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
99 /* Disk configuration info. */
100 #define IMSM_MAX_DEVICES 255
102 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
103 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
104 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
105 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
106 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
107 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
108 #define JOURNAL_DISK __cpu_to_le32(0x2000000) /* Device marked as Journaling Drive */
109 __u32 status
; /* 0xF0 - 0xF3 */
110 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
111 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
112 #define IMSM_DISK_FILLERS 3
113 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
116 /* map selector for map managment
122 /* RAID map configuration infos. */
124 __u32 pba_of_lba0_lo
; /* start address of partition */
125 __u32 blocks_per_member_lo
;/* blocks per member */
126 __u32 num_data_stripes_lo
; /* number of data stripes */
127 __u16 blocks_per_strip
;
128 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
129 #define IMSM_T_STATE_NORMAL 0
130 #define IMSM_T_STATE_UNINITIALIZED 1
131 #define IMSM_T_STATE_DEGRADED 2
132 #define IMSM_T_STATE_FAILED 3
134 #define IMSM_T_RAID0 0
135 #define IMSM_T_RAID1 1
136 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
137 __u8 num_members
; /* number of member disks */
138 __u8 num_domains
; /* number of parity domains */
139 __u8 failed_disk_num
; /* valid only when state is degraded */
141 __u32 pba_of_lba0_hi
;
142 __u32 blocks_per_member_hi
;
143 __u32 num_data_stripes_hi
;
144 __u32 filler
[4]; /* expansion area */
145 #define IMSM_ORD_REBUILD (1 << 24)
146 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
147 * top byte contains some flags
149 } __attribute__ ((packed
));
152 __u32 curr_migr_unit
;
153 __u32 checkpoint_id
; /* id to access curr_migr_unit */
154 __u8 migr_state
; /* Normal or Migrating */
156 #define MIGR_REBUILD 1
157 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
158 #define MIGR_GEN_MIGR 3
159 #define MIGR_STATE_CHANGE 4
160 #define MIGR_REPAIR 5
161 __u8 migr_type
; /* Initializing, Rebuilding, ... */
162 #define RAIDVOL_CLEAN 0
163 #define RAIDVOL_DIRTY 1
164 #define RAIDVOL_DSRECORD_VALID 2
166 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
167 __u16 verify_errors
; /* number of mismatches */
168 __u16 bad_blocks
; /* number of bad blocks during verify */
170 struct imsm_map map
[1];
171 /* here comes another one if migr_state */
172 } __attribute__ ((packed
));
175 __u8 volume
[MAX_RAID_SERIAL_LEN
];
178 #define DEV_BOOTABLE __cpu_to_le32(0x01)
179 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
180 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
181 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
182 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
183 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
184 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
185 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
186 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
187 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
188 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
189 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
190 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
191 __u32 status
; /* Persistent RaidDev status */
192 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
196 __u8 cng_master_disk
;
200 __u16 my_vol_raid_dev_num
; /* Used in Unique volume Id for this RaidDev */
206 /* Unique Volume Id of the NvCache Volume associated with this volume */
207 __u32 nvc_vol_orig_family_num
;
208 __u16 nvc_vol_raid_dev_num
;
211 #define RWH_DISTRIBUTED 1
212 #define RWH_JOURNALING_DRIVE 2
213 #define RWH_MULTIPLE_DISTRIBUTED 3
214 #define RWH_MULTIPLE_PPLS_JOURNALING_DRIVE 4
215 #define RWH_MULTIPLE_OFF 5
216 __u8 rwh_policy
; /* Raid Write Hole Policy */
217 __u8 jd_serial
[MAX_RAID_SERIAL_LEN
]; /* Journal Drive serial number */
220 #define IMSM_DEV_FILLERS 3
221 __u32 filler
[IMSM_DEV_FILLERS
];
223 } __attribute__ ((packed
));
226 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
227 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
228 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
229 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
230 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
231 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
232 __u32 attributes
; /* 0x34 - 0x37 */
233 __u8 num_disks
; /* 0x38 Number of configured disks */
234 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
235 __u8 error_log_pos
; /* 0x3A */
236 __u8 fill
[1]; /* 0x3B */
237 __u32 cache_size
; /* 0x3c - 0x40 in mb */
238 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
239 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
240 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
241 __u16 num_raid_devs_created
; /* 0x4C - 0x4D Used for generating unique
242 * volume IDs for raid_dev created in this array
245 __u16 filler1
; /* 0x4E - 0x4F */
246 #define IMSM_FILLERS 34
247 __u32 filler
[IMSM_FILLERS
]; /* 0x50 - 0xD7 RAID_MPB_FILLERS */
248 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
249 /* here comes imsm_dev[num_raid_devs] */
250 /* here comes BBM logs */
251 } __attribute__ ((packed
));
253 #define BBM_LOG_MAX_ENTRIES 254
254 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
255 #define BBM_LOG_SIGNATURE 0xabadb10c
257 struct bbm_log_block_addr
{
260 } __attribute__ ((__packed__
));
262 struct bbm_log_entry
{
263 __u8 marked_count
; /* Number of blocks marked - 1 */
264 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
265 struct bbm_log_block_addr defective_block_start
;
266 } __attribute__ ((__packed__
));
269 __u32 signature
; /* 0xABADB10C */
271 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
272 } __attribute__ ((__packed__
));
274 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
276 #define BLOCKS_PER_KB (1024/512)
278 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
280 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
282 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
283 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
284 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
287 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
288 * be recovered using srcMap */
289 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
290 * already been migrated and must
291 * be recovered from checkpoint area */
293 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of single PPL, without the header */
296 __u32 rec_status
; /* Status used to determine how to restart
297 * migration in case it aborts
299 __u32 curr_migr_unit_lo
; /* 0..numMigrUnits-1 */
300 __u32 family_num
; /* Family number of MPB
301 * containing the RaidDev
302 * that is migrating */
303 __u32 ascending_migr
; /* True if migrating in increasing
305 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
306 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
308 * advances per unit-of-operation */
309 __u32 ckpt_area_pba_lo
; /* Pba of first block of ckpt copy area */
310 __u32 dest_1st_member_lba_lo
; /* First member lba on first
311 * stripe of destination */
312 __u32 num_migr_units_lo
; /* Total num migration units-of-op */
313 __u32 post_migr_vol_cap
; /* Size of volume after
314 * migration completes */
315 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
316 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
317 * migration ckpt record was read from
318 * (for recovered migrations) */
319 __u32 curr_migr_unit_hi
; /* 0..numMigrUnits-1 high order 32 bits */
320 __u32 ckpt_area_pba_hi
; /* Pba of first block of ckpt copy area
321 * high order 32 bits */
322 __u32 dest_1st_member_lba_hi
; /* First member lba on first stripe of
323 * destination - high order 32 bits */
324 __u32 num_migr_units_hi
; /* Total num migration units-of-op
325 * high order 32 bits */
326 } __attribute__ ((__packed__
));
331 * 2: metadata does not match
339 struct md_list
*next
;
342 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
344 static __u8
migr_type(struct imsm_dev
*dev
)
346 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
347 dev
->status
& DEV_VERIFY_AND_FIX
)
350 return dev
->vol
.migr_type
;
353 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
355 /* for compatibility with older oroms convert MIGR_REPAIR, into
356 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
358 if (migr_type
== MIGR_REPAIR
) {
359 dev
->vol
.migr_type
= MIGR_VERIFY
;
360 dev
->status
|= DEV_VERIFY_AND_FIX
;
362 dev
->vol
.migr_type
= migr_type
;
363 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
367 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
369 return ROUND_UP(bytes
, sector_size
) / sector_size
;
372 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
373 unsigned int sector_size
)
375 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
379 struct imsm_dev
*dev
;
380 struct intel_dev
*next
;
385 enum sys_dev_type type
;
388 struct intel_hba
*next
;
395 /* internal representation of IMSM metadata */
398 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
399 struct imsm_super
*anchor
; /* immovable parameters */
402 void *migr_rec_buf
; /* buffer for I/O operations */
403 struct migr_record
*migr_rec
; /* migration record */
405 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
406 array, it indicates that mdmon is allowed to clean migration
408 size_t len
; /* size of the 'buf' allocation */
409 size_t extra_space
; /* extra space in 'buf' that is not used yet */
410 void *next_buf
; /* for realloc'ing buf from the manager */
412 int updates_pending
; /* count of pending updates for mdmon */
413 int current_vol
; /* index of raid device undergoing creation */
414 unsigned long long create_offset
; /* common start for 'current_vol' */
415 __u32 random
; /* random data for seeding new family numbers */
416 struct intel_dev
*devlist
;
417 unsigned int sector_size
; /* sector size of used member drives */
421 __u8 serial
[MAX_RAID_SERIAL_LEN
];
424 struct imsm_disk disk
;
427 struct extent
*e
; /* for determining freespace @ create */
428 int raiddisk
; /* slot to fill in autolayout */
430 } *disks
, *current_disk
;
431 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
433 struct dl
*missing
; /* disks removed while we weren't looking */
434 struct bbm_log
*bbm_log
;
435 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
436 const struct imsm_orom
*orom
; /* platform firmware support */
437 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
438 struct md_bb bb
; /* memory for get_bad_blocks call */
442 struct imsm_disk disk
;
443 #define IMSM_UNKNOWN_OWNER (-1)
445 struct intel_disk
*next
;
449 unsigned long long start
, size
;
452 /* definitions of reshape process types */
453 enum imsm_reshape_type
{
459 /* definition of messages passed to imsm_process_update */
460 enum imsm_update_type
{
461 update_activate_spare
,
465 update_add_remove_disk
,
466 update_reshape_container_disks
,
467 update_reshape_migration
,
469 update_general_migration_checkpoint
,
471 update_prealloc_badblocks_mem
,
475 struct imsm_update_activate_spare
{
476 enum imsm_update_type type
;
480 struct imsm_update_activate_spare
*next
;
486 unsigned long long size
;
493 enum takeover_direction
{
497 struct imsm_update_takeover
{
498 enum imsm_update_type type
;
500 enum takeover_direction direction
;
503 struct imsm_update_reshape
{
504 enum imsm_update_type type
;
508 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
511 struct imsm_update_reshape_migration
{
512 enum imsm_update_type type
;
515 /* fields for array migration changes
522 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
525 struct imsm_update_size_change
{
526 enum imsm_update_type type
;
531 struct imsm_update_general_migration_checkpoint
{
532 enum imsm_update_type type
;
533 __u32 curr_migr_unit
;
537 __u8 serial
[MAX_RAID_SERIAL_LEN
];
540 struct imsm_update_create_array
{
541 enum imsm_update_type type
;
546 struct imsm_update_kill_array
{
547 enum imsm_update_type type
;
551 struct imsm_update_rename_array
{
552 enum imsm_update_type type
;
553 __u8 name
[MAX_RAID_SERIAL_LEN
];
557 struct imsm_update_add_remove_disk
{
558 enum imsm_update_type type
;
561 struct imsm_update_prealloc_bb_mem
{
562 enum imsm_update_type type
;
565 struct imsm_update_rwh_policy
{
566 enum imsm_update_type type
;
571 static const char *_sys_dev_type
[] = {
572 [SYS_DEV_UNKNOWN
] = "Unknown",
573 [SYS_DEV_SAS
] = "SAS",
574 [SYS_DEV_SATA
] = "SATA",
575 [SYS_DEV_NVME
] = "NVMe",
576 [SYS_DEV_VMD
] = "VMD"
579 const char *get_sys_dev_type(enum sys_dev_type type
)
581 if (type
>= SYS_DEV_MAX
)
582 type
= SYS_DEV_UNKNOWN
;
584 return _sys_dev_type
[type
];
587 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
589 struct intel_hba
*result
= xmalloc(sizeof(*result
));
591 result
->type
= device
->type
;
592 result
->path
= xstrdup(device
->path
);
594 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
600 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
602 struct intel_hba
*result
;
604 for (result
= hba
; result
; result
= result
->next
) {
605 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
611 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
613 struct intel_hba
*hba
;
615 /* check if disk attached to Intel HBA */
616 hba
= find_intel_hba(super
->hba
, device
);
619 /* Check if HBA is already attached to super */
620 if (super
->hba
== NULL
) {
621 super
->hba
= alloc_intel_hba(device
);
626 /* Intel metadata allows for all disks attached to the same type HBA.
627 * Do not support HBA types mixing
629 if (device
->type
!= hba
->type
)
632 /* Multiple same type HBAs can be used if they share the same OROM */
633 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
635 if (device_orom
!= super
->orom
)
641 hba
->next
= alloc_intel_hba(device
);
645 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
647 struct sys_dev
*list
, *elem
;
650 if ((list
= find_intel_devices()) == NULL
)
654 disk_path
= (char *) devname
;
656 disk_path
= diskfd_to_devpath(fd
);
661 for (elem
= list
; elem
; elem
= elem
->next
)
662 if (path_attached_to_hba(disk_path
, elem
->path
))
665 if (disk_path
!= devname
)
671 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
674 static struct supertype
*match_metadata_desc_imsm(char *arg
)
676 struct supertype
*st
;
678 if (strcmp(arg
, "imsm") != 0 &&
679 strcmp(arg
, "default") != 0
683 st
= xcalloc(1, sizeof(*st
));
684 st
->ss
= &super_imsm
;
685 st
->max_devs
= IMSM_MAX_DEVICES
;
686 st
->minor_version
= 0;
691 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
693 return &mpb
->sig
[MPB_SIG_LEN
];
696 /* retrieve a disk directly from the anchor when the anchor is known to be
697 * up-to-date, currently only at load time
699 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
701 if (index
>= mpb
->num_disks
)
703 return &mpb
->disk
[index
];
706 /* retrieve the disk description based on a index of the disk
709 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
713 for (d
= super
->disks
; d
; d
= d
->next
)
714 if (d
->index
== index
)
719 /* retrieve a disk from the parsed metadata */
720 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
724 dl
= get_imsm_dl_disk(super
, index
);
731 /* generate a checksum directly from the anchor when the anchor is known to be
732 * up-to-date, currently only at load or write_super after coalescing
734 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
736 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
737 __u32
*p
= (__u32
*) mpb
;
741 sum
+= __le32_to_cpu(*p
);
745 return sum
- __le32_to_cpu(mpb
->check_sum
);
748 static size_t sizeof_imsm_map(struct imsm_map
*map
)
750 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
753 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
755 /* A device can have 2 maps if it is in the middle of a migration.
757 * MAP_0 - we return the first map
758 * MAP_1 - we return the second map if it exists, else NULL
759 * MAP_X - we return the second map if it exists, else the first
761 struct imsm_map
*map
= &dev
->vol
.map
[0];
762 struct imsm_map
*map2
= NULL
;
764 if (dev
->vol
.migr_state
)
765 map2
= (void *)map
+ sizeof_imsm_map(map
);
767 switch (second_map
) {
784 /* return the size of the device.
785 * migr_state increases the returned size if map[0] were to be duplicated
787 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
789 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
790 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
792 /* migrating means an additional map */
793 if (dev
->vol
.migr_state
)
794 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
796 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
801 /* retrieve disk serial number list from a metadata update */
802 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
805 struct disk_info
*inf
;
807 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
808 sizeof_imsm_dev(&update
->dev
, 0);
813 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
819 if (index
>= mpb
->num_raid_devs
)
822 /* devices start after all disks */
823 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
825 for (i
= 0; i
<= index
; i
++)
827 return _mpb
+ offset
;
829 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
834 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
836 struct intel_dev
*dv
;
838 if (index
>= super
->anchor
->num_raid_devs
)
840 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
841 if (dv
->index
== index
)
846 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
849 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
850 __le16_to_cpu(addr
->w1
));
853 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
855 struct bbm_log_block_addr addr
;
857 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
858 addr
.dw1
= __cpu_to_le32((__u32
)(sec
>> 16) & 0xffffffff);
862 /* get size of the bbm log */
863 static __u32
get_imsm_bbm_log_size(struct bbm_log
*log
)
865 if (!log
|| log
->entry_count
== 0)
868 return sizeof(log
->signature
) +
869 sizeof(log
->entry_count
) +
870 log
->entry_count
* sizeof(struct bbm_log_entry
);
873 /* check if bad block is not partially stored in bbm log */
874 static int is_stored_in_bbm(struct bbm_log
*log
, const __u8 idx
, const unsigned
875 long long sector
, const int length
, __u32
*pos
)
879 for (i
= *pos
; i
< log
->entry_count
; i
++) {
880 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
881 unsigned long long bb_start
;
882 unsigned long long bb_end
;
884 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
885 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
887 if ((entry
->disk_ordinal
== idx
) && (bb_start
>= sector
) &&
888 (bb_end
<= sector
+ length
)) {
896 /* record new bad block in bbm log */
897 static int record_new_badblock(struct bbm_log
*log
, const __u8 idx
, unsigned
898 long long sector
, int length
)
902 struct bbm_log_entry
*entry
= NULL
;
904 while (is_stored_in_bbm(log
, idx
, sector
, length
, &pos
)) {
905 struct bbm_log_entry
*e
= &log
->marked_block_entries
[pos
];
907 if ((e
->marked_count
+ 1 == BBM_LOG_MAX_LBA_ENTRY_VAL
) &&
908 (__le48_to_cpu(&e
->defective_block_start
) == sector
)) {
909 sector
+= BBM_LOG_MAX_LBA_ENTRY_VAL
;
910 length
-= BBM_LOG_MAX_LBA_ENTRY_VAL
;
919 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
920 BBM_LOG_MAX_LBA_ENTRY_VAL
;
921 entry
->defective_block_start
= __cpu_to_le48(sector
);
922 entry
->marked_count
= cnt
- 1;
929 new_bb
= ROUND_UP(length
, BBM_LOG_MAX_LBA_ENTRY_VAL
) /
930 BBM_LOG_MAX_LBA_ENTRY_VAL
;
931 if (log
->entry_count
+ new_bb
> BBM_LOG_MAX_ENTRIES
)
935 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
936 BBM_LOG_MAX_LBA_ENTRY_VAL
;
937 struct bbm_log_entry
*entry
=
938 &log
->marked_block_entries
[log
->entry_count
];
940 entry
->defective_block_start
= __cpu_to_le48(sector
);
941 entry
->marked_count
= cnt
- 1;
942 entry
->disk_ordinal
= idx
;
953 /* clear all bad blocks for given disk */
954 static void clear_disk_badblocks(struct bbm_log
*log
, const __u8 idx
)
958 while (i
< log
->entry_count
) {
959 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
961 if (entries
[i
].disk_ordinal
== idx
) {
962 if (i
< log
->entry_count
- 1)
963 entries
[i
] = entries
[log
->entry_count
- 1];
971 /* clear given bad block */
972 static int clear_badblock(struct bbm_log
*log
, const __u8 idx
, const unsigned
973 long long sector
, const int length
) {
976 while (i
< log
->entry_count
) {
977 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
979 if ((entries
[i
].disk_ordinal
== idx
) &&
980 (__le48_to_cpu(&entries
[i
].defective_block_start
) ==
981 sector
) && (entries
[i
].marked_count
+ 1 == length
)) {
982 if (i
< log
->entry_count
- 1)
983 entries
[i
] = entries
[log
->entry_count
- 1];
993 /* allocate and load BBM log from metadata */
994 static int load_bbm_log(struct intel_super
*super
)
996 struct imsm_super
*mpb
= super
->anchor
;
997 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
999 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
1000 if (!super
->bbm_log
)
1004 struct bbm_log
*log
= (void *)mpb
+
1005 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1009 if (bbm_log_size
< sizeof(log
->signature
) +
1010 sizeof(log
->entry_count
))
1013 entry_count
= __le32_to_cpu(log
->entry_count
);
1014 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
1015 (entry_count
> BBM_LOG_MAX_ENTRIES
))
1019 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
1020 entry_count
* sizeof(struct bbm_log_entry
))
1023 memcpy(super
->bbm_log
, log
, bbm_log_size
);
1025 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
1026 super
->bbm_log
->entry_count
= 0;
1032 /* checks if bad block is within volume boundaries */
1033 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
1034 const unsigned long long start_sector
,
1035 const unsigned long long size
)
1037 unsigned long long bb_start
;
1038 unsigned long long bb_end
;
1040 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1041 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1043 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1044 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1050 /* get list of bad blocks on a drive for a volume */
1051 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1052 const unsigned long long start_sector
,
1053 const unsigned long long size
,
1059 for (i
= 0; i
< log
->entry_count
; i
++) {
1060 const struct bbm_log_entry
*ent
=
1061 &log
->marked_block_entries
[i
];
1062 struct md_bb_entry
*bb
;
1064 if ((ent
->disk_ordinal
== idx
) &&
1065 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1067 if (!bbs
->entries
) {
1068 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1074 bb
= &bbs
->entries
[count
++];
1075 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1076 bb
->length
= ent
->marked_count
+ 1;
1084 * == MAP_0 get first map
1085 * == MAP_1 get second map
1086 * == MAP_X than get map according to the current migr_state
1088 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1092 struct imsm_map
*map
;
1094 map
= get_imsm_map(dev
, second_map
);
1096 /* top byte identifies disk under rebuild */
1097 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1100 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1101 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1103 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1105 return ord_to_idx(ord
);
1108 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1110 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1113 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
1118 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1119 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1120 if (ord_to_idx(ord
) == idx
)
1127 static int get_imsm_raid_level(struct imsm_map
*map
)
1129 if (map
->raid_level
== 1) {
1130 if (map
->num_members
== 2)
1136 return map
->raid_level
;
1139 static int cmp_extent(const void *av
, const void *bv
)
1141 const struct extent
*a
= av
;
1142 const struct extent
*b
= bv
;
1143 if (a
->start
< b
->start
)
1145 if (a
->start
> b
->start
)
1150 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1152 int memberships
= 0;
1155 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1156 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1157 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1159 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
1166 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1168 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
1170 if (lo
== 0 || hi
== 0)
1172 *lo
= __le32_to_cpu((unsigned)n
);
1173 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
1177 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1179 return (unsigned long long)__le32_to_cpu(lo
) |
1180 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1183 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1187 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1190 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1194 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1197 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1201 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1204 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1208 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1211 static unsigned long long imsm_dev_size(struct imsm_dev
*dev
)
1215 return join_u32(dev
->size_low
, dev
->size_high
);
1218 static unsigned long long migr_chkp_area_pba(struct migr_record
*migr_rec
)
1220 if (migr_rec
== NULL
)
1222 return join_u32(migr_rec
->ckpt_area_pba_lo
,
1223 migr_rec
->ckpt_area_pba_hi
);
1226 static unsigned long long current_migr_unit(struct migr_record
*migr_rec
)
1228 if (migr_rec
== NULL
)
1230 return join_u32(migr_rec
->curr_migr_unit_lo
,
1231 migr_rec
->curr_migr_unit_hi
);
1234 static unsigned long long migr_dest_1st_member_lba(struct migr_record
*migr_rec
)
1236 if (migr_rec
== NULL
)
1238 return join_u32(migr_rec
->dest_1st_member_lba_lo
,
1239 migr_rec
->dest_1st_member_lba_hi
);
1242 static unsigned long long get_num_migr_units(struct migr_record
*migr_rec
)
1244 if (migr_rec
== NULL
)
1246 return join_u32(migr_rec
->num_migr_units_lo
,
1247 migr_rec
->num_migr_units_hi
);
1250 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1252 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1255 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1257 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1260 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1262 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1265 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1267 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1270 static void set_imsm_dev_size(struct imsm_dev
*dev
, unsigned long long n
)
1272 split_ull(n
, &dev
->size_low
, &dev
->size_high
);
1275 static void set_migr_chkp_area_pba(struct migr_record
*migr_rec
,
1276 unsigned long long n
)
1278 split_ull(n
, &migr_rec
->ckpt_area_pba_lo
, &migr_rec
->ckpt_area_pba_hi
);
1281 static void set_current_migr_unit(struct migr_record
*migr_rec
,
1282 unsigned long long n
)
1284 split_ull(n
, &migr_rec
->curr_migr_unit_lo
,
1285 &migr_rec
->curr_migr_unit_hi
);
1288 static void set_migr_dest_1st_member_lba(struct migr_record
*migr_rec
,
1289 unsigned long long n
)
1291 split_ull(n
, &migr_rec
->dest_1st_member_lba_lo
,
1292 &migr_rec
->dest_1st_member_lba_hi
);
1295 static void set_num_migr_units(struct migr_record
*migr_rec
,
1296 unsigned long long n
)
1298 split_ull(n
, &migr_rec
->num_migr_units_lo
,
1299 &migr_rec
->num_migr_units_hi
);
1302 static unsigned long long per_dev_array_size(struct imsm_map
*map
)
1304 unsigned long long array_size
= 0;
1309 array_size
= num_data_stripes(map
) * map
->blocks_per_strip
;
1310 if (get_imsm_raid_level(map
) == 1 || get_imsm_raid_level(map
) == 10)
1316 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
1318 /* find a list of used extents on the given physical device */
1319 struct extent
*rv
, *e
;
1321 int memberships
= count_memberships(dl
, super
);
1324 /* trim the reserved area for spares, so they can join any array
1325 * regardless of whether the OROM has assigned sectors from the
1326 * IMSM_RESERVED_SECTORS region
1328 if (dl
->index
== -1)
1329 reservation
= imsm_min_reserved_sectors(super
);
1331 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1333 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
1336 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1337 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1338 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1340 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1341 e
->start
= pba_of_lba0(map
);
1342 e
->size
= per_dev_array_size(map
);
1346 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1348 /* determine the start of the metadata
1349 * when no raid devices are defined use the default
1350 * ...otherwise allow the metadata to truncate the value
1351 * as is the case with older versions of imsm
1354 struct extent
*last
= &rv
[memberships
- 1];
1355 unsigned long long remainder
;
1357 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1358 /* round down to 1k block to satisfy precision of the kernel
1362 /* make sure remainder is still sane */
1363 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1364 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1365 if (reservation
> remainder
)
1366 reservation
= remainder
;
1368 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1373 /* try to determine how much space is reserved for metadata from
1374 * the last get_extents() entry, otherwise fallback to the
1377 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1383 /* for spares just return a minimal reservation which will grow
1384 * once the spare is picked up by an array
1386 if (dl
->index
== -1)
1387 return MPB_SECTOR_CNT
;
1389 e
= get_extents(super
, dl
);
1391 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1393 /* scroll to last entry */
1394 for (i
= 0; e
[i
].size
; i
++)
1397 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1404 static int is_spare(struct imsm_disk
*disk
)
1406 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1409 static int is_configured(struct imsm_disk
*disk
)
1411 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1414 static int is_failed(struct imsm_disk
*disk
)
1416 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1419 static int is_journal(struct imsm_disk
*disk
)
1421 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1424 /* round array size down to closest MB and ensure it splits evenly
1427 static unsigned long long round_size_to_mb(unsigned long long size
, unsigned int
1431 size
= (size
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
1437 static int able_to_resync(int raid_level
, int missing_disks
)
1439 int max_missing_disks
= 0;
1441 switch (raid_level
) {
1443 max_missing_disks
= 1;
1446 max_missing_disks
= 0;
1448 return missing_disks
<= max_missing_disks
;
1451 /* try to determine how much space is reserved for metadata from
1452 * the last get_extents() entry on the smallest active disk,
1453 * otherwise fallback to the default
1455 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1459 unsigned long long min_active
;
1461 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1462 struct dl
*dl
, *dl_min
= NULL
;
1468 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1471 unsigned long long blocks
= total_blocks(&dl
->disk
);
1472 if (blocks
< min_active
|| min_active
== 0) {
1474 min_active
= blocks
;
1480 /* find last lba used by subarrays on the smallest active disk */
1481 e
= get_extents(super
, dl_min
);
1484 for (i
= 0; e
[i
].size
; i
++)
1487 remainder
= min_active
- e
[i
].start
;
1490 /* to give priority to recovery we should not require full
1491 IMSM_RESERVED_SECTORS from the spare */
1492 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1494 /* if real reservation is smaller use that value */
1495 return (remainder
< rv
) ? remainder
: rv
;
1499 * Return minimum size of a spare and sector size
1500 * that can be used in this array
1502 int get_spare_criteria_imsm(struct supertype
*st
, struct spare_criteria
*c
)
1504 struct intel_super
*super
= st
->sb
;
1508 unsigned long long size
= 0;
1515 /* find first active disk in array */
1517 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1521 /* find last lba used by subarrays */
1522 e
= get_extents(super
, dl
);
1525 for (i
= 0; e
[i
].size
; i
++)
1528 size
= e
[i
-1].start
+ e
[i
-1].size
;
1531 /* add the amount of space needed for metadata */
1532 size
+= imsm_min_reserved_sectors(super
);
1534 c
->min_size
= size
* 512;
1535 c
->sector_size
= super
->sector_size
;
1540 static int is_gen_migration(struct imsm_dev
*dev
);
1542 #define IMSM_4K_DIV 8
1544 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1545 struct imsm_dev
*dev
);
1547 static void print_imsm_dev(struct intel_super
*super
,
1548 struct imsm_dev
*dev
,
1554 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1555 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1559 printf("[%.16s]:\n", dev
->volume
);
1560 printf(" UUID : %s\n", uuid
);
1561 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1563 printf(" <-- %d", get_imsm_raid_level(map2
));
1565 printf(" Members : %d", map
->num_members
);
1567 printf(" <-- %d", map2
->num_members
);
1569 printf(" Slots : [");
1570 for (i
= 0; i
< map
->num_members
; i
++) {
1571 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1572 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1577 for (i
= 0; i
< map2
->num_members
; i
++) {
1578 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1579 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1584 printf(" Failed disk : ");
1585 if (map
->failed_disk_num
== 0xff)
1588 printf("%i", map
->failed_disk_num
);
1590 slot
= get_imsm_disk_slot(map
, disk_idx
);
1592 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1593 printf(" This Slot : %d%s\n", slot
,
1594 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1596 printf(" This Slot : ?\n");
1597 printf(" Sector Size : %u\n", super
->sector_size
);
1598 sz
= imsm_dev_size(dev
);
1599 printf(" Array Size : %llu%s\n",
1600 (unsigned long long)sz
* 512 / super
->sector_size
,
1601 human_size(sz
* 512));
1602 sz
= blocks_per_member(map
);
1603 printf(" Per Dev Size : %llu%s\n",
1604 (unsigned long long)sz
* 512 / super
->sector_size
,
1605 human_size(sz
* 512));
1606 printf(" Sector Offset : %llu\n",
1608 printf(" Num Stripes : %llu\n",
1609 num_data_stripes(map
));
1610 printf(" Chunk Size : %u KiB",
1611 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1613 printf(" <-- %u KiB",
1614 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1616 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1617 printf(" Migrate State : ");
1618 if (dev
->vol
.migr_state
) {
1619 if (migr_type(dev
) == MIGR_INIT
)
1620 printf("initialize\n");
1621 else if (migr_type(dev
) == MIGR_REBUILD
)
1622 printf("rebuild\n");
1623 else if (migr_type(dev
) == MIGR_VERIFY
)
1625 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1626 printf("general migration\n");
1627 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1628 printf("state change\n");
1629 else if (migr_type(dev
) == MIGR_REPAIR
)
1632 printf("<unknown:%d>\n", migr_type(dev
));
1635 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1636 if (dev
->vol
.migr_state
) {
1637 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1639 printf(" <-- %s", map_state_str
[map
->map_state
]);
1640 printf("\n Checkpoint : %u ",
1641 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1642 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1645 printf("(%llu)", (unsigned long long)
1646 blocks_per_migr_unit(super
, dev
));
1649 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1651 printf(" RWH Policy : ");
1652 if (dev
->rwh_policy
== RWH_OFF
|| dev
->rwh_policy
== RWH_MULTIPLE_OFF
)
1654 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1655 printf("PPL distributed\n");
1656 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1657 printf("PPL journaling drive\n");
1658 else if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
1659 printf("Multiple distributed PPLs\n");
1660 else if (dev
->rwh_policy
== RWH_MULTIPLE_PPLS_JOURNALING_DRIVE
)
1661 printf("Multiple PPLs on journaling drive\n");
1663 printf("<unknown:%d>\n", dev
->rwh_policy
);
1666 static void print_imsm_disk(struct imsm_disk
*disk
,
1669 unsigned int sector_size
) {
1670 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1673 if (index
< -1 || !disk
)
1677 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1679 printf(" Disk%02d Serial : %s\n", index
, str
);
1681 printf(" Disk Serial : %s\n", str
);
1682 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1683 is_configured(disk
) ? " active" : "",
1684 is_failed(disk
) ? " failed" : "",
1685 is_journal(disk
) ? " journal" : "");
1686 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1687 sz
= total_blocks(disk
) - reserved
;
1688 printf(" Usable Size : %llu%s\n",
1689 (unsigned long long)sz
* 512 / sector_size
,
1690 human_size(sz
* 512));
1693 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1695 struct migr_record
*migr_rec
= super
->migr_rec
;
1697 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1698 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1699 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1700 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1701 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1702 set_migr_chkp_area_pba(migr_rec
,
1703 migr_chkp_area_pba(migr_rec
) / IMSM_4K_DIV
);
1704 set_migr_dest_1st_member_lba(migr_rec
,
1705 migr_dest_1st_member_lba(migr_rec
) / IMSM_4K_DIV
);
1708 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1710 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1713 void convert_to_4k(struct intel_super
*super
)
1715 struct imsm_super
*mpb
= super
->anchor
;
1716 struct imsm_disk
*disk
;
1718 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1720 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1721 disk
= __get_imsm_disk(mpb
, i
);
1723 convert_to_4k_imsm_disk(disk
);
1725 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1726 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1727 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1729 set_imsm_dev_size(dev
, imsm_dev_size(dev
)/IMSM_4K_DIV
);
1730 dev
->vol
.curr_migr_unit
/= IMSM_4K_DIV
;
1733 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1734 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1735 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1737 if (dev
->vol
.migr_state
) {
1739 map
= get_imsm_map(dev
, MAP_1
);
1740 set_blocks_per_member(map
,
1741 blocks_per_member(map
)/IMSM_4K_DIV
);
1742 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1743 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1747 struct bbm_log
*log
= (void *)mpb
+
1748 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1751 for (i
= 0; i
< log
->entry_count
; i
++) {
1752 struct bbm_log_entry
*entry
=
1753 &log
->marked_block_entries
[i
];
1755 __u8 count
= entry
->marked_count
+ 1;
1756 unsigned long long sector
=
1757 __le48_to_cpu(&entry
->defective_block_start
);
1759 entry
->defective_block_start
=
1760 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1761 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1765 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1768 void examine_migr_rec_imsm(struct intel_super
*super
)
1770 struct migr_record
*migr_rec
= super
->migr_rec
;
1771 struct imsm_super
*mpb
= super
->anchor
;
1774 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1775 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1776 struct imsm_map
*map
;
1779 if (is_gen_migration(dev
) == 0)
1782 printf("\nMigration Record Information:");
1784 /* first map under migration */
1785 map
= get_imsm_map(dev
, MAP_0
);
1787 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1788 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1789 printf(" Empty\n ");
1790 printf("Examine one of first two disks in array\n");
1793 printf("\n Status : ");
1794 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1797 printf("Contains Data\n");
1798 printf(" Current Unit : %llu\n",
1799 current_migr_unit(migr_rec
));
1800 printf(" Family : %u\n",
1801 __le32_to_cpu(migr_rec
->family_num
));
1802 printf(" Ascending : %u\n",
1803 __le32_to_cpu(migr_rec
->ascending_migr
));
1804 printf(" Blocks Per Unit : %u\n",
1805 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1806 printf(" Dest. Depth Per Unit : %u\n",
1807 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1808 printf(" Checkpoint Area pba : %llu\n",
1809 migr_chkp_area_pba(migr_rec
));
1810 printf(" First member lba : %llu\n",
1811 migr_dest_1st_member_lba(migr_rec
));
1812 printf(" Total Number of Units : %llu\n",
1813 get_num_migr_units(migr_rec
));
1814 printf(" Size of volume : %llu\n",
1815 join_u32(migr_rec
->post_migr_vol_cap
,
1816 migr_rec
->post_migr_vol_cap_hi
));
1817 printf(" Record was read from : %u\n",
1818 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1824 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1826 struct migr_record
*migr_rec
= super
->migr_rec
;
1828 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1829 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1830 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1831 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1832 &migr_rec
->post_migr_vol_cap
,
1833 &migr_rec
->post_migr_vol_cap_hi
);
1834 set_migr_chkp_area_pba(migr_rec
,
1835 migr_chkp_area_pba(migr_rec
) * IMSM_4K_DIV
);
1836 set_migr_dest_1st_member_lba(migr_rec
,
1837 migr_dest_1st_member_lba(migr_rec
) * IMSM_4K_DIV
);
1840 void convert_from_4k(struct intel_super
*super
)
1842 struct imsm_super
*mpb
= super
->anchor
;
1843 struct imsm_disk
*disk
;
1845 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1847 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1848 disk
= __get_imsm_disk(mpb
, i
);
1850 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1853 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1854 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1855 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1857 set_imsm_dev_size(dev
, imsm_dev_size(dev
)*IMSM_4K_DIV
);
1858 dev
->vol
.curr_migr_unit
*= IMSM_4K_DIV
;
1861 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1862 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1863 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1865 if (dev
->vol
.migr_state
) {
1867 map
= get_imsm_map(dev
, MAP_1
);
1868 set_blocks_per_member(map
,
1869 blocks_per_member(map
)*IMSM_4K_DIV
);
1870 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1871 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1875 struct bbm_log
*log
= (void *)mpb
+
1876 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1879 for (i
= 0; i
< log
->entry_count
; i
++) {
1880 struct bbm_log_entry
*entry
=
1881 &log
->marked_block_entries
[i
];
1883 __u8 count
= entry
->marked_count
+ 1;
1884 unsigned long long sector
=
1885 __le48_to_cpu(&entry
->defective_block_start
);
1887 entry
->defective_block_start
=
1888 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1889 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1893 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1896 /*******************************************************************************
1897 * function: imsm_check_attributes
1898 * Description: Function checks if features represented by attributes flags
1899 * are supported by mdadm.
