2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
47 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
51 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
53 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
54 /* supports RAID CNG */
55 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
56 /* supports expanded stripe sizes of 256K, 512K and 1MB */
57 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
59 /* The OROM Support RST Caching of Volumes */
60 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
61 /* The OROM supports creating disks greater than 2TB */
62 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
63 /* The OROM supports Bad Block Management */
64 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
66 /* THe OROM Supports NVM Caching of Volumes */
67 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
68 /* The OROM supports creating volumes greater than 2TB */
69 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
70 /* originally for PMP, now it's wasted b/c. Never use this bit! */
71 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
72 /* Verify MPB contents against checksum after reading MPB */
73 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
75 /* Define all supported attributes that have to be accepted by mdadm
77 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
79 MPB_ATTRIB_2TB_DISK | \
84 MPB_ATTRIB_EXP_STRIPE_SIZE | \
87 /* Define attributes that are unused but not harmful */
88 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
90 #define MPB_SECTOR_CNT 2210
91 #define IMSM_RESERVED_SECTORS 8192
92 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2048
93 #define SECT_PER_MB_SHIFT 11
94 #define MAX_SECTOR_SIZE 4096
95 #define MULTIPLE_PPL_AREA_SIZE_IMSM (1024 * 1024) /* Size of the whole
100 * This macro let's us ensure that no-one accidentally
101 * changes the size of a struct
103 #define ASSERT_SIZE(_struct, size) \
104 static inline void __assert_size_##_struct(void) \
108 case (sizeof(struct _struct) == size): break; \
112 /* Disk configuration info. */
113 #define IMSM_MAX_DEVICES 255
115 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
116 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
117 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
118 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
119 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
120 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
121 #define JOURNAL_DISK __cpu_to_le32(0x2000000) /* Device marked as Journaling Drive */
122 __u32 status
; /* 0xF0 - 0xF3 */
123 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
124 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
125 #define IMSM_DISK_FILLERS 3
126 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
128 ASSERT_SIZE(imsm_disk
, 48)
130 /* map selector for map managment
136 /* RAID map configuration infos. */
138 __u32 pba_of_lba0_lo
; /* start address of partition */
139 __u32 blocks_per_member_lo
;/* blocks per member */
140 __u32 num_data_stripes_lo
; /* number of data stripes */
141 __u16 blocks_per_strip
;
142 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
143 #define IMSM_T_STATE_NORMAL 0
144 #define IMSM_T_STATE_UNINITIALIZED 1
145 #define IMSM_T_STATE_DEGRADED 2
146 #define IMSM_T_STATE_FAILED 3
148 #define IMSM_T_RAID0 0
149 #define IMSM_T_RAID1 1
150 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
151 __u8 num_members
; /* number of member disks */
152 __u8 num_domains
; /* number of parity domains */
153 __u8 failed_disk_num
; /* valid only when state is degraded */
155 __u32 pba_of_lba0_hi
;
156 __u32 blocks_per_member_hi
;
157 __u32 num_data_stripes_hi
;
158 __u32 filler
[4]; /* expansion area */
159 #define IMSM_ORD_REBUILD (1 << 24)
160 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
161 * top byte contains some flags
164 ASSERT_SIZE(imsm_map
, 52)
167 __u32 curr_migr_unit
;
168 __u32 checkpoint_id
; /* id to access curr_migr_unit */
169 __u8 migr_state
; /* Normal or Migrating */
171 #define MIGR_REBUILD 1
172 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
173 #define MIGR_GEN_MIGR 3
174 #define MIGR_STATE_CHANGE 4
175 #define MIGR_REPAIR 5
176 __u8 migr_type
; /* Initializing, Rebuilding, ... */
177 #define RAIDVOL_CLEAN 0
178 #define RAIDVOL_DIRTY 1
179 #define RAIDVOL_DSRECORD_VALID 2
181 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
182 __u16 verify_errors
; /* number of mismatches */
183 __u16 bad_blocks
; /* number of bad blocks during verify */
185 struct imsm_map map
[1];
186 /* here comes another one if migr_state */
188 ASSERT_SIZE(imsm_vol
, 84)
191 __u8 volume
[MAX_RAID_SERIAL_LEN
];
194 #define DEV_BOOTABLE __cpu_to_le32(0x01)
195 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
196 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
197 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
198 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
199 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
200 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
201 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
202 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
203 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
204 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
205 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
206 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
207 __u32 status
; /* Persistent RaidDev status */
208 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
212 __u8 cng_master_disk
;
216 __u16 my_vol_raid_dev_num
; /* Used in Unique volume Id for this RaidDev */
222 /* Unique Volume Id of the NvCache Volume associated with this volume */
223 __u32 nvc_vol_orig_family_num
;
224 __u16 nvc_vol_raid_dev_num
;
227 #define RWH_DISTRIBUTED 1
228 #define RWH_JOURNALING_DRIVE 2
229 #define RWH_MULTIPLE_DISTRIBUTED 3
230 #define RWH_MULTIPLE_PPLS_JOURNALING_DRIVE 4
231 #define RWH_MULTIPLE_OFF 5
232 __u8 rwh_policy
; /* Raid Write Hole Policy */
233 __u8 jd_serial
[MAX_RAID_SERIAL_LEN
]; /* Journal Drive serial number */
236 #define IMSM_DEV_FILLERS 3
237 __u32 filler
[IMSM_DEV_FILLERS
];
240 ASSERT_SIZE(imsm_dev
, 164)
243 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
244 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
245 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
246 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
247 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
248 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
249 __u32 attributes
; /* 0x34 - 0x37 */
250 __u8 num_disks
; /* 0x38 Number of configured disks */
251 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
252 __u8 error_log_pos
; /* 0x3A */
253 __u8 fill
[1]; /* 0x3B */
254 __u32 cache_size
; /* 0x3c - 0x40 in mb */
255 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
256 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
257 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
258 __u16 num_raid_devs_created
; /* 0x4C - 0x4D Used for generating unique
259 * volume IDs for raid_dev created in this array
262 __u16 filler1
; /* 0x4E - 0x4F */
263 #define IMSM_FILLERS 34
264 __u32 filler
[IMSM_FILLERS
]; /* 0x50 - 0xD7 RAID_MPB_FILLERS */
265 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
266 /* here comes imsm_dev[num_raid_devs] */
267 /* here comes BBM logs */
269 ASSERT_SIZE(imsm_super
, 264)
271 #define BBM_LOG_MAX_ENTRIES 254
272 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
273 #define BBM_LOG_SIGNATURE 0xabadb10c
275 struct bbm_log_block_addr
{
278 } __attribute__ ((__packed__
));
280 struct bbm_log_entry
{
281 __u8 marked_count
; /* Number of blocks marked - 1 */
282 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
283 struct bbm_log_block_addr defective_block_start
;
284 } __attribute__ ((__packed__
));
287 __u32 signature
; /* 0xABADB10C */
289 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
291 ASSERT_SIZE(bbm_log
, 2040)
293 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
295 #define BLOCKS_PER_KB (1024/512)
297 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
299 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
301 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
302 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
303 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
306 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
307 * be recovered using srcMap */
308 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
309 * already been migrated and must
310 * be recovered from checkpoint area */
312 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of single PPL, without the header */
315 __u32 rec_status
; /* Status used to determine how to restart
316 * migration in case it aborts
318 __u32 curr_migr_unit_lo
; /* 0..numMigrUnits-1 */
319 __u32 family_num
; /* Family number of MPB
320 * containing the RaidDev
321 * that is migrating */
322 __u32 ascending_migr
; /* True if migrating in increasing
324 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
325 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
327 * advances per unit-of-operation */
328 __u32 ckpt_area_pba_lo
; /* Pba of first block of ckpt copy area */
329 __u32 dest_1st_member_lba_lo
; /* First member lba on first
330 * stripe of destination */
331 __u32 num_migr_units_lo
; /* Total num migration units-of-op */
332 __u32 post_migr_vol_cap
; /* Size of volume after
333 * migration completes */
334 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
335 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
336 * migration ckpt record was read from
337 * (for recovered migrations) */
338 __u32 curr_migr_unit_hi
; /* 0..numMigrUnits-1 high order 32 bits */
339 __u32 ckpt_area_pba_hi
; /* Pba of first block of ckpt copy area
340 * high order 32 bits */
341 __u32 dest_1st_member_lba_hi
; /* First member lba on first stripe of
342 * destination - high order 32 bits */
343 __u32 num_migr_units_hi
; /* Total num migration units-of-op
344 * high order 32 bits */
346 ASSERT_SIZE(migr_record
, 64)
351 * 2: metadata does not match
359 struct md_list
*next
;
362 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
364 static __u8
migr_type(struct imsm_dev
*dev
)
366 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
367 dev
->status
& DEV_VERIFY_AND_FIX
)
370 return dev
->vol
.migr_type
;
373 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
375 /* for compatibility with older oroms convert MIGR_REPAIR, into
376 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
378 if (migr_type
== MIGR_REPAIR
) {
379 dev
->vol
.migr_type
= MIGR_VERIFY
;
380 dev
->status
|= DEV_VERIFY_AND_FIX
;
382 dev
->vol
.migr_type
= migr_type
;
383 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
387 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
389 return ROUND_UP(bytes
, sector_size
) / sector_size
;
392 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
393 unsigned int sector_size
)
395 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
399 struct imsm_dev
*dev
;
400 struct intel_dev
*next
;
405 enum sys_dev_type type
;
408 struct intel_hba
*next
;
415 /* internal representation of IMSM metadata */
418 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
419 struct imsm_super
*anchor
; /* immovable parameters */
422 void *migr_rec_buf
; /* buffer for I/O operations */
423 struct migr_record
*migr_rec
; /* migration record */
425 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
426 array, it indicates that mdmon is allowed to clean migration
428 size_t len
; /* size of the 'buf' allocation */
429 size_t extra_space
; /* extra space in 'buf' that is not used yet */
430 void *next_buf
; /* for realloc'ing buf from the manager */
432 int updates_pending
; /* count of pending updates for mdmon */
433 int current_vol
; /* index of raid device undergoing creation */
434 unsigned long long create_offset
; /* common start for 'current_vol' */
435 __u32 random
; /* random data for seeding new family numbers */
436 struct intel_dev
*devlist
;
437 unsigned int sector_size
; /* sector size of used member drives */
441 __u8 serial
[MAX_RAID_SERIAL_LEN
];
444 struct imsm_disk disk
;
447 struct extent
*e
; /* for determining freespace @ create */
448 int raiddisk
; /* slot to fill in autolayout */
450 } *disks
, *current_disk
;
451 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
453 struct dl
*missing
; /* disks removed while we weren't looking */
454 struct bbm_log
*bbm_log
;
455 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
456 const struct imsm_orom
*orom
; /* platform firmware support */
457 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
458 struct md_bb bb
; /* memory for get_bad_blocks call */
462 struct imsm_disk disk
;
463 #define IMSM_UNKNOWN_OWNER (-1)
465 struct intel_disk
*next
;
469 unsigned long long start
, size
;
472 /* definitions of reshape process types */
473 enum imsm_reshape_type
{
479 /* definition of messages passed to imsm_process_update */
480 enum imsm_update_type
{
481 update_activate_spare
,
485 update_add_remove_disk
,
486 update_reshape_container_disks
,
487 update_reshape_migration
,
489 update_general_migration_checkpoint
,
491 update_prealloc_badblocks_mem
,
495 struct imsm_update_activate_spare
{
496 enum imsm_update_type type
;
500 struct imsm_update_activate_spare
*next
;
506 unsigned long long size
;
513 enum takeover_direction
{
517 struct imsm_update_takeover
{
518 enum imsm_update_type type
;
520 enum takeover_direction direction
;
523 struct imsm_update_reshape
{
524 enum imsm_update_type type
;
528 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
531 struct imsm_update_reshape_migration
{
532 enum imsm_update_type type
;
535 /* fields for array migration changes
542 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
545 struct imsm_update_size_change
{
546 enum imsm_update_type type
;
551 struct imsm_update_general_migration_checkpoint
{
552 enum imsm_update_type type
;
553 __u32 curr_migr_unit
;
557 __u8 serial
[MAX_RAID_SERIAL_LEN
];
560 struct imsm_update_create_array
{
561 enum imsm_update_type type
;
566 struct imsm_update_kill_array
{
567 enum imsm_update_type type
;
571 struct imsm_update_rename_array
{
572 enum imsm_update_type type
;
573 __u8 name
[MAX_RAID_SERIAL_LEN
];
577 struct imsm_update_add_remove_disk
{
578 enum imsm_update_type type
;
581 struct imsm_update_prealloc_bb_mem
{
582 enum imsm_update_type type
;
585 struct imsm_update_rwh_policy
{
586 enum imsm_update_type type
;
591 static const char *_sys_dev_type
[] = {
592 [SYS_DEV_UNKNOWN
] = "Unknown",
593 [SYS_DEV_SAS
] = "SAS",
594 [SYS_DEV_SATA
] = "SATA",
595 [SYS_DEV_NVME
] = "NVMe",
596 [SYS_DEV_VMD
] = "VMD"
599 const char *get_sys_dev_type(enum sys_dev_type type
)
601 if (type
>= SYS_DEV_MAX
)
602 type
= SYS_DEV_UNKNOWN
;
604 return _sys_dev_type
[type
];
607 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
609 struct intel_hba
*result
= xmalloc(sizeof(*result
));
611 result
->type
= device
->type
;
612 result
->path
= xstrdup(device
->path
);
614 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
620 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
622 struct intel_hba
*result
;
624 for (result
= hba
; result
; result
= result
->next
) {
625 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
631 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
633 struct intel_hba
*hba
;
635 /* check if disk attached to Intel HBA */
636 hba
= find_intel_hba(super
->hba
, device
);
639 /* Check if HBA is already attached to super */
640 if (super
->hba
== NULL
) {
641 super
->hba
= alloc_intel_hba(device
);
646 /* Intel metadata allows for all disks attached to the same type HBA.
