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7106 /* check/set platform and metadata limits/defaults */
7107 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7108 pr_vrb("platform supports a maximum of %d disks per array\n",
7113 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7114 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7115 pr_vrb("platform does not support raid%d with %d disk%s\n",
7116 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7120 if (*chunk
== 0 || *chunk
== UnSet
)
7121 *chunk
= imsm_default_chunk(super
->orom
);
7123 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7124 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7128 if (layout
!= imsm_level_to_layout(level
)) {
7130 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7131 else if (level
== 10)
7132 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7134 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7139 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7140 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7141 pr_vrb("platform does not support a volume size over 2TB\n");
7148 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7149 * FIX ME add ahci details
7151 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7152 int layout
, int raiddisks
, int *chunk
,
7153 unsigned long long size
,
7154 unsigned long long data_offset
,
7156 unsigned long long *freesize
,
7160 struct intel_super
*super
= st
->sb
;
7161 struct imsm_super
*mpb
;
7163 unsigned long long pos
= 0;
7164 unsigned long long maxsize
;
7168 /* We must have the container info already read in. */
7172 mpb
= super
->anchor
;
7174 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7175 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7179 /* General test: make sure there is space for
7180 * 'raiddisks' device extents of size 'size' at a given
7183 unsigned long long minsize
= size
;
7184 unsigned long long start_offset
= MaxSector
;
7187 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7188 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7193 e
= get_extents(super
, dl
, 0);
7196 unsigned long long esize
;
7197 esize
= e
[i
].start
- pos
;
7198 if (esize
>= minsize
)
7200 if (found
&& start_offset
== MaxSector
) {
7203 } else if (found
&& pos
!= start_offset
) {
7207 pos
= e
[i
].start
+ e
[i
].size
;
7209 } while (e
[i
-1].size
);
7214 if (dcnt
< raiddisks
) {
7216 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7223 /* This device must be a member of the set */
7224 if (!stat_is_blkdev(dev
, &rdev
))
7226 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7227 if (dl
->major
== (int)major(rdev
) &&
7228 dl
->minor
== (int)minor(rdev
))
7233 pr_err("%s is not in the same imsm set\n", dev
);
7235 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7236 /* If a volume is present then the current creation attempt
7237 * cannot incorporate new spares because the orom may not
7238 * understand this configuration (all member disks must be
7239 * members of each array in the container).
7241 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7242 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7244 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7245 mpb
->num_disks
!= raiddisks
) {
7246 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7250 /* retrieve the largest free space block */
7251 e
= get_extents(super
, dl
, 0);
7256 unsigned long long esize
;
7258 esize
= e
[i
].start
- pos
;
7259 if (esize
>= maxsize
)
7261 pos
= e
[i
].start
+ e
[i
].size
;
7263 } while (e
[i
-1].size
);
7268 pr_err("unable to determine free space for: %s\n",
7272 if (maxsize
< size
) {
7274 pr_err("%s not enough space (%llu < %llu)\n",
7275 dev
, maxsize
, size
);
7279 /* count total number of extents for merge */
7281 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7283 i
+= dl
->extent_cnt
;
7285 maxsize
= merge_extents(super
, i
);
7287 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7288 pr_err("attempting to create a second volume with size less then remaining space.\n");
7290 if (maxsize
< size
|| maxsize
== 0) {
7293 pr_err("no free space left on device. Aborting...\n");
7295 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7301 *freesize
= maxsize
;
7304 int count
= count_volumes(super
->hba
,
7305 super
->orom
->dpa
, verbose
);
7306 if (super
->orom
->vphba
<= count
) {
7307 pr_vrb("platform does not support more than %d raid volumes.\n",
7308 super
->orom
->vphba
);
7315 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7316 unsigned long long size
, int chunk
,
7317 unsigned long long *freesize
)
7319 struct intel_super
*super
= st
->sb
;
7320 struct imsm_super
*mpb
= super
->anchor
;
7325 unsigned long long maxsize
;
7326 unsigned long long minsize
;
7330 /* find the largest common start free region of the possible disks */
7334 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7340 /* don't activate new spares if we are orom constrained
7341 * and there is already a volume active in the container
7343 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7346 e
= get_extents(super
, dl
, 0);
7349 for (i
= 1; e
[i
-1].size
; i
++)
7357 maxsize
= merge_extents(super
, extent_cnt
);
7361 minsize
= chunk
* 2;
7363 if (cnt
< raiddisks
||
7364 (super
->orom
&& used
&& used
!= raiddisks
) ||
7365 maxsize
< minsize
||
7367 pr_err("not enough devices with space to create array.\n");
7368 return 0; /* No enough free spaces large enough */
7379 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7380 pr_err("attempting to create a second volume with size less then remaining space.\n");
7382 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7384 dl
->raiddisk
= cnt
++;
7388 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7393 static int reserve_space(struct supertype
*st
, int raiddisks
,
7394 unsigned long long size
, int chunk
,
7395 unsigned long long *freesize
)
7397 struct intel_super
*super
= st
->sb
;
7402 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7405 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7407 dl
->raiddisk
= cnt
++;
7414 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7415 int raiddisks
, int *chunk
, unsigned long long size
,
7416 unsigned long long data_offset
,
7417 char *dev
, unsigned long long *freesize
,
7418 int consistency_policy
, int verbose
)
7425 * if given unused devices create a container
7426 * if given given devices in a container create a member volume
7428 if (level
== LEVEL_CONTAINER
) {
7429 /* Must be a fresh device to add to a container */
7430 return validate_geometry_imsm_container(st
, level
, layout
,
7439 * Size is given in sectors.
7441 if (size
&& (size
< 2048)) {
7442 pr_err("Given size must be greater than 1M.\n");
7443 /* Depends on algorithm in Create.c :
7444 * if container was given (dev == NULL) return -1,
7445 * if block device was given ( dev != NULL) return 0.
7447 return dev
? -1 : 0;
7452 struct intel_super
*super
= st
->sb
;
7453 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7454 raiddisks
, chunk
, size
,
7457 /* we are being asked to automatically layout a
7458 * new volume based on the current contents of
7459 * the container. If the the parameters can be
7460 * satisfied reserve_space will record the disks,
7461 * start offset, and size of the volume to be
7462 * created. add_to_super and getinfo_super
7463 * detect when autolayout is in progress.
7465 /* assuming that freesize is always given when array is
7467 if (super
->orom
&& freesize
) {
7469 count
= count_volumes(super
->hba
,
7470 super
->orom
->dpa
, verbose
);
7471 if (super
->orom
->vphba
<= count
) {
7472 pr_vrb("platform does not support more than %d raid volumes.\n",
7473 super
->orom
->vphba
);
7478 return reserve_space(st
, raiddisks
, size
,
7484 /* creating in a given container */
7485 return validate_geometry_imsm_volume(st
, level
, layout
,
7486 raiddisks
, chunk
, size
,
7488 dev
, freesize
, verbose
);
7491 /* This device needs to be a device in an 'imsm' container */
7492 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7495 pr_err("Cannot create this array on device %s\n",
7500 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7502 pr_err("Cannot open %s: %s\n",
7503 dev
, strerror(errno
));
7506 /* Well, it is in use by someone, maybe an 'imsm' container. */
7507 cfd
= open_container(fd
);
7511 pr_err("Cannot use %s: It is busy\n",
7515 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7516 if (sra
&& sra
->array
.major_version
== -1 &&
7517 strcmp(sra
->text_version
, "imsm") == 0)
7521 /* This is a member of a imsm container. Load the container
7522 * and try to create a volume
7524 struct intel_super
*super
;
7526 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7528 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7530 return validate_geometry_imsm_volume(st
, level
, layout
,
7532 size
, data_offset
, dev
,
7539 pr_err("failed container membership check\n");
7545 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7547 struct intel_super
*super
= st
->sb
;
7549 if (level
&& *level
== UnSet
)
7550 *level
= LEVEL_CONTAINER
;
7552 if (level
&& layout
&& *layout
== UnSet
)
7553 *layout
= imsm_level_to_layout(*level
);
7555 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7556 *chunk
= imsm_default_chunk(super
->orom
);
7559 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7561 static int kill_subarray_imsm(struct supertype
*st
, char *subarray_id
)
7563 /* remove the subarray currently referenced by subarray_id */
7565 struct intel_dev
**dp
;
7566 struct intel_super
*super
= st
->sb
;
7567 __u8 current_vol
= strtoul(subarray_id
, NULL
, 10);
7568 struct imsm_super
*mpb
= super
->anchor
;
7570 if (mpb
->num_raid_devs
== 0)
7573 /* block deletions that would change the uuid of active subarrays
7575 * FIXME when immutable ids are available, but note that we'll
7576 * also need to fixup the invalidated/active subarray indexes in
7579 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7582 if (i
< current_vol
)
7584 sprintf(subarray
, "%u", i
);
7585 if (is_subarray_active(subarray
, st
->devnm
)) {
7586 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7593 if (st
->update_tail
) {
7594 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7596 u
->type
= update_kill_array
;
7597 u
->dev_idx
= current_vol
;
7598 append_metadata_update(st
, u
, sizeof(*u
));
7603 for (dp
= &super
->devlist
; *dp
;)
7604 if ((*dp
)->index
== current_vol
) {
7607 handle_missing(super
, (*dp
)->dev
);
7608 if ((*dp
)->index
> current_vol
)
7613 /* no more raid devices, all active components are now spares,
7614 * but of course failed are still failed
7616 if (--mpb
->num_raid_devs
== 0) {
7619 for (d
= super
->disks
; d
; d
= d
->next
)
7624 super
->updates_pending
++;
7629 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7630 char *update
, struct mddev_ident
*ident
)
7632 /* update the subarray currently referenced by ->current_vol */
7633 struct intel_super
*super
= st
->sb
;
7634 struct imsm_super
*mpb
= super
->anchor
;
7636 if (strcmp(update
, "name") == 0) {
7637 char *name
= ident
->name
;
7641 if (is_subarray_active(subarray
, st
->devnm
)) {
7642 pr_err("Unable to update name of active subarray\n");
7646 if (!check_name(super
, name
, 0))
7649 vol
= strtoul(subarray
, &ep
, 10);
7650 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7653 if (st
->update_tail
) {
7654 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7656 u
->type
= update_rename_array
;
7658 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7659 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7660 append_metadata_update(st
, u
, sizeof(*u
));
7662 struct imsm_dev
*dev
;
7665 dev
= get_imsm_dev(super
, vol
);
7666 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7667 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7668 memcpy(dev
->volume
, name
, namelen
);
7669 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7670 dev
= get_imsm_dev(super
, i
);
7671 handle_missing(super
, dev
);
7673 super
->updates_pending
++;
7675 } else if (strcmp(update
, "ppl") == 0 ||
7676 strcmp(update
, "no-ppl") == 0) {
7679 int vol
= strtoul(subarray
, &ep
, 10);
7681 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7684 if (strcmp(update
, "ppl") == 0)
7685 new_policy
= RWH_MULTIPLE_DISTRIBUTED
;
7687 new_policy
= RWH_MULTIPLE_OFF
;
7689 if (st
->update_tail
) {
7690 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7692 u
->type
= update_rwh_policy
;
7694 u
->new_policy
= new_policy
;
7695 append_metadata_update(st
, u
, sizeof(*u
));
7697 struct imsm_dev
*dev
;
7699 dev
= get_imsm_dev(super
, vol
);
7700 dev
->rwh_policy
= new_policy
;
7701 super
->updates_pending
++;
7709 static int is_gen_migration(struct imsm_dev
*dev
)
7714 if (!dev
->vol
.migr_state
)
7717 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7723 static int is_rebuilding(struct imsm_dev
*dev
)
7725 struct imsm_map
*migr_map
;
7727 if (!dev
->vol
.migr_state
)
7730 if (migr_type(dev
) != MIGR_REBUILD
)
7733 migr_map
= get_imsm_map(dev
, MAP_1
);
7735 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7741 static int is_initializing(struct imsm_dev
*dev
)
7743 struct imsm_map
*migr_map
;
7745 if (!dev
->vol
.migr_state
)
7748 if (migr_type(dev
) != MIGR_INIT
)
7751 migr_map
= get_imsm_map(dev
, MAP_1
);
7753 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7759 static void update_recovery_start(struct intel_super
*super
,
7760 struct imsm_dev
*dev
,
7761 struct mdinfo
*array
)
7763 struct mdinfo
*rebuild
= NULL
;
7767 if (!is_rebuilding(dev
))
7770 /* Find the rebuild target, but punt on the dual rebuild case */
7771 for (d
= array
->devs
; d
; d
= d
->next
)
7772 if (d
->recovery_start
== 0) {
7779 /* (?) none of the disks are marked with
7780 * IMSM_ORD_REBUILD, so assume they are missing and the
7781 * disk_ord_tbl was not correctly updated
7783 dprintf("failed to locate out-of-sync disk\n");
7787 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7788 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7791 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7793 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7795 /* Given a container loaded by load_super_imsm_all,
7796 * extract information about all the arrays into
7798 * If 'subarray' is given, just extract info about that array.
7800 * For each imsm_dev create an mdinfo, fill it in,
7801 * then look for matching devices in super->disks
7802 * and create appropriate device mdinfo.