1901 * attributes - Attributes read from metadata
1903 * 0 - passed attributes contains unsupported features flags
1904 * 1 - all features are supported
1905 ******************************************************************************/
1906 static int imsm_check_attributes(__u32 attributes
)
1909 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1911 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1913 not_supported
&= attributes
;
1914 if (not_supported
) {
1915 pr_err("(IMSM): Unsupported attributes : %x\n",
1916 (unsigned)__le32_to_cpu(not_supported
));
1917 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1918 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1919 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1921 if (not_supported
& MPB_ATTRIB_2TB
) {
1922 dprintf("\t\tMPB_ATTRIB_2TB\n");
1923 not_supported
^= MPB_ATTRIB_2TB
;
1925 if (not_supported
& MPB_ATTRIB_RAID0
) {
1926 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1927 not_supported
^= MPB_ATTRIB_RAID0
;
1929 if (not_supported
& MPB_ATTRIB_RAID1
) {
1930 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1931 not_supported
^= MPB_ATTRIB_RAID1
;
1933 if (not_supported
& MPB_ATTRIB_RAID10
) {
1934 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1935 not_supported
^= MPB_ATTRIB_RAID10
;
1937 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1938 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1939 not_supported
^= MPB_ATTRIB_RAID1E
;
1941 if (not_supported
& MPB_ATTRIB_RAID5
) {
1942 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1943 not_supported
^= MPB_ATTRIB_RAID5
;
1945 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1946 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1947 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1949 if (not_supported
& MPB_ATTRIB_BBM
) {
1950 dprintf("\t\tMPB_ATTRIB_BBM\n");
1951 not_supported
^= MPB_ATTRIB_BBM
;
1953 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1954 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1955 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1957 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1958 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1959 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1961 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1962 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1963 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1965 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1966 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1967 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1969 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1970 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1971 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1975 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1983 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1985 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1987 struct intel_super
*super
= st
->sb
;
1988 struct imsm_super
*mpb
= super
->anchor
;
1989 char str
[MAX_SIGNATURE_LENGTH
];
1994 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1997 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
1998 str
[MPB_SIG_LEN
-1] = '\0';
1999 printf(" Magic : %s\n", str
);
2000 printf(" Version : %s\n", get_imsm_version(mpb
));
2001 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
2002 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
2003 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
2004 printf(" Attributes : ");
2005 if (imsm_check_attributes(mpb
->attributes
))
2006 printf("All supported\n");
2008 printf("not supported\n");
2009 getinfo_super_imsm(st
, &info
, NULL
);
2010 fname_from_uuid(st
, &info
, nbuf
, ':');
2011 printf(" UUID : %s\n", nbuf
+ 5);
2012 sum
= __le32_to_cpu(mpb
->check_sum
);
2013 printf(" Checksum : %08x %s\n", sum
,
2014 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
2015 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
2016 printf(" Disks : %d\n", mpb
->num_disks
);
2017 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
2018 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
2019 super
->disks
->index
, reserved
, super
->sector_size
);
2020 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
2021 struct bbm_log
*log
= super
->bbm_log
;
2024 printf("Bad Block Management Log:\n");
2025 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
2026 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
2027 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
2029 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2031 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2033 super
->current_vol
= i
;
2034 getinfo_super_imsm(st
, &info
, NULL
);
2035 fname_from_uuid(st
, &info
, nbuf
, ':');
2036 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
2038 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2039 if (i
== super
->disks
->index
)
2041 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
2042 super
->sector_size
);
2045 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2046 if (dl
->index
== -1)
2047 print_imsm_disk(&dl
->disk
, -1, reserved
,
2048 super
->sector_size
);
2050 examine_migr_rec_imsm(super
);
2053 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
2055 /* We just write a generic IMSM ARRAY entry */
2058 struct intel_super
*super
= st
->sb
;
2060 if (!super
->anchor
->num_raid_devs
) {
2061 printf("ARRAY metadata=imsm\n");
2065 getinfo_super_imsm(st
, &info
, NULL
);
2066 fname_from_uuid(st
, &info
, nbuf
, ':');
2067 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
2070 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
2072 /* We just write a generic IMSM ARRAY entry */
2076 struct intel_super
*super
= st
->sb
;
2079 if (!super
->anchor
->num_raid_devs
)
2082 getinfo_super_imsm(st
, &info
, NULL
);
2083 fname_from_uuid(st
, &info
, nbuf
, ':');
2084 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2085 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2087 super
->current_vol
= i
;
2088 getinfo_super_imsm(st
, &info
, NULL
);
2089 fname_from_uuid(st
, &info
, nbuf1
, ':');
2090 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
2091 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
2095 static void export_examine_super_imsm(struct supertype
*st
)
2097 struct intel_super
*super
= st
->sb
;
2098 struct imsm_super
*mpb
= super
->anchor
;
2102 getinfo_super_imsm(st
, &info
, NULL
);
2103 fname_from_uuid(st
, &info
, nbuf
, ':');
2104 printf("MD_METADATA=imsm\n");
2105 printf("MD_LEVEL=container\n");
2106 printf("MD_UUID=%s\n", nbuf
+5);
2107 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
2110 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
2112 /* The second last sector of the device contains
2113 * the "struct imsm_super" metadata.
2114 * This contains mpb_size which is the size in bytes of the
2115 * extended metadata. This is located immediately before
2117 * We want to read all that, plus the last sector which
2118 * may contain a migration record, and write it all
2122 unsigned long long dsize
, offset
;
2124 struct imsm_super
*sb
;
2125 struct intel_super
*super
= st
->sb
;
2126 unsigned int sector_size
= super
->sector_size
;
2127 unsigned int written
= 0;
2129 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
) != 0)
2132 if (!get_dev_size(from
, NULL
, &dsize
))
2135 if (lseek64(from
, dsize
-(2*sector_size
), 0) < 0)
2137 if ((unsigned int)read(from
, buf
, sector_size
) != sector_size
)
2140 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
2143 sectors
= mpb_sectors(sb
, sector_size
) + 2;
2144 offset
= dsize
- sectors
* sector_size
;
2145 if (lseek64(from
, offset
, 0) < 0 ||
2146 lseek64(to
, offset
, 0) < 0)
2148 while (written
< sectors
* sector_size
) {
2149 int n
= sectors
*sector_size
- written
;
2152 if (read(from
, buf
, n
) != n
)
2154 if (write(to
, buf
, n
) != n
)
2165 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
2170 getinfo_super_imsm(st
, &info
, NULL
);
2171 fname_from_uuid(st
, &info
, nbuf
, ':');
2172 printf("\n UUID : %s\n", nbuf
+ 5);
2175 static void brief_detail_super_imsm(struct supertype
*st
)
2179 getinfo_super_imsm(st
, &info
, NULL
);
2180 fname_from_uuid(st
, &info
, nbuf
, ':');
2181 printf(" UUID=%s", nbuf
+ 5);
2184 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
2185 static void fd2devname(int fd
, char *name
);
2187 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2189 /* dump an unsorted list of devices attached to AHCI Intel storage
2190 * controller, as well as non-connected ports
2192 int hba_len
= strlen(hba_path
) + 1;
2197 unsigned long port_mask
= (1 << port_count
) - 1;
2199 if (port_count
> (int)sizeof(port_mask
) * 8) {
2201 pr_err("port_count %d out of range\n", port_count
);
2205 /* scroll through /sys/dev/block looking for devices attached to
2208 dir
= opendir("/sys/dev/block");
2212 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2223 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2225 path
= devt_to_devpath(makedev(major
, minor
));
2228 if (!path_attached_to_hba(path
, hba_path
)) {
2234 /* retrieve the scsi device type */
2235 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2237 pr_err("failed to allocate 'device'\n");
2241 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2242 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2244 pr_err("failed to read device type for %s\n",
2250 type
= strtoul(buf
, NULL
, 10);
2252 /* if it's not a disk print the vendor and model */
2253 if (!(type
== 0 || type
== 7 || type
== 14)) {
2256 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2257 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2258 strncpy(vendor
, buf
, sizeof(vendor
));
2259 vendor
[sizeof(vendor
) - 1] = '\0';
2260 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2261 while (isspace(*c
) || *c
== '\0')
2265 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2266 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2267 strncpy(model
, buf
, sizeof(model
));
2268 model
[sizeof(model
) - 1] = '\0';
2269 c
= (char *) &model
[sizeof(model
) - 1];
2270 while (isspace(*c
) || *c
== '\0')
2274 if (vendor
[0] && model
[0])
2275 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2277 switch (type
) { /* numbers from hald/linux/device.c */
2278 case 1: sprintf(buf
, "tape"); break;
2279 case 2: sprintf(buf
, "printer"); break;
2280 case 3: sprintf(buf
, "processor"); break;
2282 case 5: sprintf(buf
, "cdrom"); break;
2283 case 6: sprintf(buf
, "scanner"); break;
2284 case 8: sprintf(buf
, "media_changer"); break;
2285 case 9: sprintf(buf
, "comm"); break;
2286 case 12: sprintf(buf
, "raid"); break;
2287 default: sprintf(buf
, "unknown");
2293 /* chop device path to 'host%d' and calculate the port number */
2294 c
= strchr(&path
[hba_len
], '/');
2297 pr_err("%s - invalid path name\n", path
+ hba_len
);
2302 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2303 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2307 *c
= '/'; /* repair the full string */
2308 pr_err("failed to determine port number for %s\n",
2315 /* mark this port as used */
2316 port_mask
&= ~(1 << port
);
2318 /* print out the device information */
2320 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2324 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2326 printf(" Port%d : - disk info unavailable -\n", port
);
2328 fd2devname(fd
, buf
);
2329 printf(" Port%d : %s", port
, buf
);
2330 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
2331 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
2346 for (i
= 0; i
< port_count
; i
++)
2347 if (port_mask
& (1 << i
))
2348 printf(" Port%d : - no device attached -\n", i
);
2354 static int print_vmd_attached_devs(struct sys_dev
*hba
)
2362 if (hba
->type
!= SYS_DEV_VMD
)
2365 /* scroll through /sys/dev/block looking for devices attached to
2368 dir
= opendir("/sys/bus/pci/drivers/nvme");
2372 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2375 /* is 'ent' a device? check that the 'subsystem' link exists and
2376 * that its target matches 'bus'
2378 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
2380 n
= readlink(path
, link
, sizeof(link
));
2381 if (n
< 0 || n
>= (int)sizeof(link
))
2384 c
= strrchr(link
, '/');
2387 if (strncmp("pci", c
+1, strlen("pci")) != 0)
2390 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
2392 rp
= realpath(path
, NULL
);
2396 if (path_attached_to_hba(rp
, hba
->path
)) {
2397 printf(" NVMe under VMD : %s\n", rp
);
2406 static void print_found_intel_controllers(struct sys_dev
*elem
)
2408 for (; elem
; elem
= elem
->next
) {
2409 pr_err("found Intel(R) ");
2410 if (elem
->type
== SYS_DEV_SATA
)
2411 fprintf(stderr
, "SATA ");
2412 else if (elem
->type
== SYS_DEV_SAS
)
2413 fprintf(stderr
, "SAS ");
2414 else if (elem
->type
== SYS_DEV_NVME
)
2415 fprintf(stderr
, "NVMe ");
2417 if (elem
->type
== SYS_DEV_VMD
)
2418 fprintf(stderr
, "VMD domain");
2420 fprintf(stderr
, "RAID controller");
2423 fprintf(stderr
, " at %s", elem
->pci_id
);
2424 fprintf(stderr
, ".\n");
2429 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2436 if ((dir
= opendir(hba_path
)) == NULL
)
2439 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2442 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2443 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2445 if (*port_count
== 0)
2447 else if (host
< host_base
)
2450 if (host
+ 1 > *port_count
+ host_base
)
2451 *port_count
= host
+ 1 - host_base
;
2457 static void print_imsm_capability(const struct imsm_orom
*orom
)
2459 printf(" Platform : Intel(R) ");
2460 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2461 printf("Matrix Storage Manager\n");
2462 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2463 printf("Virtual RAID on CPU\n");
2465 printf("Rapid Storage Technology%s\n",
2466 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2467 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2468 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2469 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2470 printf(" RAID Levels :%s%s%s%s%s\n",
2471 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2472 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2473 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2474 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2475 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2476 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2477 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2478 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2479 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2480 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2481 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2482 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2483 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2484 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2485 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2486 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2487 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2488 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2489 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2490 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2491 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2492 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2493 printf(" 2TB volumes :%s supported\n",
2494 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2495 printf(" 2TB disks :%s supported\n",
2496 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2497 printf(" Max Disks : %d\n", orom
->tds
);
2498 printf(" Max Volumes : %d per array, %d per %s\n",
2499 orom
->vpa
, orom
->vphba
,
2500 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2504 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2506 printf("MD_FIRMWARE_TYPE=imsm\n");
2507 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2508 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2509 orom
->hotfix_ver
, orom
->build
);
2510 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2511 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2512 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2513 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2514 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2515 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2516 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2517 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2518 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2519 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2520 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2521 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2522 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2523 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2524 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2525 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2526 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2527 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2528 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2529 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2530 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2531 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2532 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2533 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2534 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2535 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2536 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2537 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2540 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2542 /* There are two components to imsm platform support, the ahci SATA
2543 * controller and the option-rom. To find the SATA controller we
2544 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2545 * controller with the Intel vendor id is present. This approach
2546 * allows mdadm to leverage the kernel's ahci detection logic, with the
2547 * caveat that if ahci.ko is not loaded mdadm will not be able to
2548 * detect platform raid capabilities. The option-rom resides in a
2549 * platform "Adapter ROM". We scan for its signature to retrieve the
2550 * platform capabilities. If raid support is disabled in the BIOS the
2551 * option-rom capability structure will not be available.
2553 struct sys_dev
*list
, *hba
;
2558 if (enumerate_only
) {
2559 if (check_env("IMSM_NO_PLATFORM"))
2561 list
= find_intel_devices();
2564 for (hba
= list
; hba
; hba
= hba
->next
) {
2565 if (find_imsm_capability(hba
)) {
2575 list
= find_intel_devices();
2578 pr_err("no active Intel(R) RAID controller found.\n");
2580 } else if (verbose
> 0)
2581 print_found_intel_controllers(list
);
2583 for (hba
= list
; hba
; hba
= hba
->next
) {
2584 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2586 if (!find_imsm_capability(hba
)) {
2588 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2589 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2590 get_sys_dev_type(hba
->type
));
2596 if (controller_path
&& result
== 1) {
2597 pr_err("no active Intel(R) RAID controller found under %s\n",
2602 const struct orom_entry
*entry
;
2604 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2605 if (entry
->type
== SYS_DEV_VMD
) {
2606 print_imsm_capability(&entry
->orom
);
2607 printf(" 3rd party NVMe :%s supported\n",
2608 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2609 for (hba
= list
; hba
; hba
= hba
->next
) {
2610 if (hba
->type
== SYS_DEV_VMD
) {
2612 printf(" I/O Controller : %s (%s)\n",
2613 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2614 if (print_vmd_attached_devs(hba
)) {
2616 pr_err("failed to get devices attached to VMD domain.\n");
2625 print_imsm_capability(&entry
->orom
);
2626 if (entry
->type
== SYS_DEV_NVME
) {
2627 for (hba
= list
; hba
; hba
= hba
->next
) {
2628 if (hba
->type
== SYS_DEV_NVME
)
2629 printf(" NVMe Device : %s\n", hba
->path
);
2635 struct devid_list
*devid
;
2636 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2637 hba
= device_by_id(devid
->devid
);
2641 printf(" I/O Controller : %s (%s)\n",
2642 hba
->path
, get_sys_dev_type(hba
->type
));
2643 if (hba
->type
== SYS_DEV_SATA
) {
2644 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2645 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2647 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2658 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2660 struct sys_dev
*list
, *hba
;
2663 list
= find_intel_devices();
2666 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2671 for (hba
= list
; hba
; hba
= hba
->next
) {
2672 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2674 if (!find_imsm_capability(hba
) && verbose
> 0) {
2676 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2677 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2683 const struct orom_entry
*entry
;
2685 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2686 if (entry
->type
== SYS_DEV_VMD
) {
2687 for (hba
= list
; hba
; hba
= hba
->next
)
2688 print_imsm_capability_export(&entry
->orom
);
2691 print_imsm_capability_export(&entry
->orom
);
2697 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2699 /* the imsm metadata format does not specify any host
2700 * identification information. We return -1 since we can never
2701 * confirm nor deny whether a given array is "meant" for this
2702 * host. We rely on compare_super and the 'family_num' fields to
2703 * exclude member disks that do not belong, and we rely on
2704 * mdadm.conf to specify the arrays that should be assembled.
2705 * Auto-assembly may still pick up "foreign" arrays.
2711 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2713 /* The uuid returned here is used for:
2714 * uuid to put into bitmap file (Create, Grow)
2715 * uuid for backup header when saving critical section (Grow)
2716 * comparing uuids when re-adding a device into an array
2717 * In these cases the uuid required is that of the data-array,
2718 * not the device-set.
2719 * uuid to recognise same set when adding a missing device back
2720 * to an array. This is a uuid for the device-set.
2722 * For each of these we can make do with a truncated
2723 * or hashed uuid rather than the original, as long as
2725 * In each case the uuid required is that of the data-array,
2726 * not the device-set.
2728 /* imsm does not track uuid's so we synthesis one using sha1 on
2729 * - The signature (Which is constant for all imsm array, but no matter)
2730 * - the orig_family_num of the container
2731 * - the index number of the volume
2732 * - the 'serial' number of the volume.
2733 * Hopefully these are all constant.
2735 struct intel_super
*super
= st
->sb
;
2738 struct sha1_ctx ctx
;
2739 struct imsm_dev
*dev
= NULL
;
2742 /* some mdadm versions failed to set ->orig_family_num, in which
2743 * case fall back to ->family_num. orig_family_num will be
2744 * fixed up with the first metadata update.
2746 family_num
= super
->anchor
->orig_family_num
;
2747 if (family_num
== 0)
2748 family_num
= super
->anchor
->family_num
;
2749 sha1_init_ctx(&ctx
);
2750 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2751 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2752 if (super
->current_vol
>= 0)
2753 dev
= get_imsm_dev(super
, super
->current_vol
);
2755 __u32 vol
= super
->current_vol
;
2756 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2757 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2759 sha1_finish_ctx(&ctx
, buf
);
2760 memcpy(uuid
, buf
, 4*4);
2765 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2767 __u8
*v
= get_imsm_version(mpb
);
2768 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2769 char major
[] = { 0, 0, 0 };
2770 char minor
[] = { 0 ,0, 0 };
2771 char patch
[] = { 0, 0, 0 };
2772 char *ver_parse
[] = { major
, minor
, patch
};
2776 while (*v
!= '\0' && v
< end
) {
2777 if (*v
!= '.' && j
< 2)
2778 ver_parse
[i
][j
++] = *v
;
2786 *m
= strtol(minor
, NULL
, 0);
2787 *p
= strtol(patch
, NULL
, 0);
2791 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2793 /* migr_strip_size when repairing or initializing parity */
2794 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2795 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2797 switch (get_imsm_raid_level(map
)) {
2802 return 128*1024 >> 9;
2806 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2808 /* migr_strip_size when rebuilding a degraded disk, no idea why
2809 * this is different than migr_strip_size_resync(), but it's good
2812 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2813 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2815 switch (get_imsm_raid_level(map
)) {
2818 if (map
->num_members
% map
->num_domains
== 0)
2819 return 128*1024 >> 9;
2823 return max((__u32
) 64*1024 >> 9, chunk
);
2825 return 128*1024 >> 9;
2829 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2831 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2832 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2833 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2834 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2836 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2839 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2841 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2842 int level
= get_imsm_raid_level(lo
);
2844 if (level
== 1 || level
== 10) {
2845 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2847 return hi
->num_domains
;
2849 return num_stripes_per_unit_resync(dev
);
2852 static __u8
imsm_num_data_members(struct imsm_map
*map
)
2854 /* named 'imsm_' because raid0, raid1 and raid10
2855 * counter-intuitively have the same number of data disks
2857 switch (get_imsm_raid_level(map
)) {
2859 return map
->num_members
;
2863 return map
->num_members
/2;
2865 return map
->num_members
- 1;
2867 dprintf("unsupported raid level\n");
2872 static unsigned long long calc_component_size(struct imsm_map
*map
,
2873 struct imsm_dev
*dev
)
2875 unsigned long long component_size
;
2876 unsigned long long dev_size
= imsm_dev_size(dev
);
2877 unsigned long long calc_dev_size
= 0;
2878 unsigned int member_disks
= imsm_num_data_members(map
);
2880 if (member_disks
== 0)
2883 component_size
= per_dev_array_size(map
);
2884 calc_dev_size
= component_size
* member_disks
;
2886 /* Component size is rounded to 1MB so difference between size from
2887 * metadata and size calculated from num_data_stripes equals up to
2888 * 2048 blocks per each device. If the difference is higher it means
2889 * that array size was expanded and num_data_stripes was not updated.
2891 if ((unsigned int)abs(calc_dev_size
- dev_size
) >
2892 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
2893 component_size
= dev_size
/ member_disks
;
2894 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
2895 component_size
/ map
->blocks_per_strip
,
2896 num_data_stripes(map
));
2899 return component_size
;
2902 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2904 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2905 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2907 switch(get_imsm_raid_level(map
)) {
2910 return chunk
* map
->num_domains
;
2912 return chunk
* map
->num_members
;
2918 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2920 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2921 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2922 __u32 strip
= block
/ chunk
;
2924 switch (get_imsm_raid_level(map
)) {
2927 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2928 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2930 return vol_stripe
* chunk
+ block
% chunk
;
2932 __u32 stripe
= strip
/ (map
->num_members
- 1);
2934 return stripe
* chunk
+ block
% chunk
;
2941 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2942 struct imsm_dev
*dev
)
2944 /* calculate the conversion factor between per member 'blocks'
2945 * (md/{resync,rebuild}_start) and imsm migration units, return
2946 * 0 for the 'not migrating' and 'unsupported migration' cases
2948 if (!dev
->vol
.migr_state
)
2951 switch (migr_type(dev
)) {
2952 case MIGR_GEN_MIGR
: {
2953 struct migr_record
*migr_rec
= super
->migr_rec
;
2954 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2959 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2960 __u32 stripes_per_unit
;
2961 __u32 blocks_per_unit
;
2970 /* yes, this is really the translation of migr_units to
2971 * per-member blocks in the 'resync' case
2973 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2974 migr_chunk
= migr_strip_blocks_resync(dev
);
2975 disks
= imsm_num_data_members(map
);
2976 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2977 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2978 segment
= blocks_per_unit
/ stripe
;
2979 block_rel
= blocks_per_unit
- segment
* stripe
;
2980 parity_depth
= parity_segment_depth(dev
);
2981 block_map
= map_migr_block(dev
, block_rel
);
2982 return block_map
+ parity_depth
* segment
;
2984 case MIGR_REBUILD
: {
2985 __u32 stripes_per_unit
;
2988 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2989 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2990 return migr_chunk
* stripes_per_unit
;
2992 case MIGR_STATE_CHANGE
:
2998 static int imsm_level_to_layout(int level
)
3006 return ALGORITHM_LEFT_ASYMMETRIC
;
3013 /*******************************************************************************
3014 * Function: read_imsm_migr_rec
3015 * Description: Function reads imsm migration record from last sector of disk
3017 * fd : disk descriptor
3018 * super : metadata info
3022 ******************************************************************************/
3023 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
3026 unsigned int sector_size
= super
->sector_size
;
3027 unsigned long long dsize
;
3029 get_dev_size(fd
, NULL
, &dsize
);
3030 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
3032 pr_err("Cannot seek to anchor block: %s\n",
3036 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
3037 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3038 MIGR_REC_BUF_SECTORS
*sector_size
) {
3039 pr_err("Cannot read migr record block: %s\n",
3044 if (sector_size
== 4096)
3045 convert_from_4k_imsm_migr_rec(super
);
3051 static struct imsm_dev
*imsm_get_device_during_migration(
3052 struct intel_super
*super
)
3055 struct intel_dev
*dv
;
3057 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
3058 if (is_gen_migration(dv
->dev
))
3064 /*******************************************************************************
3065 * Function: load_imsm_migr_rec
3066 * Description: Function reads imsm migration record (it is stored at the last
3069 * super : imsm internal array info
3070 * info : general array info
3074 * -2 : no migration in progress
3075 ******************************************************************************/
3076 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
3083 struct imsm_dev
*dev
;
3084 struct imsm_map
*map
;
3087 /* find map under migration */
3088 dev
= imsm_get_device_during_migration(super
);
3089 /* nothing to load,no migration in progress?
3095 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
3096 /* read only from one of the first two slots */
3097 if ((sd
->disk
.raid_disk
< 0) ||
3098 (sd
->disk
.raid_disk
> 1))
3101 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3102 fd
= dev_open(nm
, O_RDONLY
);
3108 map
= get_imsm_map(dev
, MAP_0
);
3109 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3110 /* skip spare and failed disks
3114 /* read only from one of the first two slots */
3116 slot
= get_imsm_disk_slot(map
, dl
->index
);
3117 if (map
== NULL
|| slot
> 1 || slot
< 0)
3119 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3120 fd
= dev_open(nm
, O_RDONLY
);
3127 retval
= read_imsm_migr_rec(fd
, super
);
3135 /*******************************************************************************
3136 * function: imsm_create_metadata_checkpoint_update
3137 * Description: It creates update for checkpoint change.