647 * Do not support HBA types mixing
649 if (device
->type
!= hba
->type
)
652 /* Multiple same type HBAs can be used if they share the same OROM */
653 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
655 if (device_orom
!= super
->orom
)
661 hba
->next
= alloc_intel_hba(device
);
665 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
667 struct sys_dev
*list
, *elem
;
670 if ((list
= find_intel_devices()) == NULL
)
674 disk_path
= (char *) devname
;
676 disk_path
= diskfd_to_devpath(fd
);
681 for (elem
= list
; elem
; elem
= elem
->next
)
682 if (path_attached_to_hba(disk_path
, elem
->path
))
685 if (disk_path
!= devname
)
691 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
694 static struct supertype
*match_metadata_desc_imsm(char *arg
)
696 struct supertype
*st
;
698 if (strcmp(arg
, "imsm") != 0 &&
699 strcmp(arg
, "default") != 0
703 st
= xcalloc(1, sizeof(*st
));
704 st
->ss
= &super_imsm
;
705 st
->max_devs
= IMSM_MAX_DEVICES
;
706 st
->minor_version
= 0;
711 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
713 return &mpb
->sig
[MPB_SIG_LEN
];
716 /* retrieve a disk directly from the anchor when the anchor is known to be
717 * up-to-date, currently only at load time
719 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
721 if (index
>= mpb
->num_disks
)
723 return &mpb
->disk
[index
];
726 /* retrieve the disk description based on a index of the disk
729 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
733 for (d
= super
->disks
; d
; d
= d
->next
)
734 if (d
->index
== index
)
739 /* retrieve a disk from the parsed metadata */
740 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
744 dl
= get_imsm_dl_disk(super
, index
);
751 /* generate a checksum directly from the anchor when the anchor is known to be
752 * up-to-date, currently only at load or write_super after coalescing
754 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
756 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
757 __u32
*p
= (__u32
*) mpb
;
761 sum
+= __le32_to_cpu(*p
);
765 return sum
- __le32_to_cpu(mpb
->check_sum
);
768 static size_t sizeof_imsm_map(struct imsm_map
*map
)
770 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
773 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
775 /* A device can have 2 maps if it is in the middle of a migration.
777 * MAP_0 - we return the first map
778 * MAP_1 - we return the second map if it exists, else NULL
779 * MAP_X - we return the second map if it exists, else the first
781 struct imsm_map
*map
= &dev
->vol
.map
[0];
782 struct imsm_map
*map2
= NULL
;
784 if (dev
->vol
.migr_state
)
785 map2
= (void *)map
+ sizeof_imsm_map(map
);
787 switch (second_map
) {
804 /* return the size of the device.
805 * migr_state increases the returned size if map[0] were to be duplicated
807 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
809 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
810 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
812 /* migrating means an additional map */
813 if (dev
->vol
.migr_state
)
814 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
816 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
821 /* retrieve disk serial number list from a metadata update */
822 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
825 struct disk_info
*inf
;
827 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
828 sizeof_imsm_dev(&update
->dev
, 0);
833 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
839 if (index
>= mpb
->num_raid_devs
)
842 /* devices start after all disks */
843 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
845 for (i
= 0; i
<= index
; i
++)
847 return _mpb
+ offset
;
849 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
854 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
856 struct intel_dev
*dv
;
858 if (index
>= super
->anchor
->num_raid_devs
)
860 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
861 if (dv
->index
== index
)
866 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
869 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
870 __le16_to_cpu(addr
->w1
));
873 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
875 struct bbm_log_block_addr addr
;
877 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
878 addr
.dw1
= __cpu_to_le32((__u32
)(sec
>> 16) & 0xffffffff);
882 /* get size of the bbm log */
883 static __u32
get_imsm_bbm_log_size(struct bbm_log
*log
)
885 if (!log
|| log
->entry_count
== 0)
888 return sizeof(log
->signature
) +
889 sizeof(log
->entry_count
) +
890 log
->entry_count
* sizeof(struct bbm_log_entry
);
893 /* check if bad block is not partially stored in bbm log */
894 static int is_stored_in_bbm(struct bbm_log
*log
, const __u8 idx
, const unsigned
895 long long sector
, const int length
, __u32
*pos
)
899 for (i
= *pos
; i
< log
->entry_count
; i
++) {
900 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
901 unsigned long long bb_start
;
902 unsigned long long bb_end
;
904 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
905 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
907 if ((entry
->disk_ordinal
== idx
) && (bb_start
>= sector
) &&
908 (bb_end
<= sector
+ length
)) {
916 /* record new bad block in bbm log */
917 static int record_new_badblock(struct bbm_log
*log
, const __u8 idx
, unsigned
918 long long sector
, int length
)
922 struct bbm_log_entry
*entry
= NULL
;
924 while (is_stored_in_bbm(log
, idx
, sector
, length
, &pos
)) {
925 struct bbm_log_entry
*e
= &log
->marked_block_entries
[pos
];
927 if ((e
->marked_count
+ 1 == BBM_LOG_MAX_LBA_ENTRY_VAL
) &&
928 (__le48_to_cpu(&e
->defective_block_start
) == sector
)) {
929 sector
+= BBM_LOG_MAX_LBA_ENTRY_VAL
;
930 length
-= BBM_LOG_MAX_LBA_ENTRY_VAL
;
939 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
940 BBM_LOG_MAX_LBA_ENTRY_VAL
;
941 entry
->defective_block_start
= __cpu_to_le48(sector
);
942 entry
->marked_count
= cnt
- 1;
949 new_bb
= ROUND_UP(length
, BBM_LOG_MAX_LBA_ENTRY_VAL
) /
950 BBM_LOG_MAX_LBA_ENTRY_VAL
;
951 if (log
->entry_count
+ new_bb
> BBM_LOG_MAX_ENTRIES
)
955 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
956 BBM_LOG_MAX_LBA_ENTRY_VAL
;
957 struct bbm_log_entry
*entry
=
958 &log
->marked_block_entries
[log
->entry_count
];
960 entry
->defective_block_start
= __cpu_to_le48(sector
);
961 entry
->marked_count
= cnt
- 1;
962 entry
->disk_ordinal
= idx
;
973 /* clear all bad blocks for given disk */
974 static void clear_disk_badblocks(struct bbm_log
*log
, const __u8 idx
)
978 while (i
< log
->entry_count
) {
979 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
981 if (entries
[i
].disk_ordinal
== idx
) {
982 if (i
< log
->entry_count
- 1)
983 entries
[i
] = entries
[log
->entry_count
- 1];
991 /* clear given bad block */
992 static int clear_badblock(struct bbm_log
*log
, const __u8 idx
, const unsigned
993 long long sector
, const int length
) {
996 while (i
< log
->entry_count
) {
997 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
999 if ((entries
[i
].disk_ordinal
== idx
) &&
1000 (__le48_to_cpu(&entries
[i
].defective_block_start
) ==
1001 sector
) && (entries
[i
].marked_count
+ 1 == length
)) {
1002 if (i
< log
->entry_count
- 1)
1003 entries
[i
] = entries
[log
->entry_count
- 1];
1013 /* allocate and load BBM log from metadata */
1014 static int load_bbm_log(struct intel_super
*super
)
1016 struct imsm_super
*mpb
= super
->anchor
;
1017 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1019 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
1020 if (!super
->bbm_log
)
1024 struct bbm_log
*log
= (void *)mpb
+
1025 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1029 if (bbm_log_size
< sizeof(log
->signature
) +
1030 sizeof(log
->entry_count
))
1033 entry_count
= __le32_to_cpu(log
->entry_count
);
1034 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
1035 (entry_count
> BBM_LOG_MAX_ENTRIES
))
1039 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
1040 entry_count
* sizeof(struct bbm_log_entry
))
1043 memcpy(super
->bbm_log
, log
, bbm_log_size
);
1045 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
1046 super
->bbm_log
->entry_count
= 0;
1052 /* checks if bad block is within volume boundaries */
1053 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
1054 const unsigned long long start_sector
,
1055 const unsigned long long size
)
1057 unsigned long long bb_start
;
1058 unsigned long long bb_end
;
1060 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1061 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1063 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1064 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1070 /* get list of bad blocks on a drive for a volume */
1071 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1072 const unsigned long long start_sector
,
1073 const unsigned long long size
,
1079 for (i
= 0; i
< log
->entry_count
; i
++) {
1080 const struct bbm_log_entry
*ent
=
1081 &log
->marked_block_entries
[i
];
1082 struct md_bb_entry
*bb
;
1084 if ((ent
->disk_ordinal
== idx
) &&
1085 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1087 if (!bbs
->entries
) {
1088 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1094 bb
= &bbs
->entries
[count
++];
1095 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1096 bb
->length
= ent
->marked_count
+ 1;
1104 * == MAP_0 get first map
1105 * == MAP_1 get second map
1106 * == MAP_X than get map according to the current migr_state
1108 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1112 struct imsm_map
*map
;
1114 map
= get_imsm_map(dev
, second_map
);
1116 /* top byte identifies disk under rebuild */
1117 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1120 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1121 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1123 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1125 return ord_to_idx(ord
);
1128 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1130 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1133 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
1138 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1139 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1140 if (ord_to_idx(ord
) == idx
)
1147 static int get_imsm_raid_level(struct imsm_map
*map
)
1149 if (map
->raid_level
== 1) {
1150 if (map
->num_members
== 2)
1156 return map
->raid_level
;
1159 static int cmp_extent(const void *av
, const void *bv
)
1161 const struct extent
*a
= av
;
1162 const struct extent
*b
= bv
;
1163 if (a
->start
< b
->start
)
1165 if (a
->start
> b
->start
)
1170 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1172 int memberships
= 0;
1175 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1176 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1177 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1179 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
1186 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1188 static int split_ull(unsigned long long n
, void *lo
, void *hi
)
1190 if (lo
== 0 || hi
== 0)
1192 __put_unaligned32(__cpu_to_le32((__u32
)n
), lo
);
1193 __put_unaligned32(__cpu_to_le32((n
>> 32)), hi
);
1197 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1199 return (unsigned long long)__le32_to_cpu(lo
) |
1200 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1203 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1207 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1210 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1214 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1217 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1221 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1224 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1228 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1231 static unsigned long long imsm_dev_size(struct imsm_dev
*dev
)
1235 return join_u32(dev
->size_low
, dev
->size_high
);
1238 static unsigned long long migr_chkp_area_pba(struct migr_record
*migr_rec
)
1240 if (migr_rec
== NULL
)
1242 return join_u32(migr_rec
->ckpt_area_pba_lo
,
1243 migr_rec
->ckpt_area_pba_hi
);
1246 static unsigned long long current_migr_unit(struct migr_record
*migr_rec
)
1248 if (migr_rec
== NULL
)
1250 return join_u32(migr_rec
->curr_migr_unit_lo
,
1251 migr_rec
->curr_migr_unit_hi
);
1254 static unsigned long long migr_dest_1st_member_lba(struct migr_record
*migr_rec
)
1256 if (migr_rec
== NULL
)
1258 return join_u32(migr_rec
->dest_1st_member_lba_lo
,
1259 migr_rec
->dest_1st_member_lba_hi
);
1262 static unsigned long long get_num_migr_units(struct migr_record
*migr_rec
)
1264 if (migr_rec
== NULL
)
1266 return join_u32(migr_rec
->num_migr_units_lo
,
1267 migr_rec
->num_migr_units_hi
);
1270 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1272 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1275 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1277 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1280 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1282 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1285 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1287 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1290 static void set_imsm_dev_size(struct imsm_dev
*dev
, unsigned long long n
)
1292 split_ull(n
, &dev
->size_low
, &dev
->size_high
);
1295 static void set_migr_chkp_area_pba(struct migr_record
*migr_rec
,
1296 unsigned long long n
)
1298 split_ull(n
, &migr_rec
->ckpt_area_pba_lo
, &migr_rec
->ckpt_area_pba_hi
);
1301 static void set_current_migr_unit(struct migr_record
*migr_rec
,
1302 unsigned long long n
)
1304 split_ull(n
, &migr_rec
->curr_migr_unit_lo
,
1305 &migr_rec
->curr_migr_unit_hi
);
1308 static void set_migr_dest_1st_member_lba(struct migr_record
*migr_rec
,
1309 unsigned long long n
)
1311 split_ull(n
, &migr_rec
->dest_1st_member_lba_lo
,
1312 &migr_rec
->dest_1st_member_lba_hi
);
1315 static void set_num_migr_units(struct migr_record
*migr_rec
,
1316 unsigned long long n
)
1318 split_ull(n
, &migr_rec
->num_migr_units_lo
,
1319 &migr_rec
->num_migr_units_hi
);
1322 static unsigned long long per_dev_array_size(struct imsm_map
*map
)
1324 unsigned long long array_size
= 0;
1329 array_size
= num_data_stripes(map
) * map
->blocks_per_strip
;
1330 if (get_imsm_raid_level(map
) == 1 || get_imsm_raid_level(map
) == 10)
1336 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
,
1337 int get_minimal_reservation
)
1339 /* find a list of used extents on the given physical device */
1340 struct extent
*rv
, *e
;
1342 int memberships
= count_memberships(dl
, super
);
1345 /* trim the reserved area for spares, so they can join any array
1346 * regardless of whether the OROM has assigned sectors from the
1347 * IMSM_RESERVED_SECTORS region
1349 if (dl
->index
== -1 || get_minimal_reservation
)
1350 reservation
= imsm_min_reserved_sectors(super
);
1352 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1354 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
1357 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1358 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1359 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1361 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1362 e