7804 struct intel_super
*super
= st
->sb
;
7805 struct imsm_super
*mpb
= super
->anchor
;
7806 struct mdinfo
*rest
= NULL
;
7810 int spare_disks
= 0;
7811 int current_vol
= super
->current_vol
;
7813 /* do not assemble arrays when not all attributes are supported */
7814 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7816 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7819 /* count spare devices, not used in maps
7821 for (d
= super
->disks
; d
; d
= d
->next
)
7825 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7826 struct imsm_dev
*dev
;
7827 struct imsm_map
*map
;
7828 struct imsm_map
*map2
;
7829 struct mdinfo
*this;
7836 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7839 dev
= get_imsm_dev(super
, i
);
7840 map
= get_imsm_map(dev
, MAP_0
);
7841 map2
= get_imsm_map(dev
, MAP_1
);
7842 level
= get_imsm_raid_level(map
);
7844 /* do not publish arrays that are in the middle of an
7845 * unsupported migration
7847 if (dev
->vol
.migr_state
&&
7848 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7849 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7853 /* do not publish arrays that are not support by controller's
7857 this = xmalloc(sizeof(*this));
7859 super
->current_vol
= i
;
7860 getinfo_super_imsm_volume(st
, this, NULL
);
7862 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7863 /* mdadm does not support all metadata features- set the bit in all arrays state */
7864 if (!validate_geometry_imsm_orom(super
,
7865 level
, /* RAID level */
7866 imsm_level_to_layout(level
),
7867 map
->num_members
, /* raid disks */
7868 &chunk
, imsm_dev_size(dev
),
7870 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7872 this->array
.state
|=
7873 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7874 (1<<MD_SB_BLOCK_VOLUME
);
7877 /* if array has bad blocks, set suitable bit in all arrays state */
7879 this->array
.state
|=
7880 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7881 (1<<MD_SB_BLOCK_VOLUME
);
7883 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7884 unsigned long long recovery_start
;
7885 struct mdinfo
*info_d
;
7893 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7894 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7895 for (d
= super
->disks
; d
; d
= d
->next
)
7896 if (d
->index
== idx
)
7899 recovery_start
= MaxSector
;
7902 if (d
&& is_failed(&d
->disk
))
7904 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
7906 if (!(ord
& IMSM_ORD_REBUILD
))
7907 this->array
.working_disks
++;
7909 * if we skip some disks the array will be assmebled degraded;
7910 * reset resync start to avoid a dirty-degraded
7911 * situation when performing the intial sync
7916 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
7917 if ((!able_to_resync(level
, missing
) ||
7918 recovery_start
== 0))
7919 this->resync_start
= MaxSector
;
7922 * FIXME handle dirty degraded
7929 info_d
= xcalloc(1, sizeof(*info_d
));
7930 info_d
->next
= this->devs
;
7931 this->devs
= info_d
;
7933 info_d
->disk
.number
= d
->index
;
7934 info_d
->disk
.major
= d
->major
;
7935 info_d
->disk
.minor
= d
->minor
;
7936 info_d
->disk
.raid_disk
= slot
;
7937 info_d
->recovery_start
= recovery_start
;
7939 if (slot
< map2
->num_members
)
7940 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7942 this->array
.spare_disks
++;
7944 if (slot
< map
->num_members
)
7945 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7947 this->array
.spare_disks
++;
7950 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7951 info_d
->data_offset
= pba_of_lba0(map
);
7952 info_d
->component_size
= calc_component_size(map
, dev
);
7954 if (map
->raid_level
== 5) {
7955 info_d
->ppl_sector
= this->ppl_sector
;
7956 info_d
->ppl_size
= this->ppl_size
;
7957 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
7958 recovery_start
== 0)
7959 this->resync_start
= 0;
7962 info_d
->bb
.supported
= 1;
7963 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7964 info_d
->data_offset
,
7965 info_d
->component_size
,
7968 /* now that the disk list is up-to-date fixup recovery_start */
7969 update_recovery_start(super
, dev
, this);
7970 this->array
.spare_disks
+= spare_disks
;
7972 /* check for reshape */
7973 if (this->reshape_active
== 1)
7974 recover_backup_imsm(st
, this);
7978 super
->current_vol
= current_vol
;
7982 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7983 int failed
, int look_in_map
)
7985 struct imsm_map
*map
;
7987 map
= get_imsm_map(dev
, look_in_map
);
7990 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
7991 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
7993 switch (get_imsm_raid_level(map
)) {
7995 return IMSM_T_STATE_FAILED
;
7998 if (failed
< map
->num_members
)
7999 return IMSM_T_STATE_DEGRADED
;
8001 return IMSM_T_STATE_FAILED
;
8006 * check to see if any mirrors have failed, otherwise we
8007 * are degraded. Even numbered slots are mirrored on
8011 /* gcc -Os complains that this is unused */
8012 int insync
= insync
;
8014 for (i
= 0; i
< map
->num_members
; i
++) {
8015 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8016 int idx
= ord_to_idx(ord
);
8017 struct imsm_disk
*disk
;
8019 /* reset the potential in-sync count on even-numbered
8020 * slots. num_copies is always 2 for imsm raid10
8025 disk
= get_imsm_disk(super
, idx
);
8026 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8029 /* no in-sync disks left in this mirror the
8033 return IMSM_T_STATE_FAILED
;
8036 return IMSM_T_STATE_DEGRADED
;
8040 return IMSM_T_STATE_DEGRADED
;
8042 return IMSM_T_STATE_FAILED
;
8048 return map
->map_state
;
8051 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8056 struct imsm_disk
*disk
;
8057 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8058 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8059 struct imsm_map
*map_for_loop
;
8064 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8065 * disks that are being rebuilt. New failures are recorded to
8066 * map[0]. So we look through all the disks we started with and
8067 * see if any failures are still present, or if any new ones
8071 if (prev
&& (map
->num_members
< prev
->num_members
))
8072 map_for_loop
= prev
;
8074 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8076 /* when MAP_X is passed both maps failures are counted
8079 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8080 i
< prev
->num_members
) {
8081 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8082 idx_1
= ord_to_idx(ord
);
8084 disk
= get_imsm_disk(super
, idx_1
);
8085 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8088 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8089 i
< map
->num_members
) {
8090 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8091 idx
= ord_to_idx(ord
);
8094 disk
= get_imsm_disk(super
, idx
);
8095 if (!disk
|| is_failed(disk
) ||
8096 ord
& IMSM_ORD_REBUILD
)
8105 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8108 struct intel_super
*super
= c
->sb
;
8109 struct imsm_super
*mpb
= super
->anchor
;
8110 struct imsm_update_prealloc_bb_mem u
;
8112 if (atoi(inst
) >= mpb
->num_raid_devs
) {
8113 pr_err("subarry index %d, out of range\n", atoi(inst
));
8117 dprintf("imsm: open_new %s\n", inst
);
8118 a
->info
.container_member
= atoi(inst
);
8120 u
.type
= update_prealloc_badblocks_mem
;
8121 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8126 static int is_resyncing(struct imsm_dev
*dev
)
8128 struct imsm_map
*migr_map
;
8130 if (!dev
->vol
.migr_state
)
8133 if (migr_type(dev
) == MIGR_INIT
||
8134 migr_type(dev
) == MIGR_REPAIR
)
8137 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8140 migr_map
= get_imsm_map(dev
, MAP_1
);
8142 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8143 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8149 /* return true if we recorded new information */
8150 static int mark_failure(struct intel_super
*super
,
8151 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8155 struct imsm_map
*map
;
8156 char buf
[MAX_RAID_SERIAL_LEN
+3];
8157 unsigned int len
, shift
= 0;
8159 /* new failures are always set in map[0] */
8160 map
= get_imsm_map(dev
, MAP_0
);
8162 slot
= get_imsm_disk_slot(map
, idx
);
8166 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8167 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8170 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8171 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8173 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8174 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8175 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8177 disk
->status
|= FAILED_DISK
;
8178 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8179 /* mark failures in second map if second map exists and this disk
8181 * This is valid for migration, initialization and rebuild
8183 if (dev
->vol
.migr_state
) {
8184 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8185 int slot2
= get_imsm_disk_slot(map2
, idx
);
8187 if (slot2
< map2
->num_members
&& slot2
>= 0)
8188 set_imsm_ord_tbl_ent(map2
, slot2
,
8189 idx
| IMSM_ORD_REBUILD
);
8191 if (map
->failed_disk_num
== 0xff ||
8192 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8193 map
->failed_disk_num
= slot
;
8195 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8200 static void mark_missing(struct intel_super
*super
,
8201 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8203 mark_failure(super
, dev
, disk
, idx
);
8205 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8208 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8209 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8212 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8216 if (!super
->missing
)
8219 /* When orom adds replacement for missing disk it does
8220 * not remove entry of missing disk, but just updates map with
8221 * new added disk. So it is not enough just to test if there is
8222 * any missing disk, we have to look if there are any failed disks
8223 * in map to stop migration */
8225 dprintf("imsm: mark missing\n");
8226 /* end process for initialization and rebuild only
8228 if (is_gen_migration(dev
) == 0) {
8229 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8233 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8234 struct imsm_map
*map1
;
8235 int i
, ord
, ord_map1
;
8238 for (i
= 0; i
< map
->num_members
; i
++) {
8239 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8240 if (!(ord
& IMSM_ORD_REBUILD
))
8243 map1
= get_imsm_map(dev
, MAP_1
);
8247 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8248 if (ord_map1
& IMSM_ORD_REBUILD
)
8253 map_state
= imsm_check_degraded(super
, dev
,
8255 end_migration(dev
, super
, map_state
);
8259 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8260 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8261 super
->updates_pending
++;
8264 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8267 unsigned long long array_blocks
;
8268 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8269 int used_disks
= imsm_num_data_members(map
);
8271 if (used_disks
== 0) {
8272 /* when problems occures
8273 * return current array_blocks value
8275 array_blocks
= imsm_dev_size(dev
);
8277 return array_blocks
;
8280 /* set array size in metadata
8283 /* OLCE size change is caused by added disks
8285 array_blocks
= per_dev_array_size(map
) * used_disks
;
8287 /* Online Volume Size Change
8288 * Using available free space
8290 array_blocks
= new_size
;
8292 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8293 set_imsm_dev_size(dev
, array_blocks
);
8295 return array_blocks
;
8298 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8300 static void imsm_progress_container_reshape(struct intel_super
*super
)
8302 /* if no device has a migr_state, but some device has a
8303 * different number of members than the previous device, start
8304 * changing the number of devices in this device to match
8307 struct imsm_super
*mpb
= super
->anchor
;
8308 int prev_disks
= -1;
8312 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8313 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8314 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8315 struct imsm_map
*map2
;
8316 int prev_num_members
;
8318 if (dev
->vol
.migr_state
)
8321 if (prev_disks
== -1)
8322 prev_disks
= map
->num_members
;
8323 if (prev_disks
== map
->num_members
)
8326 /* OK, this array needs to enter reshape mode.
8327 * i.e it needs a migr_state
8330 copy_map_size
= sizeof_imsm_map(map
);
8331 prev_num_members
= map
->num_members
;
8332 map
->num_members
= prev_disks
;
8333 dev
->vol
.migr_state
= 1;
8334 dev
->vol
.curr_migr_unit
= 0;
8335 set_migr_type(dev
, MIGR_GEN_MIGR
);
8336 for (i
= prev_num_members
;
8337 i
< map
->num_members
; i
++)
8338 set_imsm_ord_tbl_ent(map
, i
, i
);
8339 map2
= get_imsm_map(dev
, MAP_1
);
8340 /* Copy the current map */
8341 memcpy(map2
, map
, copy_map_size
);
8342 map2
->num_members
= prev_num_members
;
8344 imsm_set_array_size(dev
, -1);
8345 super
->clean_migration_record_by_mdmon
= 1;
8346 super
->updates_pending
++;
8350 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8351 * states are handled in imsm_set_disk() with one exception, when a
8352 * resync is stopped due to a new failure this routine will set the
8353 * 'degraded' state for the array.
8355 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8357 int inst
= a
->info
.container_member
;
8358 struct intel_super
*super
= a
->container
->sb
;
8359 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8360 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8361 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8362 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8363 __u32 blocks_per_unit
;
8365 if (dev
->vol
.migr_state
&&
8366 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8367 /* array state change is blocked due to reshape action
8369 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8370 * - finish the reshape (if last_checkpoint is big and action != reshape)
8371 * - update curr_migr_unit
8373 if (a
->curr_action
== reshape
) {
8374 /* still reshaping, maybe update curr_migr_unit */
8375 goto mark_checkpoint
;
8377 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8378 /* for some reason we aborted the reshape.