3139 * super : imsm internal array info
3140 * u : pointer to prepared update
3143 * If length is equal to 0, input pointer u contains no update
3144 ******************************************************************************/
3145 static int imsm_create_metadata_checkpoint_update(
3146 struct intel_super
*super
,
3147 struct imsm_update_general_migration_checkpoint
**u
)
3150 int update_memory_size
= 0;
3152 dprintf("(enter)\n");
3158 /* size of all update data without anchor */
3159 update_memory_size
=
3160 sizeof(struct imsm_update_general_migration_checkpoint
);
3162 *u
= xcalloc(1, update_memory_size
);
3164 dprintf("error: cannot get memory\n");
3167 (*u
)->type
= update_general_migration_checkpoint
;
3168 (*u
)->curr_migr_unit
= current_migr_unit(super
->migr_rec
);
3169 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3171 return update_memory_size
;
3174 static void imsm_update_metadata_locally(struct supertype
*st
,
3175 void *buf
, int len
);
3177 /*******************************************************************************
3178 * Function: write_imsm_migr_rec
3179 * Description: Function writes imsm migration record
3180 * (at the last sector of disk)
3182 * super : imsm internal array info
3186 ******************************************************************************/
3187 static int write_imsm_migr_rec(struct supertype
*st
)
3189 struct intel_super
*super
= st
->sb
;
3190 unsigned int sector_size
= super
->sector_size
;
3191 unsigned long long dsize
;
3197 struct imsm_update_general_migration_checkpoint
*u
;
3198 struct imsm_dev
*dev
;
3199 struct imsm_map
*map
;
3201 /* find map under migration */
3202 dev
= imsm_get_device_during_migration(super
);
3203 /* if no migration, write buffer anyway to clear migr_record
3204 * on disk based on first available device
3207 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3208 super
->current_vol
);
3210 map
= get_imsm_map(dev
, MAP_0
);
3212 if (sector_size
== 4096)
3213 convert_to_4k_imsm_migr_rec(super
);
3214 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3217 /* skip failed and spare devices */
3220 /* write to 2 first slots only */
3222 slot
= get_imsm_disk_slot(map
, sd
->index
);
3223 if (map
== NULL
|| slot
> 1 || slot
< 0)
3226 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3227 fd
= dev_open(nm
, O_RDWR
);
3230 get_dev_size(fd
, NULL
, &dsize
);
3231 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3233 pr_err("Cannot seek to anchor block: %s\n",
3237 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3238 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3239 MIGR_REC_BUF_SECTORS
*sector_size
) {
3240 pr_err("Cannot write migr record block: %s\n",
3247 if (sector_size
== 4096)
3248 convert_from_4k_imsm_migr_rec(super
);
3249 /* update checkpoint information in metadata */
3250 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3252 dprintf("imsm: Cannot prepare update\n");
3255 /* update metadata locally */
3256 imsm_update_metadata_locally(st
, u
, len
);
3257 /* and possibly remotely */
3258 if (st
->update_tail
) {
3259 append_metadata_update(st
, u
, len
);
3260 /* during reshape we do all work inside metadata handler
3261 * manage_reshape(), so metadata update has to be triggered
3264 flush_metadata_updates(st
);
3265 st
->update_tail
= &st
->updates
;
3276 /* spare/missing disks activations are not allowe when
3277 * array/container performs reshape operation, because
3278 * all arrays in container works on the same disks set
3280 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3283 struct intel_dev
*i_dev
;
3284 struct imsm_dev
*dev
;
3286 /* check whole container
3288 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3290 if (is_gen_migration(dev
)) {
3291 /* No repair during any migration in container
3299 static unsigned long long imsm_component_size_alignment_check(int level
,
3301 unsigned int sector_size
,
3302 unsigned long long component_size
)
3304 unsigned int component_size_alignment
;
3306 /* check component size alignment
3308 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3310 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3311 level
, chunk_size
, component_size
,
3312 component_size_alignment
);
3314 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3315 dprintf("imsm: reported component size aligned from %llu ",
3317 component_size
-= component_size_alignment
;
3318 dprintf_cont("to %llu (%i).\n",
3319 component_size
, component_size_alignment
);
3322 return component_size
;
3325 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3327 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3328 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3330 return pba_of_lba0(map
) +
3331 (num_data_stripes(map
) * map
->blocks_per_strip
);
3334 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3336 struct intel_super
*super
= st
->sb
;
3337 struct migr_record
*migr_rec
= super
->migr_rec
;
3338 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3339 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3340 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3341 struct imsm_map
*map_to_analyse
= map
;
3343 int map_disks
= info
->array
.raid_disks
;
3345 memset(info
, 0, sizeof(*info
));
3347 map_to_analyse
= prev_map
;
3349 dl
= super
->current_disk
;
3351 info
->container_member
= super
->current_vol
;
3352 info
->array
.raid_disks
= map
->num_members
;
3353 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3354 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3355 info
->array
.md_minor
= -1;
3356 info
->array
.ctime
= 0;
3357 info
->array
.utime
= 0;
3358 info
->array
.chunk_size
=
3359 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3360 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3361 info
->custom_array_size
= imsm_dev_size(dev
);
3362 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3364 if (is_gen_migration(dev
)) {
3365 info
->reshape_active
= 1;
3366 info
->new_level
= get_imsm_raid_level(map
);
3367 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3368 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3369 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3370 if (info
->delta_disks
) {
3371 /* this needs to be applied to every array
3374 info
->reshape_active
= CONTAINER_RESHAPE
;
3376 /* We shape information that we give to md might have to be
3377 * modify to cope with md's requirement for reshaping arrays.
3378 * For example, when reshaping a RAID0, md requires it to be
3379 * presented as a degraded RAID4.
3380 * Also if a RAID0 is migrating to a RAID5 we need to specify
3381 * the array as already being RAID5, but the 'before' layout
3382 * is a RAID4-like layout.
3384 switch (info
->array
.level
) {
3386 switch(info
->new_level
) {
3388 /* conversion is happening as RAID4 */
3389 info
->array
.level
= 4;
3390 info
->array
.raid_disks
+= 1;
3393 /* conversion is happening as RAID5 */
3394 info
->array
.level
= 5;
3395 info
->array
.layout
= ALGORITHM_PARITY_N
;
3396 info
->delta_disks
-= 1;
3399 /* FIXME error message */
3400 info
->array
.level
= UnSet
;
3406 info
->new_level
= UnSet
;
3407 info
->new_layout
= UnSet
;
3408 info
->new_chunk
= info
->array
.chunk_size
;
3409 info
->delta_disks
= 0;
3413 info
->disk
.major
= dl
->major
;
3414 info
->disk
.minor
= dl
->minor
;
3415 info
->disk
.number
= dl
->index
;
3416 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3420 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3421 info
->component_size
= calc_component_size(map
, dev
);
3422 info
->component_size
= imsm_component_size_alignment_check(
3424 info
->array
.chunk_size
,
3426 info
->component_size
);
3427 info
->bb
.supported
= 1;
3429 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3430 info
->recovery_start
= MaxSector
;
3432 if (info
->array
.level
== 5 &&
3433 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3434 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3435 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3436 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3437 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3438 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3440 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3442 } else if (info
->array
.level
<= 0) {
3443 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3445 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3448 info
->reshape_progress
= 0;
3449 info
->resync_start
= MaxSector
;
3450 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3451 !(info
->array
.state
& 1)) &&
3452 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3453 info
->resync_start
= 0;
3455 if (dev
->vol
.migr_state
) {
3456 switch (migr_type(dev
)) {
3459 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3461 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3463 info
->resync_start
= blocks_per_unit
* units
;
3466 case MIGR_GEN_MIGR
: {
3467 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3469 __u64 units
= current_migr_unit(migr_rec
);
3470 unsigned long long array_blocks
;
3473 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3475 (get_num_migr_units(migr_rec
)-1)) &&
3476 (super
->migr_rec
->rec_status
==
3477 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3480 info
->reshape_progress
= blocks_per_unit
* units
;
3482 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3483 (unsigned long long)units
,
3484 (unsigned long long)blocks_per_unit
,
3485 info
->reshape_progress
);
3487 used_disks
= imsm_num_data_members(prev_map
);
3488 if (used_disks
> 0) {
3489 array_blocks
= per_dev_array_size(map
) *
3491 info
->custom_array_size
=
3492 round_size_to_mb(array_blocks
,
3498 /* we could emulate the checkpointing of
3499 * 'sync_action=check' migrations, but for now
3500 * we just immediately complete them
3503 /* this is handled by container_content_imsm() */
3504 case MIGR_STATE_CHANGE
:
3505 /* FIXME handle other migrations */
3507 /* we are not dirty, so... */
3508 info
->resync_start
= MaxSector
;
3512 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3513 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3515 info
->array
.major_version
= -1;
3516 info
->array
.minor_version
= -2;
3517 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3518 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3519 uuid_from_super_imsm(st
, info
->uuid
);
3523 for (i
=0; i
<map_disks
; i
++) {
3525 if (i
< info
->array
.raid_disks
) {
3526 struct imsm_disk
*dsk
;
3527 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3528 dsk
= get_imsm_disk(super
, j
);
3529 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3536 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3537 int failed
, int look_in_map
);
3539 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3542 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3544 if (is_gen_migration(dev
)) {
3547 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3549 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3550 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3551 if (map2
->map_state
!= map_state
) {
3552 map2
->map_state
= map_state
;
3553 super
->updates_pending
++;
3558 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3562 for (d
= super
->missing
; d
; d
= d
->next
)
3563 if (d
->index
== index
)
3568 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3570 struct intel_super
*super
= st
->sb
;
3571 struct imsm_disk
*disk
;
3572 int map_disks
= info
->array
.raid_disks
;
3573 int max_enough
= -1;
3575 struct imsm_super
*mpb
;
3577 if (super
->current_vol
>= 0) {
3578 getinfo_super_imsm_volume(st
, info
, map
);
3581 memset(info
, 0, sizeof(*info
));
3583 /* Set raid_disks to zero so that Assemble will always pull in valid
3586 info
->array
.raid_disks
= 0;
3587 info
->array
.level
= LEVEL_CONTAINER
;
3588 info
->array
.layout
= 0;
3589 info
->array
.md_minor
= -1;
3590 info
->array
.ctime
= 0; /* N/A for imsm */
3591 info
->array
.utime
= 0;
3592 info
->array
.chunk_size
= 0;
3594 info
->disk
.major
= 0;
3595 info
->disk
.minor
= 0;
3596 info
->disk
.raid_disk
= -1;
3597 info
->reshape_active
= 0;
3598 info
->array
.major_version
= -1;
3599 info
->array
.minor_version
= -2;
3600 strcpy(info
->text_version
, "imsm");
3601 info
->safe_mode_delay
= 0;
3602 info
->disk
.number
= -1;
3603 info
->disk
.state
= 0;
3605 info
->recovery_start
= MaxSector
;
3606 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3607 info
->bb
.supported
= 1;
3609 /* do we have the all the insync disks that we expect? */
3610 mpb
= super
->anchor
;
3611 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3613 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3614 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3615 int failed
, enough
, j
, missing
= 0;
3616 struct imsm_map
*map
;
3619 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3620 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3621 map
= get_imsm_map(dev
, MAP_0
);
3623 /* any newly missing disks?
3624 * (catches single-degraded vs double-degraded)
3626 for (j
= 0; j
< map
->num_members
; j
++) {
3627 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3628 __u32 idx
= ord_to_idx(ord
);
3630 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3631 info
->disk
.raid_disk
= j
;
3633 if (!(ord
& IMSM_ORD_REBUILD
) &&
3634 get_imsm_missing(super
, idx
)) {
3640 if (state
== IMSM_T_STATE_FAILED
)
3642 else if (state
== IMSM_T_STATE_DEGRADED
&&
3643 (state
!= map
->map_state
|| missing
))
3645 else /* we're normal, or already degraded */
3647 if (is_gen_migration(dev
) && missing
) {
3648 /* during general migration we need all disks
3649 * that process is running on.
3650 * No new missing disk is allowed.
3654 /* no more checks necessary
3658 /* in the missing/failed disk case check to see
3659 * if at least one array is runnable
3661 max_enough
= max(max_enough
, enough
);
3663 dprintf("enough: %d\n", max_enough
);
3664 info
->container_enough
= max_enough
;
3667 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3669 disk
= &super
->disks
->disk
;
3670 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3671 info
->component_size
= reserved
;
3672 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3673 /* we don't change info->disk.raid_disk here because
3674 * this state will be finalized in mdmon after we have
3675 * found the 'most fresh' version of the metadata
3677 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3678 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3679 0 : (1 << MD_DISK_SYNC
);
3682 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3683 * ->compare_super may have updated the 'num_raid_devs' field for spares
3685 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3686 uuid_from_super_imsm(st
, info
->uuid
);
3688 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3690 /* I don't know how to compute 'map' on imsm, so use safe default */
3693 for (i
= 0; i
< map_disks
; i
++)
3699 /* allocates memory and fills disk in mdinfo structure
3700 * for each disk in array */
3701 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3703 struct mdinfo
*mddev
;
3704 struct intel_super
*super
= st
->sb
;
3705 struct imsm_disk
*disk
;
3708 if (!super
|| !super
->disks
)
3711 mddev
= xcalloc(1, sizeof(*mddev
));
3715 tmp
= xcalloc(1, sizeof(*tmp
));
3717 tmp
->next
= mddev
->devs
;
3719 tmp
->disk
.number
= count
++;
3720 tmp
->disk
.major
= dl
->major
;
3721 tmp
->disk
.minor
= dl
->minor
;
3722 tmp
->disk
.state
= is_configured(disk
) ?
3723 (1 << MD_DISK_ACTIVE
) : 0;
3724 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3725 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3726 tmp
->disk
.raid_disk
= -1;
3732 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3733 char *update
, char *devname
, int verbose
,
3734 int uuid_set
, char *homehost
)
3736 /* For 'assemble' and 'force' we need to return non-zero if any
3737 * change was made. For others, the return value is ignored.
3738 * Update options are:
3739 * force-one : This device looks a bit old but needs to be included,
3740 * update age info appropriately.
3741 * assemble: clear any 'faulty' flag to allow this device to
3743 * force-array: Array is degraded but being forced, mark it clean
3744 * if that will be needed to assemble it.
3746 * newdev: not used ????
3747 * grow: Array has gained a new device - this is currently for
3749 * resync: mark as dirty so a resync will happen.
3750 * name: update the name - preserving the homehost
3751 * uuid: Change the uuid of the array to match watch is given
3753 * Following are not relevant for this imsm:
3754 * sparc2.2 : update from old dodgey metadata
3755 * super-minor: change the preferred_minor number
3756 * summaries: update redundant counters.
3757 * homehost: update the recorded homehost
3758 * _reshape_progress: record new reshape_progress position.
3761 struct intel_super
*super
= st
->sb
;
3762 struct imsm_super
*mpb
;
3764 /* we can only update container info */
3765 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3768 mpb
= super
->anchor
;
3770 if (strcmp(update
, "uuid") == 0) {
3771 /* We take this to mean that the family_num should be updated.
3772 * However that is much smaller than the uuid so we cannot really
3773 * allow an explicit uuid to be given. And it is hard to reliably
3775 * So if !uuid_set we know the current uuid is random and just used
3776 * the first 'int' and copy it to the other 3 positions.
3777 * Otherwise we require the 4 'int's to be the same as would be the
3778 * case if we are using a random uuid. So an explicit uuid will be
3779 * accepted as long as all for ints are the same... which shouldn't hurt
3782 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3785 if (info
->uuid
[0] != info
->uuid
[1] ||
3786 info
->uuid
[1] != info
->uuid
[2] ||
3787 info
->uuid
[2] != info
->uuid
[3])
3793 mpb
->orig_family_num
= info
->uuid
[0];
3794 } else if (strcmp(update
, "assemble") == 0)
3799 /* successful update? recompute checksum */
3801 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3806 static size_t disks_to_mpb_size(int disks
)
3810 size
= sizeof(struct imsm_super
);
3811 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3812 size
+= 2 * sizeof(struct imsm_dev
);
3813 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3814 size
+= (4 - 2) * sizeof(struct imsm_map
);
3815 /* 4 possible disk_ord_tbl's */
3816 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3817 /* maximum bbm log */
3818 size
+= sizeof(struct bbm_log
);
3823 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3824 unsigned long long data_offset
)
3826 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3829 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3832 static void free_devlist(struct intel_super
*super
)
3834 struct intel_dev
*dv
;
3836 while (super
->devlist
) {
3837 dv
= super
->devlist
->next
;
3838 free(super
->devlist
->dev
);
3839 free(super
->devlist
);
3840 super
->devlist
= dv
;
3844 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3846 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3849 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3853 * 0 same, or first was empty, and second was copied
3854 * 1 second had wrong number
3856 * 3 wrong other info
3858 struct intel_super
*first
= st
->sb
;
3859 struct intel_super
*sec
= tst
->sb
;
3866 /* in platform dependent environment test if the disks
3867 * use the same Intel hba
3868 * If not on Intel hba at all, allow anything.
3870 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3871 if (first
->hba
->type
!= sec
->hba
->type
) {
3873 "HBAs of devices do not match %s != %s\n",
3874 get_sys_dev_type(first
->hba
->type
),
3875 get_sys_dev_type(sec
->hba
->type
));
3878 if (first
->orom
!= sec
->orom
) {
3880 "HBAs of devices do not match %s != %s\n",
3881 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3886 /* if an anchor does not have num_raid_devs set then it is a free
3889 if (first
->anchor
->num_raid_devs
> 0 &&
3890 sec
->anchor
->num_raid_devs
> 0) {
3891 /* Determine if these disks might ever have been
3892 * related. Further disambiguation can only take place
3893 * in load_super_imsm_all
3895 __u32 first_family
= first
->anchor
->orig_family_num
;
3896 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3898 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3899 MAX_SIGNATURE_LENGTH
) != 0)
3902 if (first_family
== 0)
3903 first_family
= first
->anchor
->family_num
;
3904 if (sec_family
== 0)
3905 sec_family
= sec
->anchor
->family_num
;
3907 if (first_family
!= sec_family
)
3912 /* if 'first' is a spare promote it to a populated mpb with sec's
3915 if (first
->anchor
->num_raid_devs
== 0 &&
3916 sec
->anchor
->num_raid_devs
> 0) {
3918 struct intel_dev
*dv
;
3919 struct imsm_dev
*dev
;
3921 /* we need to copy raid device info from sec if an allocation
3922 * fails here we don't associate the spare
3924 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3925 dv
= xmalloc(sizeof(*dv
));
3926 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3929 dv
->next
= first
->devlist
;
3930 first
->devlist
= dv
;
3932 if (i
< sec
->anchor
->num_raid_devs
) {
3933 /* allocation failure */
3934 free_devlist(first
);
3935 pr_err("imsm: failed to associate spare\n");
3938 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3939 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3940 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3941 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3942 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3943 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3949 static void fd2devname(int fd
, char *name
)
3953 char dname
[PATH_MAX
];
3958 if (fstat(fd
, &st
) != 0)
3960 sprintf(path
, "/sys/dev/block/%d:%d",
3961 major(st
.st_rdev
), minor(st
.st_rdev
));
3963 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3968 nm
= strrchr(dname
, '/');
3971 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3975 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3978 char *name
= fd2kname(fd
);
3983 if (strncmp(name
, "nvme", 4) != 0)
3986 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3988 return load_sys(path
, buf
, buf_len
);
3991 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3993 static int imsm_read_serial(int fd
, char *devname
,
3994 __u8 serial
[MAX_RAID_SERIAL_LEN
])
4003 memset(buf
, 0, sizeof(buf
));
4005 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
4008 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
4010 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
4011 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
4012 fd2devname(fd
, (char *) serial
);
4018 pr_err("Failed to retrieve serial for %s\n",
4023 /* trim all whitespace and non-printable characters and convert
4026 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4029 /* ':' is reserved for use in placeholder serial
4030 * numbers for missing disks
4041 /* truncate leading characters */
4042 if (len
> MAX_RAID_SERIAL_LEN
) {
4043 dest
+= len
- MAX_RAID_SERIAL_LEN
;
4044 len
= MAX_RAID_SERIAL_LEN
;
4047 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
4048 memcpy(serial
, dest
, len
);
4053 static int serialcmp(__u8
*s1
, __u8
*s2
)
4055 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4058 static void serialcpy(__u8
*dest
, __u8
*src
)
4060 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4063 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4067 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4068 if (serialcmp(dl
->serial
, serial
) == 0)
4074 static struct imsm_disk
*
4075 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4079 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4080 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4082 if (serialcmp(disk
->serial
, serial
) == 0) {
4093 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4095 struct imsm_disk
*disk
;
4100 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4102 rv
= imsm_read_serial(fd
, devname
, serial
);
4107 dl
= xcalloc(1, sizeof(*dl
));
4110 dl
->major
= major(stb
.st_rdev
);
4111 dl
->minor
= minor(stb
.st_rdev
);
4112 dl
->next
= super
->disks
;
4113 dl
->fd
= keep_fd
? fd
: -1;
4114 assert(super
->disks
== NULL
);
4116 serialcpy(dl
->serial
, serial
);
4119 fd2devname(fd
, name
);
4121 dl
->devname
= xstrdup(devname
);
4123 dl
->devname
= xstrdup(name
);
4125 /* look up this disk's index in the current anchor */
4126 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4129 /* only set index on disks that are a member of a
4130 * populated contianer, i.e. one with raid_devs
4132 if (is_failed(&dl
->disk
))
4134 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4141 /* When migrating map0 contains the 'destination' state while map1
4142 * contains the current state. When not migrating map0 contains the
4143 * current state. This routine assumes that map[0].map_state is set to
4144 * the current array state before being called.
4146 * Migration is indicated by one of the following states
4147 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4148 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4149 * map1state=unitialized)
4150 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4152 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4153 * map1state=degraded)
4154 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4157 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4158 __u8 to_state
, int migr_type
)
4160 struct imsm_map
*dest
;
4161 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4163 dev
->vol
.migr_state
= 1;
4164 set_migr_type(dev
, migr_type
);
4165 dev
->vol
.curr_migr_unit
= 0;
4166 dest
= get_imsm_map(dev
, MAP_1
);
4168 /* duplicate and then set the target end state in map[0] */
4169 memcpy(dest
, src
, sizeof_imsm_map(src
));
4170 if (migr_type
== MIGR_GEN_MIGR
) {
4174 for (i
= 0; i
< src
->num_members
; i
++) {
4175 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4176 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4180 if (migr_type
== MIGR_GEN_MIGR
)
4181 /* Clear migration record */
4182 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4184 src
->map_state
= to_state
;
4187 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4190 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4191 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4195 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4196 * completed in the last migration.
4198 * FIXME add support for raid-level-migration
4200 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4201 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4202 /* when final map state is other than expected
4203 * merge maps (not for migration)
4207 for (i
= 0; i
< prev
->num_members
; i
++)
4208 for (j
= 0; j
< map
->num_members
; j
++)
4209 /* during online capacity expansion
4210 * disks position can be changed
4211 * if takeover is used
4213 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4214 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4215 map
->disk_ord_tbl
[j
] |=
4216 prev
->disk_ord_tbl
[i
];
4219 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4220 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4223 dev
->vol
.migr_state
= 0;
4224 set_migr_type(dev
, 0);
4225 dev
->vol
.curr_migr_unit
= 0;
4226 map
->map_state
= map_state
;
4229 static int parse_raid_devices(struct intel_super
*super
)
4232 struct imsm_dev
*dev_new
;
4233 size_t len
, len_migr
;
4235 size_t space_needed
= 0;
4236 struct imsm_super
*mpb
= super
->anchor
;
4238 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4239 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4240 struct intel_dev
*dv
;
4242 len
= sizeof_imsm_dev(dev_iter
, 0);
4243 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4245 space_needed
+= len_migr
- len
;
4247 dv
= xmalloc(sizeof(*dv
));
4248 if (max_len
< len_migr
)
4250 if (max_len
> len_migr
)
4251 space_needed
+= max_len
- len_migr
;
4252 dev_new
= xmalloc(max_len
);
4253 imsm_copy_dev(dev_new
, dev_iter
);
4256 dv
->next
= super
->devlist
;
4257 super
->devlist
= dv
;
4260 /* ensure that super->buf is large enough when all raid devices
4263 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4266 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4267 super
->sector_size
);
4268 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4271 memcpy(buf
, super
->buf
, super
->len
);
4272 memset(buf
+ super
->len
, 0, len
- super
->len
);
4278 super
->extra_space
+= space_needed
;
4283 /*******************************************************************************
4284 * Function: check_mpb_migr_compatibility
4285 * Description: Function checks for unsupported migration features:
4286 * - migration optimization area (pba_of_lba0)
4287 * - descending reshape (ascending_migr)
4289 * super : imsm metadata information
4291 * 0 : migration is compatible
4292 * -1 : migration is not compatible
4293 ******************************************************************************/
4294 int check_mpb_migr_compatibility(struct intel_super
*super
)
4296 struct imsm_map
*map0
, *map1
;
4297 struct migr_record
*migr_rec
= super
->migr_rec
;
4300 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4301 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4304 dev_iter
->vol
.migr_state
== 1 &&
4305 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4306 /* This device is migrating */
4307 map0
= get_imsm_map(dev_iter
, MAP_0
);
4308 map1
= get_imsm_map(dev_iter
, MAP_1
);
4309 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4310 /* migration optimization area was used */
4312 if (migr_rec
->ascending_migr
== 0 &&
4313 migr_rec
->dest_depth_per_unit
> 0)
4314 /* descending reshape not supported yet */
4321 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4323 /* load_imsm_mpb - read matrix metadata
4324 * allocates super->mpb to be freed by free_imsm
4326 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4328 unsigned long long dsize
;
4329 unsigned long long sectors
;
4330 unsigned int sector_size
= super
->sector_size
;
4332 struct imsm_super
*anchor
;
4335 get_dev_size(fd
, NULL
, &dsize
);
4336 if (dsize
< 2*sector_size
) {
4338 pr_err("%s: device to small for imsm\n",
4343 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4345 pr_err("Cannot seek to anchor block on %s: %s\n",
4346 devname
, strerror(errno
));
4350 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4352 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4355 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4357 pr_err("Cannot read anchor block on %s: %s\n",
4358 devname
, strerror(errno
));
4363 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4365 pr_err("no IMSM anchor on %s\n", devname
);
4370 __free_imsm(super
, 0);
4371 /* reload capability and hba */
4373 /* capability and hba must be updated with new super allocation */
4374 find_intel_hba_capability(fd
, super
, devname
);
4375 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4376 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4378 pr_err("unable to allocate %zu byte mpb buffer\n",
4383 memcpy(super
->buf
, anchor
, sector_size
);
4385 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4388 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4389 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4390 pr_err("could not allocate migr_rec buffer\n");
4394 super
->clean_migration_record_by_mdmon
= 0;
4397 check_sum
= __gen_imsm_checksum(super
->anchor
);
4398 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4400 pr_err("IMSM checksum %x != %x on %s\n",
4402 __le32_to_cpu(super
->anchor
->check_sum
),
4410 /* read the extended mpb */
4411 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4413 pr_err("Cannot seek to extended mpb on %s: %s\n",
4414 devname
, strerror(errno
));
4418 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4419 super
->len
- sector_size
) != super
->len
- sector_size
) {
4421 pr_err("Cannot read extended mpb on %s: %s\n",
4422 devname
, strerror(errno
));
4426 check_sum
= __gen_imsm_checksum(super
->anchor
);
4427 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4429 pr_err("IMSM checksum %x != %x on %s\n",
4430 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4438 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4440 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4441 static void clear_hi(struct intel_super
*super
)
4443 struct imsm_super
*mpb
= super
->anchor
;
4445 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4447 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4448 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4449 disk
->total_blocks_hi
= 0;
4451 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4452 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4455 for (n
= 0; n
< 2; ++n
) {
4456 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4459 map
->pba_of_lba0_hi
= 0;
4460 map
->blocks_per_member_hi
= 0;
4461 map
->num_data_stripes_hi
= 0;
4467 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4471 err
= load_imsm_mpb(fd
, super
, devname
);
4474 if (super
->sector_size
== 4096)
4475 convert_from_4k(super
);
4476 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4479 err
= parse_raid_devices(super
);
4482 err
= load_bbm_log(super
);
4487 static void __free_imsm_disk(struct dl
*d
)
4499 static void free_imsm_disks(struct intel_super
*super
)
4503 while (super
->disks
) {
4505 super
->disks
= d
->next
;
4506 __free_imsm_disk(d
);
4508 while (super
->disk_mgmt_list
) {
4509 d
= super
->disk_mgmt_list
;
4510 super
->disk_mgmt_list
= d
->next
;
4511 __free_imsm_disk(d
);
4513 while (super
->missing
) {
4515 super
->missing
= d
->next
;
4516 __free_imsm_disk(d
);
4521 /* free all the pieces hanging off of a super pointer */
4522 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4524 struct intel_hba
*elem
, *next
;
4530 /* unlink capability description */
4532 if (super
->migr_rec_buf
) {
4533 free(super
->migr_rec_buf
);
4534 super
->migr_rec_buf
= NULL
;
4537 free_imsm_disks(super
);
4538 free_devlist(super
);
4542 free((void *)elem
->path
);
4548 free(super
->bbm_log
);
4552 static void free_imsm(struct intel_super
*super
)
4554 __free_imsm(super
, 1);
4555 free(super
->bb
.entries
);
4559 static void free_super_imsm(struct supertype
*st
)
4561 struct intel_super
*super
= st
->sb
;
4570 static struct intel_super
*alloc_super(void)
4572 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4574 super
->current_vol
= -1;
4575 super
->create_offset
= ~((unsigned long long) 0);
4577 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4578 sizeof(struct md_bb_entry
));
4579 if (!super
->bb
.entries
) {
4588 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4590 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4592 struct sys_dev
*hba_name
;
4595 if (fd
>= 0 && test_partition(fd
)) {
4596 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4600 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4605 hba_name
= find_disk_attached_hba(fd
, NULL
);
4608 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4612 rv
= attach_hba_to_super(super
, hba_name
);
4615 struct intel_hba
*hba
= super
->hba
;
4617 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4618 " but the container is assigned to Intel(R) %s %s (",
4620 get_sys_dev_type(hba_name
->type
),
4621 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4622 hba_name
->pci_id
? : "Err!",
4623 get_sys_dev_type(super
->hba
->type
),
4624 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4627 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4629 fprintf(stderr
, ", ");
4632 fprintf(stderr
, ").\n"
4633 " Mixing devices attached to different controllers is not allowed.\n");
4637 super
->orom
= find_imsm_capability(hba_name
);
4644 /* find_missing - helper routine for load_super_imsm_all that identifies
4645 * disks that have disappeared from the system. This routine relies on
4646 * the mpb being uptodate, which it is at load time.