->start
= pba_of_lba0(map
);
1363 e
->size
= per_dev_array_size(map
);
1367 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1369 /* determine the start of the metadata
1370 * when no raid devices are defined use the default
1371 * ...otherwise allow the metadata to truncate the value
1372 * as is the case with older versions of imsm
1375 struct extent
*last
= &rv
[memberships
- 1];
1376 unsigned long long remainder
;
1378 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1379 /* round down to 1k block to satisfy precision of the kernel
1383 /* make sure remainder is still sane */
1384 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1385 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1386 if (reservation
> remainder
)
1387 reservation
= remainder
;
1389 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1394 /* try to determine how much space is reserved for metadata from
1395 * the last get_extents() entry, otherwise fallback to the
1398 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1404 /* for spares just return a minimal reservation which will grow
1405 * once the spare is picked up by an array
1407 if (dl
->index
== -1)
1408 return MPB_SECTOR_CNT
;
1410 e
= get_extents(super
, dl
, 0);
1412 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1414 /* scroll to last entry */
1415 for (i
= 0; e
[i
].size
; i
++)
1418 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1425 static int is_spare(struct imsm_disk
*disk
)
1427 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1430 static int is_configured(struct imsm_disk
*disk
)
1432 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1435 static int is_failed(struct imsm_disk
*disk
)
1437 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1440 static int is_journal(struct imsm_disk
*disk
)
1442 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1445 /* round array size down to closest MB and ensure it splits evenly
1448 static unsigned long long round_size_to_mb(unsigned long long size
, unsigned int
1452 size
= (size
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
1458 static int able_to_resync(int raid_level
, int missing_disks
)
1460 int max_missing_disks
= 0;
1462 switch (raid_level
) {
1464 max_missing_disks
= 1;
1467 max_missing_disks
= 0;
1469 return missing_disks
<= max_missing_disks
;
1472 /* try to determine how much space is reserved for metadata from
1473 * the last get_extents() entry on the smallest active disk,
1474 * otherwise fallback to the default
1476 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1480 unsigned long long min_active
;
1482 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1483 struct dl
*dl
, *dl_min
= NULL
;
1489 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1492 unsigned long long blocks
= total_blocks(&dl
->disk
);
1493 if (blocks
< min_active
|| min_active
== 0) {
1495 min_active
= blocks
;
1501 /* find last lba used by subarrays on the smallest active disk */
1502 e
= get_extents(super
, dl_min
, 0);
1505 for (i
= 0; e
[i
].size
; i
++)
1508 remainder
= min_active
- e
[i
].start
;
1511 /* to give priority to recovery we should not require full
1512 IMSM_RESERVED_SECTORS from the spare */
1513 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1515 /* if real reservation is smaller use that value */
1516 return (remainder
< rv
) ? remainder
: rv
;
1520 * Return minimum size of a spare and sector size
1521 * that can be used in this array
1523 int get_spare_criteria_imsm(struct supertype
*st
, struct spare_criteria
*c
)
1525 struct intel_super
*super
= st
->sb
;
1529 unsigned long long size
= 0;
1536 /* find first active disk in array */
1538 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1542 /* find last lba used by subarrays */
1543 e
= get_extents(super
, dl
, 0);
1546 for (i
= 0; e
[i
].size
; i
++)
1549 size
= e
[i
-1].start
+ e
[i
-1].size
;
1552 /* add the amount of space needed for metadata */
1553 size
+= imsm_min_reserved_sectors(super
);
1555 c
->min_size
= size
* 512;
1556 c
->sector_size
= super
->sector_size
;
1561 static int is_gen_migration(struct imsm_dev
*dev
);
1563 #define IMSM_4K_DIV 8
1565 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1566 struct imsm_dev
*dev
);
1568 static void print_imsm_dev(struct intel_super
*super
,
1569 struct imsm_dev
*dev
,
1575 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1576 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1580 printf("[%.16s]:\n", dev
->volume
);
1581 printf(" UUID : %s\n", uuid
);
1582 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1584 printf(" <-- %d", get_imsm_raid_level(map2
));
1586 printf(" Members : %d", map
->num_members
);
1588 printf(" <-- %d", map2
->num_members
);
1590 printf(" Slots : [");
1591 for (i
= 0; i
< map
->num_members
; i
++) {
1592 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1593 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1598 for (i
= 0; i
< map2
->num_members
; i
++) {
1599 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1600 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1605 printf(" Failed disk : ");
1606 if (map
->failed_disk_num
== 0xff)
1609 printf("%i", map
->failed_disk_num
);
1611 slot
= get_imsm_disk_slot(map
, disk_idx
);
1613 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1614 printf(" This Slot : %d%s\n", slot
,
1615 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1617 printf(" This Slot : ?\n");
1618 printf(" Sector Size : %u\n", super
->sector_size
);
1619 sz
= imsm_dev_size(dev
);
1620 printf(" Array Size : %llu%s\n",
1621 (unsigned long long)sz
* 512 / super
->sector_size
,
1622 human_size(sz
* 512));
1623 sz
= blocks_per_member(map
);
1624 printf(" Per Dev Size : %llu%s\n",
1625 (unsigned long long)sz
* 512 / super
->sector_size
,
1626 human_size(sz
* 512));
1627 printf(" Sector Offset : %llu\n",
1629 printf(" Num Stripes : %llu\n",
1630 num_data_stripes(map
));
1631 printf(" Chunk Size : %u KiB",
1632 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1634 printf(" <-- %u KiB",
1635 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1637 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1638 printf(" Migrate State : ");
1639 if (dev
->vol
.migr_state
) {
1640 if (migr_type(dev
) == MIGR_INIT
)
1641 printf("initialize\n");
1642 else if (migr_type(dev
) == MIGR_REBUILD
)
1643 printf("rebuild\n");
1644 else if (migr_type(dev
) == MIGR_VERIFY
)
1646 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1647 printf("general migration\n");
1648 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1649 printf("state change\n");
1650 else if (migr_type(dev
) == MIGR_REPAIR
)
1653 printf("<unknown:%d>\n", migr_type(dev
));
1656 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1657 if (dev
->vol
.migr_state
) {
1658 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1660 printf(" <-- %s", map_state_str
[map
->map_state
]);
1661 printf("\n Checkpoint : %u ",
1662 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1663 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1666 printf("(%llu)", (unsigned long long)
1667 blocks_per_migr_unit(super
, dev
));
1670 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1672 printf(" RWH Policy : ");
1673 if (dev
->rwh_policy
== RWH_OFF
|| dev
->rwh_policy
== RWH_MULTIPLE_OFF
)
1675 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1676 printf("PPL distributed\n");
1677 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1678 printf("PPL journaling drive\n");
1679 else if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
1680 printf("Multiple distributed PPLs\n");
1681 else if (dev
->rwh_policy
== RWH_MULTIPLE_PPLS_JOURNALING_DRIVE
)
1682 printf("Multiple PPLs on journaling drive\n");
1684 printf("<unknown:%d>\n", dev
->rwh_policy
);
1687 static void print_imsm_disk(struct imsm_disk
*disk
,
1690 unsigned int sector_size
) {
1691 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1694 if (index
< -1 || !disk
)
1698 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1700 printf(" Disk%02d Serial : %s\n", index
, str
);
1702 printf(" Disk Serial : %s\n", str
);
1703 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1704 is_configured(disk
) ? " active" : "",
1705 is_failed(disk
) ? " failed" : "",
1706 is_journal(disk
) ? " journal" : "");
1707 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1708 sz
= total_blocks(disk
) - reserved
;
1709 printf(" Usable Size : %llu%s\n",
1710 (unsigned long long)sz
* 512 / sector_size
,
1711 human_size(sz
* 512));
1714 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1716 struct migr_record
*migr_rec
= super
->migr_rec
;
1718 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1719 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1720 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1721 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1722 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1723 set_migr_chkp_area_pba(migr_rec
,
1724 migr_chkp_area_pba(migr_rec
) / IMSM_4K_DIV
);
1725 set_migr_dest_1st_member_lba(migr_rec
,
1726 migr_dest_1st_member_lba(migr_rec
) / IMSM_4K_DIV
);
1729 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1731 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1734 void convert_to_4k(struct intel_super
*super
)
1736 struct imsm_super
*mpb
= super
->anchor
;
1737 struct imsm_disk
*disk
;
1739 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1741 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1742 disk
= __get_imsm_disk(mpb
, i
);
1744 convert_to_4k_imsm_disk(disk
);
1746 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1747 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1748 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1750 set_imsm_dev_size(dev
, imsm_dev_size(dev
)/IMSM_4K_DIV
);
1751 dev
->vol
.curr_migr_unit
/= IMSM_4K_DIV
;
1754 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1755 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1756 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1758 if (dev
->vol
.migr_state
) {
1760 map
= get_imsm_map(dev
, MAP_1
);
1761 set_blocks_per_member(map
,
1762 blocks_per_member(map
)/IMSM_4K_DIV
);
1763 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1764 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1768 struct bbm_log
*log
= (void *)mpb
+
1769 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1772 for (i
= 0; i
< log
->entry_count
; i
++) {
1773 struct bbm_log_entry
*entry
=
1774 &log
->marked_block_entries
[i
];
1776 __u8 count
= entry
->marked_count
+ 1;
1777 unsigned long long sector
=
1778 __le48_to_cpu(&entry
->defective_block_start
);
1780 entry
->defective_block_start
=
1781 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1782 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1786 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1789 void examine_migr_rec_imsm(struct intel_super
*super
)
1791 struct migr_record
*migr_rec
= super
->migr_rec
;
1792 struct imsm_super
*mpb
= super
->anchor
;
1795 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1796 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1797 struct imsm_map
*map
;
1800 if (is_gen_migration(dev
) == 0)
1803 printf("\nMigration Record Information:");
1805 /* first map under migration */
1806 map
= get_imsm_map(dev
, MAP_0
);
1808 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1809 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1810 printf(" Empty\n ");
1811 printf("Examine one of first two disks in array\n");
1814 printf("\n Status : ");
1815 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1818 printf("Contains Data\n");
1819 printf(" Current Unit : %llu\n",
1820 current_migr_unit(migr_rec
));
1821 printf(" Family : %u\n",
1822 __le32_to_cpu(migr_rec
->family_num
));
1823 printf(" Ascending : %u\n",
1824 __le32_to_cpu(migr_rec
->ascending_migr
));
1825 printf(" Blocks Per Unit : %u\n",
1826 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1827 printf(" Dest. Depth Per Unit : %u\n",
1828 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1829 printf(" Checkpoint Area pba : %llu\n",
1830 migr_chkp_area_pba(migr_rec
));
1831 printf(" First member lba : %llu\n",
1832 migr_dest_1st_member_lba(migr_rec
));
1833 printf(" Total Number of Units : %llu\n",
1834 get_num_migr_units(migr_rec
));
1835 printf(" Size of volume : %llu\n",
1836 join_u32(migr_rec
->post_migr_vol_cap
,
1837 migr_rec
->post_migr_vol_cap_hi
));
1838 printf(" Record was read from : %u\n",
1839 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1845 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1847 struct migr_record
*migr_rec
= super
->migr_rec
;
1849 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1850 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1851 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1852 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1853 &migr_rec
->post_migr_vol_cap
,
1854 &migr_rec
->post_migr_vol_cap_hi
);
1855 set_migr_chkp_area_pba(migr_rec
,
1856 migr_chkp_area_pba(migr_rec
) * IMSM_4K_DIV
);
1857 set_migr_dest_1st_member_lba(migr_rec
,
1858 migr_dest_1st_member_lba(migr_rec
) * IMSM_4K_DIV
);
1861 void convert_from_4k(struct intel_super
*super
)
1863 struct imsm_super
*mpb
= super
->anchor
;
1864 struct imsm_disk
*disk
;
1866 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1868 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1869 disk
= __get_imsm_disk(mpb
, i
);
1871 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1874 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1875 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1876 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1878 set_imsm_dev_size(dev
, imsm_dev_size(dev
)*IMSM_4K_DIV
);
1879 dev
->vol
.curr_migr_unit
*= IMSM_4K_DIV
;
1882 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1883 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1884 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1886 if (dev
->vol
.migr_state
) {
1888 map
= get_imsm_map(dev
, MAP_1
);
1889 set_blocks_per_member(map
,
1890 blocks_per_member(map
)*IMSM_4K_DIV
);
1891 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1892 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1896 struct bbm_log
*log
= (void *)mpb
+
1897 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1900 for (i
= 0; i
< log
->entry_count
; i
++) {
1901 struct bbm_log_entry
*entry
=
1902 &log
->marked_block_entries
[i
];
1904 __u8 count
= entry
->marked_count
+ 1;
1905 unsigned long long sector
=
1906 __le48_to_cpu(&entry
->defective_block_start
);
1908 entry
->defective_block_start
=
1909 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1910 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1914 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1917 /*******************************************************************************
1918 * function: imsm_check_attributes
1919 * Description: Function checks if features represented by attributes flags
1920 * are supported by mdadm.