8380 * disable automatic metadata rollback
8381 * user action is required to recover process
8384 struct imsm_map
*map2
=
8385 get_imsm_map(dev
, MAP_1
);
8386 dev
->vol
.migr_state
= 0;
8387 set_migr_type(dev
, 0);
8388 dev
->vol
.curr_migr_unit
= 0;
8390 sizeof_imsm_map(map2
));
8391 super
->updates_pending
++;
8394 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8395 unsigned long long array_blocks
;
8399 used_disks
= imsm_num_data_members(map
);
8400 if (used_disks
> 0) {
8402 per_dev_array_size(map
) *
8405 round_size_to_mb(array_blocks
,
8407 a
->info
.custom_array_size
= array_blocks
;
8408 /* encourage manager to update array
8412 a
->check_reshape
= 1;
8414 /* finalize online capacity expansion/reshape */
8415 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8417 mdi
->disk
.raid_disk
,
8420 imsm_progress_container_reshape(super
);
8425 /* before we activate this array handle any missing disks */
8426 if (consistent
== 2)
8427 handle_missing(super
, dev
);
8429 if (consistent
== 2 &&
8430 (!is_resync_complete(&a
->info
) ||
8431 map_state
!= IMSM_T_STATE_NORMAL
||
8432 dev
->vol
.migr_state
))
8435 if (is_resync_complete(&a
->info
)) {
8436 /* complete intialization / resync,
8437 * recovery and interrupted recovery is completed in
8440 if (is_resyncing(dev
)) {
8441 dprintf("imsm: mark resync done\n");
8442 end_migration(dev
, super
, map_state
);
8443 super
->updates_pending
++;
8444 a
->last_checkpoint
= 0;
8446 } else if ((!is_resyncing(dev
) && !failed
) &&
8447 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8448 /* mark the start of the init process if nothing is failed */
8449 dprintf("imsm: mark resync start\n");
8450 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8451 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8453 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8454 super
->updates_pending
++;
8458 /* skip checkpointing for general migration,
8459 * it is controlled in mdadm
8461 if (is_gen_migration(dev
))
8462 goto skip_mark_checkpoint
;
8464 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8465 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8466 if (blocks_per_unit
) {
8470 units
= a
->last_checkpoint
/ blocks_per_unit
;
8473 /* check that we did not overflow 32-bits, and that
8474 * curr_migr_unit needs updating
8476 if (units32
== units
&&
8478 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8479 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8480 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8481 super
->updates_pending
++;
8485 skip_mark_checkpoint
:
8486 /* mark dirty / clean */
8487 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8488 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8489 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8491 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8493 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8494 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8495 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8496 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8498 super
->updates_pending
++;
8504 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8506 int inst
= a
->info
.container_member
;
8507 struct intel_super
*super
= a
->container
->sb
;
8508 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8509 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8511 if (slot
> map
->num_members
) {
8512 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8513 slot
, map
->num_members
- 1);
8520 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8523 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8525 int inst
= a
->info
.container_member
;
8526 struct intel_super
*super
= a
->container
->sb
;
8527 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8528 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8529 struct imsm_disk
*disk
;
8531 int recovery_not_finished
= 0;
8535 int rebuild_done
= 0;
8538 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8542 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8543 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8545 /* check for new failures */
8546 if (disk
&& (state
& DS_FAULTY
)) {
8547 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8548 super
->updates_pending
++;
8551 /* check if in_sync */
8552 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8553 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8555 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8557 super
->updates_pending
++;
8560 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8561 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8563 /* check if recovery complete, newly degraded, or failed */
8564 dprintf("imsm: Detected transition to state ");
8565 switch (map_state
) {
8566 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8567 dprintf("normal: ");
8568 if (is_rebuilding(dev
)) {
8569 dprintf_cont("while rebuilding");
8570 /* check if recovery is really finished */
8571 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8572 if (mdi
->recovery_start
!= MaxSector
) {
8573 recovery_not_finished
= 1;
8576 if (recovery_not_finished
) {
8578 dprintf("Rebuild has not finished yet, state not changed");
8579 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8580 a
->last_checkpoint
= mdi
->recovery_start
;
8581 super
->updates_pending
++;
8585 end_migration(dev
, super
, map_state
);
8586 map
= get_imsm_map(dev
, MAP_0
);
8587 map
->failed_disk_num
= ~0;
8588 super
->updates_pending
++;
8589 a
->last_checkpoint
= 0;
8592 if (is_gen_migration(dev
)) {
8593 dprintf_cont("while general migration");
8594 if (a
->last_checkpoint
>= a
->info
.component_size
)
8595 end_migration(dev
, super
, map_state
);
8597 map
->map_state
= map_state
;
8598 map
= get_imsm_map(dev
, MAP_0
);
8599 map
->failed_disk_num
= ~0;
8600 super
->updates_pending
++;
8604 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8605 dprintf_cont("degraded: ");
8606 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8607 dprintf_cont("mark degraded");
8608 map
->map_state
= map_state
;
8609 super
->updates_pending
++;
8610 a
->last_checkpoint
= 0;
8613 if (is_rebuilding(dev
)) {
8614 dprintf_cont("while rebuilding ");
8615 if (state
& DS_FAULTY
) {
8616 dprintf_cont("removing failed drive ");
8617 if (n
== map
->failed_disk_num
) {
8618 dprintf_cont("end migration");
8619 end_migration(dev
, super
, map_state
);
8620 a
->last_checkpoint
= 0;
8622 dprintf_cont("fail detected during rebuild, changing map state");
8623 map
->map_state
= map_state
;
8625 super
->updates_pending
++;
8631 /* check if recovery is really finished */
8632 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8633 if (mdi
->recovery_start
!= MaxSector
) {
8634 recovery_not_finished
= 1;
8637 if (recovery_not_finished
) {
8639 dprintf_cont("Rebuild has not finished yet");
8640 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8641 a
->last_checkpoint
=
8642 mdi
->recovery_start
;
8643 super
->updates_pending
++;
8648 dprintf_cont(" Rebuild done, still degraded");
8649 end_migration(dev
, super
, map_state
);
8650 a
->last_checkpoint
= 0;
8651 super
->updates_pending
++;
8653 for (i
= 0; i
< map
->num_members
; i
++) {
8654 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8656 if (idx
& IMSM_ORD_REBUILD
)
8657 map
->failed_disk_num
= i
;
8659 super
->updates_pending
++;
8662 if (is_gen_migration(dev
)) {
8663 dprintf_cont("while general migration");
8664 if (a
->last_checkpoint
>= a
->info
.component_size
)
8665 end_migration(dev
, super
, map_state
);
8667 map
->map_state
= map_state
;
8668 manage_second_map(super
, dev
);
8670 super
->updates_pending
++;
8673 if (is_initializing(dev
)) {
8674 dprintf_cont("while initialization.");
8675 map
->map_state
= map_state
;
8676 super
->updates_pending
++;
8680 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8681 dprintf_cont("failed: ");
8682 if (is_gen_migration(dev
)) {
8683 dprintf_cont("while general migration");
8684 map
->map_state
= map_state
;
8685 super
->updates_pending
++;
8688 if (map
->map_state
!= map_state
) {
8689 dprintf_cont("mark failed");
8690 end_migration(dev
, super
, map_state
);
8691 super
->updates_pending
++;
8692 a
->last_checkpoint
= 0;
8697 dprintf_cont("state %i\n", map_state
);
8702 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8705 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8706 unsigned long long dsize
;
8707 unsigned long long sectors
;
8708 unsigned int sector_size
;
8710 get_dev_sector_size(fd
, NULL
, §or_size
);
8711 get_dev_size(fd
, NULL
, &dsize
);
8713 if (mpb_size
> sector_size
) {
8714 /* -1 to account for anchor */
8715 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8717 /* write the extended mpb to the sectors preceeding the anchor */
8718 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8722 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8723 sector_size
* sectors
) != sector_size
* sectors
)
8727 /* first block is stored on second to last sector of the disk */
8728 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8731 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8737 static void imsm_sync_metadata(struct supertype
*container
)
8739 struct intel_super
*super
= container
->sb
;
8741 dprintf("sync metadata: %d\n", super
->updates_pending
);
8742 if (!super
->updates_pending
)
8745 write_super_imsm(container
, 0);
8747 super
->updates_pending
= 0;
8750 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8752 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8753 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8756 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8760 if (dl
&& is_failed(&dl
->disk
))
8764 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8769 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8770 struct active_array
*a
, int activate_new
,
8771 struct mdinfo
*additional_test_list
)
8773 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8774 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8775 struct imsm_super
*mpb
= super
->anchor
;
8776 struct imsm_map
*map
;
8777 unsigned long long pos
;
8782 __u32 array_start
= 0;
8783 __u32 array_end
= 0;
8785 struct mdinfo
*test_list
;
8787 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8788 /* If in this array, skip */
8789 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8790 if (d
->state_fd
>= 0 &&
8791 d
->disk
.major
== dl
->major
&&
8792 d
->disk
.minor
== dl
->minor
) {
8793 dprintf("%x:%x already in array\n",
8794 dl
->major
, dl
->minor
);
8799 test_list
= additional_test_list
;
8801 if (test_list
->disk
.major
== dl
->major
&&
8802 test_list
->disk
.minor
== dl
->minor
) {
8803 dprintf("%x:%x already in additional test list\n",
8804 dl
->major
, dl
->minor
);
8807 test_list
= test_list
->next
;
8812 /* skip in use or failed drives */
8813 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8815 dprintf("%x:%x status (failed: %d index: %d)\n",
8816 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8820 /* skip pure spares when we are looking for partially
8821 * assimilated drives
8823 if (dl
->index
== -1 && !activate_new
)
8826 if (!drive_validate_sector_size(super
, dl
))
8829 /* Does this unused device have the requisite free space?
8830 * It needs to be able to cover all member volumes
8832 ex
= get_extents(super
, dl
, 1);
8834 dprintf("cannot get extents\n");
8837 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8838 dev
= get_imsm_dev(super
, i
);
8839 map
= get_imsm_map(dev
, MAP_0
);
8841 /* check if this disk is already a member of
8844 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8850 array_start
= pba_of_lba0(map
);
8851 array_end
= array_start
+
8852 per_dev_array_size(map
) - 1;
8855 /* check that we can start at pba_of_lba0 with
8856 * num_data_stripes*blocks_per_stripe of space
8858 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8862 pos
= ex
[j
].start
+ ex
[j
].size
;
8864 } while (ex
[j
-1].size
);
8871 if (i
< mpb
->num_raid_devs
) {
8872 dprintf("%x:%x does not have %u to %u available\n",
8873 dl
->major
, dl
->minor
, array_start
, array_end
);
8883 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8885 struct imsm_dev
*dev2
;
8886 struct imsm_map
*map
;
8892 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8894 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8895 if (state
== IMSM_T_STATE_FAILED
) {
8896 map
= get_imsm_map(dev2
, MAP_0
);
8899 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8901 * Check if failed disks are deleted from intel
8902 * disk list or are marked to be deleted
8904 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8905 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8907 * Do not rebuild the array if failed disks
8908 * from failed sub-array are not removed from
8912 is_failed(&idisk
->disk
) &&
8913 (idisk
->action
!= DISK_REMOVE
))
8921 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8922 struct metadata_update
**updates
)
8925 * Find a device with unused free space and use it to replace a
8926 * failed/vacant region in an array. We replace failed regions one a
8927 * array at a time. The result is that a new spare disk will be added
8928 * to the first failed array and after the monitor has finished
8929 * propagating failures the remainder will be consumed.
8931 * FIXME add a capability for mdmon to request spares from another
8935 struct intel_super
*super
= a
->container
->sb
;
8936 int inst
= a
->info
.container_member
;
8937 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8938 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8939 int failed
= a
->info
.array
.raid_disks
;
8940 struct mdinfo
*rv
= NULL
;
8943 struct metadata_update
*mu
;
8945 struct imsm_update_activate_spare
*u
;
8950 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8951 if ((d
->curr_state
& DS_FAULTY
) &&
8953 /* wait for Removal to happen */
8955 if (d
->state_fd
>= 0)
8959 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8960 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8962 if (imsm_reshape_blocks_arrays_changes(super
))
8965 /* Cannot activate another spare if rebuild is in progress already
8967 if (is_rebuilding(dev
)) {
8968 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8972 if (a
->info
.array
.level
== 4)
8973 /* No repair for takeovered array
8974 * imsm doesn't support raid4
8978 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8979 IMSM_T_STATE_DEGRADED
)
8982 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8983 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8988 * If there are any failed disks check state of the other volume.
8989 * Block rebuild if the another one is failed until failed disks
8990 * are removed from container.
8993 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8994 MAX_RAID_SERIAL_LEN
, dev
->volume
);
8995 /* check if states of the other volumes allow for rebuild */
8996 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
8998 allowed
= imsm_rebuild_allowed(a
->container
,
9006 /* For each slot, if it is not working, find a spare */
9007 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
9008 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9009 if (d
->disk
.raid_disk
== i
)
9011 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
9012 if (d
&& (d
->state_fd
>= 0))
9016 * OK, this device needs recovery. Try to re-add the
9017 * previous occupant of this slot, if this fails see if
9018 * we can continue the assimilation of a spare that was
9019 * partially assimilated, finally try to activate a new
9022 dl
= imsm_readd(super
, i
, a
);
9024 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9026 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9030 /* found a usable disk with enough space */
9031 di
= xcalloc(1, sizeof(*di
));
9033 /* dl->index will be -1 in the case we are activating a
9034 * pristine spare. imsm_process_update() will create a
9035 * new index in this case. Once a disk is found to be
9036 * failed in all member arrays it is kicked from the
9039 di
->disk
.number
= dl
->index
;
9041 /* (ab)use di->devs to store a pointer to the device
9044 di
->devs
= (struct mdinfo
*) dl
;
9046 di
->disk
.raid_disk
= i
;
9047 di
->disk
.major
= dl
->major
;
9048 di
->disk
.minor
= dl
->minor
;
9050 di
->recovery_start
= 0;
9051 di
->data_offset
= pba_of_lba0(map
);
9052 di
->component_size
= a
->info
.component_size
;
9053 di
->container_member
= inst
;
9054 di
->bb
.supported
= 1;
9055 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9056 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9057 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9059 super
->random
= random32();
9063 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9064 i
, di
->data_offset
);
9068 /* No spares found */
9070 /* Now 'rv' has a list of devices to return.