4648 static int find_missing(struct intel_super
*super
)
4651 struct imsm_super
*mpb
= super
->anchor
;
4653 struct imsm_disk
*disk
;
4655 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4656 disk
= __get_imsm_disk(mpb
, i
);
4657 dl
= serial_to_dl(disk
->serial
, super
);
4661 dl
= xmalloc(sizeof(*dl
));
4665 dl
->devname
= xstrdup("missing");
4667 serialcpy(dl
->serial
, disk
->serial
);
4670 dl
->next
= super
->missing
;
4671 super
->missing
= dl
;
4677 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4679 struct intel_disk
*idisk
= disk_list
;
4682 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4684 idisk
= idisk
->next
;
4690 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4691 struct intel_super
*super
,
4692 struct intel_disk
**disk_list
)
4694 struct imsm_disk
*d
= &super
->disks
->disk
;
4695 struct imsm_super
*mpb
= super
->anchor
;
4698 for (i
= 0; i
< tbl_size
; i
++) {
4699 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4700 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4702 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4703 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4704 dprintf("mpb from %d:%d matches %d:%d\n",
4705 super
->disks
->major
,
4706 super
->disks
->minor
,
4707 table
[i
]->disks
->major
,
4708 table
[i
]->disks
->minor
);
4712 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4713 is_configured(d
) == is_configured(tbl_d
)) &&
4714 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4715 /* current version of the mpb is a
4716 * better candidate than the one in
4717 * super_table, but copy over "cross
4718 * generational" status
4720 struct intel_disk
*idisk
;
4722 dprintf("mpb from %d:%d replaces %d:%d\n",
4723 super
->disks
->major
,
4724 super
->disks
->minor
,
4725 table
[i
]->disks
->major
,
4726 table
[i
]->disks
->minor
);
4728 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4729 if (idisk
&& is_failed(&idisk
->disk
))
4730 tbl_d
->status
|= FAILED_DISK
;
4733 struct intel_disk
*idisk
;
4734 struct imsm_disk
*disk
;
4736 /* tbl_mpb is more up to date, but copy
4737 * over cross generational status before
4740 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4741 if (disk
&& is_failed(disk
))
4742 d
->status
|= FAILED_DISK
;
4744 idisk
= disk_list_get(d
->serial
, *disk_list
);
4747 if (disk
&& is_configured(disk
))
4748 idisk
->disk
.status
|= CONFIGURED_DISK
;
4751 dprintf("mpb from %d:%d prefer %d:%d\n",
4752 super
->disks
->major
,
4753 super
->disks
->minor
,
4754 table
[i
]->disks
->major
,
4755 table
[i
]->disks
->minor
);
4763 table
[tbl_size
++] = super
;
4767 /* update/extend the merged list of imsm_disk records */
4768 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4769 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4770 struct intel_disk
*idisk
;
4772 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4774 idisk
->disk
.status
|= disk
->status
;
4775 if (is_configured(&idisk
->disk
) ||
4776 is_failed(&idisk
->disk
))
4777 idisk
->disk
.status
&= ~(SPARE_DISK
);
4779 idisk
= xcalloc(1, sizeof(*idisk
));
4780 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4781 idisk
->disk
= *disk
;
4782 idisk
->next
= *disk_list
;
4786 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4793 static struct intel_super
*
4794 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4797 struct imsm_super
*mpb
= super
->anchor
;
4801 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4802 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4803 struct intel_disk
*idisk
;
4805 idisk
= disk_list_get(disk
->serial
, disk_list
);
4807 if (idisk
->owner
== owner
||
4808 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4811 dprintf("'%.16s' owner %d != %d\n",
4812 disk
->serial
, idisk
->owner
,
4815 dprintf("unknown disk %x [%d]: %.16s\n",
4816 __le32_to_cpu(mpb
->family_num
), i
,
4822 if (ok_count
== mpb
->num_disks
)
4827 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4829 struct intel_super
*s
;
4831 for (s
= super_list
; s
; s
= s
->next
) {
4832 if (family_num
!= s
->anchor
->family_num
)
4834 pr_err("Conflict, offlining family %#x on '%s'\n",
4835 __le32_to_cpu(family_num
), s
->disks
->devname
);
4839 static struct intel_super
*
4840 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4842 struct intel_super
*super_table
[len
];
4843 struct intel_disk
*disk_list
= NULL
;
4844 struct intel_super
*champion
, *spare
;
4845 struct intel_super
*s
, **del
;
4850 memset(super_table
, 0, sizeof(super_table
));
4851 for (s
= *super_list
; s
; s
= s
->next
)
4852 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4854 for (i
= 0; i
< tbl_size
; i
++) {
4855 struct imsm_disk
*d
;
4856 struct intel_disk
*idisk
;
4857 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4860 d
= &s
->disks
->disk
;
4862 /* 'd' must appear in merged disk list for its
4863 * configuration to be valid
4865 idisk
= disk_list_get(d
->serial
, disk_list
);
4866 if (idisk
&& idisk
->owner
== i
)
4867 s
= validate_members(s
, disk_list
, i
);
4872 dprintf("marking family: %#x from %d:%d offline\n",
4874 super_table
[i
]->disks
->major
,
4875 super_table
[i
]->disks
->minor
);
4879 /* This is where the mdadm implementation differs from the Windows
4880 * driver which has no strict concept of a container. We can only
4881 * assemble one family from a container, so when returning a prodigal
4882 * array member to this system the code will not be able to disambiguate
4883 * the container contents that should be assembled ("foreign" versus
4884 * "local"). It requires user intervention to set the orig_family_num
4885 * to a new value to establish a new container. The Windows driver in
4886 * this situation fixes up the volume name in place and manages the
4887 * foreign array as an independent entity.
4892 for (i
= 0; i
< tbl_size
; i
++) {
4893 struct intel_super
*tbl_ent
= super_table
[i
];
4899 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4904 if (s
&& !is_spare
) {
4905 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4907 } else if (!s
&& !is_spare
)
4920 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4921 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4923 /* collect all dl's onto 'champion', and update them to
4924 * champion's version of the status
4926 for (s
= *super_list
; s
; s
= s
->next
) {
4927 struct imsm_super
*mpb
= champion
->anchor
;
4928 struct dl
*dl
= s
->disks
;
4933 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4935 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4936 struct imsm_disk
*disk
;
4938 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4941 /* only set index on disks that are a member of
4942 * a populated contianer, i.e. one with
4945 if (is_failed(&dl
->disk
))
4947 else if (is_spare(&dl
->disk
))
4953 if (i
>= mpb
->num_disks
) {
4954 struct intel_disk
*idisk
;
4956 idisk
= disk_list_get(dl
->serial
, disk_list
);
4957 if (idisk
&& is_spare(&idisk
->disk
) &&
4958 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4966 dl
->next
= champion
->disks
;
4967 champion
->disks
= dl
;
4971 /* delete 'champion' from super_list */
4972 for (del
= super_list
; *del
; ) {
4973 if (*del
== champion
) {
4974 *del
= (*del
)->next
;
4977 del
= &(*del
)->next
;
4979 champion
->next
= NULL
;
4983 struct intel_disk
*idisk
= disk_list
;
4985 disk_list
= disk_list
->next
;
4993 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4994 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4995 int major
, int minor
, int keep_fd
);
4997 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4998 int *max
, int keep_fd
);
5000 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
5001 char *devname
, struct md_list
*devlist
,
5004 struct intel_super
*super_list
= NULL
;
5005 struct intel_super
*super
= NULL
;
5010 /* 'fd' is an opened container */
5011 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
5013 /* get super block from devlist devices */
5014 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
5017 /* all mpbs enter, maybe one leaves */
5018 super
= imsm_thunderdome(&super_list
, i
);
5024 if (find_missing(super
) != 0) {
5030 /* load migration record */
5031 err
= load_imsm_migr_rec(super
, NULL
);
5033 /* migration is in progress,
5034 * but migr_rec cannot be loaded,
5040 /* Check migration compatibility */
5041 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5042 pr_err("Unsupported migration detected");
5044 fprintf(stderr
, " on %s\n", devname
);
5046 fprintf(stderr
, " (IMSM).\n");
5055 while (super_list
) {
5056 struct intel_super
*s
= super_list
;
5058 super_list
= super_list
->next
;
5067 strcpy(st
->container_devnm
, fd2devnm(fd
));
5069 st
->container_devnm
[0] = 0;
5070 if (err
== 0 && st
->ss
== NULL
) {
5071 st
->ss
= &super_imsm
;
5072 st
->minor_version
= 0;
5073 st
->max_devs
= IMSM_MAX_DEVICES
;
5079 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5080 int *max
, int keep_fd
)
5082 struct md_list
*tmpdev
;
5086 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5087 if (tmpdev
->used
!= 1)
5089 if (tmpdev
->container
== 1) {
5091 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5093 pr_err("cannot open device %s: %s\n",
5094 tmpdev
->devname
, strerror(errno
));
5098 err
= get_sra_super_block(fd
, super_list
,
5099 tmpdev
->devname
, &lmax
,
5108 int major
= major(tmpdev
->st_rdev
);
5109 int minor
= minor(tmpdev
->st_rdev
);
5110 err
= get_super_block(super_list
,
5127 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5128 int major
, int minor
, int keep_fd
)
5130 struct intel_super
*s
;
5142 sprintf(nm
, "%d:%d", major
, minor
);
5143 dfd
= dev_open(nm
, O_RDWR
);
5149 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5150 find_intel_hba_capability(dfd
, s
, devname
);
5151 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5153 /* retry the load if we might have raced against mdmon */
5154 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5155 for (retry
= 0; retry
< 3; retry
++) {
5157 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5163 s
->next
= *super_list
;
5171 if (dfd
>= 0 && !keep_fd
)
5178 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5185 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5189 if (sra
->array
.major_version
!= -1 ||
5190 sra
->array
.minor_version
!= -2 ||
5191 strcmp(sra
->text_version
, "imsm") != 0) {
5196 devnm
= fd2devnm(fd
);
5197 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5198 if (get_super_block(super_list
, devnm
, devname
,
5199 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5210 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5212 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5215 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5217 struct intel_super
*super
;
5221 if (test_partition(fd
))
5222 /* IMSM not allowed on partitions */
5225 free_super_imsm(st
);
5227 super
= alloc_super();
5228 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5231 /* Load hba and capabilities if they exist.
5232 * But do not preclude loading metadata in case capabilities or hba are
5233 * non-compliant and ignore_hw_compat is set.
5235 rv
= find_intel_hba_capability(fd
, super
, devname
);
5236 /* no orom/efi or non-intel hba of the disk */
5237 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5239 pr_err("No OROM/EFI properties for %s\n", devname
);
5243 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5245 /* retry the load if we might have raced against mdmon */
5247 struct mdstat_ent
*mdstat
= NULL
;
5248 char *name
= fd2kname(fd
);
5251 mdstat
= mdstat_by_component(name
);
5253 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5254 for (retry
= 0; retry
< 3; retry
++) {
5256 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5262 free_mdstat(mdstat
);
5267 pr_err("Failed to load all information sections on %s\n", devname
);
5273 if (st
->ss
== NULL
) {
5274 st
->ss
= &super_imsm
;
5275 st
->minor_version
= 0;
5276 st
->max_devs
= IMSM_MAX_DEVICES
;
5279 /* load migration record */
5280 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5281 /* Check for unsupported migration features */
5282 if (check_mpb_migr_compatibility(super
) != 0) {
5283 pr_err("Unsupported migration detected");
5285 fprintf(stderr
, " on %s\n", devname
);
5287 fprintf(stderr
, " (IMSM).\n");
5295 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5297 if (info
->level
== 1)
5299 return info
->chunk_size
>> 9;
5302 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5303 unsigned long long size
)
5305 if (info
->level
== 1)
5308 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5311 static void imsm_update_version_info(struct intel_super
*super
)
5313 /* update the version and attributes */
5314 struct imsm_super
*mpb
= super
->anchor
;
5316 struct imsm_dev
*dev
;
5317 struct imsm_map
*map
;
5320 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5321 dev
= get_imsm_dev(super
, i
);
5322 map
= get_imsm_map(dev
, MAP_0
);
5323 if (__le32_to_cpu(dev
->size_high
) > 0)
5324 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5326 /* FIXME detect when an array spans a port multiplier */
5328 mpb
->attributes
|= MPB_ATTRIB_PM
;
5331 if (mpb
->num_raid_devs
> 1 ||
5332 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5333 version
= MPB_VERSION_ATTRIBS
;
5334 switch (get_imsm_raid_level(map
)) {
5335 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5336 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5337 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5338 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5341 if (map
->num_members
>= 5)
5342 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5343 else if (dev
->status
== DEV_CLONE_N_GO
)
5344 version
= MPB_VERSION_CNG
;
5345 else if (get_imsm_raid_level(map
) == 5)
5346 version
= MPB_VERSION_RAID5
;
5347 else if (map
->num_members
>= 3)
5348 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5349 else if (get_imsm_raid_level(map
) == 1)
5350 version
= MPB_VERSION_RAID1
;
5352 version
= MPB_VERSION_RAID0
;
5354 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5358 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5360 struct imsm_super
*mpb
= super
->anchor
;
5361 char *reason
= NULL
;
5363 size_t len
= strlen(name
);
5367 while (isspace(start
[len
- 1]))
5369 while (*start
&& isspace(*start
))
5371 memmove(name
, start
, len
+ 1);
5374 if (len
> MAX_RAID_SERIAL_LEN
)
5375 reason
= "must be 16 characters or less";
5377 reason
= "must be a non-empty string";
5379 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5380 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5382 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5383 reason
= "already exists";
5388 if (reason
&& !quiet
)
5389 pr_err("imsm volume name %s\n", reason
);
5394 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5395 struct shape
*s
, char *name
,
5396 char *homehost
, int *uuid
,
5397 long long data_offset
)
5399 /* We are creating a volume inside a pre-existing container.
5400 * so st->sb is already set.
5402 struct intel_super
*super
= st
->sb
;
5403 unsigned int sector_size
= super
->sector_size
;
5404 struct imsm_super
*mpb
= super
->anchor
;
5405 struct intel_dev
*dv
;
5406 struct imsm_dev
*dev
;
5407 struct imsm_vol
*vol
;
5408 struct imsm_map
*map
;
5409 int idx
= mpb
->num_raid_devs
;
5412 unsigned long long array_blocks
;
5413 size_t size_old
, size_new
;
5414 unsigned long long num_data_stripes
;
5415 unsigned int data_disks
;
5416 unsigned long long size_per_member
;
5418 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5419 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5423 /* ensure the mpb is large enough for the new data */
5424 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5425 size_new
= disks_to_mpb_size(info
->nr_disks
);
5426 if (size_new
> size_old
) {
5428 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5430 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5431 pr_err("could not allocate new mpb\n");
5434 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5435 MIGR_REC_BUF_SECTORS
*
5436 MAX_SECTOR_SIZE
) != 0) {
5437 pr_err("could not allocate migr_rec buffer\n");
5443 memcpy(mpb_new
, mpb
, size_old
);
5446 super
->anchor
= mpb_new
;
5447 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5448 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5449 super
->len
= size_round
;
5451 super
->current_vol
= idx
;
5453 /* handle 'failed_disks' by either:
5454 * a) create dummy disk entries in the table if this the first
5455 * volume in the array. We add them here as this is the only
5456 * opportunity to add them. add_to_super_imsm_volume()
5457 * handles the non-failed disks and continues incrementing
5459 * b) validate that 'failed_disks' matches the current number
5460 * of missing disks if the container is populated
5462 if (super
->current_vol
== 0) {
5464 for (i
= 0; i
< info
->failed_disks
; i
++) {
5465 struct imsm_disk
*disk
;
5468 disk
= __get_imsm_disk(mpb
, i
);
5469 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5470 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5471 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5472 "missing:%d", (__u8
)i
);
5474 find_missing(super
);
5479 for (d
= super
->missing
; d
; d
= d
->next
)
5481 if (info
->failed_disks
> missing
) {
5482 pr_err("unable to add 'missing' disk to container\n");
5487 if (!check_name(super
, name
, 0))
5489 dv
= xmalloc(sizeof(*dv
));
5490 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5492 * Explicitly allow truncating to not confuse gcc's
5493 * -Werror=stringop-truncation
5495 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5496 memcpy(dev
->volume
, name
, namelen
);
5497 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5498 info
->layout
, info
->chunk_size
,
5499 s
->size
* BLOCKS_PER_KB
);
5500 data_disks
= get_data_disks(info
->level
, info
->layout
,
5502 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5503 size_per_member
= array_blocks
/ data_disks
;
5505 set_imsm_dev_size(dev
, array_blocks
);
5506 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5508 vol
->migr_state
= 0;
5509 set_migr_type(dev
, MIGR_INIT
);
5510 vol
->dirty
= !info
->state
;
5511 vol
->curr_migr_unit
= 0;
5512 map
= get_imsm_map(dev
, MAP_0
);
5513 set_pba_of_lba0(map
, super
->create_offset
);
5514 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5515 map
->failed_disk_num
= ~0;
5516 if (info
->level
> 0)
5517 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5518 : IMSM_T_STATE_UNINITIALIZED
);
5520 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5521 IMSM_T_STATE_NORMAL
;
5524 if (info
->level
== 1 && info
->raid_disks
> 2) {
5527 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5531 map
->raid_level
= info
->level
;
5532 if (info
->level
== 10) {
5533 map
->raid_level
= 1;
5534 map
->num_domains
= info
->raid_disks
/ 2;
5535 } else if (info
->level
== 1)
5536 map
->num_domains
= info
->raid_disks
;
5538 map
->num_domains
= 1;
5540 /* info->size is only int so use the 'size' parameter instead */
5541 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5542 num_data_stripes
/= map
->num_domains
;
5543 set_num_data_stripes(map
, num_data_stripes
);
5545 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5546 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5550 map
->num_members
= info
->raid_disks
;
5551 for (i
= 0; i
< map
->num_members
; i
++) {
5552 /* initialized in add_to_super */
5553 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5555 mpb
->num_raid_devs
++;
5556 mpb
->num_raid_devs_created
++;
5557 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5559 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5560 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5561 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5562 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5566 pr_err("imsm does not support consistency policy %s\n",
5567 map_num(consistency_policies
, s
->consistency_policy
));
5572 dv
->index
= super
->current_vol
;
5573 dv
->next
= super
->devlist
;
5574 super
->devlist
= dv
;
5576 imsm_update_version_info(super
);
5581 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5582 struct shape
*s
, char *name
,
5583 char *homehost
, int *uuid
,
5584 unsigned long long data_offset
)
5586 /* This is primarily called by Create when creating a new array.
5587 * We will then get add_to_super called for each component, and then
5588 * write_init_super called to write it out to each device.
5589 * For IMSM, Create can create on fresh devices or on a pre-existing
5591 * To create on a pre-existing array a different method will be called.
5592 * This one is just for fresh drives.
5594 struct intel_super
*super
;
5595 struct imsm_super
*mpb
;
5599 if (data_offset
!= INVALID_SECTORS
) {
5600 pr_err("data-offset not supported by imsm\n");
5605 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5609 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5611 mpb_size
= MAX_SECTOR_SIZE
;
5613 super
= alloc_super();
5615 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5620 pr_err("could not allocate superblock\n");
5623 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5624 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5625 pr_err("could not allocate migr_rec buffer\n");
5630 memset(super
->buf
, 0, mpb_size
);
5632 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5636 /* zeroing superblock */
5640 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5642 version
= (char *) mpb
->sig
;
5643 strcpy(version
, MPB_SIGNATURE
);
5644 version
+= strlen(MPB_SIGNATURE
);
5645 strcpy(version
, MPB_VERSION_RAID0
);
5650 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5652 unsigned int member_sector_size
;
5655 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5659 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5661 if (member_sector_size
!= super
->sector_size
)
5666 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5667 int fd
, char *devname
)
5669 struct intel_super
*super
= st
->sb
;
5670 struct imsm_super
*mpb
= super
->anchor
;
5671 struct imsm_disk
*_disk
;
5672 struct imsm_dev
*dev
;
5673 struct imsm_map
*map
;
5677 dev
= get_imsm_dev(super
, super
->current_vol
);
5678 map
= get_imsm_map(dev
, MAP_0
);
5680 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5681 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5687 /* we're doing autolayout so grab the pre-marked (in
5688 * validate_geometry) raid_disk
5690 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5691 if (dl
->raiddisk
== dk
->raid_disk
)
5694 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5695 if (dl
->major
== dk
->major
&&
5696 dl
->minor
== dk
->minor
)
5701 pr_err("%s is not a member of the same container\n", devname
);
5705 if (mpb
->num_disks
== 0)
5706 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5707 &super
->sector_size
))
5710 if (!drive_validate_sector_size(super
, dl
)) {
5711 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5715 /* add a pristine spare to the metadata */
5716 if (dl
->index
< 0) {
5717 dl
->index
= super
->anchor
->num_disks
;
5718 super
->anchor
->num_disks
++;
5720 /* Check the device has not already been added */
5721 slot
= get_imsm_disk_slot(map
, dl
->index
);
5723 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5724 pr_err("%s has been included in this array twice\n",
5728 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5729 dl
->disk
.status
= CONFIGURED_DISK
;
5731 /* update size of 'missing' disks to be at least as large as the
5732 * largest acitve member (we only have dummy missing disks when
5733 * creating the first volume)
5735 if (super
->current_vol
== 0) {
5736 for (df
= super
->missing
; df
; df
= df
->next
) {
5737 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5738 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5739 _disk
= __get_imsm_disk(mpb
, df
->index
);
5744 /* refresh unset/failed slots to point to valid 'missing' entries */
5745 for (df
= super
->missing
; df
; df
= df
->next
)
5746 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5747 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5749 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5751 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5752 if (is_gen_migration(dev
)) {
5753 struct imsm_map
*map2
= get_imsm_map(dev
,
5755 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5756 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5757 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5760 if ((unsigned)df
->index
==
5762 set_imsm_ord_tbl_ent(map2
,
5768 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5772 /* if we are creating the first raid device update the family number */
5773 if (super
->current_vol
== 0) {
5775 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5777 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5778 if (!_dev
|| !_disk
) {
5779 pr_err("BUG mpb setup error\n");
5785 sum
+= __gen_imsm_checksum(mpb
);
5786 mpb
->family_num
= __cpu_to_le32(sum
);
5787 mpb
->orig_family_num
= mpb
->family_num
;
5789 super
->current_disk
= dl
;
5794 * Function marks disk as spare and restores disk serial
5795 * in case it was previously marked as failed by takeover operation
5797 * -1 : critical error
5798 * 0 : disk is marked as spare but serial is not set
5801 int mark_spare(struct dl
*disk
)
5803 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5810 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5811 /* Restore disk serial number, because takeover marks disk
5812 * as failed and adds to serial ':0' before it becomes
5815 serialcpy(disk
->serial
, serial
);
5816 serialcpy(disk
->disk
.serial
, serial
);
5819 disk
->disk
.status
= SPARE_DISK
;
5825 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5826 int fd
, char *devname
,
5827 unsigned long long data_offset
)
5829 struct intel_super
*super
= st
->sb
;
5831 unsigned long long size
;
5832 unsigned int member_sector_size
;
5837 /* If we are on an RAID enabled platform check that the disk is
5838 * attached to the raid controller.
5839 * We do not need to test disks attachment for container based additions,
5840 * they shall be already tested when container was created/assembled.
5842 rv
= find_intel_hba_capability(fd
, super
, devname
);
5843 /* no orom/efi or non-intel hba of the disk */
5845 dprintf("capability: %p fd: %d ret: %d\n",
5846 super
->orom
, fd
, rv
);
5850 if (super
->current_vol
>= 0)
5851 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5854 dd
= xcalloc(sizeof(*dd
), 1);
5855 dd
->major
= major(stb
.st_rdev
);
5856 dd
->minor
= minor(stb
.st_rdev
);
5857 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5860 dd
->action
= DISK_ADD
;
5861 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5863 pr_err("failed to retrieve scsi serial, aborting\n");
5869 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5870 (super
->hba
->type
== SYS_DEV_VMD
))) {
5872 char *devpath
= diskfd_to_devpath(fd
);
5873 char controller_path
[PATH_MAX
];
5876 pr_err("failed to get devpath, aborting\n");
5883 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5886 if (devpath_to_vendor(controller_path
) == 0x8086) {
5888 * If Intel's NVMe drive has serial ended with
5889 * "-A","-B","-1" or "-2" it means that this is "x8"
5890 * device (double drive on single PCIe card).
5891 * User should be warned about potential data loss.
5893 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5894 /* Skip empty character at the end */
5895 if (dd
->serial
[i
] == 0)
5898 if (((dd
->serial
[i
] == 'A') ||
5899 (dd
->serial
[i
] == 'B') ||
5900 (dd
->serial
[i
] == '1') ||
5901 (dd
->serial
[i
] == '2')) &&
5902 (dd
->serial
[i
-1] == '-'))
5903 pr_err("\tThe action you are about to take may put your data at risk.\n"
5904 "\tPlease note that x8 devices may consist of two separate x4 devices "
5905 "located on a single PCIe port.\n"
5906 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5909 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5910 !imsm_orom_has_tpv_support(super
->orom
)) {
5911 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5912 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
5919 get_dev_size(fd
, NULL
, &size
);
5920 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5922 if (super
->sector_size
== 0) {
5923 /* this a first device, so sector_size is not set yet */
5924 super
->sector_size
= member_sector_size
;
5927 /* clear migr_rec when adding disk to container */
5928 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5929 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5931 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5932 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5933 MIGR_REC_BUF_SECTORS
*member_sector_size
)
5934 perror("Write migr_rec failed");
5938 serialcpy(dd
->disk
.serial
, dd
->serial
);
5939 set_total_blocks(&dd
->disk
, size
);
5940 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5941 struct imsm_super
*mpb
= super
->anchor
;
5942 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5945 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5946 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5948 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5950 if (st
->update_tail
) {
5951 dd
->next
= super
->disk_mgmt_list
;
5952 super
->disk_mgmt_list
= dd
;
5954 dd
->next
= super
->disks
;
5956 super
->updates_pending
++;
5962 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5964 struct intel_super
*super
= st
->sb
;
5967 /* remove from super works only in mdmon - for communication
5968 * manager - monitor. Check if communication memory buffer
5971 if (!st
->update_tail
) {
5972 pr_err("shall be used in mdmon context only\n");
5975 dd
= xcalloc(1, sizeof(*dd
));
5976 dd
->major
= dk
->major
;
5977 dd
->minor
= dk
->minor
;
5980 dd
->action
= DISK_REMOVE
;
5982 dd
->next
= super
->disk_mgmt_list
;
5983 super
->disk_mgmt_list
= dd
;
5988 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5991 char buf
[MAX_SECTOR_SIZE
];
5992 struct imsm_super anchor
;
5993 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5995 /* spare records have their own family number and do not have any defined raid
5998 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
6000 struct imsm_super
*mpb
= super
->anchor
;
6001 struct imsm_super
*spare
= &spare_record
.anchor
;
6005 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
6006 spare
->generation_num
= __cpu_to_le32(1UL);
6007 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
6008 spare
->num_disks
= 1;
6009 spare
->num_raid_devs
= 0;
6010 spare
->cache_size
= mpb
->cache_size
;
6011 spare
->pwr_cycle_count
= __cpu_to_le32(1);
6013 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
6014 MPB_SIGNATURE MPB_VERSION_RAID0
);
6016 for (d
= super
->disks
; d
; d
= d
->next
) {
6020 spare
->disk
[0] = d
->disk
;
6021 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
6022 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6024 if (super
->sector_size
== 4096)
6025 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6027 sum
= __gen_imsm_checksum(spare
);
6028 spare
->family_num
= __cpu_to_le32(sum
);
6029 spare
->orig_family_num
= 0;
6030 sum
= __gen_imsm_checksum(spare
);
6031 spare
->check_sum
= __cpu_to_le32(sum
);
6033 if (store_imsm_mpb(d
->fd
, spare
)) {
6034 pr_err("failed for device %d:%d %s\n",
6035 d
->major
, d
->minor
, strerror(errno
));
6047 static int write_super_imsm(struct supertype
*st
, int doclose
)
6049 struct intel_super
*super
= st
->sb
;
6050 unsigned int sector_size
= super
->sector_size
;
6051 struct imsm_super
*mpb
= super
->anchor
;
6057 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6059 int clear_migration_record
= 1;
6062 /* 'generation' is incremented everytime the metadata is written */
6063 generation
= __le32_to_cpu(mpb
->generation_num
);
6065 mpb
->generation_num
= __cpu_to_le32(generation
);
6067 /* fix up cases where previous mdadm releases failed to set
6070 if (mpb
->orig_family_num
== 0)
6071 mpb
->orig_family_num
= mpb
->family_num
;
6073 for (d
= super
->disks
; d
; d
= d
->next
) {
6077 mpb
->disk
[d
->index
] = d
->disk
;
6081 for (d
= super
->missing
; d
; d
= d
->next
) {
6082 mpb
->disk
[d
->index
] = d
->disk
;
6085 mpb
->num_disks
= num_disks
;
6086 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6088 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6089 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6090 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6092 imsm_copy_dev(dev
, dev2
);
6093 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6095 if (is_gen_migration(dev2
))
6096 clear_migration_record
= 0;
6099 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6102 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6103 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6105 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6107 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6108 mpb_size
+= bbm_log_size
;
6109 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6112 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6115 /* recalculate checksum */
6116 sum
= __gen_imsm_checksum(mpb
);
6117 mpb
->check_sum
= __cpu_to_le32(sum
);
6119 if (super
->clean_migration_record_by_mdmon
) {
6120 clear_migration_record
= 1;
6121 super
->clean_migration_record_by_mdmon
= 0;
6123 if (clear_migration_record
)
6124 memset(super
->migr_rec_buf
, 0,
6125 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6127 if (sector_size
== 4096)
6128 convert_to_4k(super
);
6130 /* write the mpb for disks that compose raid devices */
6131 for (d
= super
->disks
; d
; d
= d
->next
) {
6132 if (d
->index
< 0 || is_failed(&d
->disk
))
6135 if (clear_migration_record
) {
6136 unsigned long long dsize
;
6138 get_dev_size(d
->fd
, NULL
, &dsize
);
6139 if (lseek64(d
->fd
, dsize
- sector_size
,
6141 if ((unsigned int)write(d
->fd
,
6142 super
->migr_rec_buf
,
6143 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6144 MIGR_REC_BUF_SECTORS
*sector_size
)
6145 perror("Write migr_rec failed");
6149 if (store_imsm_mpb(d
->fd
, mpb
))
6151 "failed for device %d:%d (fd: %d)%s\n",
6153 d
->fd
, strerror(errno
));
6162 return write_super_imsm_spares(super
, doclose
);
6167 static int create_array(struct supertype
*st
, int dev_idx
)
6170 struct imsm_update_create_array
*u
;
6171 struct intel_super
*super
= st
->sb
;
6172 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6173 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6174 struct disk_info
*inf
;
6175 struct imsm_disk
*disk
;
6178 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6179 sizeof(*inf
) * map
->num_members
;
6181 u
->type
= update_create_array
;
6182 u
->dev_idx
= dev_idx
;
6183 imsm_copy_dev(&u
->dev
, dev
);
6184 inf
= get_disk_info(u
);
6185 for (i
= 0; i
< map
->num_members
; i
++) {
6186 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6188 disk
= get_imsm_disk(super
, idx
);
6190 disk
= get_imsm_missing(super
, idx
);
6191 serialcpy(inf
[i
].serial
, disk
->serial
);
6193 append_metadata_update(st
, u
, len
);
6198 static int mgmt_disk(struct supertype
*st
)
6200 struct intel_super
*super
= st
->sb
;
6202 struct imsm_update_add_remove_disk
*u
;
6204 if (!super
->disk_mgmt_list
)
6209 u
->type
= update_add_remove_disk
;
6210 append_metadata_update(st
, u
, len
);
6215 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6217 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6219 struct ppl_header
*ppl_hdr
= buf
;
6222 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6224 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6226 perror("Failed to seek to PPL header location");
6230 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6232 perror("Write PPL header failed");
6241 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6243 struct intel_super
*super
= st
->sb
;
6245 struct ppl_header
*ppl_hdr
;
6248 /* first clear entire ppl space */
6249 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6253 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6255 pr_err("Failed to allocate PPL header buffer\n");
6259 memset(buf
, 0, PPL_HEADER_SIZE
);
6261 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6262 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6264 if (info
->mismatch_cnt
) {
6266 * We are overwriting an invalid ppl. Make one entry with wrong
6267 * checksum to prevent the kernel from skipping resync.
6269 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6270 ppl_hdr
->entries
[0].checksum
= ~0;
6273 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6279 static int is_rebuilding(struct imsm_dev
*dev
);
6281 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6282 struct mdinfo
*disk
)
6284 struct intel_super
*super
= st
->sb
;
6286 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6288 struct ppl_header
*ppl_hdr
= NULL
;
6290 struct imsm_dev
*dev
;
6293 unsigned long long ppl_offset
= 0;
6294 unsigned long long prev_gen_num
= 0;
6296 if (disk
->disk
.raid_disk
< 0)
6299 dev
= get_imsm_dev(super
, info
->container_member
);
6300 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6301 d
= get_imsm_dl_disk(super
, idx
);
6303 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6306 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6307 pr_err("Failed to allocate PPL header buffer\n");
6313 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6316 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6318 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6320 perror("Failed to seek to PPL header location");
6325 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6326 perror("Read PPL header failed");
6333 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6334 ppl_hdr
->checksum
= 0;
6336 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6337 dprintf("Wrong PPL header checksum on %s\n",
6342 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6343 /* previous was newest, it was already checked */
6347 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6348 super
->anchor
->orig_family_num
)) {
6349 dprintf("Wrong PPL header signature on %s\n",
6356 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6358 ppl_offset
+= PPL_HEADER_SIZE
;
6359 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6361 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6364 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6376 * Update metadata to use mutliple PPLs area (1MB).