1922 * attributes - Attributes read from metadata
1924 * 0 - passed attributes contains unsupported features flags
1925 * 1 - all features are supported
1926 ******************************************************************************/
1927 static int imsm_check_attributes(__u32 attributes
)
1930 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1932 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1934 not_supported
&= attributes
;
1935 if (not_supported
) {
1936 pr_err("(IMSM): Unsupported attributes : %x\n",
1937 (unsigned)__le32_to_cpu(not_supported
));
1938 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1939 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1940 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1942 if (not_supported
& MPB_ATTRIB_2TB
) {
1943 dprintf("\t\tMPB_ATTRIB_2TB\n");
1944 not_supported
^= MPB_ATTRIB_2TB
;
1946 if (not_supported
& MPB_ATTRIB_RAID0
) {
1947 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1948 not_supported
^= MPB_ATTRIB_RAID0
;
1950 if (not_supported
& MPB_ATTRIB_RAID1
) {
1951 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1952 not_supported
^= MPB_ATTRIB_RAID1
;
1954 if (not_supported
& MPB_ATTRIB_RAID10
) {
1955 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1956 not_supported
^= MPB_ATTRIB_RAID10
;
1958 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1959 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1960 not_supported
^= MPB_ATTRIB_RAID1E
;
1962 if (not_supported
& MPB_ATTRIB_RAID5
) {
1963 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1964 not_supported
^= MPB_ATTRIB_RAID5
;
1966 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1967 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1968 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1970 if (not_supported
& MPB_ATTRIB_BBM
) {
1971 dprintf("\t\tMPB_ATTRIB_BBM\n");
1972 not_supported
^= MPB_ATTRIB_BBM
;
1974 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1975 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1976 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1978 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1979 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1980 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1982 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1983 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1984 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1986 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1987 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1988 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1990 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1991 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1992 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1996 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
2004 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
2006 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
2008 struct intel_super
*super
= st
->sb
;
2009 struct imsm_super
*mpb
= super
->anchor
;
2010 char str
[MAX_SIGNATURE_LENGTH
];
2015 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2018 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
2019 str
[MPB_SIG_LEN
-1] = '\0';
2020 printf(" Magic : %s\n", str
);
2021 printf(" Version : %s\n", get_imsm_version(mpb
));
2022 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
2023 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
2024 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
2025 printf(" Attributes : ");
2026 if (imsm_check_attributes(mpb
->attributes
))
2027 printf("All supported\n");
2029 printf("not supported\n");
2030 getinfo_super_imsm(st
, &info
, NULL
);
2031 fname_from_uuid(st
, &info
, nbuf
, ':');
2032 printf(" UUID : %s\n", nbuf
+ 5);
2033 sum
= __le32_to_cpu(mpb
->check_sum
);
2034 printf(" Checksum : %08x %s\n", sum
,
2035 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
2036 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
2037 printf(" Disks : %d\n", mpb
->num_disks
);
2038 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
2039 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
2040 super
->disks
->index
, reserved
, super
->sector_size
);
2041 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
2042 struct bbm_log
*log
= super
->bbm_log
;
2045 printf("Bad Block Management Log:\n");
2046 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
2047 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
2048 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
2050 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2052 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2054 super
->current_vol
= i
;
2055 getinfo_super_imsm(st
, &info
, NULL
);
2056 fname_from_uuid(st
, &info
, nbuf
, ':');
2057 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
2059 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2060 if (i
== super
->disks
->index
)
2062 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
2063 super
->sector_size
);
2066 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2067 if (dl
->index
== -1)
2068 print_imsm_disk(&dl
->disk
, -1, reserved
,
2069 super
->sector_size
);
2071 examine_migr_rec_imsm(super
);
2074 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
2076 /* We just write a generic IMSM ARRAY entry */
2079 struct intel_super
*super
= st
->sb
;
2081 if (!super
->anchor
->num_raid_devs
) {
2082 printf("ARRAY metadata=imsm\n");
2086 getinfo_super_imsm(st
, &info
, NULL
);
2087 fname_from_uuid(st
, &info
, nbuf
, ':');
2088 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
2091 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
2093 /* We just write a generic IMSM ARRAY entry */
2097 struct intel_super
*super
= st
->sb
;
2100 if (!super
->anchor
->num_raid_devs
)
2103 getinfo_super_imsm(st
, &info
, NULL
);
2104 fname_from_uuid(st
, &info
, nbuf
, ':');
2105 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2106 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2108 super
->current_vol
= i
;
2109 getinfo_super_imsm(st
, &info
, NULL
);
2110 fname_from_uuid(st
, &info
, nbuf1
, ':');
2111 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
2112 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
2116 static void export_examine_super_imsm(struct supertype
*st
)
2118 struct intel_super
*super
= st
->sb
;
2119 struct imsm_super
*mpb
= super
->anchor
;
2123 getinfo_super_imsm(st
, &info
, NULL
);
2124 fname_from_uuid(st
, &info
, nbuf
, ':');
2125 printf("MD_METADATA=imsm\n");
2126 printf("MD_LEVEL=container\n");
2127 printf("MD_UUID=%s\n", nbuf
+5);
2128 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
2131 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
2133 /* The second last sector of the device contains
2134 * the "struct imsm_super" metadata.
2135 * This contains mpb_size which is the size in bytes of the
2136 * extended metadata. This is located immediately before
2138 * We want to read all that, plus the last sector which
2139 * may contain a migration record, and write it all
2143 unsigned long long dsize
, offset
;
2145 struct imsm_super
*sb
;
2146 struct intel_super
*super
= st
->sb
;
2147 unsigned int sector_size
= super
->sector_size
;
2148 unsigned int written
= 0;
2150 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
) != 0)
2153 if (!get_dev_size(from
, NULL
, &dsize
))
2156 if (lseek64(from
, dsize
-(2*sector_size
), 0) < 0)
2158 if ((unsigned int)read(from
, buf
, sector_size
) != sector_size
)
2161 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
2164 sectors
= mpb_sectors(sb
, sector_size
) + 2;
2165 offset
= dsize
- sectors
* sector_size
;
2166 if (lseek64(from
, offset
, 0) < 0 ||
2167 lseek64(to
, offset
, 0) < 0)
2169 while (written
< sectors
* sector_size
) {
2170 int n
= sectors
*sector_size
- written
;
2173 if (read(from
, buf
, n
) != n
)
2175 if (write(to
, buf
, n
) != n
)
2186 static void detail_super_imsm(struct supertype
*st
, char *homehost
,
2191 struct intel_super
*super
= st
->sb
;
2192 int temp_vol
= super
->current_vol
;
2195 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2197 getinfo_super_imsm(st
, &info
, NULL
);
2198 fname_from_uuid(st
, &info
, nbuf
, ':');
2199 printf("\n UUID : %s\n", nbuf
+ 5);
2201 super
->current_vol
= temp_vol
;
2204 static void brief_detail_super_imsm(struct supertype
*st
, char *subarray
)
2208 struct intel_super
*super
= st
->sb
;
2209 int temp_vol
= super
->current_vol
;
2212 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2214 getinfo_super_imsm(st
, &info
, NULL
);
2215 fname_from_uuid(st
, &info
, nbuf
, ':');
2216 printf(" UUID=%s", nbuf
+ 5);
2218 super
->current_vol
= temp_vol
;
2221 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
,
2222 size_t serial_buf_len
);
2223 static void fd2devname(int fd
, char *name
);
2225 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2227 /* dump an unsorted list of devices attached to AHCI Intel storage
2228 * controller, as well as non-connected ports
2230 int hba_len
= strlen(hba_path
) + 1;
2235 unsigned long port_mask
= (1 << port_count
) - 1;
2237 if (port_count
> (int)sizeof(port_mask
) * 8) {
2239 pr_err("port_count %d out of range\n", port_count
);
2243 /* scroll through /sys/dev/block looking for devices attached to
2246 dir
= opendir("/sys/dev/block");
2250 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2261 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2263 path
= devt_to_devpath(makedev(major
, minor
));
2266 if (!path_attached_to_hba(path
, hba_path
)) {
2272 /* retrieve the scsi device type */
2273 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2275 pr_err("failed to allocate 'device'\n");
2279 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2280 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2282 pr_err("failed to read device type for %s\n",
2288 type
= strtoul(buf
, NULL
, 10);
2290 /* if it's not a disk print the vendor and model */
2291 if (!(type
== 0 || type
== 7 || type
== 14)) {
2294 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2295 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2296 strncpy(vendor
, buf
, sizeof(vendor
));
2297 vendor
[sizeof(vendor
) - 1] = '\0';
2298 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2299 while (isspace(*c
) || *c
== '\0')
2303 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2304 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2305 strncpy(model
, buf
, sizeof(model
));
2306 model
[sizeof(model
) - 1] = '\0';
2307 c
= (char *) &model
[sizeof(model
) - 1];
2308 while (isspace(*c
) || *c
== '\0')
2312 if (vendor
[0] && model
[0])
2313 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2315 switch (type
) { /* numbers from hald/linux/device.c */
2316 case 1: sprintf(buf
, "tape"); break;
2317 case 2: sprintf(buf
, "printer"); break;
2318 case 3: sprintf(buf
, "processor"); break;
2320 case 5: sprintf(buf
, "cdrom"); break;
2321 case 6: sprintf(buf
, "scanner"); break;
2322 case 8: sprintf(buf
, "media_changer"); break;
2323 case 9: sprintf(buf
, "comm"); break;
2324 case 12: sprintf(buf
, "raid"); break;
2325 default: sprintf(buf
, "unknown");
2331 /* chop device path to 'host%d' and calculate the port number */
2332 c
= strchr(&path
[hba_len
], '/');
2335 pr_err("%s - invalid path name\n", path
+ hba_len
);
2340 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2341 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2345 *c
= '/'; /* repair the full string */
2346 pr_err("failed to determine port number for %s\n",
2353 /* mark this port as used */
2354 port_mask
&= ~(1 << port
);
2356 /* print out the device information */
2358 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2362 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2364 printf(" Port%d : - disk info unavailable -\n", port
);
2366 fd2devname(fd
, buf
);
2367 printf(" Port%d : %s", port
, buf
);
2368 if (imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2370 printf(" (%s)\n", buf
);
2385 for (i
= 0; i
< port_count
; i
++)
2386 if (port_mask
& (1 << i
))
2387 printf(" Port%d : - no device attached -\n", i
);
2393 static int print_nvme_info(struct sys_dev
*hba
)
2401 dir
= opendir("/sys/block/");
2405 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2406 if (strstr(ent
->d_name
, "nvme")) {
2407 sprintf(buf
, "/sys/block/%s", ent
->d_name
);
2408 rp
= realpath(buf
, NULL
);
2411 if (path_attached_to_hba(rp
, hba
->path
)) {
2412 fd
= open_dev(ent
->d_name
);
2418 fd2devname(fd
, buf
);
2419 if (hba
->type
== SYS_DEV_VMD
)
2420 printf(" NVMe under VMD : %s", buf
);
2421 else if (hba
->type
== SYS_DEV_NVME
)
2422 printf(" NVMe Device : %s", buf
);
2423 if (!imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2425 printf(" (%s)\n", buf
);
2438 static void print_found_intel_controllers(struct sys_dev
*elem
)
2440 for (; elem
; elem
= elem
->next
) {
2441 pr_err("found Intel(R) ");
2442 if (elem
->type
== SYS_DEV_SATA
)
2443 fprintf(stderr
, "SATA ");
2444 else if (elem
->type
== SYS_DEV_SAS
)
2445 fprintf(stderr
, "SAS ");
2446 else if (elem
->type
== SYS_DEV_NVME
)
2447 fprintf(stderr
, "NVMe ");
2449 if (elem
->type
== SYS_DEV_VMD
)
2450 fprintf(stderr
, "VMD domain");
2452 fprintf(stderr
, "RAID controller");
2455 fprintf(stderr
, " at %s", elem
->pci_id
);
2456 fprintf(stderr
, ".\n");
2461 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2468 if ((dir
= opendir(hba_path
)) == NULL
)
2471 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2474 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2475 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2477 if (*port_count
== 0)
2479 else if (host
< host_base
)
2482 if (host
+ 1 > *port_count
+ host_base
)
2483 *port_count
= host
+ 1 - host_base
;
2489 static void print_imsm_capability(const struct imsm_orom
*orom
)
2491 printf(" Platform : Intel(R) ");
2492 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2493 printf("Matrix Storage Manager\n");
2494 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2495 printf("Virtual RAID on CPU\n");
2497 printf("Rapid Storage Technology%s\n",
2498 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2499 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2500 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2501 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2502 printf(" RAID Levels :%s%s%s%s%s\n",
2503 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2504 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2505 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2506 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2507 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2508 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2509 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2510 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2511 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2512 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2513 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2514 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2515 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2516 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2517 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2518 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2519 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2520 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2521 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2522 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2523 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2524 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2525 printf(" 2TB volumes :%s supported\n",
2526 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2527 printf(" 2TB disks :%s supported\n",
2528 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2529 printf(" Max Disks : %d\n", orom
->tds
);
2530 printf(" Max Volumes : %d per array, %d per %s\n",
2531 orom
->vpa
, orom
->vphba
,
2532 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2536 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2538 printf("MD_FIRMWARE_TYPE=imsm\n");
2539 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2540 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2541 orom
->hotfix_ver
, orom
->build
);
2542 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2543 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2544 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2545 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2546 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2547 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2548 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2549 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2550 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2551 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2552 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2553 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2554 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2555 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2556 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2557 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2558 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2559 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2560 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2561 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2562 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2563 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2564 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2565 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2566 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2567 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2568 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2569 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2572 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2574 /* There are two components to imsm platform support, the ahci SATA
2575 * controller and the option-rom. To find the SATA controller we
2576 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2577 * controller with the Intel vendor id is present. This approach
2578 * allows mdadm to leverage the kernel's ahci detection logic, with the
2579 * caveat that if ahci.ko is not loaded mdadm will not be able to
2580 * detect platform raid capabilities. The option-rom resides in a
2581 * platform "Adapter ROM". We scan for its signature to retrieve the
2582 * platform capabilities. If raid support is disabled in the BIOS the
2583 * option-rom capability structure will not be available.