9071 * Create a metadata_update record to update the
9072 * disk_ord_tbl for the array
9074 mu
= xmalloc(sizeof(*mu
));
9075 mu
->buf
= xcalloc(num_spares
,
9076 sizeof(struct imsm_update_activate_spare
));
9078 mu
->space_list
= NULL
;
9079 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9080 mu
->next
= *updates
;
9081 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9083 for (di
= rv
; di
; di
= di
->next
) {
9084 u
->type
= update_activate_spare
;
9085 u
->dl
= (struct dl
*) di
->devs
;
9087 u
->slot
= di
->disk
.raid_disk
;
9098 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9100 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9101 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9102 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9103 struct disk_info
*inf
= get_disk_info(u
);
9104 struct imsm_disk
*disk
;
9108 for (i
= 0; i
< map
->num_members
; i
++) {
9109 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9110 for (j
= 0; j
< new_map
->num_members
; j
++)
9111 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9118 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9122 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9123 if (dl
->major
== major
&& dl
->minor
== minor
)
9128 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9134 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9135 if (dl
->major
== major
&& dl
->minor
== minor
) {
9138 prev
->next
= dl
->next
;
9140 super
->disks
= dl
->next
;
9142 __free_imsm_disk(dl
);
9143 dprintf("removed %x:%x\n", major
, minor
);
9151 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9153 static int add_remove_disk_update(struct intel_super
*super
)
9155 int check_degraded
= 0;
9158 /* add/remove some spares to/from the metadata/contrainer */
9159 while (super
->disk_mgmt_list
) {
9160 struct dl
*disk_cfg
;
9162 disk_cfg
= super
->disk_mgmt_list
;
9163 super
->disk_mgmt_list
= disk_cfg
->next
;
9164 disk_cfg
->next
= NULL
;
9166 if (disk_cfg
->action
== DISK_ADD
) {
9167 disk_cfg
->next
= super
->disks
;
9168 super
->disks
= disk_cfg
;
9170 dprintf("added %x:%x\n",
9171 disk_cfg
->major
, disk_cfg
->minor
);
9172 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9173 dprintf("Disk remove action processed: %x.%x\n",
9174 disk_cfg
->major
, disk_cfg
->minor
);
9175 disk
= get_disk_super(super
,
9179 /* store action status */
9180 disk
->action
= DISK_REMOVE
;
9181 /* remove spare disks only */
9182 if (disk
->index
== -1) {
9183 remove_disk_super(super
,
9187 disk_cfg
->fd
= disk
->fd
;
9191 /* release allocate disk structure */
9192 __free_imsm_disk(disk_cfg
);
9195 return check_degraded
;
9198 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9199 struct intel_super
*super
,
9202 struct intel_dev
*id
;
9203 void **tofree
= NULL
;
9206 dprintf("(enter)\n");
9207 if (u
->subdev
< 0 || u
->subdev
> 1) {
9208 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9211 if (space_list
== NULL
|| *space_list
== NULL
) {
9212 dprintf("imsm: Error: Memory is not allocated\n");
9216 for (id
= super
->devlist
; id
; id
= id
->next
) {
9217 if (id
->index
== (unsigned)u
->subdev
) {
9218 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9219 struct imsm_map
*map
;
9220 struct imsm_dev
*new_dev
=
9221 (struct imsm_dev
*)*space_list
;
9222 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9224 struct dl
*new_disk
;
9226 if (new_dev
== NULL
)
9228 *space_list
= **space_list
;
9229 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9230 map
= get_imsm_map(new_dev
, MAP_0
);
9232 dprintf("imsm: Error: migration in progress");
9236 to_state
= map
->map_state
;
9237 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9239 /* this should not happen */
9240 if (u
->new_disks
[0] < 0) {
9241 map
->failed_disk_num
=
9242 map
->num_members
- 1;
9243 to_state
= IMSM_T_STATE_DEGRADED
;
9245 to_state
= IMSM_T_STATE_NORMAL
;
9247 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9248 if (u
->new_level
> -1)
9249 map
->raid_level
= u
->new_level
;
9250 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9251 if ((u
->new_level
== 5) &&
9252 (migr_map
->raid_level
== 0)) {
9253 int ord
= map
->num_members
- 1;
9254 migr_map
->num_members
--;
9255 if (u
->new_disks
[0] < 0)
9256 ord
|= IMSM_ORD_REBUILD
;
9257 set_imsm_ord_tbl_ent(map
,
9258 map
->num_members
- 1,
9262 tofree
= (void **)dev
;
9264 /* update chunk size
9266 if (u
->new_chunksize
> 0) {
9267 unsigned long long num_data_stripes
;
9268 struct imsm_map
*dest_map
=
9269 get_imsm_map(dev
, MAP_0
);
9271 imsm_num_data_members(dest_map
);
9273 if (used_disks
== 0)
9276 map
->blocks_per_strip
=
9277 __cpu_to_le16(u
->new_chunksize
* 2);
9279 imsm_dev_size(dev
) / used_disks
;
9280 num_data_stripes
/= map
->blocks_per_strip
;
9281 num_data_stripes
/= map
->num_domains
;
9282 set_num_data_stripes(map
, num_data_stripes
);
9285 /* ensure blocks_per_member has valid value
9287 set_blocks_per_member(map
,
9288 per_dev_array_size(map
) +
9289 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9293 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9294 migr_map
->raid_level
== map
->raid_level
)
9297 if (u
->new_disks
[0] >= 0) {
9300 new_disk
= get_disk_super(super
,
9301 major(u
->new_disks
[0]),
9302 minor(u
->new_disks
[0]));
9303 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9304 major(u
->new_disks
[0]),
9305 minor(u
->new_disks
[0]),
9306 new_disk
, new_disk
->index
);
9307 if (new_disk
== NULL
)
9308 goto error_disk_add
;
9310 new_disk
->index
= map
->num_members
- 1;
9311 /* slot to fill in autolayout
9313 new_disk
->raiddisk
= new_disk
->index
;
9314 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9315 new_disk
->disk
.status
&= ~SPARE_DISK
;
9317 goto error_disk_add
;
9320 *tofree
= *space_list
;
9321 /* calculate new size
9323 imsm_set_array_size(new_dev
, -1);
9330 *space_list
= tofree
;
9334 dprintf("Error: imsm: Cannot find disk.\n");
9338 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9339 struct intel_super
*super
)
9341 struct intel_dev
*id
;
9344 dprintf("(enter)\n");
9345 if (u
->subdev
< 0 || u
->subdev
> 1) {
9346 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9350 for (id
= super
->devlist
; id
; id
= id
->next
) {
9351 if (id
->index
== (unsigned)u
->subdev
) {
9352 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9353 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9354 int used_disks
= imsm_num_data_members(map
);
9355 unsigned long long blocks_per_member
;
9356 unsigned long long num_data_stripes
;
9357 unsigned long long new_size_per_disk
;
9359 if (used_disks
== 0)
9362 /* calculate new size
9364 new_size_per_disk
= u
->new_size
/ used_disks
;
9365 blocks_per_member
= new_size_per_disk
+
9366 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9367 num_data_stripes
= new_size_per_disk
/
9368 map
->blocks_per_strip
;
9369 num_data_stripes
/= map
->num_domains
;
9370 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9371 u
->new_size
, new_size_per_disk
,
9373 set_blocks_per_member(map
, blocks_per_member
);
9374 set_num_data_stripes(map
, num_data_stripes
);
9375 imsm_set_array_size(dev
, u
->new_size
);
9385 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9386 struct intel_super
*super
,
9387 struct active_array
*active_array
)
9389 struct imsm_super
*mpb
= super
->anchor
;
9390 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9391 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9392 struct imsm_map
*migr_map
;
9393 struct active_array
*a
;
9394 struct imsm_disk
*disk
;
9401 int second_map_created
= 0;
9403 for (; u
; u
= u
->next
) {
9404 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9409 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9414 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9419 /* count failures (excluding rebuilds and the victim)
9420 * to determine map[0] state
9423 for (i
= 0; i
< map
->num_members
; i
++) {
9426 disk
= get_imsm_disk(super
,
9427 get_imsm_disk_idx(dev
, i
, MAP_X
));
9428 if (!disk
|| is_failed(disk
))
9432 /* adding a pristine spare, assign a new index */
9433 if (dl
->index
< 0) {
9434 dl
->index
= super
->anchor
->num_disks
;
9435 super
->anchor
->num_disks
++;
9438 disk
->status
|= CONFIGURED_DISK
;
9439 disk
->status
&= ~SPARE_DISK
;
9442 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9443 if (!second_map_created
) {
9444 second_map_created
= 1;
9445 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9446 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9448 map
->map_state
= to_state
;
9449 migr_map
= get_imsm_map(dev
, MAP_1
);
9450 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9451 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9452 dl
->index
| IMSM_ORD_REBUILD
);
9454 /* update the family_num to mark a new container
9455 * generation, being careful to record the existing
9456 * family_num in orig_family_num to clean up after
9457 * earlier mdadm versions that neglected to set it.
9459 if (mpb
->orig_family_num
== 0)
9460 mpb
->orig_family_num
= mpb
->family_num
;
9461 mpb
->family_num
+= super
->random
;
9463 /* count arrays using the victim in the metadata */
9465 for (a
= active_array
; a
; a
= a
->next
) {
9466 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9467 map
= get_imsm_map(dev
, MAP_0
);
9469 if (get_imsm_disk_slot(map
, victim
) >= 0)
9473 /* delete the victim if it is no longer being
9479 /* We know that 'manager' isn't touching anything,
9480 * so it is safe to delete
9482 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9483 if ((*dlp
)->index
== victim
)
9486 /* victim may be on the missing list */
9488 for (dlp
= &super
->missing
; *dlp
;
9489 dlp
= &(*dlp
)->next
)
9490 if ((*dlp
)->index
== victim
)
9492 imsm_delete(super
, dlp
, victim
);
9499 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9500 struct intel_super
*super
,
9503 struct dl
*new_disk
;
9504 struct intel_dev
*id
;
9506 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9507 int disk_count
= u
->old_raid_disks
;
9508 void **tofree
= NULL
;
9509 int devices_to_reshape
= 1;
9510 struct imsm_super
*mpb
= super
->anchor
;
9512 unsigned int dev_id
;
9514 dprintf("(enter)\n");
9516 /* enable spares to use in array */
9517 for (i
= 0; i
< delta_disks
; i
++) {
9518 new_disk
= get_disk_super(super
,
9519 major(u
->new_disks
[i
]),
9520 minor(u
->new_disks
[i
]));
9521 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9522 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9523 new_disk
, new_disk
->index
);
9524 if (new_disk
== NULL
||
9525 (new_disk
->index
>= 0 &&
9526 new_disk
->index
< u
->old_raid_disks
))
9527 goto update_reshape_exit
;
9528 new_disk
->index
= disk_count
++;
9529 /* slot to fill in autolayout
9531 new_disk
->raiddisk
= new_disk
->index
;
9532 new_disk
->disk
.status
|=
9534 new_disk
->disk
.status
&= ~SPARE_DISK
;
9537 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9538 mpb
->num_raid_devs
);
9539 /* manage changes in volume
9541 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9542 void **sp
= *space_list
;
9543 struct imsm_dev
*newdev
;
9544 struct imsm_map
*newmap
, *oldmap
;
9546 for (id
= super
->devlist
; id
; id
= id
->next
) {
9547 if (id
->index
== dev_id
)
9556 /* Copy the dev, but not (all of) the map */
9557 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9558 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9559 newmap
= get_imsm_map(newdev
, MAP_0
);
9560 /* Copy the current map */
9561 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9562 /* update one device only
9564 if (devices_to_reshape
) {
9565 dprintf("imsm: modifying subdev: %i\n",
9567 devices_to_reshape
--;
9568 newdev
->vol
.migr_state
= 1;
9569 newdev
->vol
.curr_migr_unit
= 0;
9570 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9571 newmap
->num_members
= u
->new_raid_disks
;
9572 for (i
= 0; i
< delta_disks
; i
++) {
9573 set_imsm_ord_tbl_ent(newmap
,
9574 u
->old_raid_disks
+ i
,
9575 u
->old_raid_disks
+ i
);
9577 /* New map is correct, now need to save old map
9579 newmap
= get_imsm_map(newdev
, MAP_1
);
9580 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9582 imsm_set_array_size(newdev
, -1);
9585 sp
= (void **)id
->dev
;
9590 /* Clear migration record */
9591 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9594 *space_list
= tofree
;
9597 update_reshape_exit
:
9602 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9603 struct intel_super
*super
,
9606 struct imsm_dev
*dev
= NULL
;
9607 struct intel_dev
*dv
;
9608 struct imsm_dev
*dev_new
;
9609 struct imsm_map
*map
;
9613 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9614 if (dv
->index
== (unsigned int)u
->subarray
) {
9622 map
= get_imsm_map(dev
, MAP_0
);
9624 if (u
->direction
== R10_TO_R0
) {
9625 unsigned long long num_data_stripes
;
9627 /* Number of failed disks must be half of initial disk number */
9628 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9629 (map
->num_members
/ 2))
9632 /* iterate through devices to mark removed disks as spare */
9633 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9634 if (dm
->disk
.status
& FAILED_DISK
) {
9635 int idx
= dm
->index
;
9636 /* update indexes on the disk list */
9637 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9638 the index values will end up being correct.... NB */
9639 for (du
= super
->disks
; du
; du
= du
->next
)
9640 if (du
->index
> idx
)
9642 /* mark as spare disk */
9647 map
->num_members
= map
->num_members
/ 2;
9648 map
->map_state
= IMSM_T_STATE_NORMAL
;
9649 map
->num_domains
= 1;
9650 map
->raid_level
= 0;
9651 map
->failed_disk_num
= -1;
9652 num_data_stripes
= imsm_dev_size(dev
) / 2;
9653 num_data_stripes
/= map
->blocks_per_strip
;
9654 set_num_data_stripes(map
, num_data_stripes
);
9657 if (u
->direction
== R0_TO_R10
) {
9659 unsigned long long num_data_stripes
;
9661 /* update slots in current disk list */
9662 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9666 /* create new *missing* disks */
9667 for (i
= 0; i
< map
->num_members
; i
++) {
9668 space
= *space_list
;
9671 *space_list
= *space
;
9673 memcpy(du
, super
->disks
, sizeof(*du
));
9677 du
->index
= (i
* 2) + 1;
9678 sprintf((char *)du
->disk
.serial
,
9679 " MISSING_%d", du
->index
);
9680 sprintf((char *)du
->serial
,
9681 "MISSING_%d", du
->index
);
9682 du
->next
= super
->missing
;
9683 super
->missing
= du
;
9685 /* create new dev and map */
9686 space
= *space_list
;
9689 *space_list
= *space
;
9690 dev_new
= (void *)space
;
9691 memcpy(dev_new
, dev
, sizeof(*dev
));
9692 /* update new map */
9693 map
= get_imsm_map(dev_new
, MAP_0
);
9694 map
->num_members
= map
->num_members
* 2;
9695 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9696 map
->num_domains
= 2;
9697 map
->raid_level
= 1;
9698 num_data_stripes
= imsm_dev_size(dev
) / 2;
9699 num_data_stripes
/= map
->blocks_per_strip
;
9700 num_data_stripes
/= map
->num_domains
;
9701 set_num_data_stripes(map
, num_data_stripes
);
9703 /* replace dev<->dev_new */
9706 /* update disk order table */
9707 for (du
= super
->disks
; du
; du
= du
->next
)
9709 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9710 for (du
= super
->missing
; du
; du
= du
->next
)
9711 if (du
->index
>= 0) {
9712 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9713 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9719 static void imsm_process_update(struct supertype
*st
,
9720 struct metadata_update
*update
)
9723 * crack open the metadata_update envelope to find the update record
9724 * update can be one of:
9725 * update_reshape_container_disks - all the arrays in the container
9726 * are being reshaped to have more devices. We need to mark
9727 * the arrays for general migration and convert selected spares
9728 * into active devices.
9729 * update_activate_spare - a spare device has replaced a failed
9730 * device in an array, update the disk_ord_tbl. If this disk is
9731 * present in all member arrays then also clear the SPARE_DISK
9733 * update_create_array
9735 * update_rename_array
9736 * update_add_remove_disk
9738 struct intel_super
*super
= st
->sb
;
9739 struct imsm_super
*mpb
;
9740 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9742 /* update requires a larger buf but the allocation failed */
9743 if (super
->next_len
&& !super
->next_buf
) {
9744 super
->next_len
= 0;
9748 if (super
->next_buf
) {
9749 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9751 super
->len
= super
->next_len
;
9752 super
->buf
= super
->next_buf
;
9754 super
->next_len
= 0;
9755 super
->next_buf
= NULL
;
9758 mpb
= super
->anchor
;
9761 case update_general_migration_checkpoint
: {
9762 struct intel_dev
*id
;
9763 struct imsm_update_general_migration_checkpoint
*u
=
9764 (void *)update
->buf
;
9766 dprintf("called for update_general_migration_checkpoint\n");
9768 /* find device under general migration */
9769 for (id
= super
->devlist
; id
; id
= id
->next
) {
9770 if (is_gen_migration(id
->dev
)) {
9771 id
->dev
->vol
.curr_migr_unit
=
9772 __cpu_to_le32(u
->curr_migr_unit
);
9773 super
->updates_pending
++;
9778 case update_takeover
: {
9779 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9780 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9781 imsm_update_version_info(super
);
9782 super
->updates_pending
++;
9787 case update_reshape_container_disks
: {
9788 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9789 if (apply_reshape_container_disks_update(
9790 u
, super
, &update
->space_list
))
9791 super
->updates_pending
++;
9794 case update_reshape_migration
: {
9795 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9796 if (apply_reshape_migration_update(
9797 u
, super
, &update
->space_list
))
9798 super
->updates_pending
++;
9801 case update_size_change
: {
9802 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9803 if (apply_size_change_update(u
, super
))
9804 super
->updates_pending
++;
9807 case update_activate_spare
: {
9808 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9809 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9810 super
->updates_pending
++;
9813 case update_create_array
: {
9814 /* someone wants to create a new array, we need to be aware of
9815 * a few races/collisions:
9816 * 1/ 'Create' called by two separate instances of mdadm
9817 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9818 * devices that have since been assimilated via
9820 * In the event this update can not be carried out mdadm will
9821 * (FIX ME) notice that its update did not take hold.