6377 * This is done once for all RAID members
6379 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6380 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6382 struct mdinfo
*member_dev
;
6384 sprintf(subarray
, "%d", info
->container_member
);
6386 if (mdmon_running(st
->container_devnm
))
6387 st
->update_tail
= &st
->updates
;
6389 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6390 pr_err("Failed to update subarray %s\n",
6393 if (st
->update_tail
)
6394 flush_metadata_updates(st
);
6396 st
->ss
->sync_metadata(st
);
6397 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6398 for (member_dev
= info
->devs
; member_dev
;
6399 member_dev
= member_dev
->next
)
6400 member_dev
->ppl_size
=
6401 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6406 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6408 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6409 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6410 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6411 (is_rebuilding(dev
) &&
6412 dev
->vol
.curr_migr_unit
== 0 &&
6413 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6414 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6416 info
->mismatch_cnt
++;
6417 } else if (ret
== 0 &&
6418 ppl_hdr
->entries_count
== 0 &&
6419 is_rebuilding(dev
) &&
6420 info
->resync_start
== 0) {
6422 * The header has no entries - add a single empty entry and
6423 * rewrite the header to prevent the kernel from going into
6424 * resync after an interrupted rebuild.
6426 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6427 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6435 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6437 struct intel_super
*super
= st
->sb
;
6441 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6442 info
->array
.level
!= 5)
6445 for (d
= super
->disks
; d
; d
= d
->next
) {
6446 if (d
->index
< 0 || is_failed(&d
->disk
))
6449 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6457 static int write_init_super_imsm(struct supertype
*st
)
6459 struct intel_super
*super
= st
->sb
;
6460 int current_vol
= super
->current_vol
;
6464 getinfo_super_imsm(st
, &info
, NULL
);
6466 /* we are done with current_vol reset it to point st at the container */
6467 super
->current_vol
= -1;
6469 if (st
->update_tail
) {
6470 /* queue the recently created array / added disk
6471 * as a metadata update */
6473 /* determine if we are creating a volume or adding a disk */
6474 if (current_vol
< 0) {
6475 /* in the mgmt (add/remove) disk case we are running
6476 * in mdmon context, so don't close fd's
6480 rv
= write_init_ppl_imsm_all(st
, &info
);
6482 rv
= create_array(st
, current_vol
);
6486 for (d
= super
->disks
; d
; d
= d
->next
)
6487 Kill(d
->devname
, NULL
, 0, -1, 1);
6488 if (current_vol
>= 0)
6489 rv
= write_init_ppl_imsm_all(st
, &info
);
6491 rv
= write_super_imsm(st
, 1);
6497 static int store_super_imsm(struct supertype
*st
, int fd
)
6499 struct intel_super
*super
= st
->sb
;
6500 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6505 if (super
->sector_size
== 4096)
6506 convert_to_4k(super
);
6507 return store_imsm_mpb(fd
, mpb
);
6510 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6511 int layout
, int raiddisks
, int chunk
,
6512 unsigned long long size
,
6513 unsigned long long data_offset
,
6515 unsigned long long *freesize
,
6519 unsigned long long ldsize
;
6520 struct intel_super
*super
;
6523 if (level
!= LEVEL_CONTAINER
)
6528 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6531 pr_err("imsm: Cannot open %s: %s\n",
6532 dev
, strerror(errno
));
6535 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6540 /* capabilities retrieve could be possible
6541 * note that there is no fd for the disks in array.
6543 super
= alloc_super();
6548 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6554 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6558 fd2devname(fd
, str
);
6559 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6560 fd
, str
, super
->orom
, rv
, raiddisks
);
6562 /* no orom/efi or non-intel hba of the disk */
6569 if (raiddisks
> super
->orom
->tds
) {
6571 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6572 raiddisks
, super
->orom
->tds
);
6576 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6577 (ldsize
>> 9) >> 32 > 0) {
6579 pr_err("%s exceeds maximum platform supported size\n", dev
);
6585 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6591 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6593 const unsigned long long base_start
= e
[*idx
].start
;
6594 unsigned long long end
= base_start
+ e
[*idx
].size
;
6597 if (base_start
== end
)
6601 for (i
= *idx
; i
< num_extents
; i
++) {
6602 /* extend overlapping extents */
6603 if (e
[i
].start
>= base_start
&&
6604 e
[i
].start
<= end
) {
6607 if (e
[i
].start
+ e
[i
].size
> end
)
6608 end
= e
[i
].start
+ e
[i
].size
;
6609 } else if (e
[i
].start
> end
) {
6615 return end
- base_start
;
6618 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6620 /* build a composite disk with all known extents and generate a new
6621 * 'maxsize' given the "all disks in an array must share a common start
6622 * offset" constraint
6624 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6628 unsigned long long pos
;
6629 unsigned long long start
= 0;
6630 unsigned long long maxsize
;
6631 unsigned long reserve
;
6633 /* coalesce and sort all extents. also, check to see if we need to
6634 * reserve space between member arrays
6637 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6640 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6643 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6648 while (i
< sum_extents
) {
6649 e
[j
].start
= e
[i
].start
;
6650 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6652 if (e
[j
-1].size
== 0)
6661 unsigned long long esize
;
6663 esize
= e
[i
].start
- pos
;
6664 if (esize
>= maxsize
) {
6669 pos
= e
[i
].start
+ e
[i
].size
;
6671 } while (e
[i
-1].size
);
6677 /* FIXME assumes volume at offset 0 is the first volume in a
6680 if (start_extent
> 0)
6681 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6685 if (maxsize
< reserve
)
6688 super
->create_offset
= ~((unsigned long long) 0);
6689 if (start
+ reserve
> super
->create_offset
)
6690 return 0; /* start overflows create_offset */
6691 super
->create_offset
= start
+ reserve
;
6693 return maxsize
- reserve
;
6696 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6698 if (level
< 0 || level
== 6 || level
== 4)
6701 /* if we have an orom prevent invalid raid levels */
6704 case 0: return imsm_orom_has_raid0(orom
);
6707 return imsm_orom_has_raid1e(orom
);
6708 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6709 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6710 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6713 return 1; /* not on an Intel RAID platform so anything goes */
6719 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6720 int dpa
, int verbose
)
6722 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6723 struct mdstat_ent
*memb
;
6729 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6730 if (memb
->metadata_version
&&
6731 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6732 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6733 !is_subarray(memb
->metadata_version
+9) &&
6735 struct dev_member
*dev
= memb
->members
;
6737 while(dev
&& (fd
< 0)) {
6738 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6739 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6741 fd
= open(path
, O_RDONLY
, 0);
6742 if (num
<= 0 || fd
< 0) {
6743 pr_vrb("Cannot open %s: %s\n",
6744 dev
->name
, strerror(errno
));
6750 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6751 struct mdstat_ent
*vol
;
6752 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6753 if (vol
->active
> 0 &&
6754 vol
->metadata_version
&&
6755 is_container_member(vol
, memb
->devnm
)) {
6760 if (*devlist
&& (found
< dpa
)) {
6761 dv
= xcalloc(1, sizeof(*dv
));
6762 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6763 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6766 dv
->next
= *devlist
;
6774 free_mdstat(mdstat
);
6779 static struct md_list
*
6780 get_loop_devices(void)
6783 struct md_list
*devlist
= NULL
;
6786 for(i
= 0; i
< 12; i
++) {
6787 dv
= xcalloc(1, sizeof(*dv
));
6788 dv
->devname
= xmalloc(40);
6789 sprintf(dv
->devname
, "/dev/loop%d", i
);
6797 static struct md_list
*
6798 get_devices(const char *hba_path
)
6800 struct md_list
*devlist
= NULL
;
6807 devlist
= get_loop_devices();
6810 /* scroll through /sys/dev/block looking for devices attached to
6813 dir
= opendir("/sys/dev/block");
6814 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6819 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6821 path
= devt_to_devpath(makedev(major
, minor
));
6824 if (!path_attached_to_hba(path
, hba_path
)) {
6831 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6833 fd2devname(fd
, buf
);
6836 pr_err("cannot open device: %s\n",
6841 dv
= xcalloc(1, sizeof(*dv
));
6842 dv
->devname
= xstrdup(buf
);
6849 devlist
= devlist
->next
;
6859 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6860 int verbose
, int *found
)
6862 struct md_list
*tmpdev
;
6864 struct supertype
*st
;
6866 /* first walk the list of devices to find a consistent set
6867 * that match the criterea, if that is possible.
6868 * We flag the ones we like with 'used'.
6871 st
= match_metadata_desc_imsm("imsm");
6873 pr_vrb("cannot allocate memory for imsm supertype\n");
6877 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6878 char *devname
= tmpdev
->devname
;
6880 struct supertype
*tst
;
6882 if (tmpdev
->used
> 1)
6884 tst
= dup_super(st
);
6886 pr_vrb("cannot allocate memory for imsm supertype\n");
6889 tmpdev
->container
= 0;
6890 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6892 dprintf("cannot open device %s: %s\n",
6893 devname
, strerror(errno
));
6895 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
6897 } else if (must_be_container(dfd
)) {
6898 struct supertype
*cst
;
6899 cst
= super_by_fd(dfd
, NULL
);
6901 dprintf("cannot recognize container type %s\n",
6904 } else if (tst
->ss
!= st
->ss
) {
6905 dprintf("non-imsm container - ignore it: %s\n",
6908 } else if (!tst
->ss
->load_container
||
6909 tst
->ss
->load_container(tst
, dfd
, NULL
))
6912 tmpdev
->container
= 1;
6915 cst
->ss
->free_super(cst
);
6917 tmpdev
->st_rdev
= rdev
;
6918 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6919 dprintf("no RAID superblock on %s\n",
6922 } else if (tst
->ss
->compare_super
== NULL
) {
6923 dprintf("Cannot assemble %s metadata on %s\n",
6924 tst
->ss
->name
, devname
);
6930 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6931 /* Ignore unrecognised devices during auto-assembly */
6936 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6938 if (st
->minor_version
== -1)
6939 st
->minor_version
= tst
->minor_version
;
6941 if (memcmp(info
.uuid
, uuid_zero
,
6942 sizeof(int[4])) == 0) {
6943 /* this is a floating spare. It cannot define
6944 * an array unless there are no more arrays of
6945 * this type to be found. It can be included
6946 * in an array of this type though.
6952 if (st
->ss
!= tst
->ss
||
6953 st
->minor_version
!= tst
->minor_version
||
6954 st
->ss
->compare_super(st
, tst
) != 0) {
6955 /* Some mismatch. If exactly one array matches this host,
6956 * we can resolve on that one.
6957 * Or, if we are auto assembling, we just ignore the second
6960 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6966 dprintf("found: devname: %s\n", devname
);
6970 tst
->ss
->free_super(tst
);
6974 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6975 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6976 for (iter
= head
; iter
; iter
= iter
->next
) {
6977 dprintf("content->text_version: %s vol\n",
6978 iter
->text_version
);
6979 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6980 /* do not assemble arrays with unsupported
6982 dprintf("Cannot activate member %s.\n",
6983 iter
->text_version
);
6990 dprintf("No valid super block on device list: err: %d %p\n",
6994 dprintf("no more devices to examine\n");
6997 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6998 if (tmpdev
->used
== 1 && tmpdev
->found
) {
7000 if (count
< tmpdev
->found
)
7003 count
-= tmpdev
->found
;
7006 if (tmpdev
->used
== 1)
7011 st
->ss
->free_super(st
);
7015 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
7018 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
7020 const struct orom_entry
*entry
;
7021 struct devid_list
*dv
, *devid_list
;
7026 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7027 if (strstr(idev
->path
, hba_path
))
7031 if (!idev
|| !idev
->dev_id
)
7034 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7036 if (!entry
|| !entry
->devid_list
)
7039 devid_list
= entry
->devid_list
;
7040 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7041 struct md_list
*devlist
;
7042 struct sys_dev
*device
= NULL
;
7047 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7049 device
= device_by_id(dv
->devid
);
7052 hpath
= device
->path
;
7056 devlist
= get_devices(hpath
);
7057 /* if no intel devices return zero volumes */
7058 if (devlist
== NULL
)
7061 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7063 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7064 if (devlist
== NULL
)
7068 count
+= count_volumes_list(devlist
,
7072 dprintf("found %d count: %d\n", found
, count
);
7075 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7078 struct md_list
*dv
= devlist
;
7079 devlist
= devlist
->next
;
7087 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7091 if (hba
->type
== SYS_DEV_VMD
) {
7092 struct sys_dev
*dev
;
7095 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7096 if (dev
->type
== SYS_DEV_VMD
)
7097 count
+= __count_volumes(dev
->path
, dpa
,
7102 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7105 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7107 /* up to 512 if the plaform supports it, otherwise the platform max.
7108 * 128 if no platform detected
7110 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7112 return min(512, (1 << fs
));
7116 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7117 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7119 /* check/set platform and metadata limits/defaults */
7120 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7121 pr_vrb("platform supports a maximum of %d disks per array\n",
7126 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7127 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7128 pr_vrb("platform does not support raid%d with %d disk%s\n",
7129 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7133 if (*chunk
== 0 || *chunk
== UnSet
)
7134 *chunk
= imsm_default_chunk(super
->orom
);
7136 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7137 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7141 if (layout
!= imsm_level_to_layout(level
)) {
7143 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7144 else if (level
== 10)
7145 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7147 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7152 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7153 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7154 pr_vrb("platform does not support a volume size over 2TB\n");
7161 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7162 * FIX ME add ahci details
7164 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7165 int layout
, int raiddisks
, int *chunk
,
7166 unsigned long long size
,
7167 unsigned long long data_offset
,
7169 unsigned long long *freesize
,
7173 struct intel_super
*super
= st
->sb
;
7174 struct imsm_super
*mpb
;
7176 unsigned long long pos
= 0;
7177 unsigned long long maxsize
;
7181 /* We must have the container info already read in. */
7185 mpb
= super
->anchor
;
7187 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7188 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7192 /* General test: make sure there is space for
7193 * 'raiddisks' device extents of size 'size' at a given
7196 unsigned long long minsize
= size
;
7197 unsigned long long start_offset
= MaxSector
;
7200 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7201 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7206 e
= get_extents(super
, dl
);
7209 unsigned long long esize
;
7210 esize
= e
[i
].start
- pos
;
7211 if (esize
>= minsize
)
7213 if (found
&& start_offset
== MaxSector
) {
7216 } else if (found
&& pos
!= start_offset
) {
7220 pos
= e
[i
].start
+ e
[i
].size
;
7222 } while (e
[i
-1].size
);
7227 if (dcnt
< raiddisks
) {
7229 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7236 /* This device must be a member of the set */
7237 if (!stat_is_blkdev(dev
, &rdev
))
7239 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7240 if (dl
->major
== (int)major(rdev
) &&
7241 dl
->minor
== (int)minor(rdev
))
7246 pr_err("%s is not in the same imsm set\n", dev
);
7248 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7249 /* If a volume is present then the current creation attempt
7250 * cannot incorporate new spares because the orom may not
7251 * understand this configuration (all member disks must be
7252 * members of each array in the container).
7254 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7255 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7257 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7258 mpb
->num_disks
!= raiddisks
) {
7259 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7263 /* retrieve the largest free space block */
7264 e
= get_extents(super
, dl
);
7269 unsigned long long esize
;
7271 esize
= e
[i
].start
- pos
;
7272 if (esize
>= maxsize
)
7274 pos
= e
[i
].start
+ e
[i
].size
;
7276 } while (e
[i
-1].size
);
7281 pr_err("unable to determine free space for: %s\n",
7285 if (maxsize
< size
) {
7287 pr_err("%s not enough space (%llu < %llu)\n",
7288 dev
, maxsize
, size
);
7292 /* count total number of extents for merge */
7294 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7296 i
+= dl
->extent_cnt
;
7298 maxsize
= merge_extents(super
, i
);
7300 if (!check_env("IMSM_NO_PLATFORM") &&
7301 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7302 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7306 if (maxsize
< size
|| maxsize
== 0) {
7309 pr_err("no free space left on device. Aborting...\n");
7311 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7317 *freesize
= maxsize
;
7320 int count
= count_volumes(super
->hba
,
7321 super
->orom
->dpa
, verbose
);
7322 if (super
->orom
->vphba
<= count
) {
7323 pr_vrb("platform does not support more than %d raid volumes.\n",
7324 super
->orom
->vphba
);
7331 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7332 unsigned long long size
, int chunk
,
7333 unsigned long long *freesize
)
7335 struct intel_super
*super
= st
->sb
;
7336 struct imsm_super
*mpb
= super
->anchor
;
7341 unsigned long long maxsize
;
7342 unsigned long long minsize
;
7346 /* find the largest common start free region of the possible disks */
7350 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7356 /* don't activate new spares if we are orom constrained
7357 * and there is already a volume active in the container
7359 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7362 e
= get_extents(super
, dl
);
7365 for (i
= 1; e
[i
-1].size
; i
++)
7373 maxsize
= merge_extents(super
, extent_cnt
);
7377 minsize
= chunk
* 2;
7379 if (cnt
< raiddisks
||
7380 (super
->orom
&& used
&& used
!= raiddisks
) ||
7381 maxsize
< minsize
||
7383 pr_err("not enough devices with space to create array.\n");
7384 return 0; /* No enough free spaces large enough */
7395 if (!check_env("IMSM_NO_PLATFORM") &&
7396 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7397 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7401 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7403 dl
->raiddisk
= cnt
++;
7407 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7412 static int reserve_space(struct supertype
*st
, int raiddisks
,
7413 unsigned long long size
, int chunk
,
7414 unsigned long long *freesize
)
7416 struct intel_super
*super
= st
->sb
;
7421 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7424 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7426 dl
->raiddisk
= cnt
++;
7433 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7434 int raiddisks
, int *chunk
, unsigned long long size
,
7435 unsigned long long data_offset
,
7436 char *dev
, unsigned long long *freesize
,
7437 int consistency_policy
, int verbose
)
7444 * if given unused devices create a container
7445 * if given given devices in a container create a member volume
7447 if (level
== LEVEL_CONTAINER
) {
7448 /* Must be a fresh device to add to a container */
7449 return validate_geometry_imsm_container(st
, level
, layout
,
7457 if (size
&& ((size
< 1024) || (*chunk
!= UnSet
&&
7458 size
< (unsigned long long) *chunk
))) {
7459 pr_err("Given size must be greater than 1M and chunk size.\n");
7460 /* Depends on algorithm in Create.c :
7461 * if container was given (dev == NULL) return -1,
7462 * if block device was given ( dev != NULL) return 0.
7464 return dev
? -1 : 0;
7469 struct intel_super
*super
= st
->sb
;
7470 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7471 raiddisks
, chunk
, size
,
7474 /* we are being asked to automatically layout a
7475 * new volume based on the current contents of
7476 * the container. If the the parameters can be
7477 * satisfied reserve_space will record the disks,
7478 * start offset, and size of the volume to be
7479 * created. add_to_super and getinfo_super
7480 * detect when autolayout is in progress.
7482 /* assuming that freesize is always given when array is
7484 if (super
->orom
&& freesize
) {
7486 count
= count_volumes(super
->hba
,
7487 super
->orom
->dpa
, verbose
);
7488 if (super
->orom
->vphba
<= count
) {
7489 pr_vrb("platform does not support more than %d raid volumes.\n",
7490 super
->orom
->vphba
);
7495 return reserve_space(st
, raiddisks
, size
,
7501 /* creating in a given container */
7502 return validate_geometry_imsm_volume(st
, level
, layout
,
7503 raiddisks
, chunk
, size
,
7505 dev
, freesize
, verbose
);
7508 /* This device needs to be a device in an 'imsm' container */
7509 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7512 pr_err("Cannot create this array on device %s\n",
7517 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7519 pr_err("Cannot open %s: %s\n",
7520 dev
, strerror(errno
));
7523 /* Well, it is in use by someone, maybe an 'imsm' container. */
7524 cfd
= open_container(fd
);
7528 pr_err("Cannot use %s: It is busy\n",
7532 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7533 if (sra
&& sra
->array
.major_version
== -1 &&
7534 strcmp(sra
->text_version
, "imsm") == 0)
7538 /* This is a member of a imsm container. Load the container
7539 * and try to create a volume
7541 struct intel_super
*super
;
7543 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7545 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7547 return validate_geometry_imsm_volume(st
, level
, layout
,
7549 size
, data_offset
, dev
,
7556 pr_err("failed container membership check\n");
7562 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7564 struct intel_super
*super
= st
->sb
;
7566 if (level
&& *level
== UnSet
)
7567 *level
= LEVEL_CONTAINER
;
7569 if (level
&& layout
&& *layout
== UnSet
)
7570 *layout
= imsm_level_to_layout(*level
);
7572 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7573 *chunk
= imsm_default_chunk(super
->orom
);
7576 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7578 static int kill_subarray_imsm(struct supertype
*st
)
7580 /* remove the subarray currently referenced by ->current_vol */
7582 struct intel_dev
**dp
;
7583 struct intel_super
*super
= st
->sb
;
7584 __u8 current_vol
= super
->current_vol
;
7585 struct imsm_super
*mpb
= super
->anchor
;
7587 if (super
->current_vol
< 0)
7589 super
->current_vol
= -1; /* invalidate subarray cursor */
7591 /* block deletions that would change the uuid of active subarrays
7593 * FIXME when immutable ids are available, but note that we'll
7594 * also need to fixup the invalidated/active subarray indexes in
7597 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7600 if (i
< current_vol
)
7602 sprintf(subarray
, "%u", i
);
7603 if (is_subarray_active(subarray
, st
->devnm
)) {
7604 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7611 if (st
->update_tail
) {
7612 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7614 u
->type
= update_kill_array
;
7615 u
->dev_idx
= current_vol
;
7616 append_metadata_update(st
, u
, sizeof(*u
));
7621 for (dp
= &super
->devlist
; *dp
;)
7622 if ((*dp
)->index
== current_vol
) {
7625 handle_missing(super
, (*dp
)->dev
);
7626 if ((*dp
)->index
> current_vol
)
7631 /* no more raid devices, all active components are now spares,
7632 * but of course failed are still failed
7634 if (--mpb
->num_raid_devs
== 0) {
7637 for (d
= super
->disks
; d
; d
= d
->next
)
7642 super
->updates_pending
++;
7647 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7648 char *update
, struct mddev_ident
*ident
)
7650 /* update the subarray currently referenced by ->current_vol */
7651 struct intel_super
*super
= st
->sb
;
7652 struct imsm_super
*mpb
= super
->anchor
;
7654 if (strcmp(update
, "name") == 0) {
7655 char *name
= ident
->name
;
7659 if (is_subarray_active(subarray
, st
->devnm
)) {
7660 pr_err("Unable to update name of active subarray\n");
7664 if (!check_name(super
, name
, 0))
7667 vol
= strtoul(subarray
, &ep
, 10);
7668 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7671 if (st
->update_tail
) {
7672 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7674 u
->type
= update_rename_array
;
7676 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7677 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7678 append_metadata_update(st
, u
, sizeof(*u
));
7680 struct imsm_dev
*dev
;
7683 dev
= get_imsm_dev(super
, vol
);
7684 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7685 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7686 memcpy(dev
->volume
, name
, namelen
);
7687 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7688 dev
= get_imsm_dev(super
, i
);
7689 handle_missing(super
, dev
);
7691 super
->updates_pending
++;
7693 } else if (strcmp(update
, "ppl") == 0 ||
7694 strcmp(update
, "no-ppl") == 0) {
7697 int vol
= strtoul(subarray
, &ep
, 10);
7699 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7702 if (strcmp(update
, "ppl") == 0)
7703 new_policy
= RWH_MULTIPLE_DISTRIBUTED
;
7705 new_policy
= RWH_MULTIPLE_OFF
;
7707 if (st
->update_tail
) {
7708 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7710 u
->type
= update_rwh_policy
;
7712 u
->new_policy
= new_policy
;
7713 append_metadata_update(st
, u
, sizeof(*u
));
7715 struct imsm_dev
*dev
;
7717 dev
= get_imsm_dev(super
, vol
);
7718 dev
->rwh_policy
= new_policy
;
7719 super
->updates_pending
++;
7727 static int is_gen_migration(struct imsm_dev
*dev
)
7732 if (!dev
->vol
.migr_state
)
7735 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7741 static int is_rebuilding(struct imsm_dev
*dev
)
7743 struct imsm_map
*migr_map
;
7745 if (!dev
->vol
.migr_state
)
7748 if (migr_type(dev
) != MIGR_REBUILD
)
7751 migr_map
= get_imsm_map(dev
, MAP_1
);
7753 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7759 static int is_initializing(struct imsm_dev
*dev
)
7761 struct imsm_map
*migr_map
;
7763 if (!dev
->vol
.migr_state
)
7766 if (migr_type(dev
) != MIGR_INIT
)
7769 migr_map
= get_imsm_map(dev
, MAP_1
);
7771 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7777 static void update_recovery_start(struct intel_super
*super
,
7778 struct imsm_dev
*dev
,
7779 struct mdinfo
*array
)
7781 struct mdinfo
*rebuild
= NULL
;
7785 if (!is_rebuilding(dev
))
7788 /* Find the rebuild target, but punt on the dual rebuild case */
7789 for (d
= array
->devs
; d
; d
= d
->next
)
7790 if (d
->recovery_start
== 0) {
7797 /* (?) none of the disks are marked with
7798 * IMSM_ORD_REBUILD, so assume they are missing and the
7799 * disk_ord_tbl was not correctly updated
7801 dprintf("failed to locate out-of-sync disk\n");
7805 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7806 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7809 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7811 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7813 /* Given a container loaded by load_super_imsm_all,
7814 * extract information about all the arrays into
7816 * If 'subarray' is given, just extract info about that array.
7818 * For each imsm_dev create an mdinfo, fill it in,
7819 * then look for matching devices in super->disks
7820 * and create appropriate device mdinfo.
7822 struct intel_super
*super
= st
->sb
;
7823 struct imsm_super
*mpb
= super
->anchor
;
7824 struct mdinfo
*rest
= NULL
;
7828 int spare_disks
= 0;
7830 /* do not assemble arrays when not all attributes are supported */
7831 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7833 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7836 /* count spare devices, not used in maps
7838 for (d
= super
->disks
; d
; d
= d
->next
)
7842 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7843 struct imsm_dev
*dev
;
7844 struct imsm_map
*map
;
7845 struct imsm_map
*map2
;
7846 struct mdinfo
*this;
7853 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7856 dev
= get_imsm_dev(super
, i
);
7857 map
= get_imsm_map(dev
, MAP_0
);
7858 map2
= get_imsm_map(dev
, MAP_1
);
7859 level
= get_imsm_raid_level(map
);
7861 /* do not publish arrays that are in the middle of an
7862 * unsupported migration
7864 if (dev
->vol
.migr_state
&&
7865 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7866 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7870 /* do not publish arrays that are not support by controller's
7874 this = xmalloc(sizeof(*this));
7876 super
->current_vol
= i
;
7877 getinfo_super_imsm_volume(st
, this, NULL
);
7879 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7880 /* mdadm does not support all metadata features- set the bit in all arrays state */
7881 if (!validate_geometry_imsm_orom(super
,
7882 level
, /* RAID level */
7883 imsm_level_to_layout(level
),
7884 map
->num_members
, /* raid disks */
7885 &chunk
, imsm_dev_size(dev
),
7887 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7889 this->array
.state
|=
7890 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7891 (1<<MD_SB_BLOCK_VOLUME
);
7894 /* if array has bad blocks, set suitable bit in all arrays state */
7896 this->array
.state
|=
7897 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7898 (1<<MD_SB_BLOCK_VOLUME
);
7900 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7901 unsigned long long recovery_start
;
7902 struct mdinfo
*info_d
;
7910 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7911 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7912 for (d
= super
->disks
; d
; d
= d
->next
)
7913 if (d
->index
== idx
)
7916 recovery_start
= MaxSector
;
7919 if (d
&& is_failed(&d
->disk
))
7921 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
7925 * if we skip some disks the array will be assmebled degraded;
7926 * reset resync start to avoid a dirty-degraded
7927 * situation when performing the intial sync
7932 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
7933 if ((!able_to_resync(level
, missing
) ||
7934 recovery_start
== 0))
7935 this->resync_start
= MaxSector
;
7938 * FIXME handle dirty degraded
7945 info_d
= xcalloc(1, sizeof(*info_d
));
7946 info_d
->next
= this->devs
;
7947 this->devs
= info_d
;
7949 info_d
->disk
.number
= d
->index
;
7950 info_d
->disk
.major
= d
->major
;
7951 info_d
->disk
.minor
= d
->minor
;
7952 info_d
->disk
.raid_disk
= slot
;
7953 info_d
->recovery_start
= recovery_start
;
7955 if (slot
< map2
->num_members
)
7956 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7958 this->array
.spare_disks
++;
7960 if (slot
< map
->num_members
)
7961 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7963 this->array
.spare_disks
++;
7965 if (info_d
->recovery_start
== MaxSector
)
7966 this->array
.working_disks
++;
7968 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7969 info_d
->data_offset
= pba_of_lba0(map
);
7970 info_d
->component_size
= calc_component_size(map
, dev
);
7972 if (map
->raid_level
== 5) {
7973 info_d
->ppl_sector
= this->ppl_sector
;
7974 info_d
->ppl_size
= this->ppl_size
;
7975 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
7976 recovery_start
== 0)
7977 this->resync_start
= 0;
7980 info_d
->bb
.supported
= 1;
7981 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7982 info_d
->data_offset
,
7983 info_d
->component_size
,
7986 /* now that the disk list is up-to-date fixup recovery_start */
7987 update_recovery_start(super
, dev
, this);
7988 this->array
.spare_disks
+= spare_disks
;
7990 /* check for reshape */
7991 if (this->reshape_active
== 1)
7992 recover_backup_imsm(st
, this);
7999 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
8000 int failed
, int look_in_map
)
8002 struct imsm_map
*map
;
8004 map
= get_imsm_map(dev
, look_in_map
);
8007 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
8008 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
8010 switch (get_imsm_raid_level(map
)) {
8012 return IMSM_T_STATE_FAILED
;
8015 if (failed
< map
->num_members
)
8016 return IMSM_T_STATE_DEGRADED
;
8018 return IMSM_T_STATE_FAILED
;
8023 * check to see if any mirrors have failed, otherwise we
8024 * are degraded. Even numbered slots are mirrored on
8028 /* gcc -Os complains that this is unused */
8029 int insync
= insync
;
8031 for (i
= 0; i
< map
->num_members
; i
++) {
8032 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8033 int idx
= ord_to_idx(ord
);
8034 struct imsm_disk
*disk
;
8036 /* reset the potential in-sync count on even-numbered
8037 * slots. num_copies is always 2 for imsm raid10
8042 disk
= get_imsm_disk(super
, idx
);
8043 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8046 /* no in-sync disks left in this mirror the
8050 return IMSM_T_STATE_FAILED
;
8053 return IMSM_T_STATE_DEGRADED
;
8057 return IMSM_T_STATE_DEGRADED
;
8059 return IMSM_T_STATE_FAILED
;
8065 return map
->map_state
;
8068 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8073 struct imsm_disk
*disk
;
8074 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8075 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8076 struct imsm_map
*map_for_loop
;
8081 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8082 * disks that are being rebuilt. New failures are recorded to
8083 * map[0]. So we look through all the disks we started with and
8084 * see if any failures are still present, or if any new ones
8088 if (prev
&& (map
->num_members
< prev
->num_members
))
8089 map_for_loop
= prev
;
8091 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8093 /* when MAP_X is passed both maps failures are counted
8096 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8097 i
< prev
->num_members
) {
8098 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8099 idx_1
= ord_to_idx(ord
);
8101 disk
= get_imsm_disk(super
, idx_1
);
8102 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8105 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8106 i
< map
->num_members
) {
8107 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8108 idx
= ord_to_idx(ord
);
8111 disk
= get_imsm_disk(super
, idx
);
8112 if (!disk
|| is_failed(disk
) ||
8113 ord
& IMSM_ORD_REBUILD
)
8122 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8125 struct intel_super
*super
= c
->sb
;
8126 struct imsm_super
*mpb
= super
->anchor
;
8127 struct imsm_update_prealloc_bb_mem u
;
8129 if (atoi(inst
) >= mpb
->num_raid_devs
) {
8130 pr_err("subarry index %d, out of range\n", atoi(inst
));
8134 dprintf("imsm: open_new %s\n", inst
);
8135 a
->info
.container_member
= atoi(inst
);
8137 u
.type
= update_prealloc_badblocks_mem
;
8138 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8143 static int is_resyncing(struct imsm_dev
*dev
)
8145 struct imsm_map
*migr_map
;
8147 if (!dev
->vol
.migr_state
)
8150 if (migr_type(dev
) == MIGR_INIT
||
8151 migr_type(dev
) == MIGR_REPAIR
)
8154 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8157 migr_map
= get_imsm_map(dev
, MAP_1
);
8159 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8160 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8166 /* return true if we recorded new information */
8167 static int mark_failure(struct intel_super
*super
,
8168 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8172 struct imsm_map
*map
;
8173 char buf
[MAX_RAID_SERIAL_LEN
+3];
8174 unsigned int len
, shift
= 0;
8176 /* new failures are always set in map[0] */
8177 map
= get_imsm_map(dev
, MAP_0
);
8179 slot
= get_imsm_disk_slot(map
, idx
);
8183 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8184 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8187 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8188 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8190 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8191 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8192 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8194 disk
->status
|= FAILED_DISK
;
8195 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8196 /* mark failures in second map if second map exists and this disk
8198 * This is valid for migration, initialization and rebuild
8200 if (dev
->vol
.migr_state
) {
8201 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8202 int slot2
= get_imsm_disk_slot(map2
, idx
);
8204 if (slot2
< map2
->num_members
&& slot2
>= 0)
8205 set_imsm_ord_tbl_ent(map2
, slot2
,
8206 idx
| IMSM_ORD_REBUILD
);
8208 if (map
->failed_disk_num
== 0xff ||
8209 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8210 map
->failed_disk_num
= slot
;
8212 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8217 static void mark_missing(struct intel_super
*super
,
8218 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8220 mark_failure(super
, dev
, disk
, idx
);
8222 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8225 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8226 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8229 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8233 if (!super
->missing
)
8236 /* When orom adds replacement for missing disk it does
8237 * not remove entry of missing disk, but just updates map with
8238 * new added disk. So it is not enough just to test if there is
8239 * any missing disk, we have to look if there are any failed disks
8240 * in map to stop migration */
8242 dprintf("imsm: mark missing\n");
8243 /* end process for initialization and rebuild only
8245 if (is_gen_migration(dev
) == 0) {
8246 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8250 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8251 struct imsm_map
*map1
;
8252 int i
, ord
, ord_map1
;
8255 for (i
= 0; i
< map
->num_members
; i
++) {
8256 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8257 if (!(ord
& IMSM_ORD_REBUILD
))
8260 map1
= get_imsm_map(dev
, MAP_1
);
8264 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8265 if (ord_map1
& IMSM_ORD_REBUILD
)
8270 map_state
= imsm_check_degraded(super
, dev
,
8272 end_migration(dev
, super
, map_state
);
8276 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8277 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8278 super
->updates_pending
++;
8281 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8284 unsigned long long array_blocks
;
8285 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8286 int used_disks
= imsm_num_data_members(map
);
8288 if (used_disks
== 0) {
8289 /* when problems occures
8290 * return current array_blocks value
8292 array_blocks
= imsm_dev_size(dev
);
8294 return array_blocks
;
8297 /* set array size in metadata
8300 /* OLCE size change is caused by added disks
8302 array_blocks
= per_dev_array_size(map
) * used_disks
;
8304 /* Online Volume Size Change
8305 * Using available free space
8307 array_blocks
= new_size
;
8309 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8310 set_imsm_dev_size(dev
, array_blocks
);
8312 return array_blocks
;
8315 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8317 static void imsm_progress_container_reshape(struct intel_super
*super
)
8319 /* if no device has a migr_state, but some device has a
8320 * different number of members than the previous device, start
8321 * changing the number of devices in this device to match
8324 struct imsm_super
*mpb
= super
->anchor
;
8325 int prev_disks
= -1;
8329 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8330 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8331 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8332 struct imsm_map
*map2
;
8333 int prev_num_members
;
8335 if (dev
->vol
.migr_state
)
8338 if (prev_disks
== -1)
8339 prev_disks
= map
->num_members
;
8340 if (prev_disks
== map
->num_members
)
8343 /* OK, this array needs to enter reshape mode.