2585 struct sys_dev
*list
, *hba
;
2590 if (enumerate_only
) {
2591 if (check_env("IMSM_NO_PLATFORM"))
2593 list
= find_intel_devices();
2596 for (hba
= list
; hba
; hba
= hba
->next
) {
2597 if (find_imsm_capability(hba
)) {
2607 list
= find_intel_devices();
2610 pr_err("no active Intel(R) RAID controller found.\n");
2612 } else if (verbose
> 0)
2613 print_found_intel_controllers(list
);
2615 for (hba
= list
; hba
; hba
= hba
->next
) {
2616 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2618 if (!find_imsm_capability(hba
)) {
2620 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2621 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2622 get_sys_dev_type(hba
->type
));
2628 if (controller_path
&& result
== 1) {
2629 pr_err("no active Intel(R) RAID controller found under %s\n",
2634 const struct orom_entry
*entry
;
2636 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2637 if (entry
->type
== SYS_DEV_VMD
) {
2638 print_imsm_capability(&entry
->orom
);
2639 printf(" 3rd party NVMe :%s supported\n",
2640 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2641 for (hba
= list
; hba
; hba
= hba
->next
) {
2642 if (hba
->type
== SYS_DEV_VMD
) {
2644 printf(" I/O Controller : %s (%s)\n",
2645 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2646 if (print_nvme_info(hba
)) {
2648 pr_err("failed to get devices attached to VMD domain.\n");
2657 print_imsm_capability(&entry
->orom
);
2658 if (entry
->type
== SYS_DEV_NVME
) {
2659 for (hba
= list
; hba
; hba
= hba
->next
) {
2660 if (hba
->type
== SYS_DEV_NVME
)
2661 print_nvme_info(hba
);
2667 struct devid_list
*devid
;
2668 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2669 hba
= device_by_id(devid
->devid
);
2673 printf(" I/O Controller : %s (%s)\n",
2674 hba
->path
, get_sys_dev_type(hba
->type
));
2675 if (hba
->type
== SYS_DEV_SATA
) {
2676 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2677 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2679 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2690 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2692 struct sys_dev
*list
, *hba
;
2695 list
= find_intel_devices();
2698 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2703 for (hba
= list
; hba
; hba
= hba
->next
) {
2704 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2706 if (!find_imsm_capability(hba
) && verbose
> 0) {
2708 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2709 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2715 const struct orom_entry
*entry
;
2717 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2718 if (entry
->type
== SYS_DEV_VMD
) {
2719 for (hba
= list
; hba
; hba
= hba
->next
)
2720 print_imsm_capability_export(&entry
->orom
);
2723 print_imsm_capability_export(&entry
->orom
);
2729 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2731 /* the imsm metadata format does not specify any host
2732 * identification information. We return -1 since we can never
2733 * confirm nor deny whether a given array is "meant" for this
2734 * host. We rely on compare_super and the 'family_num' fields to
2735 * exclude member disks that do not belong, and we rely on
2736 * mdadm.conf to specify the arrays that should be assembled.
2737 * Auto-assembly may still pick up "foreign" arrays.
2743 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2745 /* The uuid returned here is used for:
2746 * uuid to put into bitmap file (Create, Grow)
2747 * uuid for backup header when saving critical section (Grow)
2748 * comparing uuids when re-adding a device into an array
2749 * In these cases the uuid required is that of the data-array,
2750 * not the device-set.
2751 * uuid to recognise same set when adding a missing device back
2752 * to an array. This is a uuid for the device-set.
2754 * For each of these we can make do with a truncated
2755 * or hashed uuid rather than the original, as long as
2757 * In each case the uuid required is that of the data-array,
2758 * not the device-set.
2760 /* imsm does not track uuid's so we synthesis one using sha1 on
2761 * - The signature (Which is constant for all imsm array, but no matter)
2762 * - the orig_family_num of the container
2763 * - the index number of the volume
2764 * - the 'serial' number of the volume.
2765 * Hopefully these are all constant.
2767 struct intel_super
*super
= st
->sb
;
2770 struct sha1_ctx ctx
;
2771 struct imsm_dev
*dev
= NULL
;
2774 /* some mdadm versions failed to set ->orig_family_num, in which
2775 * case fall back to ->family_num. orig_family_num will be
2776 * fixed up with the first metadata update.
2778 family_num
= super
->anchor
->orig_family_num
;
2779 if (family_num
== 0)
2780 family_num
= super
->anchor
->family_num
;
2781 sha1_init_ctx(&ctx
);
2782 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2783 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2784 if (super
->current_vol
>= 0)
2785 dev
= get_imsm_dev(super
, super
->current_vol
);
2787 __u32 vol
= super
->current_vol
;
2788 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2789 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2791 sha1_finish_ctx(&ctx
, buf
);
2792 memcpy(uuid
, buf
, 4*4);
2797 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2799 __u8
*v
= get_imsm_version(mpb
);
2800 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2801 char major
[] = { 0, 0, 0 };
2802 char minor
[] = { 0 ,0, 0 };
2803 char patch
[] = { 0, 0, 0 };
2804 char *ver_parse
[] = { major
, minor
, patch
};
2808 while (*v
!= '\0' && v
< end
) {
2809 if (*v
!= '.' && j
< 2)
2810 ver_parse
[i
][j
++] = *v
;
2818 *m
= strtol(minor
, NULL
, 0);
2819 *p
= strtol(patch
, NULL
, 0);
2823 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2825 /* migr_strip_size when repairing or initializing parity */
2826 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2827 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2829 switch (get_imsm_raid_level(map
)) {
2834 return 128*1024 >> 9;
2838 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2840 /* migr_strip_size when rebuilding a degraded disk, no idea why
2841 * this is different than migr_strip_size_resync(), but it's good
2844 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2845 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2847 switch (get_imsm_raid_level(map
)) {
2850 if (map
->num_members
% map
->num_domains
== 0)
2851 return 128*1024 >> 9;
2855 return max((__u32
) 64*1024 >> 9, chunk
);
2857 return 128*1024 >> 9;
2861 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2863 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2864 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2865 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2866 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2868 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2871 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2873 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2874 int level
= get_imsm_raid_level(lo
);
2876 if (level
== 1 || level
== 10) {
2877 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2879 return hi
->num_domains
;
2881 return num_stripes_per_unit_resync(dev
);
2884 static __u8
imsm_num_data_members(struct imsm_map
*map
)
2886 /* named 'imsm_' because raid0, raid1 and raid10
2887 * counter-intuitively have the same number of data disks
2889 switch (get_imsm_raid_level(map
)) {
2891 return map
->num_members
;
2895 return map
->num_members
/2;
2897 return map
->num_members
- 1;
2899 dprintf("unsupported raid level\n");
2904 static unsigned long long calc_component_size(struct imsm_map
*map
,
2905 struct imsm_dev
*dev
)
2907 unsigned long long component_size
;
2908 unsigned long long dev_size
= imsm_dev_size(dev
);
2909 long long calc_dev_size
= 0;
2910 unsigned int member_disks
= imsm_num_data_members(map
);
2912 if (member_disks
== 0)
2915 component_size
= per_dev_array_size(map
);
2916 calc_dev_size
= component_size
* member_disks
;
2918 /* Component size is rounded to 1MB so difference between size from
2919 * metadata and size calculated from num_data_stripes equals up to
2920 * 2048 blocks per each device. If the difference is higher it means
2921 * that array size was expanded and num_data_stripes was not updated.
2923 if (llabs(calc_dev_size
- (long long)dev_size
) >
2924 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
2925 component_size
= dev_size
/ member_disks
;
2926 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
2927 component_size
/ map
->blocks_per_strip
,
2928 num_data_stripes(map
));
2931 return component_size
;
2934 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2936 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2937 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2939 switch(get_imsm_raid_level(map
)) {
2942 return chunk
* map
->num_domains
;
2944 return chunk
* map
->num_members
;
2950 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2952 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2953 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2954 __u32 strip
= block
/ chunk
;
2956 switch (get_imsm_raid_level(map
)) {
2959 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2960 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2962 return vol_stripe
* chunk
+ block
% chunk
;
2964 __u32 stripe
= strip
/ (map
->num_members
- 1);
2966 return stripe
* chunk
+ block
% chunk
;
2973 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2974 struct imsm_dev
*dev
)
2976 /* calculate the conversion factor between per member 'blocks'
2977 * (md/{resync,rebuild}_start) and imsm migration units, return
2978 * 0 for the 'not migrating' and 'unsupported migration' cases
2980 if (!dev
->vol
.migr_state
)
2983 switch (migr_type(dev
)) {
2984 case MIGR_GEN_MIGR
: {
2985 struct migr_record
*migr_rec
= super
->migr_rec
;
2986 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2991 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2992 __u32 stripes_per_unit
;
2993 __u32 blocks_per_unit
;
3002 /* yes, this is really the translation of migr_units to
3003 * per-member blocks in the 'resync' case
3005 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
3006 migr_chunk
= migr_strip_blocks_resync(dev
);
3007 disks
= imsm_num_data_members(map
);
3008 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
3009 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
3010 segment
= blocks_per_unit
/ stripe
;
3011 block_rel
= blocks_per_unit
- segment
* stripe
;
3012 parity_depth
= parity_segment_depth(dev
);
3013 block_map
= map_migr_block(dev
, block_rel
);
3014 return block_map
+ parity_depth
* segment
;
3016 case MIGR_REBUILD
: {
3017 __u32 stripes_per_unit
;
3020 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
3021 migr_chunk
= migr_strip_blocks_rebuild(dev
);
3022 return migr_chunk
* stripes_per_unit
;
3024 case MIGR_STATE_CHANGE
:
3030 static int imsm_level_to_layout(int level
)
3038 return ALGORITHM_LEFT_ASYMMETRIC
;
3045 /*******************************************************************************
3046 * Function: read_imsm_migr_rec
3047 * Description: Function reads imsm migration record from last sector of disk
3049 * fd : disk descriptor
3050 * super : metadata info
3054 ******************************************************************************/
3055 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
3058 unsigned int sector_size
= super
->sector_size
;
3059 unsigned long long dsize
;
3061 get_dev_size(fd
, NULL
, &dsize
);
3062 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
3064 pr_err("Cannot seek to anchor block: %s\n",
3068 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
3069 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3070 MIGR_REC_BUF_SECTORS
*sector_size
) {
3071 pr_err("Cannot read migr record block: %s\n",
3076 if (sector_size
== 4096)
3077 convert_from_4k_imsm_migr_rec(super
);
3083 static struct imsm_dev
*imsm_get_device_during_migration(
3084 struct intel_super
*super
)
3087 struct intel_dev
*dv
;
3089 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
3090 if (is_gen_migration(dv
->dev
))
3096 /*******************************************************************************
3097 * Function: load_imsm_migr_rec
3098 * Description: Function reads imsm migration record (it is stored at the last
3101 * super : imsm internal array info
3102 * info : general array info
3106 * -2 : no migration in progress
3107 ******************************************************************************/
3108 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
3115 struct imsm_dev
*dev
;
3116 struct imsm_map
*map
;
3119 /* find map under migration */
3120 dev
= imsm_get_device_during_migration(super
);
3121 /* nothing to load,no migration in progress?