9823 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9824 struct intel_dev
*dv
;
9825 struct imsm_dev
*dev
;
9826 struct imsm_map
*map
, *new_map
;
9827 unsigned long long start
, end
;
9828 unsigned long long new_start
, new_end
;
9830 struct disk_info
*inf
;
9833 /* handle racing creates: first come first serve */
9834 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9835 dprintf("subarray %d already defined\n", u
->dev_idx
);
9839 /* check update is next in sequence */
9840 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9841 dprintf("can not create array %d expected index %d\n",
9842 u
->dev_idx
, mpb
->num_raid_devs
);
9846 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9847 new_start
= pba_of_lba0(new_map
);
9848 new_end
= new_start
+ per_dev_array_size(new_map
);
9849 inf
= get_disk_info(u
);
9851 /* handle activate_spare versus create race:
9852 * check to make sure that overlapping arrays do not include
9855 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9856 dev
= get_imsm_dev(super
, i
);
9857 map
= get_imsm_map(dev
, MAP_0
);
9858 start
= pba_of_lba0(map
);
9859 end
= start
+ per_dev_array_size(map
);
9860 if ((new_start
>= start
&& new_start
<= end
) ||
9861 (start
>= new_start
&& start
<= new_end
))
9866 if (disks_overlap(super
, i
, u
)) {
9867 dprintf("arrays overlap\n");
9872 /* check that prepare update was successful */
9873 if (!update
->space
) {
9874 dprintf("prepare update failed\n");
9878 /* check that all disks are still active before committing
9879 * changes. FIXME: could we instead handle this by creating a
9880 * degraded array? That's probably not what the user expects,
9881 * so better to drop this update on the floor.
9883 for (i
= 0; i
< new_map
->num_members
; i
++) {
9884 dl
= serial_to_dl(inf
[i
].serial
, super
);
9886 dprintf("disk disappeared\n");
9891 super
->updates_pending
++;
9893 /* convert spares to members and fixup ord_tbl */
9894 for (i
= 0; i
< new_map
->num_members
; i
++) {
9895 dl
= serial_to_dl(inf
[i
].serial
, super
);
9896 if (dl
->index
== -1) {
9897 dl
->index
= mpb
->num_disks
;
9899 dl
->disk
.status
|= CONFIGURED_DISK
;
9900 dl
->disk
.status
&= ~SPARE_DISK
;
9902 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9907 update
->space
= NULL
;
9908 imsm_copy_dev(dev
, &u
->dev
);
9909 dv
->index
= u
->dev_idx
;
9910 dv
->next
= super
->devlist
;
9911 super
->devlist
= dv
;
9912 mpb
->num_raid_devs
++;
9914 imsm_update_version_info(super
);
9917 /* mdmon knows how to release update->space, but not
9918 * ((struct intel_dev *) update->space)->dev
9920 if (update
->space
) {
9926 case update_kill_array
: {
9927 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9928 int victim
= u
->dev_idx
;
9929 struct active_array
*a
;
9930 struct intel_dev
**dp
;
9931 struct imsm_dev
*dev
;
9933 /* sanity check that we are not affecting the uuid of
9934 * active arrays, or deleting an active array
9936 * FIXME when immutable ids are available, but note that
9937 * we'll also need to fixup the invalidated/active
9938 * subarray indexes in mdstat
9940 for (a
= st
->arrays
; a
; a
= a
->next
)
9941 if (a
->info
.container_member
>= victim
)
9943 /* by definition if mdmon is running at least one array
9944 * is active in the container, so checking
9945 * mpb->num_raid_devs is just extra paranoia
9947 dev
= get_imsm_dev(super
, victim
);
9948 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9949 dprintf("failed to delete subarray-%d\n", victim
);
9953 for (dp
= &super
->devlist
; *dp
;)
9954 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9957 if ((*dp
)->index
> (unsigned)victim
)
9961 mpb
->num_raid_devs
--;
9962 super
->updates_pending
++;
9965 case update_rename_array
: {
9966 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9967 char name
[MAX_RAID_SERIAL_LEN
+1];
9968 int target
= u
->dev_idx
;
9969 struct active_array
*a
;
9970 struct imsm_dev
*dev
;
9972 /* sanity check that we are not affecting the uuid of
9975 memset(name
, 0, sizeof(name
));
9976 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9977 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9978 for (a
= st
->arrays
; a
; a
= a
->next
)
9979 if (a
->info
.container_member
== target
)
9981 dev
= get_imsm_dev(super
, u
->dev_idx
);
9982 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9983 dprintf("failed to rename subarray-%d\n", target
);
9987 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
9988 super
->updates_pending
++;
9991 case update_add_remove_disk
: {
9992 /* we may be able to repair some arrays if disks are
9993 * being added, check the status of add_remove_disk
9994 * if discs has been added.
9996 if (add_remove_disk_update(super
)) {
9997 struct active_array
*a
;
9999 super
->updates_pending
++;
10000 for (a
= st
->arrays
; a
; a
= a
->next
)
10001 a
->check_degraded
= 1;
10005 case update_prealloc_badblocks_mem
:
10007 case update_rwh_policy
: {
10008 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
10009 int target
= u
->dev_idx
;
10010 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
10012 dprintf("could not find subarray-%d\n", target
);
10016 if (dev
->rwh_policy
!= u
->new_policy
) {
10017 dev
->rwh_policy
= u
->new_policy
;
10018 super
->updates_pending
++;
10023 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10027 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10029 static int imsm_prepare_update(struct supertype
*st
,
10030 struct metadata_update
*update
)
10033 * Allocate space to hold new disk entries, raid-device entries or a new
10034 * mpb if necessary. The manager synchronously waits for updates to
10035 * complete in the monitor, so new mpb buffers allocated here can be
10036 * integrated by the monitor thread without worrying about live pointers
10037 * in the manager thread.
10039 enum imsm_update_type type
;
10040 struct intel_super
*super
= st
->sb
;
10041 unsigned int sector_size
= super
->sector_size
;
10042 struct imsm_super
*mpb
= super
->anchor
;
10046 if (update
->len
< (int)sizeof(type
))
10049 type
= *(enum imsm_update_type
*) update
->buf
;
10052 case update_general_migration_checkpoint
:
10053 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10055 dprintf("called for update_general_migration_checkpoint\n");
10057 case update_takeover
: {
10058 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10059 if (update
->len
< (int)sizeof(*u
))
10061 if (u
->direction
== R0_TO_R10
) {
10062 void **tail
= (void **)&update
->space_list
;
10063 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10064 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10065 int num_members
= map
->num_members
;
10068 /* allocate memory for added disks */
10069 for (i
= 0; i
< num_members
; i
++) {
10070 size
= sizeof(struct dl
);
10071 space
= xmalloc(size
);
10076 /* allocate memory for new device */
10077 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10078 (num_members
* sizeof(__u32
));
10079 space
= xmalloc(size
);
10083 len
= disks_to_mpb_size(num_members
* 2);
10088 case update_reshape_container_disks
: {
10089 /* Every raid device in the container is about to
10090 * gain some more devices, and we will enter a
10092 * So each 'imsm_map' will be bigger, and the imsm_vol
10093 * will now hold 2 of them.
10094 * Thus we need new 'struct imsm_dev' allocations sized
10095 * as sizeof_imsm_dev but with more devices in both maps.
10097 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10098 struct intel_dev
*dl
;
10099 void **space_tail
= (void**)&update
->space_list
;
10101 if (update
->len
< (int)sizeof(*u
))
10104 dprintf("for update_reshape\n");
10106 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10107 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10109 if (u
->new_raid_disks
> u
->old_raid_disks
)
10110 size
+= sizeof(__u32
)*2*
10111 (u
->new_raid_disks
- u
->old_raid_disks
);
10115 *space_tail
= NULL
;
10118 len
= disks_to_mpb_size(u
->new_raid_disks
);
10119 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10122 case update_reshape_migration
: {
10123 /* for migration level 0->5 we need to add disks
10124 * so the same as for container operation we will copy
10125 * device to the bigger location.
10126 * in memory prepared device and new disk area are prepared
10127 * for usage in process update
10129 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10130 struct intel_dev
*id
;
10131 void **space_tail
= (void **)&update
->space_list
;
10134 int current_level
= -1;
10136 if (update
->len
< (int)sizeof(*u
))
10139 dprintf("for update_reshape\n");
10141 /* add space for bigger array in update
10143 for (id
= super
->devlist
; id
; id
= id
->next
) {
10144 if (id
->index
== (unsigned)u
->subdev
) {
10145 size
= sizeof_imsm_dev(id
->dev
, 1);
10146 if (u
->new_raid_disks
> u
->old_raid_disks
)
10147 size
+= sizeof(__u32
)*2*
10148 (u
->new_raid_disks
- u
->old_raid_disks
);
10152 *space_tail
= NULL
;
10156 if (update
->space_list
== NULL
)
10159 /* add space for disk in update
10161 size
= sizeof(struct dl
);
10165 *space_tail
= NULL
;
10167 /* add spare device to update
10169 for (id
= super
->devlist
; id
; id
= id
->next
)
10170 if (id
->index
== (unsigned)u
->subdev
) {
10171 struct imsm_dev
*dev
;
10172 struct imsm_map
*map
;
10174 dev
= get_imsm_dev(super
, u
->subdev
);
10175 map
= get_imsm_map(dev
, MAP_0
);
10176 current_level
= map
->raid_level
;
10179 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10180 struct mdinfo
*spares
;
10182 spares
= get_spares_for_grow(st
);
10185 struct mdinfo
*dev
;
10187 dev
= spares
->devs
;
10190 makedev(dev
->disk
.major
,
10192 dl
= get_disk_super(super
,
10195 dl
->index
= u
->old_raid_disks
;
10198 sysfs_free(spares
);
10201 len
= disks_to_mpb_size(u
->new_raid_disks
);
10202 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10205 case update_size_change
: {
10206 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10210 case update_activate_spare
: {
10211 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10215 case update_create_array
: {
10216 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10217 struct intel_dev
*dv
;
10218 struct imsm_dev
*dev
= &u
->dev
;
10219 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10221 struct disk_info
*inf
;
10225 if (update
->len
< (int)sizeof(*u
))
10228 inf
= get_disk_info(u
);
10229 len
= sizeof_imsm_dev(dev
, 1);
10230 /* allocate a new super->devlist entry */
10231 dv
= xmalloc(sizeof(*dv
));
10232 dv
->dev
= xmalloc(len
);
10233 update
->space
= dv
;
10235 /* count how many spares will be converted to members */
10236 for (i
= 0; i
< map
->num_members
; i
++) {
10237 dl
= serial_to_dl(inf
[i
].serial
, super
);
10239 /* hmm maybe it failed?, nothing we can do about
10244 if (count_memberships(dl
, super
) == 0)
10247 len
+= activate
* sizeof(struct imsm_disk
);
10250 case update_kill_array
: {
10251 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10255 case update_rename_array
: {
10256 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10260 case update_add_remove_disk
:
10261 /* no update->len needed */
10263 case update_prealloc_badblocks_mem
:
10264 super
->extra_space
+= sizeof(struct bbm_log
) -
10265 get_imsm_bbm_log_size(super
->bbm_log
);
10267 case update_rwh_policy
: {
10268 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10276 /* check if we need a larger metadata buffer */
10277 if (super
->next_buf
)
10278 buf_len
= super
->next_len
;
10280 buf_len
= super
->len
;
10282 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10283 /* ok we need a larger buf than what is currently allocated
10284 * if this allocation fails process_update will notice that
10285 * ->next_len is set and ->next_buf is NULL
10287 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10288 super
->extra_space
+ len
, sector_size
);
10289 if (super
->next_buf
)
10290 free(super
->next_buf
);
10292 super
->next_len
= buf_len
;
10293 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10294 memset(super
->next_buf
, 0, buf_len
);
10296 super
->next_buf
= NULL
;
10301 /* must be called while manager is quiesced */
10302 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10304 struct imsm_super
*mpb
= super
->anchor
;
10306 struct imsm_dev
*dev
;
10307 struct imsm_map
*map
;
10308 unsigned int i
, j
, num_members
;
10309 __u32 ord
, ord_map0
;
10310 struct bbm_log
*log
= super
->bbm_log
;
10312 dprintf("deleting device[%d] from imsm_super\n", index
);
10314 /* shift all indexes down one */
10315 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10316 if (iter
->index
> (int)index
)
10318 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10319 if (iter
->index
> (int)index
)
10322 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10323 dev
= get_imsm_dev(super
, i
);
10324 map
= get_imsm_map(dev
, MAP_0
);
10325 num_members
= map
->num_members
;
10326 for (j
= 0; j
< num_members
; j
++) {
10327 /* update ord entries being careful not to propagate
10328 * ord-flags to the first map
10330 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10331 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10333 if (ord_to_idx(ord
) <= index
)
10336 map
= get_imsm_map(dev
, MAP_0
);
10337 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10338 map
= get_imsm_map(dev
, MAP_1
);
10340 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10344 for (i
= 0; i
< log
->entry_count
; i
++) {
10345 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10347 if (entry
->disk_ordinal
<= index
)
10349 entry
->disk_ordinal
--;
10353 super
->updates_pending
++;
10355 struct dl
*dl
= *dlp
;
10357 *dlp
= (*dlp
)->next
;
10358 __free_imsm_disk(dl
);
10362 static void close_targets(int *targets
, int new_disks
)
10369 for (i
= 0; i
< new_disks
; i
++) {
10370 if (targets
[i
] >= 0) {
10377 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10378 struct intel_super
*super
,
10379 struct imsm_dev
*dev
)
10385 struct imsm_map
*map
;
10388 ret_val
= raid_disks
/2;
10389 /* check map if all disks pairs not failed
10392 map
= get_imsm_map(dev
, MAP_0
);
10393 for (i
= 0; i
< ret_val
; i
++) {
10394 int degradation
= 0;
10395 if (get_imsm_disk(super
, i
) == NULL
)
10397 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10399 if (degradation
== 2)
10402 map
= get_imsm_map(dev
, MAP_1
);
10403 /* if there is no second map
10404 * result can be returned
10408 /* check degradation in second map
10410 for (i
= 0; i
< ret_val
; i
++) {
10411 int degradation
= 0;
10412 if (get_imsm_disk(super
, i
) == NULL
)
10414 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10416 if (degradation
== 2)
10430 /*******************************************************************************
10431 * Function: open_backup_targets
10432 * Description: Function opens file descriptors for all devices given in
10435 * info : general array info
10436 * raid_disks : number of disks
10437 * raid_fds : table of device's file descriptors
10438 * super : intel super for raid10 degradation check
10439 * dev : intel device for raid10 degradation check
10443 ******************************************************************************/
10444 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10445 struct intel_super
*super
, struct imsm_dev
*dev
)
10451 for (i
= 0; i
< raid_disks
; i
++)
10454 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10457 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10458 dprintf("disk is faulty!!\n");
10462 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10465 dn
= map_dev(sd
->disk
.major
,
10466 sd
->disk
.minor
, 1);
10467 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10468 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10469 pr_err("cannot open component\n");
10474 /* check if maximum array degradation level is not exceeded
10476 if ((raid_disks
- opened
) >
10477 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10479 pr_err("Not enough disks can be opened.\n");
10480 close_targets(raid_fds
, raid_disks
);
10486 /*******************************************************************************
10487 * Function: validate_container_imsm
10488 * Description: This routine validates container after assemble,
10489 * eg. if devices in container are under the same controller.