8344 * i.e it needs a migr_state
8347 copy_map_size
= sizeof_imsm_map(map
);
8348 prev_num_members
= map
->num_members
;
8349 map
->num_members
= prev_disks
;
8350 dev
->vol
.migr_state
= 1;
8351 dev
->vol
.curr_migr_unit
= 0;
8352 set_migr_type(dev
, MIGR_GEN_MIGR
);
8353 for (i
= prev_num_members
;
8354 i
< map
->num_members
; i
++)
8355 set_imsm_ord_tbl_ent(map
, i
, i
);
8356 map2
= get_imsm_map(dev
, MAP_1
);
8357 /* Copy the current map */
8358 memcpy(map2
, map
, copy_map_size
);
8359 map2
->num_members
= prev_num_members
;
8361 imsm_set_array_size(dev
, -1);
8362 super
->clean_migration_record_by_mdmon
= 1;
8363 super
->updates_pending
++;
8367 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8368 * states are handled in imsm_set_disk() with one exception, when a
8369 * resync is stopped due to a new failure this routine will set the
8370 * 'degraded' state for the array.
8372 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8374 int inst
= a
->info
.container_member
;
8375 struct intel_super
*super
= a
->container
->sb
;
8376 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8377 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8378 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8379 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8380 __u32 blocks_per_unit
;
8382 if (dev
->vol
.migr_state
&&
8383 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8384 /* array state change is blocked due to reshape action
8386 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8387 * - finish the reshape (if last_checkpoint is big and action != reshape)
8388 * - update curr_migr_unit
8390 if (a
->curr_action
== reshape
) {
8391 /* still reshaping, maybe update curr_migr_unit */
8392 goto mark_checkpoint
;
8394 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8395 /* for some reason we aborted the reshape.
8397 * disable automatic metadata rollback
8398 * user action is required to recover process
8401 struct imsm_map
*map2
=
8402 get_imsm_map(dev
, MAP_1
);
8403 dev
->vol
.migr_state
= 0;
8404 set_migr_type(dev
, 0);
8405 dev
->vol
.curr_migr_unit
= 0;
8407 sizeof_imsm_map(map2
));
8408 super
->updates_pending
++;
8411 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8412 unsigned long long array_blocks
;
8416 used_disks
= imsm_num_data_members(map
);
8417 if (used_disks
> 0) {
8419 per_dev_array_size(map
) *
8422 round_size_to_mb(array_blocks
,
8424 a
->info
.custom_array_size
= array_blocks
;
8425 /* encourage manager to update array
8429 a
->check_reshape
= 1;
8431 /* finalize online capacity expansion/reshape */
8432 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8434 mdi
->disk
.raid_disk
,
8437 imsm_progress_container_reshape(super
);
8442 /* before we activate this array handle any missing disks */
8443 if (consistent
== 2)
8444 handle_missing(super
, dev
);
8446 if (consistent
== 2 &&
8447 (!is_resync_complete(&a
->info
) ||
8448 map_state
!= IMSM_T_STATE_NORMAL
||
8449 dev
->vol
.migr_state
))
8452 if (is_resync_complete(&a
->info
)) {
8453 /* complete intialization / resync,
8454 * recovery and interrupted recovery is completed in
8457 if (is_resyncing(dev
)) {
8458 dprintf("imsm: mark resync done\n");
8459 end_migration(dev
, super
, map_state
);
8460 super
->updates_pending
++;
8461 a
->last_checkpoint
= 0;
8463 } else if ((!is_resyncing(dev
) && !failed
) &&
8464 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8465 /* mark the start of the init process if nothing is failed */
8466 dprintf("imsm: mark resync start\n");
8467 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8468 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8470 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8471 super
->updates_pending
++;
8475 /* skip checkpointing for general migration,
8476 * it is controlled in mdadm
8478 if (is_gen_migration(dev
))
8479 goto skip_mark_checkpoint
;
8481 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8482 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8483 if (blocks_per_unit
) {
8487 units
= a
->last_checkpoint
/ blocks_per_unit
;
8490 /* check that we did not overflow 32-bits, and that
8491 * curr_migr_unit needs updating
8493 if (units32
== units
&&
8495 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8496 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8497 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8498 super
->updates_pending
++;
8502 skip_mark_checkpoint
:
8503 /* mark dirty / clean */
8504 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8505 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8506 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8508 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8510 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8511 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8512 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8513 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8515 super
->updates_pending
++;
8521 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8523 int inst
= a
->info
.container_member
;
8524 struct intel_super
*super
= a
->container
->sb
;
8525 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8526 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8528 if (slot
> map
->num_members
) {
8529 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8530 slot
, map
->num_members
- 1);
8537 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8540 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8542 int inst
= a
->info
.container_member
;
8543 struct intel_super
*super
= a
->container
->sb
;
8544 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8545 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8546 struct imsm_disk
*disk
;
8548 int recovery_not_finished
= 0;
8552 int rebuild_done
= 0;
8555 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8559 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8560 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8562 /* check for new failures */
8563 if (state
& DS_FAULTY
) {
8564 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8565 super
->updates_pending
++;
8568 /* check if in_sync */
8569 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8570 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8572 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8574 super
->updates_pending
++;
8577 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8578 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8580 /* check if recovery complete, newly degraded, or failed */
8581 dprintf("imsm: Detected transition to state ");
8582 switch (map_state
) {
8583 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8584 dprintf("normal: ");
8585 if (is_rebuilding(dev
)) {
8586 dprintf_cont("while rebuilding");
8587 /* check if recovery is really finished */
8588 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8589 if (mdi
->recovery_start
!= MaxSector
) {
8590 recovery_not_finished
= 1;
8593 if (recovery_not_finished
) {
8595 dprintf("Rebuild has not finished yet, state not changed");
8596 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8597 a
->last_checkpoint
= mdi
->recovery_start
;
8598 super
->updates_pending
++;
8602 end_migration(dev
, super
, map_state
);
8603 map
= get_imsm_map(dev
, MAP_0
);
8604 map
->failed_disk_num
= ~0;
8605 super
->updates_pending
++;
8606 a
->last_checkpoint
= 0;
8609 if (is_gen_migration(dev
)) {
8610 dprintf_cont("while general migration");
8611 if (a
->last_checkpoint
>= a
->info
.component_size
)
8612 end_migration(dev
, super
, map_state
);
8614 map
->map_state
= map_state
;
8615 map
= get_imsm_map(dev
, MAP_0
);
8616 map
->failed_disk_num
= ~0;
8617 super
->updates_pending
++;
8621 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8622 dprintf_cont("degraded: ");
8623 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8624 dprintf_cont("mark degraded");
8625 map
->map_state
= map_state
;
8626 super
->updates_pending
++;
8627 a
->last_checkpoint
= 0;
8630 if (is_rebuilding(dev
)) {
8631 dprintf_cont("while rebuilding ");
8632 if (state
& DS_FAULTY
) {
8633 dprintf_cont("removing failed drive ");
8634 if (n
== map
->failed_disk_num
) {
8635 dprintf_cont("end migration");
8636 end_migration(dev
, super
, map_state
);
8637 a
->last_checkpoint
= 0;
8639 dprintf_cont("fail detected during rebuild, changing map state");
8640 map
->map_state
= map_state
;
8642 super
->updates_pending
++;
8648 /* check if recovery is really finished */
8649 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8650 if (mdi
->recovery_start
!= MaxSector
) {
8651 recovery_not_finished
= 1;
8654 if (recovery_not_finished
) {
8656 dprintf_cont("Rebuild has not finished yet");
8657 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8658 a
->last_checkpoint
=
8659 mdi
->recovery_start
;
8660 super
->updates_pending
++;
8665 dprintf_cont(" Rebuild done, still degraded");
8666 end_migration(dev
, super
, map_state
);
8667 a
->last_checkpoint
= 0;
8668 super
->updates_pending
++;
8670 for (i
= 0; i
< map
->num_members
; i
++) {
8671 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8673 if (idx
& IMSM_ORD_REBUILD
)
8674 map
->failed_disk_num
= i
;
8676 super
->updates_pending
++;
8679 if (is_gen_migration(dev
)) {
8680 dprintf_cont("while general migration");
8681 if (a
->last_checkpoint
>= a
->info
.component_size
)
8682 end_migration(dev
, super
, map_state
);
8684 map
->map_state
= map_state
;
8685 manage_second_map(super
, dev
);
8687 super
->updates_pending
++;
8690 if (is_initializing(dev
)) {
8691 dprintf_cont("while initialization.");
8692 map
->map_state
= map_state
;
8693 super
->updates_pending
++;
8697 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8698 dprintf_cont("failed: ");
8699 if (is_gen_migration(dev
)) {
8700 dprintf_cont("while general migration");
8701 map
->map_state
= map_state
;
8702 super
->updates_pending
++;
8705 if (map
->map_state
!= map_state
) {
8706 dprintf_cont("mark failed");
8707 end_migration(dev
, super
, map_state
);
8708 super
->updates_pending
++;
8709 a
->last_checkpoint
= 0;
8714 dprintf_cont("state %i\n", map_state
);
8719 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8722 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8723 unsigned long long dsize
;
8724 unsigned long long sectors
;
8725 unsigned int sector_size
;
8727 get_dev_sector_size(fd
, NULL
, §or_size
);
8728 get_dev_size(fd
, NULL
, &dsize
);
8730 if (mpb_size
> sector_size
) {
8731 /* -1 to account for anchor */
8732 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8734 /* write the extended mpb to the sectors preceeding the anchor */
8735 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8739 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8740 sector_size
* sectors
) != sector_size
* sectors
)
8744 /* first block is stored on second to last sector of the disk */
8745 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8748 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8754 static void imsm_sync_metadata(struct supertype
*container
)
8756 struct intel_super
*super
= container
->sb
;
8758 dprintf("sync metadata: %d\n", super
->updates_pending
);
8759 if (!super
->updates_pending
)
8762 write_super_imsm(container
, 0);
8764 super
->updates_pending
= 0;
8767 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8769 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8770 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8773 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8777 if (dl
&& is_failed(&dl
->disk
))
8781 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8786 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8787 struct active_array
*a
, int activate_new
,
8788 struct mdinfo
*additional_test_list
)
8790 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8791 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8792 struct imsm_super
*mpb
= super
->anchor
;
8793 struct imsm_map
*map
;
8794 unsigned long long pos
;
8799 __u32 array_start
= 0;
8800 __u32 array_end
= 0;
8802 struct mdinfo
*test_list
;
8804 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8805 /* If in this array, skip */
8806 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8807 if (d
->state_fd
>= 0 &&
8808 d
->disk
.major
== dl
->major
&&
8809 d
->disk
.minor
== dl
->minor
) {
8810 dprintf("%x:%x already in array\n",
8811 dl
->major
, dl
->minor
);
8816 test_list
= additional_test_list
;
8818 if (test_list
->disk
.major
== dl
->major
&&
8819 test_list
->disk
.minor
== dl
->minor
) {
8820 dprintf("%x:%x already in additional test list\n",
8821 dl
->major
, dl
->minor
);
8824 test_list
= test_list
->next
;
8829 /* skip in use or failed drives */
8830 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8832 dprintf("%x:%x status (failed: %d index: %d)\n",
8833 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8837 /* skip pure spares when we are looking for partially
8838 * assimilated drives
8840 if (dl
->index
== -1 && !activate_new
)
8843 if (!drive_validate_sector_size(super
, dl
))
8846 /* Does this unused device have the requisite free space?
8847 * It needs to be able to cover all member volumes
8849 ex
= get_extents(super
, dl
);
8851 dprintf("cannot get extents\n");
8854 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8855 dev
= get_imsm_dev(super
, i
);
8856 map
= get_imsm_map(dev
, MAP_0
);
8858 /* check if this disk is already a member of
8861 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8867 array_start
= pba_of_lba0(map
);
8868 array_end
= array_start
+
8869 per_dev_array_size(map
) - 1;
8872 /* check that we can start at pba_of_lba0 with
8873 * num_data_stripes*blocks_per_stripe of space
8875 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8879 pos
= ex
[j
].start
+ ex
[j
].size
;
8881 } while (ex
[j
-1].size
);
8888 if (i
< mpb
->num_raid_devs
) {
8889 dprintf("%x:%x does not have %u to %u available\n",
8890 dl
->major
, dl
->minor
, array_start
, array_end
);
8900 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8902 struct imsm_dev
*dev2
;
8903 struct imsm_map
*map
;
8909 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8911 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8912 if (state
== IMSM_T_STATE_FAILED
) {
8913 map
= get_imsm_map(dev2
, MAP_0
);
8916 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8918 * Check if failed disks are deleted from intel
8919 * disk list or are marked to be deleted
8921 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8922 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8924 * Do not rebuild the array if failed disks
8925 * from failed sub-array are not removed from
8929 is_failed(&idisk
->disk
) &&
8930 (idisk
->action
!= DISK_REMOVE
))
8938 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8939 struct metadata_update
**updates
)
8942 * Find a device with unused free space and use it to replace a
8943 * failed/vacant region in an array. We replace failed regions one a
8944 * array at a time. The result is that a new spare disk will be added
8945 * to the first failed array and after the monitor has finished
8946 * propagating failures the remainder will be consumed.
8948 * FIXME add a capability for mdmon to request spares from another
8952 struct intel_super
*super
= a
->container
->sb
;
8953 int inst
= a
->info
.container_member
;
8954 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8955 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8956 int failed
= a
->info
.array
.raid_disks
;
8957 struct mdinfo
*rv
= NULL
;
8960 struct metadata_update
*mu
;
8962 struct imsm_update_activate_spare
*u
;
8967 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8968 if ((d
->curr_state
& DS_FAULTY
) &&
8970 /* wait for Removal to happen */
8972 if (d
->state_fd
>= 0)
8976 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8977 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8979 if (imsm_reshape_blocks_arrays_changes(super
))
8982 /* Cannot activate another spare if rebuild is in progress already
8984 if (is_rebuilding(dev
)) {
8985 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8989 if (a
->info
.array
.level
== 4)
8990 /* No repair for takeovered array
8991 * imsm doesn't support raid4
8995 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8996 IMSM_T_STATE_DEGRADED
)
8999 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
9000 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
9005 * If there are any failed disks check state of the other volume.
9006 * Block rebuild if the another one is failed until failed disks
9007 * are removed from container.
9010 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
9011 MAX_RAID_SERIAL_LEN
, dev
->volume
);
9012 /* check if states of the other volumes allow for rebuild */
9013 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
9015 allowed
= imsm_rebuild_allowed(a
->container
,
9023 /* For each slot, if it is not working, find a spare */
9024 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
9025 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9026 if (d
->disk
.raid_disk
== i
)
9028 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
9029 if (d
&& (d
->state_fd
>= 0))
9033 * OK, this device needs recovery. Try to re-add the
9034 * previous occupant of this slot, if this fails see if
9035 * we can continue the assimilation of a spare that was
9036 * partially assimilated, finally try to activate a new
9039 dl
= imsm_readd(super
, i
, a
);
9041 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9043 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9047 /* found a usable disk with enough space */
9048 di
= xcalloc(1, sizeof(*di
));
9050 /* dl->index will be -1 in the case we are activating a
9051 * pristine spare. imsm_process_update() will create a
9052 * new index in this case. Once a disk is found to be
9053 * failed in all member arrays it is kicked from the
9056 di
->disk
.number
= dl
->index
;
9058 /* (ab)use di->devs to store a pointer to the device
9061 di
->devs
= (struct mdinfo
*) dl
;
9063 di
->disk
.raid_disk
= i
;
9064 di
->disk
.major
= dl
->major
;
9065 di
->disk
.minor
= dl
->minor
;
9067 di
->recovery_start
= 0;
9068 di
->data_offset
= pba_of_lba0(map
);
9069 di
->component_size
= a
->info
.component_size
;
9070 di
->container_member
= inst
;
9071 di
->bb
.supported
= 1;
9072 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9073 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9074 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9076 super
->random
= random32();
9080 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9081 i
, di
->data_offset
);
9085 /* No spares found */
9087 /* Now 'rv' has a list of devices to return.
9088 * Create a metadata_update record to update the
9089 * disk_ord_tbl for the array
9091 mu
= xmalloc(sizeof(*mu
));
9092 mu
->buf
= xcalloc(num_spares
,
9093 sizeof(struct imsm_update_activate_spare
));
9095 mu
->space_list
= NULL
;
9096 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9097 mu
->next
= *updates
;
9098 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9100 for (di
= rv
; di
; di
= di
->next
) {
9101 u
->type
= update_activate_spare
;
9102 u
->dl
= (struct dl
*) di
->devs
;
9104 u
->slot
= di
->disk
.raid_disk
;
9115 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9117 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9118 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9119 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9120 struct disk_info
*inf
= get_disk_info(u
);
9121 struct imsm_disk
*disk
;
9125 for (i
= 0; i
< map
->num_members
; i
++) {
9126 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9127 for (j
= 0; j
< new_map
->num_members
; j
++)
9128 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9135 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9139 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9140 if (dl
->major
== major
&& dl
->minor
== minor
)
9145 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9151 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9152 if (dl
->major
== major
&& dl
->minor
== minor
) {
9155 prev
->next
= dl
->next
;
9157 super
->disks
= dl
->next
;
9159 __free_imsm_disk(dl
);
9160 dprintf("removed %x:%x\n", major
, minor
);
9168 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9170 static int add_remove_disk_update(struct intel_super
*super
)
9172 int check_degraded
= 0;
9175 /* add/remove some spares to/from the metadata/contrainer */
9176 while (super
->disk_mgmt_list
) {
9177 struct dl
*disk_cfg
;
9179 disk_cfg
= super
->disk_mgmt_list
;
9180 super
->disk_mgmt_list
= disk_cfg
->next
;
9181 disk_cfg
->next
= NULL
;
9183 if (disk_cfg
->action
== DISK_ADD
) {
9184 disk_cfg
->next
= super
->disks
;
9185 super
->disks
= disk_cfg
;
9187 dprintf("added %x:%x\n",
9188 disk_cfg
->major
, disk_cfg
->minor
);
9189 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9190 dprintf("Disk remove action processed: %x.%x\n",
9191 disk_cfg
->major
, disk_cfg
->minor
);
9192 disk
= get_disk_super(super
,
9196 /* store action status */
9197 disk
->action
= DISK_REMOVE
;
9198 /* remove spare disks only */
9199 if (disk
->index
== -1) {
9200 remove_disk_super(super
,
9205 /* release allocate disk structure */
9206 __free_imsm_disk(disk_cfg
);
9209 return check_degraded
;
9212 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9213 struct intel_super
*super
,
9216 struct intel_dev
*id
;
9217 void **tofree
= NULL
;
9220 dprintf("(enter)\n");
9221 if (u
->subdev
< 0 || u
->subdev
> 1) {
9222 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9225 if (space_list
== NULL
|| *space_list
== NULL
) {
9226 dprintf("imsm: Error: Memory is not allocated\n");
9230 for (id
= super
->devlist
; id
; id
= id
->next
) {
9231 if (id
->index
== (unsigned)u
->subdev
) {
9232 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9233 struct imsm_map
*map
;
9234 struct imsm_dev
*new_dev
=
9235 (struct imsm_dev
*)*space_list
;
9236 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9238 struct dl
*new_disk
;
9240 if (new_dev
== NULL
)
9242 *space_list
= **space_list
;
9243 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9244 map
= get_imsm_map(new_dev
, MAP_0
);
9246 dprintf("imsm: Error: migration in progress");
9250 to_state
= map
->map_state
;
9251 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9253 /* this should not happen */
9254 if (u
->new_disks
[0] < 0) {
9255 map
->failed_disk_num
=
9256 map
->num_members
- 1;
9257 to_state
= IMSM_T_STATE_DEGRADED
;
9259 to_state
= IMSM_T_STATE_NORMAL
;
9261 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9262 if (u
->new_level
> -1)
9263 map
->raid_level
= u
->new_level
;
9264 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9265 if ((u
->new_level
== 5) &&
9266 (migr_map
->raid_level
== 0)) {
9267 int ord
= map
->num_members
- 1;
9268 migr_map
->num_members
--;
9269 if (u
->new_disks
[0] < 0)
9270 ord
|= IMSM_ORD_REBUILD
;
9271 set_imsm_ord_tbl_ent(map
,
9272 map
->num_members
- 1,
9276 tofree
= (void **)dev
;
9278 /* update chunk size
9280 if (u
->new_chunksize
> 0) {
9281 unsigned long long num_data_stripes
;
9282 struct imsm_map
*dest_map
=
9283 get_imsm_map(dev
, MAP_0
);
9285 imsm_num_data_members(dest_map
);
9287 if (used_disks
== 0)
9290 map
->blocks_per_strip
=
9291 __cpu_to_le16(u
->new_chunksize
* 2);
9293 imsm_dev_size(dev
) / used_disks
;
9294 num_data_stripes
/= map
->blocks_per_strip
;
9295 num_data_stripes
/= map
->num_domains
;
9296 set_num_data_stripes(map
, num_data_stripes
);
9299 /* ensure blocks_per_member has valid value
9301 set_blocks_per_member(map
,
9302 per_dev_array_size(map
) +
9303 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9307 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9308 migr_map
->raid_level
== map
->raid_level
)
9311 if (u
->new_disks
[0] >= 0) {
9314 new_disk
= get_disk_super(super
,
9315 major(u
->new_disks
[0]),
9316 minor(u
->new_disks
[0]));
9317 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9318 major(u
->new_disks
[0]),
9319 minor(u
->new_disks
[0]),
9320 new_disk
, new_disk
->index
);
9321 if (new_disk
== NULL
)
9322 goto error_disk_add
;
9324 new_disk
->index
= map
->num_members
- 1;
9325 /* slot to fill in autolayout
9327 new_disk
->raiddisk
= new_disk
->index
;
9328 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9329 new_disk
->disk
.status
&= ~SPARE_DISK
;
9331 goto error_disk_add
;
9334 *tofree
= *space_list
;
9335 /* calculate new size
9337 imsm_set_array_size(new_dev
, -1);
9344 *space_list
= tofree
;
9348 dprintf("Error: imsm: Cannot find disk.\n");
9352 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9353 struct intel_super
*super
)
9355 struct intel_dev
*id
;
9358 dprintf("(enter)\n");
9359 if (u
->subdev
< 0 || u
->subdev
> 1) {
9360 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9364 for (id
= super
->devlist
; id
; id
= id
->next
) {
9365 if (id
->index
== (unsigned)u
->subdev
) {
9366 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9367 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9368 int used_disks
= imsm_num_data_members(map
);
9369 unsigned long long blocks_per_member
;
9370 unsigned long long num_data_stripes
;
9371 unsigned long long new_size_per_disk
;
9373 if (used_disks
== 0)
9376 /* calculate new size
9378 new_size_per_disk
= u
->new_size
/ used_disks
;
9379 blocks_per_member
= new_size_per_disk
+
9380 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9381 num_data_stripes
= new_size_per_disk
/
9382 map
->blocks_per_strip
;
9383 num_data_stripes
/= map
->num_domains
;
9384 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9385 u
->new_size
, new_size_per_disk
,
9387 set_blocks_per_member(map
, blocks_per_member
);
9388 set_num_data_stripes(map
, num_data_stripes
);
9389 imsm_set_array_size(dev
, u
->new_size
);
9399 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9400 struct intel_super
*super
,
9401 struct active_array
*active_array
)
9403 struct imsm_super
*mpb
= super
->anchor
;
9404 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9405 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9406 struct imsm_map
*migr_map
;
9407 struct active_array
*a
;
9408 struct imsm_disk
*disk
;
9415 int second_map_created
= 0;
9417 for (; u
; u
= u
->next
) {
9418 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9423 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9428 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9433 /* count failures (excluding rebuilds and the victim)
9434 * to determine map[0] state
9437 for (i
= 0; i
< map
->num_members
; i
++) {
9440 disk
= get_imsm_disk(super
,
9441 get_imsm_disk_idx(dev
, i
, MAP_X
));
9442 if (!disk
|| is_failed(disk
))
9446 /* adding a pristine spare, assign a new index */
9447 if (dl
->index
< 0) {
9448 dl
->index
= super
->anchor
->num_disks
;
9449 super
->anchor
->num_disks
++;
9452 disk
->status
|= CONFIGURED_DISK
;
9453 disk
->status
&= ~SPARE_DISK
;
9456 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9457 if (!second_map_created
) {
9458 second_map_created
= 1;
9459 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9460 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9462 map
->map_state
= to_state
;
9463 migr_map
= get_imsm_map(dev
, MAP_1
);
9464 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9465 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9466 dl
->index
| IMSM_ORD_REBUILD
);
9468 /* update the family_num to mark a new container
9469 * generation, being careful to record the existing
9470 * family_num in orig_family_num to clean up after
9471 * earlier mdadm versions that neglected to set it.