3127 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
3128 /* read only from one of the first two slots */
3129 if ((sd
->disk
.raid_disk
< 0) ||
3130 (sd
->disk
.raid_disk
> 1))
3133 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3134 fd
= dev_open(nm
, O_RDONLY
);
3140 map
= get_imsm_map(dev
, MAP_0
);
3141 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3142 /* skip spare and failed disks
3146 /* read only from one of the first two slots */
3148 slot
= get_imsm_disk_slot(map
, dl
->index
);
3149 if (map
== NULL
|| slot
> 1 || slot
< 0)
3151 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3152 fd
= dev_open(nm
, O_RDONLY
);
3159 retval
= read_imsm_migr_rec(fd
, super
);
3167 /*******************************************************************************
3168 * function: imsm_create_metadata_checkpoint_update
3169 * Description: It creates update for checkpoint change.
3171 * super : imsm internal array info
3172 * u : pointer to prepared update
3175 * If length is equal to 0, input pointer u contains no update
3176 ******************************************************************************/
3177 static int imsm_create_metadata_checkpoint_update(
3178 struct intel_super
*super
,
3179 struct imsm_update_general_migration_checkpoint
**u
)
3182 int update_memory_size
= 0;
3184 dprintf("(enter)\n");
3190 /* size of all update data without anchor */
3191 update_memory_size
=
3192 sizeof(struct imsm_update_general_migration_checkpoint
);
3194 *u
= xcalloc(1, update_memory_size
);
3196 dprintf("error: cannot get memory\n");
3199 (*u
)->type
= update_general_migration_checkpoint
;
3200 (*u
)->curr_migr_unit
= current_migr_unit(super
->migr_rec
);
3201 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3203 return update_memory_size
;
3206 static void imsm_update_metadata_locally(struct supertype
*st
,
3207 void *buf
, int len
);
3209 /*******************************************************************************
3210 * Function: write_imsm_migr_rec
3211 * Description: Function writes imsm migration record
3212 * (at the last sector of disk)
3214 * super : imsm internal array info
3218 ******************************************************************************/
3219 static int write_imsm_migr_rec(struct supertype
*st
)
3221 struct intel_super
*super
= st
->sb
;
3222 unsigned int sector_size
= super
->sector_size
;
3223 unsigned long long dsize
;
3229 struct imsm_update_general_migration_checkpoint
*u
;
3230 struct imsm_dev
*dev
;
3231 struct imsm_map
*map
;
3233 /* find map under migration */
3234 dev
= imsm_get_device_during_migration(super
);
3235 /* if no migration, write buffer anyway to clear migr_record
3236 * on disk based on first available device
3239 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3240 super
->current_vol
);
3242 map
= get_imsm_map(dev
, MAP_0
);
3244 if (sector_size
== 4096)
3245 convert_to_4k_imsm_migr_rec(super
);
3246 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3249 /* skip failed and spare devices */
3252 /* write to 2 first slots only */
3254 slot
= get_imsm_disk_slot(map
, sd
->index
);
3255 if (map
== NULL
|| slot
> 1 || slot
< 0)
3258 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3259 fd
= dev_open(nm
, O_RDWR
);
3262 get_dev_size(fd
, NULL
, &dsize
);
3263 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3265 pr_err("Cannot seek to anchor block: %s\n",
3269 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3270 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3271 MIGR_REC_BUF_SECTORS
*sector_size
) {
3272 pr_err("Cannot write migr record block: %s\n",
3279 if (sector_size
== 4096)
3280 convert_from_4k_imsm_migr_rec(super
);
3281 /* update checkpoint information in metadata */
3282 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3284 dprintf("imsm: Cannot prepare update\n");
3287 /* update metadata locally */
3288 imsm_update_metadata_locally(st
, u
, len
);
3289 /* and possibly remotely */
3290 if (st
->update_tail
) {
3291 append_metadata_update(st
, u
, len
);
3292 /* during reshape we do all work inside metadata handler
3293 * manage_reshape(), so metadata update has to be triggered
3296 flush_metadata_updates(st
);
3297 st
->update_tail
= &st
->updates
;
3308 /* spare/missing disks activations are not allowe when
3309 * array/container performs reshape operation, because
3310 * all arrays in container works on the same disks set
3312 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3315 struct intel_dev
*i_dev
;
3316 struct imsm_dev
*dev
;
3318 /* check whole container
3320 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3322 if (is_gen_migration(dev
)) {
3323 /* No repair during any migration in container
3331 static unsigned long long imsm_component_size_alignment_check(int level
,
3333 unsigned int sector_size
,
3334 unsigned long long component_size
)
3336 unsigned int component_size_alignment
;
3338 /* check component size alignment
3340 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3342 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3343 level
, chunk_size
, component_size
,
3344 component_size_alignment
);
3346 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3347 dprintf("imsm: reported component size aligned from %llu ",
3349 component_size
-= component_size_alignment
;
3350 dprintf_cont("to %llu (%i).\n",
3351 component_size
, component_size_alignment
);
3354 return component_size
;
3357 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3359 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3360 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3362 return pba_of_lba0(map
) +
3363 (num_data_stripes(map
) * map
->blocks_per_strip
);
3366 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3368 struct intel_super
*super
= st
->sb
;
3369 struct migr_record
*migr_rec
= super
->migr_rec
;
3370 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3371 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3372 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3373 struct imsm_map
*map_to_analyse
= map
;
3375 int map_disks
= info
->array
.raid_disks
;
3377 memset(info
, 0, sizeof(*info
));
3379 map_to_analyse
= prev_map
;
3381 dl
= super
->current_disk
;
3383 info
->container_member
= super
->current_vol
;
3384 info
->array
.raid_disks
= map
->num_members
;
3385 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3386 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3387 info
->array
.md_minor
= -1;
3388 info
->array
.ctime
= 0;
3389 info
->array
.utime
= 0;
3390 info
->array
.chunk_size
=
3391 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3392 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3393 info
->custom_array_size
= imsm_dev_size(dev
);
3394 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3396 if (is_gen_migration(dev
)) {
3397 info
->reshape_active
= 1;
3398 info
->new_level
= get_imsm_raid_level(map
);
3399 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3400 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3401 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3402 if (info
->delta_disks
) {
3403 /* this needs to be applied to every array
3406 info
->reshape_active
= CONTAINER_RESHAPE
;
3408 /* We shape information that we give to md might have to be
3409 * modify to cope with md's requirement for reshaping arrays.
3410 * For example, when reshaping a RAID0, md requires it to be
3411 * presented as a degraded RAID4.
3412 * Also if a RAID0 is migrating to a RAID5 we need to specify
3413 * the array as already being RAID5, but the 'before' layout
3414 * is a RAID4-like layout.
3416 switch (info
->array
.level
) {
3418 switch(info
->new_level
) {
3420 /* conversion is happening as RAID4 */
3421 info
->array
.level
= 4;
3422 info
->array
.raid_disks
+= 1;
3425 /* conversion is happening as RAID5 */
3426 info
->array
.level
= 5;
3427 info
->array
.layout
= ALGORITHM_PARITY_N
;
3428 info
->delta_disks
-= 1;
3431 /* FIXME error message */
3432 info
->array
.level
= UnSet
;
3438 info
->new_level
= UnSet
;
3439 info
->new_layout
= UnSet
;
3440 info
->new_chunk
= info
->array
.chunk_size
;
3441 info
->delta_disks
= 0;
3445 info
->disk
.major
= dl
->major
;
3446 info
->disk
.minor
= dl
->minor
;
3447 info
->disk
.number
= dl
->index
;
3448 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3452 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3453 info
->component_size
= calc_component_size(map
, dev
);
3454 info
->component_size
= imsm_component_size_alignment_check(
3456 info
->array
.chunk_size
,
3458 info
->component_size
);
3459 info
->bb
.supported
= 1;
3461 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3462 info
->recovery_start
= MaxSector
;
3464 if (info
->array
.level
== 5 &&
3465 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3466 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3467 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3468 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3469 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3470 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3472 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3474 } else if (info
->array
.level
<= 0) {
3475 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3477 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3480 info
->reshape_progress
= 0;
3481 info
->resync_start
= MaxSector
;
3482 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3483 !(info
->array
.state
& 1)) &&
3484 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3485 info
->resync_start
= 0;
3487 if (dev
->vol
.migr_state
) {
3488 switch (migr_type(dev
)) {
3491 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3493 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3495 info
->resync_start
= blocks_per_unit
* units
;
3498 case MIGR_GEN_MIGR
: {
3499 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3501 __u64 units
= current_migr_unit(migr_rec
);
3502 unsigned long long array_blocks
;
3505 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3507 (get_num_migr_units(migr_rec
)-1)) &&
3508 (super
->migr_rec
->rec_status
==
3509 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3512 info
->reshape_progress
= blocks_per_unit
* units
;
3514 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3515 (unsigned long long)units
,
3516 (unsigned long long)blocks_per_unit
,
3517 info
->reshape_progress
);
3519 used_disks
= imsm_num_data_members(prev_map
);
3520 if (used_disks
> 0) {
3521 array_blocks
= per_dev_array_size(map
) *
3523 info
->custom_array_size
=
3524 round_size_to_mb(array_blocks
,
3530 /* we could emulate the checkpointing of
3531 * 'sync_action=check' migrations, but for now
3532 * we just immediately complete them
3535 /* this is handled by container_content_imsm() */
3536 case MIGR_STATE_CHANGE
:
3537 /* FIXME handle other migrations */
3539 /* we are not dirty, so... */
3540 info
->resync_start
= MaxSector
;
3544 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3545 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3547 info
->array
.major_version
= -1;
3548 info
->array
.minor_version
= -2;
3549 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3550 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3551 uuid_from_super_imsm(st
, info
->uuid
);
3555 for (i
=0; i
<map_disks
; i
++) {
3557 if (i
< info
->array
.raid_disks
) {
3558 struct imsm_disk
*dsk
;
3559 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3560 dsk
= get_imsm_disk(super
, j
);
3561 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3568 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3569 int failed
, int look_in_map
);
3571 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3574 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3576 if (is_gen_migration(dev
)) {
3579 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3581 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3582 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3583 if (map2
->map_state
!= map_state
) {
3584 map2
->map_state
= map_state
;
3585 super
->updates_pending
++;
3590 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3594 for (d
= super
->missing
; d
; d
= d
->next
)
3595 if (d
->index
== index
)
3600 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3602 struct intel_super
*super
= st
->sb
;
3603 struct imsm_disk
*disk
;
3604 int map_disks
= info
->array
.raid_disks
;
3605 int max_enough
= -1;
3607 struct imsm_super
*mpb
;
3609 if (super
->current_vol
>= 0) {
3610 getinfo_super_imsm_volume(st
, info
, map
);
3613 memset(info
, 0, sizeof(*info
));
3615 /* Set raid_disks to zero so that Assemble will always pull in valid
3618 info
->array
.raid_disks
= 0;
3619 info
->array
.level
= LEVEL_CONTAINER
;
3620 info
->array
.layout
= 0;
3621 info
->array
.md_minor
= -1;
3622 info
->array
.ctime
= 0; /* N/A for imsm */
3623 info
->array
.utime
= 0;
3624 info
->array
.chunk_size
= 0;
3626 info
->disk
.major
= 0;
3627 info
->disk
.minor
= 0;
3628 info
->disk
.raid_disk
= -1;
3629 info
->reshape_active
= 0;
3630 info
->array
.major_version
= -1;
3631 info
->array
.minor_version
= -2;
3632 strcpy(info
->text_version
, "imsm");
3633 info
->safe_mode_delay
= 0;
3634 info
->disk
.number
= -1;
3635 info
->disk
.state
= 0;
3637 info
->recovery_start
= MaxSector
;
3638 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3639 info
->bb
.supported
= 1;
3641 /* do we have the all the insync disks that we expect? */
3642 mpb
= super
->anchor
;
3643 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3645 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3646 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3647 int failed
, enough
, j
, missing
= 0;
3648 struct imsm_map
*map
;
3651 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3652 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3653 map
= get_imsm_map(dev
, MAP_0
);
3655 /* any newly missing disks?
3656 * (catches single-degraded vs double-degraded)
3658 for (j
= 0; j
< map
->num_members
; j
++) {
3659 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3660 __u32 idx
= ord_to_idx(ord
);
3662 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3663 info
->disk
.raid_disk
= j
;
3665 if (!(ord
& IMSM_ORD_REBUILD
) &&
3666 get_imsm_missing(super
, idx
)) {
3672 if (state
== IMSM_T_STATE_FAILED
)
3674 else if (state
== IMSM_T_STATE_DEGRADED
&&
3675 (state
!= map
->map_state
|| missing
))
3677 else /* we're normal, or already degraded */
3679 if (is_gen_migration(dev
) && missing
) {
3680 /* during general migration we need all disks
3681 * that process is running on.
3682 * No new missing disk is allowed.