10492 * info : linked list with info about devices used in array
10496 ******************************************************************************/
10497 int validate_container_imsm(struct mdinfo
*info
)
10499 if (check_env("IMSM_NO_PLATFORM"))
10502 struct sys_dev
*idev
;
10503 struct sys_dev
*hba
= NULL
;
10504 struct sys_dev
*intel_devices
= find_intel_devices();
10505 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10506 info
->disk
.minor
));
10508 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10509 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10518 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10519 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10523 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10524 struct mdinfo
*dev
;
10526 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10527 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10529 struct sys_dev
*hba2
= NULL
;
10530 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10531 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10539 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10540 get_orom_by_device_id(hba2
->dev_id
);
10542 if (hba2
&& hba
->type
!= hba2
->type
) {
10543 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10544 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10548 if (orom
!= orom2
) {
10549 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10550 " This operation is not supported and can lead to data loss.\n");
10555 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10556 " This operation is not supported and can lead to data loss.\n");
10564 /*******************************************************************************
10565 * Function: imsm_record_badblock
10566 * Description: This routine stores new bad block record in BBM log
10569 * a : array containing a bad block
10570 * slot : disk number containing a bad block
10571 * sector : bad block sector
10572 * length : bad block sectors range
10576 ******************************************************************************/
10577 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10578 unsigned long long sector
, int length
)
10580 struct intel_super
*super
= a
->container
->sb
;
10584 ord
= imsm_disk_slot_to_ord(a
, slot
);
10588 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10591 super
->updates_pending
++;
10595 /*******************************************************************************
10596 * Function: imsm_clear_badblock
10597 * Description: This routine clears bad block record from BBM log
10600 * a : array containing a bad block
10601 * slot : disk number containing a bad block
10602 * sector : bad block sector
10603 * length : bad block sectors range
10607 ******************************************************************************/
10608 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10609 unsigned long long sector
, int length
)
10611 struct intel_super
*super
= a
->container
->sb
;
10615 ord
= imsm_disk_slot_to_ord(a
, slot
);
10619 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10621 super
->updates_pending
++;
10625 /*******************************************************************************
10626 * Function: imsm_get_badblocks
10627 * Description: This routine get list of bad blocks for an array
10631 * slot : disk number
10633 * bb : structure containing bad blocks
10635 ******************************************************************************/
10636 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10638 int inst
= a
->info
.container_member
;
10639 struct intel_super
*super
= a
->container
->sb
;
10640 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10641 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10644 ord
= imsm_disk_slot_to_ord(a
, slot
);
10648 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10649 per_dev_array_size(map
), &super
->bb
);
10653 /*******************************************************************************
10654 * Function: examine_badblocks_imsm
10655 * Description: Prints list of bad blocks on a disk to the standard output
10658 * st : metadata handler
10659 * fd : open file descriptor for device
10660 * devname : device name
10664 ******************************************************************************/
10665 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10667 struct intel_super
*super
= st
->sb
;
10668 struct bbm_log
*log
= super
->bbm_log
;
10669 struct dl
*d
= NULL
;
10672 for (d
= super
->disks
; d
; d
= d
->next
) {
10673 if (strcmp(d
->devname
, devname
) == 0)
10677 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10678 pr_err("%s doesn't appear to be part of a raid array\n",
10685 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10687 for (i
= 0; i
< log
->entry_count
; i
++) {
10688 if (entry
[i
].disk_ordinal
== d
->index
) {
10689 unsigned long long sector
= __le48_to_cpu(
10690 &entry
[i
].defective_block_start
);
10691 int cnt
= entry
[i
].marked_count
+ 1;
10694 printf("Bad-blocks on %s:\n", devname
);
10698 printf("%20llu for %d sectors\n", sector
, cnt
);
10704 printf("No bad-blocks list configured on %s\n", devname
);
10708 /*******************************************************************************
10709 * Function: init_migr_record_imsm
10710 * Description: Function inits imsm migration record
10712 * super : imsm internal array info
10713 * dev : device under migration
10714 * info : general array info to find the smallest device
10717 ******************************************************************************/
10718 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10719 struct mdinfo
*info
)
10721 struct intel_super
*super
= st
->sb
;
10722 struct migr_record
*migr_rec
= super
->migr_rec
;
10723 int new_data_disks
;
10724 unsigned long long dsize
, dev_sectors
;
10725 long long unsigned min_dev_sectors
= -1LLU;
10729 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10730 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10731 unsigned long long num_migr_units
;
10732 unsigned long long array_blocks
;
10734 memset(migr_rec
, 0, sizeof(struct migr_record
));
10735 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10737 /* only ascending reshape supported now */
10738 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10740 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10741 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10742 migr_rec
->dest_depth_per_unit
*=
10743 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10744 new_data_disks
= imsm_num_data_members(map_dest
);
10745 migr_rec
->blocks_per_unit
=
10746 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10747 migr_rec
->dest_depth_per_unit
=
10748 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10749 array_blocks
= info
->component_size
* new_data_disks
;
10751 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10753 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10755 set_num_migr_units(migr_rec
, num_migr_units
);
10757 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10758 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10760 /* Find the smallest dev */
10761 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10762 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10763 fd
= dev_open(nm
, O_RDONLY
);
10766 get_dev_size(fd
, NULL
, &dsize
);
10767 dev_sectors
= dsize
/ 512;
10768 if (dev_sectors
< min_dev_sectors
)
10769 min_dev_sectors
= dev_sectors
;
10772 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
10773 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10775 write_imsm_migr_rec(st
);
10780 /*******************************************************************************
10781 * Function: save_backup_imsm
10782 * Description: Function saves critical data stripes to Migration Copy Area
10783 * and updates the current migration unit status.
10784 * Use restore_stripes() to form a destination stripe,
10785 * and to write it to the Copy Area.
10787 * st : supertype information
10788 * dev : imsm device that backup is saved for
10789 * info : general array info
10790 * buf : input buffer
10791 * length : length of data to backup (blocks_per_unit)
10795 ******************************************************************************/
10796 int save_backup_imsm(struct supertype
*st
,
10797 struct imsm_dev
*dev
,
10798 struct mdinfo
*info
,
10803 struct intel_super
*super
= st
->sb
;
10804 unsigned long long *target_offsets
;
10807 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10808 int new_disks
= map_dest
->num_members
;
10809 int dest_layout
= 0;
10811 unsigned long long start
;
10812 int data_disks
= imsm_num_data_members(map_dest
);
10814 targets
= xmalloc(new_disks
* sizeof(int));
10816 for (i
= 0; i
< new_disks
; i
++)
10819 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10821 start
= info
->reshape_progress
* 512;
10822 for (i
= 0; i
< new_disks
; i
++) {
10823 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
10824 /* move back copy area adderss, it will be moved forward
10825 * in restore_stripes() using start input variable
10827 target_offsets
[i
] -= start
/data_disks
;
10830 if (open_backup_targets(info
, new_disks
, targets
,
10834 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10835 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10837 if (restore_stripes(targets
, /* list of dest devices */
10838 target_offsets
, /* migration record offsets */
10841 map_dest
->raid_level
,
10843 -1, /* source backup file descriptor */
10844 0, /* input buf offset
10845 * always 0 buf is already offseted */
10849 pr_err("Error restoring stripes\n");
10857 close_targets(targets
, new_disks
);
10860 free(target_offsets
);
10865 /*******************************************************************************
10866 * Function: save_checkpoint_imsm
10867 * Description: Function called for current unit status update
10868 * in the migration record. It writes it to disk.
10870 * super : imsm internal array info
10871 * info : general array info
10875 * 2: failure, means no valid migration record
10876 * / no general migration in progress /
10877 ******************************************************************************/
10878 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10880 struct intel_super
*super
= st
->sb
;
10881 unsigned long long blocks_per_unit
;
10882 unsigned long long curr_migr_unit
;
10884 if (load_imsm_migr_rec(super
, info
) != 0) {
10885 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10889 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10890 if (blocks_per_unit
== 0) {
10891 dprintf("imsm: no migration in progress.\n");
10894 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10895 /* check if array is alligned to copy area
10896 * if it is not alligned, add one to current migration unit value
10897 * this can happend on array reshape finish only
10899 if (info
->reshape_progress
% blocks_per_unit
)
10902 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
10903 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10904 set_migr_dest_1st_member_lba(super
->migr_rec
,
10905 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
10907 if (write_imsm_migr_rec(st
) < 0) {
10908 dprintf("imsm: Cannot write migration record outside backup area\n");
10915 /*******************************************************************************
10916 * Function: recover_backup_imsm
10917 * Description: Function recovers critical data from the Migration Copy Area
10918 * while assembling an array.
10920 * super : imsm internal array info
10921 * info : general array info
10923 * 0 : success (or there is no data to recover)
10925 ******************************************************************************/
10926 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10928 struct intel_super
*super
= st
->sb
;
10929 struct migr_record
*migr_rec
= super
->migr_rec
;
10930 struct imsm_map
*map_dest
;
10931 struct intel_dev
*id
= NULL
;
10932 unsigned long long read_offset
;
10933 unsigned long long write_offset
;
10935 int *targets
= NULL
;
10936 int new_disks
, i
, err
;
10939 unsigned int sector_size
= super
->sector_size
;
10940 unsigned long curr_migr_unit
= current_migr_unit(migr_rec
);
10941 unsigned long num_migr_units
= get_num_migr_units(migr_rec
);
10943 int skipped_disks
= 0;
10945 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10949 /* recover data only during assemblation */
10950 if (strncmp(buffer
, "inactive", 8) != 0)
10952 /* no data to recover */
10953 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10955 if (curr_migr_unit
>= num_migr_units
)
10958 /* find device during reshape */
10959 for (id
= super
->devlist
; id
; id
= id
->next
)
10960 if (is_gen_migration(id
->dev
))
10965 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10966 new_disks
= map_dest
->num_members
;
10968 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
10970 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
10971 pba_of_lba0(map_dest
)) * 512;
10973 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10974 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10976 targets
= xcalloc(new_disks
, sizeof(int));
10978 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10979 pr_err("Cannot open some devices belonging to array.\n");
10983 for (i
= 0; i
< new_disks
; i
++) {
10984 if (targets
[i
] < 0) {
10988 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
10989 pr_err("Cannot seek to block: %s\n",
10994 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
10995 pr_err("Cannot read copy area block: %s\n",
11000 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
11001 pr_err("Cannot seek to block: %s\n",
11006 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
11007 pr_err("Cannot restore block: %s\n",
11014 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11018 pr_err("Cannot restore data from backup. Too many failed disks\n");
11022 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11023 /* ignore error == 2, this can mean end of reshape here
11025 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11031 for (i
= 0; i
< new_disks
; i
++)
11040 static char disk_by_path
[] = "/dev/disk/by-path/";
11042 static const char *imsm_get_disk_controller_domain(const char *path
)
11044 char disk_path
[PATH_MAX
];
11048 strcpy(disk_path
, disk_by_path
);
11049 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
11050 if (stat(disk_path
, &st
) == 0) {
11051 struct sys_dev
* hba
;
11054 path
= devt_to_devpath(st
.st_rdev
);
11057 hba
= find_disk_attached_hba(-1, path
);
11058 if (hba
&& hba
->type
== SYS_DEV_SAS
)
11060 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
11062 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
11064 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
11068 dprintf("path: %s hba: %s attached: %s\n",
11069 path
, (hba
) ? hba
->path
: "NULL", drv
);
11075 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11077 static char devnm
[32];
11078 char subdev_name
[20];
11079 struct mdstat_ent
*mdstat
;
11081 sprintf(subdev_name
, "%d", subdev
);
11082 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11086 strcpy(devnm
, mdstat
->devnm
);
11087 free_mdstat(mdstat
);
11091 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11092 struct geo_params
*geo
,
11093 int *old_raid_disks
,
11096 /* currently we only support increasing the number of devices
11097 * for a container. This increases the number of device for each
11098 * member array. They must all be RAID0 or RAID5.
11101 struct mdinfo
*info
, *member
;
11102 int devices_that_can_grow
= 0;
11104 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11106 if (geo
->size
> 0 ||
11107 geo
->level
!= UnSet
||
11108 geo
->layout
!= UnSet
||
11109 geo
->chunksize
!= 0 ||
11110 geo
->raid_disks
== UnSet
) {
11111 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11115 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11116 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11120 info
= container_content_imsm(st
, NULL
);
11121 for (member
= info
; member
; member
= member
->next
) {
11124 dprintf("imsm: checking device_num: %i\n",
11125 member
->container_member
);
11127 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11128 /* we work on container for Online Capacity Expansion
11129 * only so raid_disks has to grow
11131 dprintf("imsm: for container operation raid disks increase is required\n");
11135 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11136 /* we cannot use this container with other raid level
11138 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11139 info
->array
.level
);
11142 /* check for platform support
11143 * for this raid level configuration
11145 struct intel_super
*super
= st
->sb
;
11146 if (!is_raid_level_supported(super
->orom
,
11147 member
->array
.level
,
11148 geo
->raid_disks
)) {
11149 dprintf("platform does not support raid%d with %d disk%s\n",
11152 geo
->raid_disks
> 1 ? "s" : "");
11155 /* check if component size is aligned to chunk size
11157 if (info
->component_size
%
11158 (info
->array
.chunk_size
/512)) {
11159 dprintf("Component size is not aligned to chunk size\n");
11164 if (*old_raid_disks
&&
11165 info
->array
.raid_disks
!= *old_raid_disks
)
11167 *old_raid_disks
= info
->array
.raid_disks
;
11169 /* All raid5 and raid0 volumes in container
11170 * have to be ready for Online Capacity Expansion
11171 * so they need to be assembled. We have already
11172 * checked that no recovery etc is happening.