9473 if (mpb
->orig_family_num
== 0)
9474 mpb
->orig_family_num
= mpb
->family_num
;
9475 mpb
->family_num
+= super
->random
;
9477 /* count arrays using the victim in the metadata */
9479 for (a
= active_array
; a
; a
= a
->next
) {
9480 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9481 map
= get_imsm_map(dev
, MAP_0
);
9483 if (get_imsm_disk_slot(map
, victim
) >= 0)
9487 /* delete the victim if it is no longer being
9493 /* We know that 'manager' isn't touching anything,
9494 * so it is safe to delete
9496 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9497 if ((*dlp
)->index
== victim
)
9500 /* victim may be on the missing list */
9502 for (dlp
= &super
->missing
; *dlp
;
9503 dlp
= &(*dlp
)->next
)
9504 if ((*dlp
)->index
== victim
)
9506 imsm_delete(super
, dlp
, victim
);
9513 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9514 struct intel_super
*super
,
9517 struct dl
*new_disk
;
9518 struct intel_dev
*id
;
9520 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9521 int disk_count
= u
->old_raid_disks
;
9522 void **tofree
= NULL
;
9523 int devices_to_reshape
= 1;
9524 struct imsm_super
*mpb
= super
->anchor
;
9526 unsigned int dev_id
;
9528 dprintf("(enter)\n");
9530 /* enable spares to use in array */
9531 for (i
= 0; i
< delta_disks
; i
++) {
9532 new_disk
= get_disk_super(super
,
9533 major(u
->new_disks
[i
]),
9534 minor(u
->new_disks
[i
]));
9535 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9536 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9537 new_disk
, new_disk
->index
);
9538 if (new_disk
== NULL
||
9539 (new_disk
->index
>= 0 &&
9540 new_disk
->index
< u
->old_raid_disks
))
9541 goto update_reshape_exit
;
9542 new_disk
->index
= disk_count
++;
9543 /* slot to fill in autolayout
9545 new_disk
->raiddisk
= new_disk
->index
;
9546 new_disk
->disk
.status
|=
9548 new_disk
->disk
.status
&= ~SPARE_DISK
;
9551 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9552 mpb
->num_raid_devs
);
9553 /* manage changes in volume
9555 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9556 void **sp
= *space_list
;
9557 struct imsm_dev
*newdev
;
9558 struct imsm_map
*newmap
, *oldmap
;
9560 for (id
= super
->devlist
; id
; id
= id
->next
) {
9561 if (id
->index
== dev_id
)
9570 /* Copy the dev, but not (all of) the map */
9571 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9572 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9573 newmap
= get_imsm_map(newdev
, MAP_0
);
9574 /* Copy the current map */
9575 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9576 /* update one device only
9578 if (devices_to_reshape
) {
9579 dprintf("imsm: modifying subdev: %i\n",
9581 devices_to_reshape
--;
9582 newdev
->vol
.migr_state
= 1;
9583 newdev
->vol
.curr_migr_unit
= 0;
9584 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9585 newmap
->num_members
= u
->new_raid_disks
;
9586 for (i
= 0; i
< delta_disks
; i
++) {
9587 set_imsm_ord_tbl_ent(newmap
,
9588 u
->old_raid_disks
+ i
,
9589 u
->old_raid_disks
+ i
);
9591 /* New map is correct, now need to save old map
9593 newmap
= get_imsm_map(newdev
, MAP_1
);
9594 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9596 imsm_set_array_size(newdev
, -1);
9599 sp
= (void **)id
->dev
;
9604 /* Clear migration record */
9605 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9608 *space_list
= tofree
;
9611 update_reshape_exit
:
9616 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9617 struct intel_super
*super
,
9620 struct imsm_dev
*dev
= NULL
;
9621 struct intel_dev
*dv
;
9622 struct imsm_dev
*dev_new
;
9623 struct imsm_map
*map
;
9627 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9628 if (dv
->index
== (unsigned int)u
->subarray
) {
9636 map
= get_imsm_map(dev
, MAP_0
);
9638 if (u
->direction
== R10_TO_R0
) {
9639 unsigned long long num_data_stripes
;
9641 /* Number of failed disks must be half of initial disk number */
9642 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9643 (map
->num_members
/ 2))
9646 /* iterate through devices to mark removed disks as spare */
9647 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9648 if (dm
->disk
.status
& FAILED_DISK
) {
9649 int idx
= dm
->index
;
9650 /* update indexes on the disk list */
9651 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9652 the index values will end up being correct.... NB */
9653 for (du
= super
->disks
; du
; du
= du
->next
)
9654 if (du
->index
> idx
)
9656 /* mark as spare disk */
9661 map
->num_members
= map
->num_members
/ 2;
9662 map
->map_state
= IMSM_T_STATE_NORMAL
;
9663 map
->num_domains
= 1;
9664 map
->raid_level
= 0;
9665 map
->failed_disk_num
= -1;
9666 num_data_stripes
= imsm_dev_size(dev
) / 2;
9667 num_data_stripes
/= map
->blocks_per_strip
;
9668 set_num_data_stripes(map
, num_data_stripes
);
9671 if (u
->direction
== R0_TO_R10
) {
9673 unsigned long long num_data_stripes
;
9675 /* update slots in current disk list */
9676 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9680 /* create new *missing* disks */
9681 for (i
= 0; i
< map
->num_members
; i
++) {
9682 space
= *space_list
;
9685 *space_list
= *space
;
9687 memcpy(du
, super
->disks
, sizeof(*du
));
9691 du
->index
= (i
* 2) + 1;
9692 sprintf((char *)du
->disk
.serial
,
9693 " MISSING_%d", du
->index
);
9694 sprintf((char *)du
->serial
,
9695 "MISSING_%d", du
->index
);
9696 du
->next
= super
->missing
;
9697 super
->missing
= du
;
9699 /* create new dev and map */
9700 space
= *space_list
;
9703 *space_list
= *space
;
9704 dev_new
= (void *)space
;
9705 memcpy(dev_new
, dev
, sizeof(*dev
));
9706 /* update new map */
9707 map
= get_imsm_map(dev_new
, MAP_0
);
9708 map
->num_members
= map
->num_members
* 2;
9709 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9710 map
->num_domains
= 2;
9711 map
->raid_level
= 1;
9712 num_data_stripes
= imsm_dev_size(dev
) / 2;
9713 num_data_stripes
/= map
->blocks_per_strip
;
9714 num_data_stripes
/= map
->num_domains
;
9715 set_num_data_stripes(map
, num_data_stripes
);
9717 /* replace dev<->dev_new */
9720 /* update disk order table */
9721 for (du
= super
->disks
; du
; du
= du
->next
)
9723 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9724 for (du
= super
->missing
; du
; du
= du
->next
)
9725 if (du
->index
>= 0) {
9726 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9727 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9733 static void imsm_process_update(struct supertype
*st
,
9734 struct metadata_update
*update
)
9737 * crack open the metadata_update envelope to find the update record
9738 * update can be one of:
9739 * update_reshape_container_disks - all the arrays in the container
9740 * are being reshaped to have more devices. We need to mark
9741 * the arrays for general migration and convert selected spares
9742 * into active devices.
9743 * update_activate_spare - a spare device has replaced a failed
9744 * device in an array, update the disk_ord_tbl. If this disk is
9745 * present in all member arrays then also clear the SPARE_DISK
9747 * update_create_array
9749 * update_rename_array
9750 * update_add_remove_disk
9752 struct intel_super
*super
= st
->sb
;
9753 struct imsm_super
*mpb
;
9754 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9756 /* update requires a larger buf but the allocation failed */
9757 if (super
->next_len
&& !super
->next_buf
) {
9758 super
->next_len
= 0;
9762 if (super
->next_buf
) {
9763 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9765 super
->len
= super
->next_len
;
9766 super
->buf
= super
->next_buf
;
9768 super
->next_len
= 0;
9769 super
->next_buf
= NULL
;
9772 mpb
= super
->anchor
;
9775 case update_general_migration_checkpoint
: {
9776 struct intel_dev
*id
;
9777 struct imsm_update_general_migration_checkpoint
*u
=
9778 (void *)update
->buf
;
9780 dprintf("called for update_general_migration_checkpoint\n");
9782 /* find device under general migration */
9783 for (id
= super
->devlist
; id
; id
= id
->next
) {
9784 if (is_gen_migration(id
->dev
)) {
9785 id
->dev
->vol
.curr_migr_unit
=
9786 __cpu_to_le32(u
->curr_migr_unit
);
9787 super
->updates_pending
++;
9792 case update_takeover
: {
9793 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9794 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9795 imsm_update_version_info(super
);
9796 super
->updates_pending
++;
9801 case update_reshape_container_disks
: {
9802 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9803 if (apply_reshape_container_disks_update(
9804 u
, super
, &update
->space_list
))
9805 super
->updates_pending
++;
9808 case update_reshape_migration
: {
9809 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9810 if (apply_reshape_migration_update(
9811 u
, super
, &update
->space_list
))
9812 super
->updates_pending
++;
9815 case update_size_change
: {
9816 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9817 if (apply_size_change_update(u
, super
))
9818 super
->updates_pending
++;
9821 case update_activate_spare
: {
9822 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9823 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9824 super
->updates_pending
++;
9827 case update_create_array
: {
9828 /* someone wants to create a new array, we need to be aware of
9829 * a few races/collisions:
9830 * 1/ 'Create' called by two separate instances of mdadm
9831 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9832 * devices that have since been assimilated via
9834 * In the event this update can not be carried out mdadm will
9835 * (FIX ME) notice that its update did not take hold.
9837 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9838 struct intel_dev
*dv
;
9839 struct imsm_dev
*dev
;
9840 struct imsm_map
*map
, *new_map
;
9841 unsigned long long start
, end
;
9842 unsigned long long new_start
, new_end
;
9844 struct disk_info
*inf
;
9847 /* handle racing creates: first come first serve */
9848 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9849 dprintf("subarray %d already defined\n", u
->dev_idx
);
9853 /* check update is next in sequence */
9854 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9855 dprintf("can not create array %d expected index %d\n",
9856 u
->dev_idx
, mpb
->num_raid_devs
);
9860 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9861 new_start
= pba_of_lba0(new_map
);
9862 new_end
= new_start
+ per_dev_array_size(new_map
);
9863 inf
= get_disk_info(u
);
9865 /* handle activate_spare versus create race:
9866 * check to make sure that overlapping arrays do not include
9869 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9870 dev
= get_imsm_dev(super
, i
);
9871 map
= get_imsm_map(dev
, MAP_0
);
9872 start
= pba_of_lba0(map
);
9873 end
= start
+ per_dev_array_size(map
);
9874 if ((new_start
>= start
&& new_start
<= end
) ||
9875 (start
>= new_start
&& start
<= new_end
))
9880 if (disks_overlap(super
, i
, u
)) {
9881 dprintf("arrays overlap\n");
9886 /* check that prepare update was successful */
9887 if (!update
->space
) {
9888 dprintf("prepare update failed\n");
9892 /* check that all disks are still active before committing
9893 * changes. FIXME: could we instead handle this by creating a
9894 * degraded array? That's probably not what the user expects,
9895 * so better to drop this update on the floor.
9897 for (i
= 0; i
< new_map
->num_members
; i
++) {
9898 dl
= serial_to_dl(inf
[i
].serial
, super
);
9900 dprintf("disk disappeared\n");
9905 super
->updates_pending
++;
9907 /* convert spares to members and fixup ord_tbl */
9908 for (i
= 0; i
< new_map
->num_members
; i
++) {
9909 dl
= serial_to_dl(inf
[i
].serial
, super
);
9910 if (dl
->index
== -1) {
9911 dl
->index
= mpb
->num_disks
;
9913 dl
->disk
.status
|= CONFIGURED_DISK
;
9914 dl
->disk
.status
&= ~SPARE_DISK
;
9916 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9921 update
->space
= NULL
;
9922 imsm_copy_dev(dev
, &u
->dev
);
9923 dv
->index
= u
->dev_idx
;
9924 dv
->next
= super
->devlist
;
9925 super
->devlist
= dv
;
9926 mpb
->num_raid_devs
++;
9928 imsm_update_version_info(super
);
9931 /* mdmon knows how to release update->space, but not
9932 * ((struct intel_dev *) update->space)->dev
9934 if (update
->space
) {
9940 case update_kill_array
: {
9941 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9942 int victim
= u
->dev_idx
;
9943 struct active_array
*a
;
9944 struct intel_dev
**dp
;
9945 struct imsm_dev
*dev
;
9947 /* sanity check that we are not affecting the uuid of
9948 * active arrays, or deleting an active array
9950 * FIXME when immutable ids are available, but note that
9951 * we'll also need to fixup the invalidated/active
9952 * subarray indexes in mdstat
9954 for (a
= st
->arrays
; a
; a
= a
->next
)
9955 if (a
->info
.container_member
>= victim
)
9957 /* by definition if mdmon is running at least one array
9958 * is active in the container, so checking
9959 * mpb->num_raid_devs is just extra paranoia
9961 dev
= get_imsm_dev(super
, victim
);
9962 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9963 dprintf("failed to delete subarray-%d\n", victim
);
9967 for (dp
= &super
->devlist
; *dp
;)
9968 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9971 if ((*dp
)->index
> (unsigned)victim
)
9975 mpb
->num_raid_devs
--;
9976 super
->updates_pending
++;
9979 case update_rename_array
: {
9980 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9981 char name
[MAX_RAID_SERIAL_LEN
+1];
9982 int target
= u
->dev_idx
;
9983 struct active_array
*a
;
9984 struct imsm_dev
*dev
;
9986 /* sanity check that we are not affecting the uuid of
9989 memset(name
, 0, sizeof(name
));
9990 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9991 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9992 for (a
= st
->arrays
; a
; a
= a
->next
)
9993 if (a
->info
.container_member
== target
)
9995 dev
= get_imsm_dev(super
, u
->dev_idx
);
9996 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9997 dprintf("failed to rename subarray-%d\n", target
);
10001 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
10002 super
->updates_pending
++;
10005 case update_add_remove_disk
: {
10006 /* we may be able to repair some arrays if disks are
10007 * being added, check the status of add_remove_disk
10008 * if discs has been added.
10010 if (add_remove_disk_update(super
)) {
10011 struct active_array
*a
;
10013 super
->updates_pending
++;
10014 for (a
= st
->arrays
; a
; a
= a
->next
)
10015 a
->check_degraded
= 1;
10019 case update_prealloc_badblocks_mem
:
10021 case update_rwh_policy
: {
10022 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
10023 int target
= u
->dev_idx
;
10024 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
10026 dprintf("could not find subarray-%d\n", target
);
10030 if (dev
->rwh_policy
!= u
->new_policy
) {
10031 dev
->rwh_policy
= u
->new_policy
;
10032 super
->updates_pending
++;
10037 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10041 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10043 static int imsm_prepare_update(struct supertype
*st
,
10044 struct metadata_update
*update
)
10047 * Allocate space to hold new disk entries, raid-device entries or a new
10048 * mpb if necessary. The manager synchronously waits for updates to
10049 * complete in the monitor, so new mpb buffers allocated here can be
10050 * integrated by the monitor thread without worrying about live pointers
10051 * in the manager thread.
10053 enum imsm_update_type type
;
10054 struct intel_super
*super
= st
->sb
;
10055 unsigned int sector_size
= super
->sector_size
;
10056 struct imsm_super
*mpb
= super
->anchor
;
10060 if (update
->len
< (int)sizeof(type
))
10063 type
= *(enum imsm_update_type
*) update
->buf
;
10066 case update_general_migration_checkpoint
:
10067 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10069 dprintf("called for update_general_migration_checkpoint\n");
10071 case update_takeover
: {
10072 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10073 if (update
->len
< (int)sizeof(*u
))
10075 if (u
->direction
== R0_TO_R10
) {
10076 void **tail
= (void **)&update
->space_list
;
10077 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10078 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10079 int num_members
= map
->num_members
;
10082 /* allocate memory for added disks */
10083 for (i
= 0; i
< num_members
; i
++) {
10084 size
= sizeof(struct dl
);
10085 space
= xmalloc(size
);
10090 /* allocate memory for new device */
10091 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10092 (num_members
* sizeof(__u32
));
10093 space
= xmalloc(size
);
10097 len
= disks_to_mpb_size(num_members
* 2);
10102 case update_reshape_container_disks
: {
10103 /* Every raid device in the container is about to
10104 * gain some more devices, and we will enter a
10106 * So each 'imsm_map' will be bigger, and the imsm_vol
10107 * will now hold 2 of them.
10108 * Thus we need new 'struct imsm_dev' allocations sized
10109 * as sizeof_imsm_dev but with more devices in both maps.
10111 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10112 struct intel_dev
*dl
;
10113 void **space_tail
= (void**)&update
->space_list
;
10115 if (update
->len
< (int)sizeof(*u
))
10118 dprintf("for update_reshape\n");
10120 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10121 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10123 if (u
->new_raid_disks
> u
->old_raid_disks
)
10124 size
+= sizeof(__u32
)*2*
10125 (u
->new_raid_disks
- u
->old_raid_disks
);
10129 *space_tail
= NULL
;
10132 len
= disks_to_mpb_size(u
->new_raid_disks
);
10133 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10136 case update_reshape_migration
: {
10137 /* for migration level 0->5 we need to add disks
10138 * so the same as for container operation we will copy
10139 * device to the bigger location.
10140 * in memory prepared device and new disk area are prepared
10141 * for usage in process update
10143 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10144 struct intel_dev
*id
;
10145 void **space_tail
= (void **)&update
->space_list
;
10148 int current_level
= -1;
10150 if (update
->len
< (int)sizeof(*u
))
10153 dprintf("for update_reshape\n");
10155 /* add space for bigger array in update
10157 for (id
= super
->devlist
; id
; id
= id
->next
) {
10158 if (id
->index
== (unsigned)u
->subdev
) {
10159 size
= sizeof_imsm_dev(id
->dev
, 1);
10160 if (u
->new_raid_disks
> u
->old_raid_disks
)
10161 size
+= sizeof(__u32
)*2*
10162 (u
->new_raid_disks
- u
->old_raid_disks
);
10166 *space_tail
= NULL
;
10170 if (update
->space_list
== NULL
)
10173 /* add space for disk in update
10175 size
= sizeof(struct dl
);
10179 *space_tail
= NULL
;
10181 /* add spare device to update
10183 for (id
= super
->devlist
; id
; id
= id
->next
)
10184 if (id
->index
== (unsigned)u
->subdev
) {
10185 struct imsm_dev
*dev
;
10186 struct imsm_map
*map
;
10188 dev
= get_imsm_dev(super
, u
->subdev
);
10189 map
= get_imsm_map(dev
, MAP_0
);
10190 current_level
= map
->raid_level
;
10193 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10194 struct mdinfo
*spares
;
10196 spares
= get_spares_for_grow(st
);
10199 struct mdinfo
*dev
;
10201 dev
= spares
->devs
;
10204 makedev(dev
->disk
.major
,
10206 dl
= get_disk_super(super
,
10209 dl
->index
= u
->old_raid_disks
;
10212 sysfs_free(spares
);
10215 len
= disks_to_mpb_size(u
->new_raid_disks
);
10216 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10219 case update_size_change
: {
10220 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10224 case update_activate_spare
: {
10225 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10229 case update_create_array
: {
10230 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10231 struct intel_dev
*dv
;
10232 struct imsm_dev
*dev
= &u
->dev
;
10233 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10235 struct disk_info
*inf
;
10239 if (update
->len
< (int)sizeof(*u
))
10242 inf
= get_disk_info(u
);
10243 len
= sizeof_imsm_dev(dev
, 1);
10244 /* allocate a new super->devlist entry */
10245 dv
= xmalloc(sizeof(*dv
));
10246 dv
->dev
= xmalloc(len
);
10247 update
->space
= dv
;
10249 /* count how many spares will be converted to members */
10250 for (i
= 0; i
< map
->num_members
; i
++) {
10251 dl
= serial_to_dl(inf
[i
].serial
, super
);
10253 /* hmm maybe it failed?, nothing we can do about
10258 if (count_memberships(dl
, super
) == 0)
10261 len
+= activate
* sizeof(struct imsm_disk
);
10264 case update_kill_array
: {
10265 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10269 case update_rename_array
: {
10270 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10274 case update_add_remove_disk
:
10275 /* no update->len needed */
10277 case update_prealloc_badblocks_mem
:
10278 super
->extra_space
+= sizeof(struct bbm_log
) -
10279 get_imsm_bbm_log_size(super
->bbm_log
);
10281 case update_rwh_policy
: {
10282 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10290 /* check if we need a larger metadata buffer */
10291 if (super
->next_buf
)
10292 buf_len
= super
->next_len
;
10294 buf_len
= super
->len
;
10296 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10297 /* ok we need a larger buf than what is currently allocated
10298 * if this allocation fails process_update will notice that
10299 * ->next_len is set and ->next_buf is NULL
10301 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10302 super
->extra_space
+ len
, sector_size
);
10303 if (super
->next_buf
)
10304 free(super
->next_buf
);
10306 super
->next_len
= buf_len
;
10307 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10308 memset(super
->next_buf
, 0, buf_len
);
10310 super
->next_buf
= NULL
;
10315 /* must be called while manager is quiesced */
10316 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10318 struct imsm_super
*mpb
= super
->anchor
;
10320 struct imsm_dev
*dev
;
10321 struct imsm_map
*map
;
10322 unsigned int i
, j
, num_members
;
10323 __u32 ord
, ord_map0
;
10324 struct bbm_log
*log
= super
->bbm_log
;
10326 dprintf("deleting device[%d] from imsm_super\n", index
);
10328 /* shift all indexes down one */
10329 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10330 if (iter
->index
> (int)index
)
10332 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10333 if (iter
->index
> (int)index
)
10336 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10337 dev
= get_imsm_dev(super
, i
);
10338 map
= get_imsm_map(dev
, MAP_0
);
10339 num_members
= map
->num_members
;
10340 for (j
= 0; j
< num_members
; j
++) {
10341 /* update ord entries being careful not to propagate
10342 * ord-flags to the first map
10344 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10345 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10347 if (ord_to_idx(ord
) <= index
)
10350 map
= get_imsm_map(dev
, MAP_0
);
10351 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10352 map
= get_imsm_map(dev
, MAP_1
);
10354 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10358 for (i
= 0; i
< log
->entry_count
; i
++) {
10359 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10361 if (entry
->disk_ordinal
<= index
)
10363 entry
->disk_ordinal
--;
10367 super
->updates_pending
++;
10369 struct dl
*dl
= *dlp
;
10371 *dlp
= (*dlp
)->next
;
10372 __free_imsm_disk(dl
);
10376 static void close_targets(int *targets
, int new_disks
)
10383 for (i
= 0; i
< new_disks
; i
++) {
10384 if (targets
[i
] >= 0) {
10391 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10392 struct intel_super
*super
,
10393 struct imsm_dev
*dev
)
10399 struct imsm_map
*map
;
10402 ret_val
= raid_disks
/2;
10403 /* check map if all disks pairs not failed
10406 map
= get_imsm_map(dev
, MAP_0
);
10407 for (i
= 0; i
< ret_val
; i
++) {
10408 int degradation
= 0;
10409 if (get_imsm_disk(super
, i
) == NULL
)
10411 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10413 if (degradation
== 2)
10416 map
= get_imsm_map(dev
, MAP_1
);
10417 /* if there is no second map
10418 * result can be returned
10422 /* check degradation in second map
10424 for (i
= 0; i
< ret_val
; i
++) {
10425 int degradation
= 0;
10426 if (get_imsm_disk(super
, i
) == NULL
)
10428 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10430 if (degradation
== 2)
10444 /*******************************************************************************
10445 * Function: open_backup_targets
10446 * Description: Function opens file descriptors for all devices given in
10449 * info : general array info
10450 * raid_disks : number of disks
10451 * raid_fds : table of device's file descriptors
10452 * super : intel super for raid10 degradation check
10453 * dev : intel device for raid10 degradation check
10457 ******************************************************************************/
10458 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10459 struct intel_super
*super
, struct imsm_dev
*dev
)
10465 for (i
= 0; i
< raid_disks
; i
++)
10468 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10471 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10472 dprintf("disk is faulty!!\n");
10476 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10479 dn
= map_dev(sd
->disk
.major
,
10480 sd
->disk
.minor
, 1);
10481 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10482 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10483 pr_err("cannot open component\n");
10488 /* check if maximum array degradation level is not exceeded
10490 if ((raid_disks
- opened
) >
10491 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10493 pr_err("Not enough disks can be opened.\n");
10494 close_targets(raid_fds
, raid_disks
);
10500 /*******************************************************************************
10501 * Function: validate_container_imsm
10502 * Description: This routine validates container after assemble,
10503 * eg. if devices in container are under the same controller.
10506 * info : linked list with info about devices used in array
10510 ******************************************************************************/
10511 int validate_container_imsm(struct mdinfo
*info
)
10513 if (check_env("IMSM_NO_PLATFORM"))
10516 struct sys_dev
*idev
;
10517 struct sys_dev
*hba
= NULL
;
10518 struct sys_dev
*intel_devices
= find_intel_devices();
10519 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10520 info
->disk
.minor
));
10522 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10523 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10532 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10533 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10537 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10538 struct mdinfo
*dev
;
10540 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10541 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10543 struct sys_dev
*hba2
= NULL
;
10544 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10545 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10553 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10554 get_orom_by_device_id(hba2
->dev_id
);
10556 if (hba2
&& hba
->type
!= hba2
->type
) {
10557 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10558 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10562 if (orom
!= orom2
) {
10563 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10564 " This operation is not supported and can lead to data loss.\n");
10569 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10570 " This operation is not supported and can lead to data loss.\n");
10578 /*******************************************************************************
10579 * Function: imsm_record_badblock
10580 * Description: This routine stores new bad block record in BBM log
10583 * a : array containing a bad block
10584 * slot : disk number containing a bad block
10585 * sector : bad block sector
10586 * length : bad block sectors range
10590 ******************************************************************************/
10591 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10592 unsigned long long sector
, int length
)
10594 struct intel_super
*super
= a
->container
->sb
;
10598 ord
= imsm_disk_slot_to_ord(a
, slot
);
10602 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10605 super
->updates_pending
++;
10609 /*******************************************************************************
10610 * Function: imsm_clear_badblock
10611 * Description: This routine clears bad block record from BBM log
10614 * a : array containing a bad block
10615 * slot : disk number containing a bad block
10616 * sector : bad block sector
10617 * length : bad block sectors range
10621 ******************************************************************************/
10622 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10623 unsigned long long sector
, int length
)
10625 struct intel_super
*super
= a
->container
->sb
;
10629 ord
= imsm_disk_slot_to_ord(a
, slot
);
10633 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10635 super
->updates_pending
++;
10639 /*******************************************************************************
10640 * Function: imsm_get_badblocks
10641 * Description: This routine get list of bad blocks for an array
10645 * slot : disk number
10647 * bb : structure containing bad blocks
10649 ******************************************************************************/
10650 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10652 int inst
= a
->info
.container_member
;
10653 struct intel_super
*super
= a
->container
->sb
;
10654 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10655 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10658 ord
= imsm_disk_slot_to_ord(a
, slot
);
10662 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10663 per_dev_array_size(map
), &super
->bb
);
10667 /*******************************************************************************
10668 * Function: examine_badblocks_imsm
10669 * Description: Prints list of bad blocks on a disk to the standard output
10672 * st : metadata handler
10673 * fd : open file descriptor for device
10674 * devname : device name
10678 ******************************************************************************/
10679 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10681 struct intel_super
*super
= st
->sb
;
10682 struct bbm_log
*log
= super
->bbm_log
;
10683 struct dl
*d
= NULL
;
10686 for (d
= super
->disks
; d
; d
= d
->next
) {
10687 if (strcmp(d
->devname
, devname
) == 0)
10691 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10692 pr_err("%s doesn't appear to be part of a raid array\n",
10699 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10701 for (i
= 0; i
< log
->entry_count
; i
++) {
10702 if (entry
[i
].disk_ordinal
== d
->index
) {
10703 unsigned long long sector
= __le48_to_cpu(
10704 &entry
[i
].defective_block_start
);
10705 int cnt
= entry
[i
].marked_count
+ 1;
10708 printf("Bad-blocks on %s:\n", devname
);
10712 printf("%20llu for %d sectors\n", sector
, cnt
);
10718 printf("No bad-blocks list configured on %s\n", devname
);
10722 /*******************************************************************************
10723 * Function: init_migr_record_imsm
10724 * Description: Function inits imsm migration record
10726 * super : imsm internal array info
10727 * dev : device under migration
10728 * info : general array info to find the smallest device
10731 ******************************************************************************/
10732 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10733 struct mdinfo
*info
)
10735 struct intel_super
*super
= st
->sb
;
10736 struct migr_record
*migr_rec
= super
->migr_rec
;
10737 int new_data_disks
;
10738 unsigned long long dsize
, dev_sectors
;
10739 long long unsigned min_dev_sectors
= -1LLU;
10743 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10744 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10745 unsigned long long num_migr_units
;
10746 unsigned long long array_blocks
;
10748 memset(migr_rec
, 0, sizeof(struct migr_record
));
10749 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10751 /* only ascending reshape supported now */
10752 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10754 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10755 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10756 migr_rec
->dest_depth_per_unit
*=
10757 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10758 new_data_disks
= imsm_num_data_members(map_dest
);
10759 migr_rec
->blocks_per_unit
=
10760 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10761 migr_rec
->dest_depth_per_unit
=
10762 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10763 array_blocks
= info
->component_size
* new_data_disks
;
10765 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10767 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10769 set_num_migr_units(migr_rec
, num_migr_units
);
10771 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10772 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10774 /* Find the smallest dev */
10775 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10776 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10777 fd
= dev_open(nm
, O_RDONLY
);
10780 get_dev_size(fd
, NULL
, &dsize
);
10781 dev_sectors
= dsize
/ 512;
10782 if (dev_sectors
< min_dev_sectors
)
10783 min_dev_sectors
= dev_sectors
;
10786 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
10787 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10789 write_imsm_migr_rec(st
);
10794 /*******************************************************************************
10795 * Function: save_backup_imsm
10796 * Description: Function saves critical data stripes to Migration Copy Area
10797 * and updates the current migration unit status.
10798 * Use restore_stripes() to form a destination stripe,
10799 * and to write it to the Copy Area.
10801 * st : supertype information
10802 * dev : imsm device that backup is saved for
10803 * info : general array info
10804 * buf : input buffer
10805 * length : length of data to backup (blocks_per_unit)
10809 ******************************************************************************/
10810 int save_backup_imsm(struct supertype
*st
,
10811 struct imsm_dev
*dev
,
10812 struct mdinfo
*info
,
10817 struct intel_super
*super
= st
->sb
;
10818 unsigned long long *target_offsets
;
10821 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10822 int new_disks
= map_dest
->num_members
;
10823 int dest_layout
= 0;
10825 unsigned long long start
;
10826 int data_disks
= imsm_num_data_members(map_dest
);
10828 targets
= xmalloc(new_disks
* sizeof(int));
10830 for (i
= 0; i
< new_disks
; i
++)
10833 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10835 start
= info
->reshape_progress
* 512;
10836 for (i
= 0; i
< new_disks
; i
++) {
10837 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
10838 /* move back copy area adderss, it will be moved forward
10839 * in restore_stripes() using start input variable
10841 target_offsets
[i
] -= start
/data_disks
;
10844 if (open_backup_targets(info
, new_disks
, targets
,
10848 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10849 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10851 if (restore_stripes(targets
, /* list of dest devices */
10852 target_offsets
, /* migration record offsets */
10855 map_dest
->raid_level
,
10857 -1, /* source backup file descriptor */
10858 0, /* input buf offset
10859 * always 0 buf is already offseted */
10863 pr_err("Error restoring stripes\n");
10871 close_targets(targets
, new_disks
);
10874 free(target_offsets
);
10879 /*******************************************************************************
10880 * Function: save_checkpoint_imsm
10881 * Description: Function called for current unit status update
10882 * in the migration record. It writes it to disk.
10884 * super : imsm internal array info
10885 * info : general array info
10889 * 2: failure, means no valid migration record
10890 * / no general migration in progress /
10891 ******************************************************************************/
10892 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10894 struct intel_super
*super
= st
->sb
;
10895 unsigned long long blocks_per_unit
;
10896 unsigned long long curr_migr_unit
;
10898 if (load_imsm_migr_rec(super
, info
) != 0) {
10899 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10903 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10904 if (blocks_per_unit
== 0) {
10905 dprintf("imsm: no migration in progress.\n");
10908 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10909 /* check if array is alligned to copy area
10910 * if it is not alligned, add one to current migration unit value
10911 * this can happend on array reshape finish only
10913 if (info
->reshape_progress
% blocks_per_unit
)
10916 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
10917 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10918 set_migr_dest_1st_member_lba(super
->migr_rec
,
10919 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
10921 if (write_imsm_migr_rec(st
) < 0) {
10922 dprintf("imsm: Cannot write migration record outside backup area\n");
10929 /*******************************************************************************
10930 * Function: recover_backup_imsm
10931 * Description: Function recovers critical data from the Migration Copy Area
10932 * while assembling an array.