3686 /* no more checks necessary
3690 /* in the missing/failed disk case check to see
3691 * if at least one array is runnable
3693 max_enough
= max(max_enough
, enough
);
3695 dprintf("enough: %d\n", max_enough
);
3696 info
->container_enough
= max_enough
;
3699 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3701 disk
= &super
->disks
->disk
;
3702 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3703 info
->component_size
= reserved
;
3704 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3705 /* we don't change info->disk.raid_disk here because
3706 * this state will be finalized in mdmon after we have
3707 * found the 'most fresh' version of the metadata
3709 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3710 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3711 0 : (1 << MD_DISK_SYNC
);
3714 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3715 * ->compare_super may have updated the 'num_raid_devs' field for spares
3717 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3718 uuid_from_super_imsm(st
, info
->uuid
);
3720 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3722 /* I don't know how to compute 'map' on imsm, so use safe default */
3725 for (i
= 0; i
< map_disks
; i
++)
3731 /* allocates memory and fills disk in mdinfo structure
3732 * for each disk in array */
3733 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3735 struct mdinfo
*mddev
;
3736 struct intel_super
*super
= st
->sb
;
3737 struct imsm_disk
*disk
;
3740 if (!super
|| !super
->disks
)
3743 mddev
= xcalloc(1, sizeof(*mddev
));
3747 tmp
= xcalloc(1, sizeof(*tmp
));
3749 tmp
->next
= mddev
->devs
;
3751 tmp
->disk
.number
= count
++;
3752 tmp
->disk
.major
= dl
->major
;
3753 tmp
->disk
.minor
= dl
->minor
;
3754 tmp
->disk
.state
= is_configured(disk
) ?
3755 (1 << MD_DISK_ACTIVE
) : 0;
3756 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3757 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3758 tmp
->disk
.raid_disk
= -1;
3764 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3765 char *update
, char *devname
, int verbose
,
3766 int uuid_set
, char *homehost
)
3768 /* For 'assemble' and 'force' we need to return non-zero if any
3769 * change was made. For others, the return value is ignored.
3770 * Update options are:
3771 * force-one : This device looks a bit old but needs to be included,
3772 * update age info appropriately.
3773 * assemble: clear any 'faulty' flag to allow this device to
3775 * force-array: Array is degraded but being forced, mark it clean
3776 * if that will be needed to assemble it.
3778 * newdev: not used ????
3779 * grow: Array has gained a new device - this is currently for
3781 * resync: mark as dirty so a resync will happen.
3782 * name: update the name - preserving the homehost
3783 * uuid: Change the uuid of the array to match watch is given
3785 * Following are not relevant for this imsm:
3786 * sparc2.2 : update from old dodgey metadata
3787 * super-minor: change the preferred_minor number
3788 * summaries: update redundant counters.
3789 * homehost: update the recorded homehost
3790 * _reshape_progress: record new reshape_progress position.
3793 struct intel_super
*super
= st
->sb
;
3794 struct imsm_super
*mpb
;
3796 /* we can only update container info */
3797 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3800 mpb
= super
->anchor
;
3802 if (strcmp(update
, "uuid") == 0) {
3803 /* We take this to mean that the family_num should be updated.
3804 * However that is much smaller than the uuid so we cannot really
3805 * allow an explicit uuid to be given. And it is hard to reliably
3807 * So if !uuid_set we know the current uuid is random and just used
3808 * the first 'int' and copy it to the other 3 positions.
3809 * Otherwise we require the 4 'int's to be the same as would be the
3810 * case if we are using a random uuid. So an explicit uuid will be
3811 * accepted as long as all for ints are the same... which shouldn't hurt
3814 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3817 if (info
->uuid
[0] != info
->uuid
[1] ||
3818 info
->uuid
[1] != info
->uuid
[2] ||
3819 info
->uuid
[2] != info
->uuid
[3])
3825 mpb
->orig_family_num
= info
->uuid
[0];
3826 } else if (strcmp(update
, "assemble") == 0)
3831 /* successful update? recompute checksum */
3833 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3838 static size_t disks_to_mpb_size(int disks
)
3842 size
= sizeof(struct imsm_super
);
3843 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3844 size
+= 2 * sizeof(struct imsm_dev
);
3845 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3846 size
+= (4 - 2) * sizeof(struct imsm_map
);
3847 /* 4 possible disk_ord_tbl's */
3848 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3849 /* maximum bbm log */
3850 size
+= sizeof(struct bbm_log
);
3855 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3856 unsigned long long data_offset
)
3858 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3861 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3864 static void free_devlist(struct intel_super
*super
)
3866 struct intel_dev
*dv
;
3868 while (super
->devlist
) {
3869 dv
= super
->devlist
->next
;
3870 free(super
->devlist
->dev
);
3871 free(super
->devlist
);
3872 super
->devlist
= dv
;
3876 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3878 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3881 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3885 * 0 same, or first was empty, and second was copied
3886 * 1 second had wrong number
3888 * 3 wrong other info
3890 struct intel_super
*first
= st
->sb
;
3891 struct intel_super
*sec
= tst
->sb
;
3898 /* in platform dependent environment test if the disks
3899 * use the same Intel hba
3900 * If not on Intel hba at all, allow anything.
3902 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3903 if (first
->hba
->type
!= sec
->hba
->type
) {
3905 "HBAs of devices do not match %s != %s\n",
3906 get_sys_dev_type(first
->hba
->type
),
3907 get_sys_dev_type(sec
->hba
->type
));
3910 if (first
->orom
!= sec
->orom
) {
3912 "HBAs of devices do not match %s != %s\n",
3913 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3918 /* if an anchor does not have num_raid_devs set then it is a free
3921 if (first
->anchor
->num_raid_devs
> 0 &&
3922 sec
->anchor
->num_raid_devs
> 0) {
3923 /* Determine if these disks might ever have been
3924 * related. Further disambiguation can only take place
3925 * in load_super_imsm_all
3927 __u32 first_family
= first
->anchor
->orig_family_num
;
3928 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3930 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3931 MAX_SIGNATURE_LENGTH
) != 0)
3934 if (first_family
== 0)
3935 first_family
= first
->anchor
->family_num
;
3936 if (sec_family
== 0)
3937 sec_family
= sec
->anchor
->family_num
;
3939 if (first_family
!= sec_family
)
3944 /* if 'first' is a spare promote it to a populated mpb with sec's
3947 if (first
->anchor
->num_raid_devs
== 0 &&
3948 sec
->anchor
->num_raid_devs
> 0) {
3950 struct intel_dev
*dv
;
3951 struct imsm_dev
*dev
;
3953 /* we need to copy raid device info from sec if an allocation
3954 * fails here we don't associate the spare
3956 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3957 dv
= xmalloc(sizeof(*dv
));
3958 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3961 dv
->next
= first
->devlist
;
3962 first
->devlist
= dv
;
3964 if (i
< sec
->anchor
->num_raid_devs
) {
3965 /* allocation failure */
3966 free_devlist(first
);
3967 pr_err("imsm: failed to associate spare\n");
3970 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3971 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3972 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3973 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3974 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3975 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3981 static void fd2devname(int fd
, char *name
)
3985 char dname
[PATH_MAX
];
3990 if (fstat(fd
, &st
) != 0)
3992 sprintf(path
, "/sys/dev/block/%d:%d",
3993 major(st
.st_rdev
), minor(st
.st_rdev
));
3995 rv
= readlink(path
, dname
, sizeof(dname
)-1);
4000 nm
= strrchr(dname
, '/');
4003 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
4007 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
4010 char *name
= fd2kname(fd
);
4015 if (strncmp(name
, "nvme", 4) != 0)
4018 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
4020 return load_sys(path
, buf
, buf_len
);
4023 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
4025 static int imsm_read_serial(int fd
, char *devname
,
4026 __u8
*serial
, size_t serial_buf_len
)
4035 memset(buf
, 0, sizeof(buf
));
4037 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
4040 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
4042 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
4043 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
4044 fd2devname(fd
, (char *) serial
);
4050 pr_err("Failed to retrieve serial for %s\n",
4055 /* trim all whitespace and non-printable characters and convert
4058 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4061 /* ':' is reserved for use in placeholder serial
4062 * numbers for missing disks
4073 if (len
> serial_buf_len
) {
4074 /* truncate leading characters */
4075 dest
+= len
- serial_buf_len
;
4076 len
= serial_buf_len
;
4079 memset(serial
, 0, serial_buf_len
);
4080 memcpy(serial
, dest
, len
);
4085 static int serialcmp(__u8
*s1
, __u8
*s2
)
4087 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4090 static void serialcpy(__u8
*dest
, __u8
*src
)
4092 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4095 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4099 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4100 if (serialcmp(dl
->serial
, serial
) == 0)
4106 static struct imsm_disk
*
4107 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4111 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4112 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4114 if (serialcmp(disk
->serial
, serial
) == 0) {
4125 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4127 struct imsm_disk
*disk
;
4132 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4134 rv
= imsm_read_serial(fd
, devname
, serial
, MAX_RAID_SERIAL_LEN
);
4139 dl
= xcalloc(1, sizeof(*dl
));
4142 dl
->major
= major(stb
.st_rdev
);
4143 dl
->minor
= minor(stb
.st_rdev
);
4144 dl
->next
= super
->disks
;
4145 dl
->fd
= keep_fd
? fd
: -1;
4146 assert(super
->disks
== NULL
);
4148 serialcpy(dl
->serial
, serial
);
4151 fd2devname(fd
, name
);
4153 dl
->devname
= xstrdup(devname
);
4155 dl
->devname
= xstrdup(name
);
4157 /* look up this disk's index in the current anchor */
4158 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4161 /* only set index on disks that are a member of a
4162 * populated contianer, i.e. one with raid_devs
4164 if (is_failed(&dl
->disk
))
4166 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4173 /* When migrating map0 contains the 'destination' state while map1
4174 * contains the current state. When not migrating map0 contains the
4175 * current state. This routine assumes that map[0].map_state is set to
4176 * the current array state before being called.
4178 * Migration is indicated by one of the following states
4179 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4180 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4181 * map1state=unitialized)
4182 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4184 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4185 * map1state=degraded)
4186 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4189 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4190 __u8 to_state
, int migr_type
)
4192 struct imsm_map
*dest
;
4193 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4195 dev
->vol
.migr_state
= 1;
4196 set_migr_type(dev
, migr_type
);
4197 dev
->vol
.curr_migr_unit
= 0;
4198 dest
= get_imsm_map(dev
, MAP_1
);
4200 /* duplicate and then set the target end state in map[0] */
4201 memcpy(dest
, src
, sizeof_imsm_map(src
));
4202 if (migr_type
== MIGR_GEN_MIGR
) {
4206 for (i
= 0; i
< src
->num_members
; i
++) {
4207 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4208 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4212 if (migr_type
== MIGR_GEN_MIGR
)
4213 /* Clear migration record */
4214 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4216 src
->map_state
= to_state
;
4219 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4222 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4223 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4227 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4228 * completed in the last migration.