11174 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11175 st
->container_devnm
);
11176 if (result
== NULL
) {
11177 dprintf("imsm: cannot find array\n");
11180 devices_that_can_grow
++;
11183 if (!member
&& devices_that_can_grow
)
11187 dprintf("Container operation allowed\n");
11189 dprintf("Error: %i\n", ret_val
);
11194 /* Function: get_spares_for_grow
11195 * Description: Allocates memory and creates list of spare devices
11196 * avaliable in container. Checks if spare drive size is acceptable.
11197 * Parameters: Pointer to the supertype structure
11198 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11201 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11203 struct spare_criteria sc
;
11205 get_spare_criteria_imsm(st
, &sc
);
11206 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11209 /******************************************************************************
11210 * function: imsm_create_metadata_update_for_reshape
11211 * Function creates update for whole IMSM container.
11213 ******************************************************************************/
11214 static int imsm_create_metadata_update_for_reshape(
11215 struct supertype
*st
,
11216 struct geo_params
*geo
,
11217 int old_raid_disks
,
11218 struct imsm_update_reshape
**updatep
)
11220 struct intel_super
*super
= st
->sb
;
11221 struct imsm_super
*mpb
= super
->anchor
;
11222 int update_memory_size
;
11223 struct imsm_update_reshape
*u
;
11224 struct mdinfo
*spares
;
11227 struct mdinfo
*dev
;
11229 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11231 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11233 /* size of all update data without anchor */
11234 update_memory_size
= sizeof(struct imsm_update_reshape
);
11236 /* now add space for spare disks that we need to add. */
11237 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11239 u
= xcalloc(1, update_memory_size
);
11240 u
->type
= update_reshape_container_disks
;
11241 u
->old_raid_disks
= old_raid_disks
;
11242 u
->new_raid_disks
= geo
->raid_disks
;
11244 /* now get spare disks list
11246 spares
= get_spares_for_grow(st
);
11248 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11249 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11254 /* we have got spares
11255 * update disk list in imsm_disk list table in anchor
11257 dprintf("imsm: %i spares are available.\n\n",
11258 spares
->array
.spare_disks
);
11260 dev
= spares
->devs
;
11261 for (i
= 0; i
< delta_disks
; i
++) {
11266 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11268 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11269 dl
->index
= mpb
->num_disks
;
11277 sysfs_free(spares
);
11279 dprintf("imsm: reshape update preparation :");
11280 if (i
== delta_disks
) {
11281 dprintf_cont(" OK\n");
11283 return update_memory_size
;
11286 dprintf_cont(" Error\n");
11291 /******************************************************************************
11292 * function: imsm_create_metadata_update_for_size_change()
11293 * Creates update for IMSM array for array size change.
11295 ******************************************************************************/
11296 static int imsm_create_metadata_update_for_size_change(
11297 struct supertype
*st
,
11298 struct geo_params
*geo
,
11299 struct imsm_update_size_change
**updatep
)
11301 struct intel_super
*super
= st
->sb
;
11302 int update_memory_size
;
11303 struct imsm_update_size_change
*u
;
11305 dprintf("(enter) New size = %llu\n", geo
->size
);
11307 /* size of all update data without anchor */
11308 update_memory_size
= sizeof(struct imsm_update_size_change
);
11310 u
= xcalloc(1, update_memory_size
);
11311 u
->type
= update_size_change
;
11312 u
->subdev
= super
->current_vol
;
11313 u
->new_size
= geo
->size
;
11315 dprintf("imsm: reshape update preparation : OK\n");
11318 return update_memory_size
;
11321 /******************************************************************************
11322 * function: imsm_create_metadata_update_for_migration()
11323 * Creates update for IMSM array.
11325 ******************************************************************************/
11326 static int imsm_create_metadata_update_for_migration(
11327 struct supertype
*st
,
11328 struct geo_params
*geo
,
11329 struct imsm_update_reshape_migration
**updatep
)
11331 struct intel_super
*super
= st
->sb
;
11332 int update_memory_size
;
11333 struct imsm_update_reshape_migration
*u
;
11334 struct imsm_dev
*dev
;
11335 int previous_level
= -1;
11337 dprintf("(enter) New Level = %i\n", geo
->level
);
11339 /* size of all update data without anchor */
11340 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11342 u
= xcalloc(1, update_memory_size
);
11343 u
->type
= update_reshape_migration
;
11344 u
->subdev
= super
->current_vol
;
11345 u
->new_level
= geo
->level
;
11346 u
->new_layout
= geo
->layout
;
11347 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11348 u
->new_disks
[0] = -1;
11349 u
->new_chunksize
= -1;
11351 dev
= get_imsm_dev(super
, u
->subdev
);
11353 struct imsm_map
*map
;
11355 map
= get_imsm_map(dev
, MAP_0
);
11357 int current_chunk_size
=
11358 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11360 if (geo
->chunksize
!= current_chunk_size
) {
11361 u
->new_chunksize
= geo
->chunksize
/ 1024;
11362 dprintf("imsm: chunk size change from %i to %i\n",
11363 current_chunk_size
, u
->new_chunksize
);
11365 previous_level
= map
->raid_level
;
11368 if (geo
->level
== 5 && previous_level
== 0) {
11369 struct mdinfo
*spares
= NULL
;
11371 u
->new_raid_disks
++;
11372 spares
= get_spares_for_grow(st
);
11373 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11375 sysfs_free(spares
);
11376 update_memory_size
= 0;
11377 pr_err("cannot get spare device for requested migration\n");
11380 sysfs_free(spares
);
11382 dprintf("imsm: reshape update preparation : OK\n");
11385 return update_memory_size
;
11388 static void imsm_update_metadata_locally(struct supertype
*st
,
11389 void *buf
, int len
)
11391 struct metadata_update mu
;
11396 mu
.space_list
= NULL
;
11398 if (imsm_prepare_update(st
, &mu
))
11399 imsm_process_update(st
, &mu
);
11401 while (mu
.space_list
) {
11402 void **space
= mu
.space_list
;
11403 mu
.space_list
= *space
;
11408 /***************************************************************************
11409 * Function: imsm_analyze_change
11410 * Description: Function analyze change for single volume
11411 * and validate if transition is supported
11412 * Parameters: Geometry parameters, supertype structure,
11413 * metadata change direction (apply/rollback)
11414 * Returns: Operation type code on success, -1 if fail
11415 ****************************************************************************/
11416 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11417 struct geo_params
*geo
,
11420 struct mdinfo info
;
11422 int check_devs
= 0;
11424 /* number of added/removed disks in operation result */
11425 int devNumChange
= 0;
11426 /* imsm compatible layout value for array geometry verification */
11427 int imsm_layout
= -1;
11429 struct imsm_dev
*dev
;
11430 struct imsm_map
*map
;
11431 struct intel_super
*super
;
11432 unsigned long long current_size
;
11433 unsigned long long free_size
;
11434 unsigned long long max_size
;
11437 getinfo_super_imsm_volume(st
, &info
, NULL
);
11438 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11439 geo
->level
!= UnSet
) {
11440 switch (info
.array
.level
) {
11442 if (geo
->level
== 5) {
11443 change
= CH_MIGRATION
;
11444 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11445 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11447 goto analyse_change_exit
;
11449 imsm_layout
= geo
->layout
;
11451 devNumChange
= 1; /* parity disk added */
11452 } else if (geo
->level
== 10) {
11453 change
= CH_TAKEOVER
;
11455 devNumChange
= 2; /* two mirrors added */
11456 imsm_layout
= 0x102; /* imsm supported layout */
11461 if (geo
->level
== 0) {
11462 change
= CH_TAKEOVER
;
11464 devNumChange
= -(geo
->raid_disks
/2);
11465 imsm_layout
= 0; /* imsm raid0 layout */
11469 if (change
== -1) {
11470 pr_err("Error. Level Migration from %d to %d not supported!\n",
11471 info
.array
.level
, geo
->level
);
11472 goto analyse_change_exit
;
11475 geo
->level
= info
.array
.level
;
11477 if (geo
->layout
!= info
.array
.layout
&&
11478 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11479 change
= CH_MIGRATION
;
11480 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11481 geo
->layout
== 5) {
11482 /* reshape 5 -> 4 */
11483 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11484 geo
->layout
== 0) {
11485 /* reshape 4 -> 5 */
11489 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11490 info
.array
.layout
, geo
->layout
);
11492 goto analyse_change_exit
;
11495 geo
->layout
= info
.array
.layout
;
11496 if (imsm_layout
== -1)
11497 imsm_layout
= info
.array
.layout
;
11500 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11501 geo
->chunksize
!= info
.array
.chunk_size
) {
11502 if (info
.array
.level
== 10) {
11503 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11505 goto analyse_change_exit
;
11506 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11507 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11508 geo
->chunksize
/1024, info
.component_size
/2);
11510 goto analyse_change_exit
;
11512 change
= CH_MIGRATION
;
11514 geo
->chunksize
= info
.array
.chunk_size
;
11517 chunk
= geo
->chunksize
/ 1024;
11520 dev
= get_imsm_dev(super
, super
->current_vol
);
11521 map
= get_imsm_map(dev
, MAP_0
);
11522 data_disks
= imsm_num_data_members(map
);
11523 /* compute current size per disk member
11525 current_size
= info
.custom_array_size
/ data_disks
;
11527 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11528 /* align component size
11530 geo
->size
= imsm_component_size_alignment_check(
11531 get_imsm_raid_level(dev
->vol
.map
),
11532 chunk
* 1024, super
->sector_size
,
11534 if (geo
->size
== 0) {
11535 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11537 goto analyse_change_exit
;
11541 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11542 if (change
!= -1) {
11543 pr_err("Error. Size change should be the only one at a time.\n");
11545 goto analyse_change_exit
;
11547 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11548 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11549 super
->current_vol
, st
->devnm
);
11550 goto analyse_change_exit
;
11552 /* check the maximum available size
11554 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11555 0, chunk
, &free_size
);
11557 /* Cannot find maximum available space
11561 max_size
= free_size
+ current_size
;
11562 /* align component size
11564 max_size
= imsm_component_size_alignment_check(
11565 get_imsm_raid_level(dev
->vol
.map
),
11566 chunk
* 1024, super
->sector_size
,
11569 if (geo
->size
== MAX_SIZE
) {
11570 /* requested size change to the maximum available size
11572 if (max_size
== 0) {
11573 pr_err("Error. Cannot find maximum available space.\n");
11575 goto analyse_change_exit
;
11577 geo
->size
= max_size
;
11580 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11581 /* accept size for rollback only
11584 /* round size due to metadata compatibility
11586 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11587 << SECT_PER_MB_SHIFT
;
11588 dprintf("Prepare update for size change to %llu\n",
11590 if (current_size
>= geo
->size
) {
11591 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11592 current_size
, geo
->size
);
11593 goto analyse_change_exit
;
11595 if (max_size
&& geo
->size
> max_size
) {
11596 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11597 max_size
, geo
->size
);
11598 goto analyse_change_exit
;
11601 geo
->size
*= data_disks
;
11602 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11603 change
= CH_ARRAY_SIZE
;
11605 if (!validate_geometry_imsm(st
,
11608 geo
->raid_disks
+ devNumChange
,
11610 geo
->size
, INVALID_SECTORS
,
11611 0, 0, info
.consistency_policy
, 1))
11615 struct intel_super
*super
= st
->sb
;
11616 struct imsm_super
*mpb
= super
->anchor
;
11618 if (mpb
->num_raid_devs
> 1) {
11619 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11625 analyse_change_exit
:
11626 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11627 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11628 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11634 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11636 struct intel_super
*super
= st
->sb
;
11637 struct imsm_update_takeover
*u
;
11639 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11641 u
->type
= update_takeover
;
11642 u
->subarray
= super
->current_vol
;
11644 /* 10->0 transition */
11645 if (geo
->level
== 0)
11646 u
->direction
= R10_TO_R0
;
11648 /* 0->10 transition */
11649 if (geo
->level
== 10)
11650 u
->direction
= R0_TO_R10
;
11652 /* update metadata locally */
11653 imsm_update_metadata_locally(st
, u
,
11654 sizeof(struct imsm_update_takeover
));
11655 /* and possibly remotely */
11656 if (st
->update_tail
)
11657 append_metadata_update(st
, u
,
11658 sizeof(struct imsm_update_takeover
));
11665 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11667 int layout
, int chunksize
, int raid_disks
,
11668 int delta_disks
, char *backup
, char *dev
,
11669 int direction
, int verbose
)
11672 struct geo_params geo
;
11674 dprintf("(enter)\n");
11676 memset(&geo
, 0, sizeof(struct geo_params
));
11678 geo
.dev_name
= dev
;
11679 strcpy(geo
.devnm
, st
->devnm
);
11682 geo
.layout
= layout
;
11683 geo
.chunksize
= chunksize
;
11684 geo
.raid_disks
= raid_disks
;
11685 if (delta_disks
!= UnSet
)
11686 geo
.raid_disks
+= delta_disks
;
11688 dprintf("for level : %i\n", geo
.level
);
11689 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11691 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11692 /* On container level we can only increase number of devices. */
11693 dprintf("imsm: info: Container operation\n");
11694 int old_raid_disks
= 0;
11696 if (imsm_reshape_is_allowed_on_container(
11697 st
, &geo
, &old_raid_disks
, direction
)) {
11698 struct imsm_update_reshape
*u
= NULL
;
11701 len
= imsm_create_metadata_update_for_reshape(
11702 st
, &geo
, old_raid_disks
, &u
);
11705 dprintf("imsm: Cannot prepare update\n");
11706 goto exit_imsm_reshape_super
;
11710 /* update metadata locally */
11711 imsm_update_metadata_locally(st
, u
, len
);
11712 /* and possibly remotely */
11713 if (st
->update_tail
)
11714 append_metadata_update(st
, u
, len
);
11719 pr_err("(imsm) Operation is not allowed on this container\n");
11722 /* On volume level we support following operations
11723 * - takeover: raid10 -> raid0; raid0 -> raid10
11724 * - chunk size migration
11725 * - migration: raid5 -> raid0; raid0 -> raid5
11727 struct intel_super
*super
= st
->sb
;
11728 struct intel_dev
*dev
= super
->devlist
;
11730 dprintf("imsm: info: Volume operation\n");
11731 /* find requested device */
11734 imsm_find_array_devnm_by_subdev(
11735 dev
->index
, st
->container_devnm
);
11736 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11741 pr_err("Cannot find %s (%s) subarray\n",
11742 geo
.dev_name
, geo
.devnm
);
11743 goto exit_imsm_reshape_super
;
11745 super
->current_vol
= dev
->index
;
11746 change
= imsm_analyze_change(st
, &geo
, direction
);
11749 ret_val
= imsm_takeover(st
, &geo
);
11751 case CH_MIGRATION
: {
11752 struct imsm_update_reshape_migration
*u
= NULL
;
11754 imsm_create_metadata_update_for_migration(
11757 dprintf("imsm: Cannot prepare update\n");
11761 /* update metadata locally */
11762 imsm_update_metadata_locally(st
, u
, len
);
11763 /* and possibly remotely */
11764 if (st
->update_tail
)
11765 append_metadata_update(st
, u
, len
);
11770 case CH_ARRAY_SIZE
: {
11771 struct imsm_update_size_change
*u
= NULL
;
11773 imsm_create_metadata_update_for_size_change(
11776 dprintf("imsm: Cannot prepare update\n");
11780 /* update metadata locally */
11781 imsm_update_metadata_locally(st
, u
, len
);
11782 /* and possibly remotely */
11783 if (st
->update_tail
)
11784 append_metadata_update(st
, u
, len
);
11794 exit_imsm_reshape_super
:
11795 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11799 #define COMPLETED_OK 0
11800 #define COMPLETED_NONE 1
11801 #define COMPLETED_DELAYED 2
11803 static int read_completed(int fd
, unsigned long long *val
)
11808 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11812 ret
= COMPLETED_OK
;
11813 if (strncmp(buf
, "none", 4) == 0) {
11814 ret
= COMPLETED_NONE
;
11815 } else if (strncmp(buf
, "delayed", 7) == 0) {
11816 ret
= COMPLETED_DELAYED
;
11819 *val
= strtoull(buf
, &ep
, 0);
11820 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11826 /*******************************************************************************
11827 * Function: wait_for_reshape_imsm
11828 * Description: Function writes new sync_max value and waits until
11829 * reshape process reach new position
11831 * sra : general array info
11832 * ndata : number of disks in new array's layout
11835 * 1 : there is no reshape in progress,
11837 ******************************************************************************/
11838 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11840 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11842 unsigned long long completed
;
11843 /* to_complete : new sync_max position */
11844 unsigned long long to_complete
= sra
->reshape_progress
;
11845 unsigned long long position_to_set
= to_complete
/ ndata
;
11848 dprintf("cannot open reshape_position\n");
11853 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11855 dprintf("cannot read reshape_position (no reshape in progres)\n");
11864 if (completed
> position_to_set
) {
11865 dprintf("wrong next position to set %llu (%llu)\n",
11866 to_complete
, position_to_set
);
11870 dprintf("Position set: %llu\n", position_to_set
);
11871 if (sysfs_set_num(sra
, NULL
, "sync_max",
11872 position_to_set
) != 0) {
11873 dprintf("cannot set reshape position to %llu\n",
11882 int timeout
= 3000;
11884 sysfs_wait(fd
, &timeout
);
11885 if (sysfs_get_str(sra
, NULL
, "sync_action",
11887 strncmp(action
, "reshape", 7) != 0) {
11888 if (strncmp(action
, "idle", 4) == 0)
11894 rc
= read_completed(fd
, &completed
);
11896 dprintf("cannot read reshape_position (in loop)\n");
11899 } else if (rc
== COMPLETED_NONE
)
11901 } while (completed
< position_to_set
);
11907 /*******************************************************************************
11908 * Function: check_degradation_change
11909 * Description: Check that array hasn't become failed.