10934 * super : imsm internal array info
10935 * info : general array info
10937 * 0 : success (or there is no data to recover)
10939 ******************************************************************************/
10940 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10942 struct intel_super
*super
= st
->sb
;
10943 struct migr_record
*migr_rec
= super
->migr_rec
;
10944 struct imsm_map
*map_dest
;
10945 struct intel_dev
*id
= NULL
;
10946 unsigned long long read_offset
;
10947 unsigned long long write_offset
;
10949 int *targets
= NULL
;
10950 int new_disks
, i
, err
;
10953 unsigned int sector_size
= super
->sector_size
;
10954 unsigned long curr_migr_unit
= current_migr_unit(migr_rec
);
10955 unsigned long num_migr_units
= get_num_migr_units(migr_rec
);
10957 int skipped_disks
= 0;
10959 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10963 /* recover data only during assemblation */
10964 if (strncmp(buffer
, "inactive", 8) != 0)
10966 /* no data to recover */
10967 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10969 if (curr_migr_unit
>= num_migr_units
)
10972 /* find device during reshape */
10973 for (id
= super
->devlist
; id
; id
= id
->next
)
10974 if (is_gen_migration(id
->dev
))
10979 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10980 new_disks
= map_dest
->num_members
;
10982 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
10984 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
10985 pba_of_lba0(map_dest
)) * 512;
10987 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10988 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10990 targets
= xcalloc(new_disks
, sizeof(int));
10992 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10993 pr_err("Cannot open some devices belonging to array.\n");
10997 for (i
= 0; i
< new_disks
; i
++) {
10998 if (targets
[i
] < 0) {
11002 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
11003 pr_err("Cannot seek to block: %s\n",
11008 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
11009 pr_err("Cannot read copy area block: %s\n",
11014 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
11015 pr_err("Cannot seek to block: %s\n",
11020 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
11021 pr_err("Cannot restore block: %s\n",
11028 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11032 pr_err("Cannot restore data from backup. Too many failed disks\n");
11036 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11037 /* ignore error == 2, this can mean end of reshape here
11039 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11045 for (i
= 0; i
< new_disks
; i
++)
11054 static char disk_by_path
[] = "/dev/disk/by-path/";
11056 static const char *imsm_get_disk_controller_domain(const char *path
)
11058 char disk_path
[PATH_MAX
];
11062 strcpy(disk_path
, disk_by_path
);
11063 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
11064 if (stat(disk_path
, &st
) == 0) {
11065 struct sys_dev
* hba
;
11068 path
= devt_to_devpath(st
.st_rdev
);
11071 hba
= find_disk_attached_hba(-1, path
);
11072 if (hba
&& hba
->type
== SYS_DEV_SAS
)
11074 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
11076 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
11078 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
11082 dprintf("path: %s hba: %s attached: %s\n",
11083 path
, (hba
) ? hba
->path
: "NULL", drv
);
11089 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11091 static char devnm
[32];
11092 char subdev_name
[20];
11093 struct mdstat_ent
*mdstat
;
11095 sprintf(subdev_name
, "%d", subdev
);
11096 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11100 strcpy(devnm
, mdstat
->devnm
);
11101 free_mdstat(mdstat
);
11105 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11106 struct geo_params
*geo
,
11107 int *old_raid_disks
,
11110 /* currently we only support increasing the number of devices
11111 * for a container. This increases the number of device for each
11112 * member array. They must all be RAID0 or RAID5.
11115 struct mdinfo
*info
, *member
;
11116 int devices_that_can_grow
= 0;
11118 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11120 if (geo
->size
> 0 ||
11121 geo
->level
!= UnSet
||
11122 geo
->layout
!= UnSet
||
11123 geo
->chunksize
!= 0 ||
11124 geo
->raid_disks
== UnSet
) {
11125 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11129 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11130 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11134 info
= container_content_imsm(st
, NULL
);
11135 for (member
= info
; member
; member
= member
->next
) {
11138 dprintf("imsm: checking device_num: %i\n",
11139 member
->container_member
);
11141 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11142 /* we work on container for Online Capacity Expansion
11143 * only so raid_disks has to grow
11145 dprintf("imsm: for container operation raid disks increase is required\n");
11149 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11150 /* we cannot use this container with other raid level
11152 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11153 info
->array
.level
);
11156 /* check for platform support
11157 * for this raid level configuration
11159 struct intel_super
*super
= st
->sb
;
11160 if (!is_raid_level_supported(super
->orom
,
11161 member
->array
.level
,
11162 geo
->raid_disks
)) {
11163 dprintf("platform does not support raid%d with %d disk%s\n",
11166 geo
->raid_disks
> 1 ? "s" : "");
11169 /* check if component size is aligned to chunk size
11171 if (info
->component_size
%
11172 (info
->array
.chunk_size
/512)) {
11173 dprintf("Component size is not aligned to chunk size\n");
11178 if (*old_raid_disks
&&
11179 info
->array
.raid_disks
!= *old_raid_disks
)
11181 *old_raid_disks
= info
->array
.raid_disks
;
11183 /* All raid5 and raid0 volumes in container
11184 * have to be ready for Online Capacity Expansion
11185 * so they need to be assembled. We have already
11186 * checked that no recovery etc is happening.
11188 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11189 st
->container_devnm
);
11190 if (result
== NULL
) {
11191 dprintf("imsm: cannot find array\n");
11194 devices_that_can_grow
++;
11197 if (!member
&& devices_that_can_grow
)
11201 dprintf("Container operation allowed\n");
11203 dprintf("Error: %i\n", ret_val
);
11208 /* Function: get_spares_for_grow
11209 * Description: Allocates memory and creates list of spare devices
11210 * avaliable in container. Checks if spare drive size is acceptable.
11211 * Parameters: Pointer to the supertype structure
11212 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11215 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11217 struct spare_criteria sc
;
11219 get_spare_criteria_imsm(st
, &sc
);
11220 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11223 /******************************************************************************
11224 * function: imsm_create_metadata_update_for_reshape
11225 * Function creates update for whole IMSM container.
11227 ******************************************************************************/
11228 static int imsm_create_metadata_update_for_reshape(
11229 struct supertype
*st
,
11230 struct geo_params
*geo
,
11231 int old_raid_disks
,
11232 struct imsm_update_reshape
**updatep
)
11234 struct intel_super
*super
= st
->sb
;
11235 struct imsm_super
*mpb
= super
->anchor
;
11236 int update_memory_size
;
11237 struct imsm_update_reshape
*u
;
11238 struct mdinfo
*spares
;
11241 struct mdinfo
*dev
;
11243 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11245 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11247 /* size of all update data without anchor */
11248 update_memory_size
= sizeof(struct imsm_update_reshape
);
11250 /* now add space for spare disks that we need to add. */
11251 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11253 u
= xcalloc(1, update_memory_size
);
11254 u
->type
= update_reshape_container_disks
;
11255 u
->old_raid_disks
= old_raid_disks
;
11256 u
->new_raid_disks
= geo
->raid_disks
;
11258 /* now get spare disks list
11260 spares
= get_spares_for_grow(st
);
11262 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11263 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11268 /* we have got spares
11269 * update disk list in imsm_disk list table in anchor
11271 dprintf("imsm: %i spares are available.\n\n",
11272 spares
->array
.spare_disks
);
11274 dev
= spares
->devs
;
11275 for (i
= 0; i
< delta_disks
; i
++) {
11280 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11282 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11283 dl
->index
= mpb
->num_disks
;
11291 sysfs_free(spares
);
11293 dprintf("imsm: reshape update preparation :");
11294 if (i
== delta_disks
) {
11295 dprintf_cont(" OK\n");
11297 return update_memory_size
;
11300 dprintf_cont(" Error\n");
11305 /******************************************************************************
11306 * function: imsm_create_metadata_update_for_size_change()
11307 * Creates update for IMSM array for array size change.
11309 ******************************************************************************/
11310 static int imsm_create_metadata_update_for_size_change(
11311 struct supertype
*st
,
11312 struct geo_params
*geo
,
11313 struct imsm_update_size_change
**updatep
)
11315 struct intel_super
*super
= st
->sb
;
11316 int update_memory_size
;
11317 struct imsm_update_size_change
*u
;
11319 dprintf("(enter) New size = %llu\n", geo
->size
);
11321 /* size of all update data without anchor */
11322 update_memory_size
= sizeof(struct imsm_update_size_change
);
11324 u
= xcalloc(1, update_memory_size
);
11325 u
->type
= update_size_change
;
11326 u
->subdev
= super
->current_vol
;
11327 u
->new_size
= geo
->size
;
11329 dprintf("imsm: reshape update preparation : OK\n");
11332 return update_memory_size
;
11335 /******************************************************************************
11336 * function: imsm_create_metadata_update_for_migration()
11337 * Creates update for IMSM array.
11339 ******************************************************************************/
11340 static int imsm_create_metadata_update_for_migration(
11341 struct supertype
*st
,
11342 struct geo_params
*geo
,
11343 struct imsm_update_reshape_migration
**updatep
)
11345 struct intel_super
*super
= st
->sb
;
11346 int update_memory_size
;
11347 struct imsm_update_reshape_migration
*u
;
11348 struct imsm_dev
*dev
;
11349 int previous_level
= -1;
11351 dprintf("(enter) New Level = %i\n", geo
->level
);
11353 /* size of all update data without anchor */
11354 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11356 u
= xcalloc(1, update_memory_size
);
11357 u
->type
= update_reshape_migration
;
11358 u
->subdev
= super
->current_vol
;
11359 u
->new_level
= geo
->level
;
11360 u
->new_layout
= geo
->layout
;
11361 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11362 u
->new_disks
[0] = -1;
11363 u
->new_chunksize
= -1;
11365 dev
= get_imsm_dev(super
, u
->subdev
);
11367 struct imsm_map
*map
;
11369 map
= get_imsm_map(dev
, MAP_0
);
11371 int current_chunk_size
=
11372 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11374 if (geo
->chunksize
!= current_chunk_size
) {
11375 u
->new_chunksize
= geo
->chunksize
/ 1024;
11376 dprintf("imsm: chunk size change from %i to %i\n",
11377 current_chunk_size
, u
->new_chunksize
);
11379 previous_level
= map
->raid_level
;
11382 if (geo
->level
== 5 && previous_level
== 0) {
11383 struct mdinfo
*spares
= NULL
;
11385 u
->new_raid_disks
++;
11386 spares
= get_spares_for_grow(st
);
11387 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11389 sysfs_free(spares
);
11390 update_memory_size
= 0;
11391 pr_err("cannot get spare device for requested migration\n");
11394 sysfs_free(spares
);
11396 dprintf("imsm: reshape update preparation : OK\n");
11399 return update_memory_size
;
11402 static void imsm_update_metadata_locally(struct supertype
*st
,
11403 void *buf
, int len
)
11405 struct metadata_update mu
;
11410 mu
.space_list
= NULL
;
11412 if (imsm_prepare_update(st
, &mu
))
11413 imsm_process_update(st
, &mu
);
11415 while (mu
.space_list
) {
11416 void **space
= mu
.space_list
;
11417 mu
.space_list
= *space
;
11422 /***************************************************************************
11423 * Function: imsm_analyze_change
11424 * Description: Function analyze change for single volume
11425 * and validate if transition is supported
11426 * Parameters: Geometry parameters, supertype structure,
11427 * metadata change direction (apply/rollback)
11428 * Returns: Operation type code on success, -1 if fail
11429 ****************************************************************************/
11430 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11431 struct geo_params
*geo
,
11434 struct mdinfo info
;
11436 int check_devs
= 0;
11438 /* number of added/removed disks in operation result */
11439 int devNumChange
= 0;
11440 /* imsm compatible layout value for array geometry verification */
11441 int imsm_layout
= -1;
11443 struct imsm_dev
*dev
;
11444 struct imsm_map
*map
;
11445 struct intel_super
*super
;
11446 unsigned long long current_size
;
11447 unsigned long long free_size
;
11448 unsigned long long max_size
;
11451 getinfo_super_imsm_volume(st
, &info
, NULL
);
11452 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11453 geo
->level
!= UnSet
) {
11454 switch (info
.array
.level
) {
11456 if (geo
->level
== 5) {
11457 change
= CH_MIGRATION
;
11458 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11459 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11461 goto analyse_change_exit
;
11463 imsm_layout
= geo
->layout
;
11465 devNumChange
= 1; /* parity disk added */
11466 } else if (geo
->level
== 10) {
11467 change
= CH_TAKEOVER
;
11469 devNumChange
= 2; /* two mirrors added */
11470 imsm_layout
= 0x102; /* imsm supported layout */
11475 if (geo
->level
== 0) {
11476 change
= CH_TAKEOVER
;
11478 devNumChange
= -(geo
->raid_disks
/2);
11479 imsm_layout
= 0; /* imsm raid0 layout */
11483 if (change
== -1) {
11484 pr_err("Error. Level Migration from %d to %d not supported!\n",
11485 info
.array
.level
, geo
->level
);
11486 goto analyse_change_exit
;
11489 geo
->level
= info
.array
.level
;
11491 if (geo
->layout
!= info
.array
.layout
&&
11492 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11493 change
= CH_MIGRATION
;
11494 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11495 geo
->layout
== 5) {
11496 /* reshape 5 -> 4 */
11497 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11498 geo
->layout
== 0) {
11499 /* reshape 4 -> 5 */
11503 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11504 info
.array
.layout
, geo
->layout
);
11506 goto analyse_change_exit
;
11509 geo
->layout
= info
.array
.layout
;
11510 if (imsm_layout
== -1)
11511 imsm_layout
= info
.array
.layout
;
11514 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11515 geo
->chunksize
!= info
.array
.chunk_size
) {
11516 if (info
.array
.level
== 10) {
11517 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11519 goto analyse_change_exit
;
11520 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11521 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11522 geo
->chunksize
/1024, info
.component_size
/2);
11524 goto analyse_change_exit
;
11526 change
= CH_MIGRATION
;
11528 geo
->chunksize
= info
.array
.chunk_size
;
11531 chunk
= geo
->chunksize
/ 1024;
11534 dev
= get_imsm_dev(super
, super
->current_vol
);
11535 map
= get_imsm_map(dev
, MAP_0
);
11536 data_disks
= imsm_num_data_members(map
);
11537 /* compute current size per disk member
11539 current_size
= info
.custom_array_size
/ data_disks
;
11541 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11542 /* align component size
11544 geo
->size
= imsm_component_size_alignment_check(
11545 get_imsm_raid_level(dev
->vol
.map
),
11546 chunk
* 1024, super
->sector_size
,
11548 if (geo
->size
== 0) {
11549 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11551 goto analyse_change_exit
;
11555 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11556 if (change
!= -1) {
11557 pr_err("Error. Size change should be the only one at a time.\n");
11559 goto analyse_change_exit
;
11561 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11562 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11563 super
->current_vol
, st
->devnm
);
11564 goto analyse_change_exit
;
11566 /* check the maximum available size
11568 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11569 0, chunk
, &free_size
);
11571 /* Cannot find maximum available space
11575 max_size
= free_size
+ current_size
;
11576 /* align component size
11578 max_size
= imsm_component_size_alignment_check(
11579 get_imsm_raid_level(dev
->vol
.map
),
11580 chunk
* 1024, super
->sector_size
,
11583 if (geo
->size
== MAX_SIZE
) {
11584 /* requested size change to the maximum available size
11586 if (max_size
== 0) {
11587 pr_err("Error. Cannot find maximum available space.\n");
11589 goto analyse_change_exit
;
11591 geo
->size
= max_size
;
11594 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11595 /* accept size for rollback only
11598 /* round size due to metadata compatibility
11600 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11601 << SECT_PER_MB_SHIFT
;
11602 dprintf("Prepare update for size change to %llu\n",
11604 if (current_size
>= geo
->size
) {
11605 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11606 current_size
, geo
->size
);
11607 goto analyse_change_exit
;
11609 if (max_size
&& geo
->size
> max_size
) {
11610 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11611 max_size
, geo
->size
);
11612 goto analyse_change_exit
;
11615 geo
->size
*= data_disks
;
11616 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11617 change
= CH_ARRAY_SIZE
;
11619 if (!validate_geometry_imsm(st
,
11622 geo
->raid_disks
+ devNumChange
,
11624 geo
->size
, INVALID_SECTORS
,
11625 0, 0, info
.consistency_policy
, 1))
11629 struct intel_super
*super
= st
->sb
;
11630 struct imsm_super
*mpb
= super
->anchor
;
11632 if (mpb
->num_raid_devs
> 1) {
11633 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11639 analyse_change_exit
:
11640 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11641 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11642 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11648 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11650 struct intel_super
*super
= st
->sb
;
11651 struct imsm_update_takeover
*u
;
11653 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11655 u
->type
= update_takeover
;
11656 u
->subarray
= super
->current_vol
;
11658 /* 10->0 transition */
11659 if (geo
->level
== 0)
11660 u
->direction
= R10_TO_R0
;
11662 /* 0->10 transition */
11663 if (geo
->level
== 10)
11664 u
->direction
= R0_TO_R10
;
11666 /* update metadata locally */
11667 imsm_update_metadata_locally(st
, u
,
11668 sizeof(struct imsm_update_takeover
));
11669 /* and possibly remotely */
11670 if (st
->update_tail
)
11671 append_metadata_update(st
, u
,
11672 sizeof(struct imsm_update_takeover
));
11679 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11681 int layout
, int chunksize
, int raid_disks
,
11682 int delta_disks
, char *backup
, char *dev
,
11683 int direction
, int verbose
)
11686 struct geo_params geo
;
11688 dprintf("(enter)\n");
11690 memset(&geo
, 0, sizeof(struct geo_params
));
11692 geo
.dev_name
= dev
;
11693 strcpy(geo
.devnm
, st
->devnm
);
11696 geo
.layout
= layout
;
11697 geo
.chunksize
= chunksize
;
11698 geo
.raid_disks
= raid_disks
;
11699 if (delta_disks
!= UnSet
)
11700 geo
.raid_disks
+= delta_disks
;
11702 dprintf("for level : %i\n", geo
.level
);
11703 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11705 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11706 /* On container level we can only increase number of devices. */
11707 dprintf("imsm: info: Container operation\n");
11708 int old_raid_disks
= 0;
11710 if (imsm_reshape_is_allowed_on_container(
11711 st
, &geo
, &old_raid_disks
, direction
)) {
11712 struct imsm_update_reshape
*u
= NULL
;
11715 len
= imsm_create_metadata_update_for_reshape(
11716 st
, &geo
, old_raid_disks
, &u
);
11719 dprintf("imsm: Cannot prepare update\n");
11720 goto exit_imsm_reshape_super
;
11724 /* update metadata locally */
11725 imsm_update_metadata_locally(st
, u
, len
);
11726 /* and possibly remotely */
11727 if (st
->update_tail
)
11728 append_metadata_update(st
, u
, len
);
11733 pr_err("(imsm) Operation is not allowed on this container\n");
11736 /* On volume level we support following operations
11737 * - takeover: raid10 -> raid0; raid0 -> raid10
11738 * - chunk size migration
11739 * - migration: raid5 -> raid0; raid0 -> raid5
11741 struct intel_super
*super
= st
->sb
;
11742 struct intel_dev
*dev
= super
->devlist
;
11744 dprintf("imsm: info: Volume operation\n");
11745 /* find requested device */
11748 imsm_find_array_devnm_by_subdev(
11749 dev
->index
, st
->container_devnm
);
11750 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11755 pr_err("Cannot find %s (%s) subarray\n",
11756 geo
.dev_name
, geo
.devnm
);
11757 goto exit_imsm_reshape_super
;
11759 super
->current_vol
= dev
->index
;
11760 change
= imsm_analyze_change(st
, &geo
, direction
);
11763 ret_val
= imsm_takeover(st
, &geo
);
11765 case CH_MIGRATION
: {
11766 struct imsm_update_reshape_migration
*u
= NULL
;
11768 imsm_create_metadata_update_for_migration(
11771 dprintf("imsm: Cannot prepare update\n");
11775 /* update metadata locally */
11776 imsm_update_metadata_locally(st
, u
, len
);
11777 /* and possibly remotely */
11778 if (st
->update_tail
)
11779 append_metadata_update(st
, u
, len
);
11784 case CH_ARRAY_SIZE
: {
11785 struct imsm_update_size_change
*u
= NULL
;
11787 imsm_create_metadata_update_for_size_change(
11790 dprintf("imsm: Cannot prepare update\n");
11794 /* update metadata locally */
11795 imsm_update_metadata_locally(st
, u
, len
);
11796 /* and possibly remotely */
11797 if (st
->update_tail
)
11798 append_metadata_update(st
, u
, len
);
11808 exit_imsm_reshape_super
:
11809 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11813 #define COMPLETED_OK 0
11814 #define COMPLETED_NONE 1
11815 #define COMPLETED_DELAYED 2
11817 static int read_completed(int fd
, unsigned long long *val
)
11822 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11826 ret
= COMPLETED_OK
;
11827 if (strncmp(buf
, "none", 4) == 0) {
11828 ret
= COMPLETED_NONE
;
11829 } else if (strncmp(buf
, "delayed", 7) == 0) {
11830 ret
= COMPLETED_DELAYED
;
11833 *val
= strtoull(buf
, &ep
, 0);
11834 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11840 /*******************************************************************************
11841 * Function: wait_for_reshape_imsm
11842 * Description: Function writes new sync_max value and waits until
11843 * reshape process reach new position
11845 * sra : general array info
11846 * ndata : number of disks in new array's layout
11849 * 1 : there is no reshape in progress,
11851 ******************************************************************************/
11852 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11854 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11856 unsigned long long completed
;
11857 /* to_complete : new sync_max position */
11858 unsigned long long to_complete
= sra
->reshape_progress
;
11859 unsigned long long position_to_set
= to_complete
/ ndata
;
11862 dprintf("cannot open reshape_position\n");
11867 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11869 dprintf("cannot read reshape_position (no reshape in progres)\n");
11878 if (completed
> position_to_set
) {
11879 dprintf("wrong next position to set %llu (%llu)\n",
11880 to_complete
, position_to_set
);
11884 dprintf("Position set: %llu\n", position_to_set
);
11885 if (sysfs_set_num(sra
, NULL
, "sync_max",
11886 position_to_set
) != 0) {
11887 dprintf("cannot set reshape position to %llu\n",
11896 int timeout
= 3000;
11898 sysfs_wait(fd
, &timeout
);
11899 if (sysfs_get_str(sra
, NULL
, "sync_action",
11901 strncmp(action
, "reshape", 7) != 0) {
11902 if (strncmp(action
, "idle", 4) == 0)
11908 rc
= read_completed(fd
, &completed
);
11910 dprintf("cannot read reshape_position (in loop)\n");
11913 } else if (rc
== COMPLETED_NONE
)
11915 } while (completed
< position_to_set
);
11921 /*******************************************************************************
11922 * Function: check_degradation_change
11923 * Description: Check that array hasn't become failed.
11925 * info : for sysfs access
11926 * sources : source disks descriptors
11927 * degraded: previous degradation level
11929 * degradation level
11930 ******************************************************************************/
11931 int check_degradation_change(struct mdinfo
*info
,
11935 unsigned long long new_degraded
;
11938 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11939 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11940 /* check each device to ensure it is still working */
11943 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11944 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11946 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11949 if (sysfs_get_str(info
,
11950 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11951 strstr(sbuf
, "faulty") ||
11952 strstr(sbuf
, "in_sync") == NULL
) {
11953 /* this device is dead */
11954 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11955 if (sd
->disk
.raid_disk
>= 0 &&
11956 sources
[sd
->disk
.raid_disk
] >= 0) {
11958 sd
->disk
.raid_disk
]);
11959 sources
[sd
->disk
.raid_disk
] =
11968 return new_degraded
;
11971 /*******************************************************************************
11972 * Function: imsm_manage_reshape
11973 * Description: Function finds array under reshape and it manages reshape
11974 * process. It creates stripes backups (if required) and sets
11977 * afd : Backup handle (nattive) - not used
11978 * sra : general array info
11979 * reshape : reshape parameters - not used
11980 * st : supertype structure
11981 * blocks : size of critical section [blocks]
11982 * fds : table of source device descriptor
11983 * offsets : start of array (offest per devices)
11985 * destfd : table of destination device descriptor
11986 * destoffsets : table of destination offsets (per device)
11988 * 1 : success, reshape is done
11990 ******************************************************************************/
11991 static int imsm_manage_reshape(
11992 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
11993 struct supertype
*st
, unsigned long backup_blocks
,
11994 int *fds
, unsigned long long *offsets
,
11995 int dests
, int *destfd
, unsigned long long *destoffsets
)
11998 struct intel_super
*super
= st
->sb
;
11999 struct intel_dev
*dv
;
12000 unsigned int sector_size
= super
->sector_size
;
12001 struct imsm_dev
*dev
= NULL
;
12002 struct imsm_map
*map_src
, *map_dest
;
12003 int migr_vol_qan
= 0;
12004 int ndata
, odata
; /* [bytes] */
12005 int chunk
; /* [bytes] */
12006 struct migr_record
*migr_rec
;
12008 unsigned int buf_size
; /* [bytes] */
12009 unsigned long long max_position
; /* array size [bytes] */
12010 unsigned long long next_step
; /* [blocks]/[bytes] */
12011 unsigned long long old_data_stripe_length
;
12012 unsigned long long start_src
; /* [bytes] */
12013 unsigned long long start
; /* [bytes] */
12014 unsigned long long start_buf_shift
; /* [bytes] */
12016 int source_layout
= 0;
12021 if (!fds
|| !offsets
)
12024 /* Find volume during the reshape */
12025 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
12026 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12027 dv
->dev
->vol
.migr_state
== 1) {
12032 /* Only one volume can migrate at the same time */
12033 if (migr_vol_qan
!= 1) {
12034 pr_err("%s", migr_vol_qan
?
12035 "Number of migrating volumes greater than 1\n" :
12036 "There is no volume during migrationg\n");
12040 map_dest
= get_imsm_map(dev
, MAP_0
);
12041 map_src
= get_imsm_map(dev
, MAP_1
);
12042 if (map_src
== NULL
)
12045 ndata
= imsm_num_data_members(map_dest
);
12046 odata
= imsm_num_data_members(map_src
);
12048 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12049 old_data_stripe_length
= odata
* chunk
;
12051 migr_rec
= super
->migr_rec
;
12053 /* initialize migration record for start condition */
12054 if (sra
->reshape_progress
== 0)
12055 init_migr_record_imsm(st
, dev
, sra
);
12057 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12058 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12061 /* Save checkpoint to update migration record for current
12062 * reshape position (in md). It can be farther than current
12063 * reshape position in metadata.
12065 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12066 /* ignore error == 2, this can mean end of reshape here
12068 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12073 /* size for data */
12074 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12075 /* extend buffer size for parity disk */
12076 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12077 /* add space for stripe alignment */
12078 buf_size
+= old_data_stripe_length
;
12079 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12080 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12084 max_position
= sra
->component_size
* ndata
;
12085 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12087 while (current_migr_unit(migr_rec
) <
12088 get_num_migr_units(migr_rec
)) {
12089 /* current reshape position [blocks] */
12090 unsigned long long current_position
=
12091 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12092 * current_migr_unit(migr_rec
);
12093 unsigned long long border
;
12095 /* Check that array hasn't become failed.
12097 degraded
= check_degradation_change(sra
, fds
, degraded
);
12098 if (degraded
> 1) {
12099 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12103 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12105 if ((current_position
+ next_step
) > max_position
)
12106 next_step
= max_position
- current_position
;
12108 start
= current_position
* 512;
12110 /* align reading start to old geometry */
12111 start_buf_shift
= start
% old_data_stripe_length
;
12112 start_src
= start
- start_buf_shift
;
12114 border
= (start_src
/ odata
) - (start
/ ndata
);
12116 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12117 /* save critical stripes to buf
12118 * start - start address of current unit
12119 * to backup [bytes]
12120 * start_src - start address of current unit
12121 * to backup alligned to source array
12124 unsigned long long next_step_filler
;
12125 unsigned long long copy_length
= next_step
* 512;
12127 /* allign copy area length to stripe in old geometry */
12128 next_step_filler
= ((copy_length
+ start_buf_shift
)
12129 % old_data_stripe_length
);
12130 if (next_step_filler
)
12131 next_step_filler
= (old_data_stripe_length
12132 - next_step_filler
);
12133 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12134 start
, start_src
, copy_length
,
12135 start_buf_shift
, next_step_filler
);
12137 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12138 chunk
, map_src
->raid_level
,
12139 source_layout
, 0, NULL
, start_src
,
12141 next_step_filler
+ start_buf_shift
,
12143 dprintf("imsm: Cannot save stripes to buffer\n");
12146 /* Convert data to destination format and store it
12147 * in backup general migration area
12149 if (save_backup_imsm(st
, dev
, sra
,
12150 buf
+ start_buf_shift
, copy_length
)) {
12151 dprintf("imsm: Cannot save stripes to target devices\n");
12154 if (save_checkpoint_imsm(st
, sra
,
12155 UNIT_SRC_IN_CP_AREA
)) {
12156 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12160 /* set next step to use whole border area */
12161 border
/= next_step
;
12163 next_step
*= border
;
12165 /* When data backed up, checkpoint stored,
12166 * kick the kernel to reshape unit of data
12168 next_step
= next_step
+ sra
->reshape_progress
;
12169 /* limit next step to array max position */
12170 if (next_step
> max_position
)
12171 next_step
= max_position
;
12172 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12173 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12174 sra
->reshape_progress
= next_step
;
12176 /* wait until reshape finish */
12177 if (wait_for_reshape_imsm(sra
, ndata
)) {
12178 dprintf("wait_for_reshape_imsm returned error!\n");
12184 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12185 /* ignore error == 2, this can mean end of reshape here
12187 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12193 /* clear migr_rec on disks after successful migration */
12196 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12197 for (d
= super
->disks
; d
; d
= d
->next
) {
12198 if (d
->index
< 0 || is_failed(&d
->disk
))
12200 unsigned long long dsize
;
12202 get_dev_size(d
->fd
, NULL
, &dsize
);
12203 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12205 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12206 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12207 MIGR_REC_BUF_SECTORS
*sector_size
)
12208 perror("Write migr_rec failed");
12212 /* return '1' if done */
12216 /* See Grow.c: abort_reshape() for further explanation */
12217 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12218 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12219 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12224 struct superswitch super_imsm
= {
12225 .examine_super
= examine_super_imsm
,
12226 .brief_examine_super
= brief_examine_super_imsm
,
12227 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12228 .export_examine_super
= export_examine_super_imsm
,
12229 .detail_super
= detail_super_imsm
,
12230 .brief_detail_super
= brief_detail_super_imsm
,
12231 .write_init_super
= write_init_super_imsm
,
12232 .validate_geometry
= validate_geometry_imsm
,
12233 .add_to_super
= add_to_super_imsm
,
12234 .remove_from_super
= remove_from_super_imsm
,
12235 .detail_platform
= detail_platform_imsm
,
12236 .export_detail_platform
= export_detail_platform_imsm
,
12237 .kill_subarray
= kill_subarray_imsm
,
12238 .update_subarray
= update_subarray_imsm
,
12239 .load_container
= load_container_imsm
,
12240 .default_geometry
= default_geometry_imsm
,
12241 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12242 .reshape_super
= imsm_reshape_super
,
12243 .manage_reshape
= imsm_manage_reshape
,
12244 .recover_backup
= recover_backup_imsm
,
12245 .copy_metadata
= copy_metadata_imsm
,
12246 .examine_badblocks
= examine_badblocks_imsm
,
12247 .match_home
= match_home_imsm
,
12248 .uuid_from_super
= uuid_from_super_imsm
,
12249 .getinfo_super
= getinfo_super_imsm
,
12250 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12251 .update_super
= update_super_imsm
,
12253 .avail_size
= avail_size_imsm
,
12254 .get_spare_criteria
= get_spare_criteria_imsm
,
12256 .compare_super
= compare_super_imsm
,
12258 .load_super
= load_super_imsm
,
12259 .init_super
= init_super_imsm
,
12260 .store_super
= store_super_imsm
,
12261 .free_super
= free_super_imsm
,
12262 .match_metadata_desc
= match_metadata_desc_imsm
,
12263 .container_content
= container_content_imsm
,
12264 .validate_container
= validate_container_imsm
,
12266 .write_init_ppl
= write_init_ppl_imsm
,
12267 .validate_ppl
= validate_ppl_imsm
,
12273 .open_new
= imsm_open_new
,
12274 .set_array_state
= imsm_set_array_state
,
12275 .set_disk
= imsm_set_disk
,
12276 .sync_metadata
= imsm_sync_metadata
,
12277 .activate_spare
= imsm_activate_spare
,
12278 .process_update
= imsm_process_update
,
12279 .prepare_update
= imsm_prepare_update
,
12280 .record_bad_block
= imsm_record_badblock
,
12281 .clear_bad_block
= imsm_clear_badblock
,
12282 .get_bad_blocks
= imsm_get_badblocks
,