11911 * info : for sysfs access
11912 * sources : source disks descriptors
11913 * degraded: previous degradation level
11915 * degradation level
11916 ******************************************************************************/
11917 int check_degradation_change(struct mdinfo
*info
,
11921 unsigned long long new_degraded
;
11924 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11925 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11926 /* check each device to ensure it is still working */
11929 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11930 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11932 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11935 if (sysfs_get_str(info
,
11936 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11937 strstr(sbuf
, "faulty") ||
11938 strstr(sbuf
, "in_sync") == NULL
) {
11939 /* this device is dead */
11940 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11941 if (sd
->disk
.raid_disk
>= 0 &&
11942 sources
[sd
->disk
.raid_disk
] >= 0) {
11944 sd
->disk
.raid_disk
]);
11945 sources
[sd
->disk
.raid_disk
] =
11954 return new_degraded
;
11957 /*******************************************************************************
11958 * Function: imsm_manage_reshape
11959 * Description: Function finds array under reshape and it manages reshape
11960 * process. It creates stripes backups (if required) and sets
11963 * afd : Backup handle (nattive) - not used
11964 * sra : general array info
11965 * reshape : reshape parameters - not used
11966 * st : supertype structure
11967 * blocks : size of critical section [blocks]
11968 * fds : table of source device descriptor
11969 * offsets : start of array (offest per devices)
11971 * destfd : table of destination device descriptor
11972 * destoffsets : table of destination offsets (per device)
11974 * 1 : success, reshape is done
11976 ******************************************************************************/
11977 static int imsm_manage_reshape(
11978 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
11979 struct supertype
*st
, unsigned long backup_blocks
,
11980 int *fds
, unsigned long long *offsets
,
11981 int dests
, int *destfd
, unsigned long long *destoffsets
)
11984 struct intel_super
*super
= st
->sb
;
11985 struct intel_dev
*dv
;
11986 unsigned int sector_size
= super
->sector_size
;
11987 struct imsm_dev
*dev
= NULL
;
11988 struct imsm_map
*map_src
, *map_dest
;
11989 int migr_vol_qan
= 0;
11990 int ndata
, odata
; /* [bytes] */
11991 int chunk
; /* [bytes] */
11992 struct migr_record
*migr_rec
;
11994 unsigned int buf_size
; /* [bytes] */
11995 unsigned long long max_position
; /* array size [bytes] */
11996 unsigned long long next_step
; /* [blocks]/[bytes] */
11997 unsigned long long old_data_stripe_length
;
11998 unsigned long long start_src
; /* [bytes] */
11999 unsigned long long start
; /* [bytes] */
12000 unsigned long long start_buf_shift
; /* [bytes] */
12002 int source_layout
= 0;
12007 if (!fds
|| !offsets
)
12010 /* Find volume during the reshape */
12011 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
12012 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12013 dv
->dev
->vol
.migr_state
== 1) {
12018 /* Only one volume can migrate at the same time */
12019 if (migr_vol_qan
!= 1) {
12020 pr_err("%s", migr_vol_qan
?
12021 "Number of migrating volumes greater than 1\n" :
12022 "There is no volume during migrationg\n");
12026 map_dest
= get_imsm_map(dev
, MAP_0
);
12027 map_src
= get_imsm_map(dev
, MAP_1
);
12028 if (map_src
== NULL
)
12031 ndata
= imsm_num_data_members(map_dest
);
12032 odata
= imsm_num_data_members(map_src
);
12034 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12035 old_data_stripe_length
= odata
* chunk
;
12037 migr_rec
= super
->migr_rec
;
12039 /* initialize migration record for start condition */
12040 if (sra
->reshape_progress
== 0)
12041 init_migr_record_imsm(st
, dev
, sra
);
12043 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12044 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12047 /* Save checkpoint to update migration record for current
12048 * reshape position (in md). It can be farther than current
12049 * reshape position in metadata.
12051 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12052 /* ignore error == 2, this can mean end of reshape here
12054 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12059 /* size for data */
12060 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12061 /* extend buffer size for parity disk */
12062 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12063 /* add space for stripe alignment */
12064 buf_size
+= old_data_stripe_length
;
12065 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12066 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12070 max_position
= sra
->component_size
* ndata
;
12071 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12073 while (current_migr_unit(migr_rec
) <
12074 get_num_migr_units(migr_rec
)) {
12075 /* current reshape position [blocks] */
12076 unsigned long long current_position
=
12077 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12078 * current_migr_unit(migr_rec
);
12079 unsigned long long border
;
12081 /* Check that array hasn't become failed.
12083 degraded
= check_degradation_change(sra
, fds
, degraded
);
12084 if (degraded
> 1) {
12085 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12089 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12091 if ((current_position
+ next_step
) > max_position
)
12092 next_step
= max_position
- current_position
;
12094 start
= current_position
* 512;
12096 /* align reading start to old geometry */
12097 start_buf_shift
= start
% old_data_stripe_length
;
12098 start_src
= start
- start_buf_shift
;
12100 border
= (start_src
/ odata
) - (start
/ ndata
);
12102 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12103 /* save critical stripes to buf
12104 * start - start address of current unit
12105 * to backup [bytes]
12106 * start_src - start address of current unit
12107 * to backup alligned to source array
12110 unsigned long long next_step_filler
;
12111 unsigned long long copy_length
= next_step
* 512;
12113 /* allign copy area length to stripe in old geometry */
12114 next_step_filler
= ((copy_length
+ start_buf_shift
)
12115 % old_data_stripe_length
);
12116 if (next_step_filler
)
12117 next_step_filler
= (old_data_stripe_length
12118 - next_step_filler
);
12119 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12120 start
, start_src
, copy_length
,
12121 start_buf_shift
, next_step_filler
);
12123 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12124 chunk
, map_src
->raid_level
,
12125 source_layout
, 0, NULL
, start_src
,
12127 next_step_filler
+ start_buf_shift
,
12129 dprintf("imsm: Cannot save stripes to buffer\n");
12132 /* Convert data to destination format and store it
12133 * in backup general migration area
12135 if (save_backup_imsm(st
, dev
, sra
,
12136 buf
+ start_buf_shift
, copy_length
)) {
12137 dprintf("imsm: Cannot save stripes to target devices\n");
12140 if (save_checkpoint_imsm(st
, sra
,
12141 UNIT_SRC_IN_CP_AREA
)) {
12142 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12146 /* set next step to use whole border area */
12147 border
/= next_step
;
12149 next_step
*= border
;
12151 /* When data backed up, checkpoint stored,
12152 * kick the kernel to reshape unit of data
12154 next_step
= next_step
+ sra
->reshape_progress
;
12155 /* limit next step to array max position */
12156 if (next_step
> max_position
)
12157 next_step
= max_position
;
12158 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12159 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12160 sra
->reshape_progress
= next_step
;
12162 /* wait until reshape finish */
12163 if (wait_for_reshape_imsm(sra
, ndata
)) {
12164 dprintf("wait_for_reshape_imsm returned error!\n");
12170 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12171 /* ignore error == 2, this can mean end of reshape here
12173 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12179 /* clear migr_rec on disks after successful migration */
12182 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12183 for (d
= super
->disks
; d
; d
= d
->next
) {
12184 if (d
->index
< 0 || is_failed(&d
->disk
))
12186 unsigned long long dsize
;
12188 get_dev_size(d
->fd
, NULL
, &dsize
);
12189 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12191 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12192 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12193 MIGR_REC_BUF_SECTORS
*sector_size
)
12194 perror("Write migr_rec failed");
12198 /* return '1' if done */
12202 /* See Grow.c: abort_reshape() for further explanation */
12203 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12204 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12205 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12210 struct superswitch super_imsm
= {
12211 .examine_super
= examine_super_imsm
,
12212 .brief_examine_super
= brief_examine_super_imsm
,
12213 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12214 .export_examine_super
= export_examine_super_imsm
,
12215 .detail_super
= detail_super_imsm
,
12216 .brief_detail_super
= brief_detail_super_imsm
,
12217 .write_init_super
= write_init_super_imsm
,
12218 .validate_geometry
= validate_geometry_imsm
,
12219 .add_to_super
= add_to_super_imsm
,
12220 .remove_from_super
= remove_from_super_imsm
,
12221 .detail_platform
= detail_platform_imsm
,
12222 .export_detail_platform
= export_detail_platform_imsm
,
12223 .kill_subarray
= kill_subarray_imsm
,
12224 .update_subarray
= update_subarray_imsm
,
12225 .load_container
= load_container_imsm
,
12226 .default_geometry
= default_geometry_imsm
,
12227 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12228 .reshape_super
= imsm_reshape_super
,
12229 .manage_reshape
= imsm_manage_reshape
,
12230 .recover_backup
= recover_backup_imsm
,
12231 .examine_badblocks
= examine_badblocks_imsm
,
12232 .match_home
= match_home_imsm
,
12233 .uuid_from_super
= uuid_from_super_imsm
,
12234 .getinfo_super
= getinfo_super_imsm
,
12235 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12236 .update_super
= update_super_imsm
,
12238 .avail_size
= avail_size_imsm
,
12239 .get_spare_criteria
= get_spare_criteria_imsm
,
12241 .compare_super
= compare_super_imsm
,
12243 .load_super
= load_super_imsm
,
12244 .init_super
= init_super_imsm
,
12245 .store_super
= store_super_imsm
,
12246 .free_super
= free_super_imsm
,
12247 .match_metadata_desc
= match_metadata_desc_imsm
,
12248 .container_content
= container_content_imsm
,
12249 .validate_container
= validate_container_imsm
,
12251 .write_init_ppl
= write_init_ppl_imsm
,
12252 .validate_ppl
= validate_ppl_imsm
,
12258 .open_new
= imsm_open_new
,
12259 .set_array_state
= imsm_set_array_state
,
12260 .set_disk
= imsm_set_disk
,
12261 .sync_metadata
= imsm_sync_metadata
,
12262 .activate_spare
= imsm_activate_spare
,
12263 .process_update
= imsm_process_update
,
12264 .prepare_update
= imsm_prepare_update
,
12265 .record_bad_block
= imsm_record_badblock
,
12266 .clear_bad_block
= imsm_clear_badblock
,
12267 .get_bad_blocks
= imsm_get_badblocks
,