4230 * FIXME add support for raid-level-migration
4232 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4233 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4234 /* when final map state is other than expected
4235 * merge maps (not for migration)
4239 for (i
= 0; i
< prev
->num_members
; i
++)
4240 for (j
= 0; j
< map
->num_members
; j
++)
4241 /* during online capacity expansion
4242 * disks position can be changed
4243 * if takeover is used
4245 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4246 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4247 map
->disk_ord_tbl
[j
] |=
4248 prev
->disk_ord_tbl
[i
];
4251 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4252 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4255 dev
->vol
.migr_state
= 0;
4256 set_migr_type(dev
, 0);
4257 dev
->vol
.curr_migr_unit
= 0;
4258 map
->map_state
= map_state
;
4261 static int parse_raid_devices(struct intel_super
*super
)
4264 struct imsm_dev
*dev_new
;
4265 size_t len
, len_migr
;
4267 size_t space_needed
= 0;
4268 struct imsm_super
*mpb
= super
->anchor
;
4270 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4271 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4272 struct intel_dev
*dv
;
4274 len
= sizeof_imsm_dev(dev_iter
, 0);
4275 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4277 space_needed
+= len_migr
- len
;
4279 dv
= xmalloc(sizeof(*dv
));
4280 if (max_len
< len_migr
)
4282 if (max_len
> len_migr
)
4283 space_needed
+= max_len
- len_migr
;
4284 dev_new
= xmalloc(max_len
);
4285 imsm_copy_dev(dev_new
, dev_iter
);
4288 dv
->next
= super
->devlist
;
4289 super
->devlist
= dv
;
4292 /* ensure that super->buf is large enough when all raid devices
4295 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4298 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4299 super
->sector_size
);
4300 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4303 memcpy(buf
, super
->buf
, super
->len
);
4304 memset(buf
+ super
->len
, 0, len
- super
->len
);
4310 super
->extra_space
+= space_needed
;
4315 /*******************************************************************************
4316 * Function: check_mpb_migr_compatibility
4317 * Description: Function checks for unsupported migration features:
4318 * - migration optimization area (pba_of_lba0)
4319 * - descending reshape (ascending_migr)
4321 * super : imsm metadata information
4323 * 0 : migration is compatible
4324 * -1 : migration is not compatible
4325 ******************************************************************************/
4326 int check_mpb_migr_compatibility(struct intel_super
*super
)
4328 struct imsm_map
*map0
, *map1
;
4329 struct migr_record
*migr_rec
= super
->migr_rec
;
4332 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4333 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4336 dev_iter
->vol
.migr_state
== 1 &&
4337 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4338 /* This device is migrating */
4339 map0
= get_imsm_map(dev_iter
, MAP_0
);
4340 map1
= get_imsm_map(dev_iter
, MAP_1
);
4341 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4342 /* migration optimization area was used */
4344 if (migr_rec
->ascending_migr
== 0 &&
4345 migr_rec
->dest_depth_per_unit
> 0)
4346 /* descending reshape not supported yet */
4353 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4355 /* load_imsm_mpb - read matrix metadata
4356 * allocates super->mpb to be freed by free_imsm
4358 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4360 unsigned long long dsize
;
4361 unsigned long long sectors
;
4362 unsigned int sector_size
= super
->sector_size
;
4364 struct imsm_super
*anchor
;
4367 get_dev_size(fd
, NULL
, &dsize
);
4368 if (dsize
< 2*sector_size
) {
4370 pr_err("%s: device to small for imsm\n",
4375 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4377 pr_err("Cannot seek to anchor block on %s: %s\n",
4378 devname
, strerror(errno
));
4382 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4384 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4387 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4389 pr_err("Cannot read anchor block on %s: %s\n",
4390 devname
, strerror(errno
));
4395 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4397 pr_err("no IMSM anchor on %s\n", devname
);
4402 __free_imsm(super
, 0);
4403 /* reload capability and hba */
4405 /* capability and hba must be updated with new super allocation */
4406 find_intel_hba_capability(fd
, super
, devname
);
4407 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4408 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4410 pr_err("unable to allocate %zu byte mpb buffer\n",
4415 memcpy(super
->buf
, anchor
, sector_size
);
4417 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4420 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4421 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4422 pr_err("could not allocate migr_rec buffer\n");
4426 super
->clean_migration_record_by_mdmon
= 0;
4429 check_sum
= __gen_imsm_checksum(super
->anchor
);
4430 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4432 pr_err("IMSM checksum %x != %x on %s\n",
4434 __le32_to_cpu(super
->anchor
->check_sum
),
4442 /* read the extended mpb */
4443 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4445 pr_err("Cannot seek to extended mpb on %s: %s\n",
4446 devname
, strerror(errno
));
4450 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4451 super
->len
- sector_size
) != super
->len
- sector_size
) {
4453 pr_err("Cannot read extended mpb on %s: %s\n",
4454 devname
, strerror(errno
));
4458 check_sum
= __gen_imsm_checksum(super
->anchor
);
4459 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4461 pr_err("IMSM checksum %x != %x on %s\n",
4462 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4470 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4472 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4473 static void clear_hi(struct intel_super
*super
)
4475 struct imsm_super
*mpb
= super
->anchor
;
4477 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4479 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4480 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4481 disk
->total_blocks_hi
= 0;
4483 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4484 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4487 for (n
= 0; n
< 2; ++n
) {
4488 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4491 map
->pba_of_lba0_hi
= 0;
4492 map
->blocks_per_member_hi
= 0;
4493 map
->num_data_stripes_hi
= 0;
4499 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4503 err
= load_imsm_mpb(fd
, super
, devname
);
4506 if (super
->sector_size
== 4096)
4507 convert_from_4k(super
);
4508 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4511 err
= parse_raid_devices(super
);
4514 err
= load_bbm_log(super
);
4519 static void __free_imsm_disk(struct dl
*d
)
4531 static void free_imsm_disks(struct intel_super
*super
)
4535 while (super
->disks
) {
4537 super
->disks
= d
->next
;
4538 __free_imsm_disk(d
);
4540 while (super
->disk_mgmt_list
) {
4541 d
= super
->disk_mgmt_list
;
4542 super
->disk_mgmt_list
= d
->next
;
4543 __free_imsm_disk(d
);
4545 while (super
->missing
) {
4547 super
->missing
= d
->next
;
4548 __free_imsm_disk(d
);
4553 /* free all the pieces hanging off of a super pointer */
4554 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4556 struct intel_hba
*elem
, *next
;
4562 /* unlink capability description */
4564 if (super
->migr_rec_buf
) {
4565 free(super
->migr_rec_buf
);
4566 super
->migr_rec_buf
= NULL
;
4569 free_imsm_disks(super
);
4570 free_devlist(super
);
4574 free((void *)elem
->path
);
4580 free(super
->bbm_log
);
4584 static void free_imsm(struct intel_super
*super
)
4586 __free_imsm(super
, 1);
4587 free(super
->bb
.entries
);
4591 static void free_super_imsm(struct supertype
*st
)
4593 struct intel_super
*super
= st
->sb
;
4602 static struct intel_super
*alloc_super(void)
4604 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4606 super
->current_vol
= -1;
4607 super
->create_offset
= ~((unsigned long long) 0);
4609 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4610 sizeof(struct md_bb_entry
));
4611 if (!super
->bb
.entries
) {
4620 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4622 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4624 struct sys_dev
*hba_name
;
4627 if (fd
>= 0 && test_partition(fd
)) {
4628 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4632 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4637 hba_name
= find_disk_attached_hba(fd
, NULL
);
4640 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4644 rv
= attach_hba_to_super(super
, hba_name
);
4647 struct intel_hba
*hba
= super
->hba
;
4649 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4650 " but the container is assigned to Intel(R) %s %s (",
4652 get_sys_dev_type(hba_name
->type
),
4653 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4654 hba_name
->pci_id
? : "Err!",
4655 get_sys_dev_type(super
->hba
->type
),
4656 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4659 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4661 fprintf(stderr
, ", ");
4664 fprintf(stderr
, ").\n"
4665 " Mixing devices attached to different controllers is not allowed.\n");
4669 super
->orom
= find_imsm_capability(hba_name
);
4676 /* find_missing - helper routine for load_super_imsm_all that identifies
4677 * disks that have disappeared from the system. This routine relies on
4678 * the mpb being uptodate, which it is at load time.
4680 static int find_missing(struct intel_super
*super
)
4683 struct imsm_super
*mpb
= super
->anchor
;
4685 struct imsm_disk
*disk
;
4687 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4688 disk
= __get_imsm_disk(mpb
, i
);
4689 dl
= serial_to_dl(disk
->serial
, super
);
4693 dl
= xmalloc(sizeof(*dl
));
4697 dl
->devname
= xstrdup("missing");
4699 serialcpy(dl
->serial
, disk
->serial
);
4702 dl
->next
= super
->missing
;
4703 super
->missing
= dl
;
4709 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4711 struct intel_disk
*idisk
= disk_list
;
4714 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4716 idisk
= idisk
->next
;
4722 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4723 struct intel_super
*super
,
4724 struct intel_disk
**disk_list
)
4726 struct imsm_disk
*d
= &super
->disks
->disk
;
4727 struct imsm_super
*mpb
= super
->anchor
;
4730 for (i
= 0; i
< tbl_size
; i
++) {
4731 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4732 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4734 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4735 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4736 dprintf("mpb from %d:%d matches %d:%d\n",
4737 super
->disks
->major
,
4738 super
->disks
->minor
,
4739 table
[i
]->disks
->major
,
4740 table
[i
]->disks
->minor
);
4744 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4745 is_configured(d
) == is_configured(tbl_d
)) &&
4746 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4747 /* current version of the mpb is a
4748 * better candidate than the one in
4749 * super_table, but copy over "cross
4750 * generational" status
4752 struct intel_disk
*idisk
;
4754 dprintf("mpb from %d:%d replaces %d:%d\n",
4755 super
->disks
->major
,
4756 super
->disks
->minor
,
4757 table
[i
]->disks
->major
,
4758 table
[i
]->disks
->minor
);
4760 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4761 if (idisk
&& is_failed(&idisk
->disk
))
4762 tbl_d
->status
|= FAILED_DISK
;
4765 struct intel_disk
*idisk
;
4766 struct imsm_disk
*disk
;
4768 /* tbl_mpb is more up to date, but copy
4769 * over cross generational status before
4772 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4773 if (disk
&& is_failed(disk
))
4774 d
->status
|= FAILED_DISK
;
4776 idisk
= disk_list_get(d
->serial
, *disk_list
);
4779 if (disk
&& is_configured(disk
))
4780 idisk
->disk
.status
|= CONFIGURED_DISK
;
4783 dprintf("mpb from %d:%d prefer %d:%d\n",
4784 super
->disks
->major
,
4785 super
->disks
->minor
,
4786 table
[i
]->disks
->major
,
4787 table
[i
]->disks
->minor
);
4795 table
[tbl_size
++] = super
;
4799 /* update/extend the merged list of imsm_disk records */
4800 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4801 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4802 struct intel_disk
*idisk
;
4804 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4806 idisk
->disk
.status
|= disk
->status
;
4807 if (is_configured(&idisk
->disk
) ||
4808 is_failed(&idisk
->disk
))
4809 idisk
->disk
.status
&= ~(SPARE_DISK
);
4811 idisk
= xcalloc(1, sizeof(*idisk
));
4812 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4813 idisk
->disk
= *disk
;
4814 idisk
->next
= *disk_list
;
4818 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4825 static struct intel_super
*
4826 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4829 struct imsm_super
*mpb
= super
->anchor
;
4833 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4834 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4835 struct intel_disk
*idisk
;
4837 idisk
= disk_list_get(disk
->serial
, disk_list
);
4839 if (idisk
->owner
== owner
||
4840 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4843 dprintf("'%.16s' owner %d != %d\n",
4844 disk
->serial
, idisk
->owner
,
4847 dprintf("unknown disk %x [%d]: %.16s\n",
4848 __le32_to_cpu(mpb
->family_num
), i
,
4854 if (ok_count
== mpb
->num_disks
)
4859 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4861 struct intel_super
*s
;
4863 for (s
= super_list
; s
; s
= s
->next
) {
4864 if (family_num
!= s
->anchor
->family_num
)
4866 pr_err("Conflict, offlining family %#x on '%s'\n",
4867 __le32_to_cpu(family_num
), s
->disks
->devname
);
4871 static struct intel_super
*
4872 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4874 struct intel_super
*super_table
[len
];
4875 struct intel_disk
*disk_list
= NULL
;
4876 struct intel_super
*champion
, *spare
;
4877 struct intel_super
*s
, **del
;
4882 memset(super_table
, 0, sizeof(super_table
));
4883 for (s
= *super_list
; s
; s
= s
->next
)
4884 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4886 for (i
= 0; i
< tbl_size
; i
++) {
4887 struct imsm_disk
*d
;
4888 struct intel_disk
*idisk
;
4889 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4892 d
= &s
->disks
->disk
;
4894 /* 'd' must appear in merged disk list for its
4895 * configuration to be valid
4897 idisk
= disk_list_get(d
->serial
, disk_list
);
4898 if (idisk
&& idisk
->owner
== i
)
4899 s
= validate_members(s
, disk_list
, i
);
4904 dprintf("marking family: %#x from %d:%d offline\n",
4906 super_table
[i
]->disks
->major
,
4907 super_table
[i
]->disks
->minor
);
4911 /* This is where the mdadm implementation differs from the Windows
4912 * driver which has no strict concept of a container. We can only
4913 * assemble one family from a container, so when returning a prodigal
4914 * array member to this system the code will not be able to disambiguate
4915 * the container contents that should be assembled ("foreign" versus
4916 * "local"). It requires user intervention to set the orig_family_num
4917 * to a new value to establish a new container. The Windows driver in
4918 * this situation fixes up the volume name in place and manages the
4919 * foreign array as an independent entity.
4924 for (i
= 0; i
< tbl_size
; i
++) {
4925 struct intel_super
*tbl_ent
= super_table
[i
];
4931 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4936 if (s
&& !is_spare
) {
4937 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4939 } else if (!s
&& !is_spare
)
4952 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4953 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4955 /* collect all dl's onto 'champion', and update them to
4956 * champion's version of the status
4958 for (s
= *super_list
; s
; s
= s
->next
) {
4959 struct imsm_super
*mpb
= champion
->anchor
;
4960 struct dl
*dl
= s
->disks
;
4965 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4967 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4968 struct imsm_disk
*disk
;
4970 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4973 /* only set index on disks that are a member of
4974 * a populated contianer, i.e. one with
4977 if (is_failed(&dl
->disk
))
4979 else if (is_spare(&dl
->disk
))
4985 if (i
>= mpb
->num_disks
) {
4986 struct intel_disk
*idisk
;
4988 idisk
= disk_list_get(dl
->serial
, disk_list
);
4989 if (idisk
&& is_spare(&idisk
->disk
) &&
4990 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4998 dl
->next
= champion
->disks
;
4999 champion
->disks
= dl
;
5003 /* delete 'champion' from super_list */
5004 for (del
= super_list
; *del
; ) {
5005 if (*del
== champion
) {
5006 *del
= (*del
)->next
;
5009 del
= &(*del
)->next
;
5011 champion
->next
= NULL
;
5015 struct intel_disk
*idisk
= disk_list
;
5017 disk_list
= disk_list
->next
;
5025 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
5026 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5027 int major
, int minor
, int keep_fd
);
5029 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5030 int *max
, int keep_fd
);
5032 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
5033 char *devname
, struct md_list
*devlist
,
5036 struct intel_super
*super_list
= NULL
;
5037 struct intel_super
*super
= NULL
;
5042 /* 'fd' is an opened container */
5043 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
5045 /* get super block from devlist devices */
5046 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
5049 /* all mpbs enter, maybe one leaves */
5050 super
= imsm_thunderdome(&super_list
, i
);
5056 if (find_missing(super
) != 0) {
5062 /* load migration record */
5063 err
= load_imsm_migr_rec(super
, NULL
);
5065 /* migration is in progress,
5066 * but migr_rec cannot be loaded,
5072 /* Check migration compatibility */
5073 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5074 pr_err("Unsupported migration detected");
5076 fprintf(stderr
, " on %s\n", devname
);
5078 fprintf(stderr
, " (IMSM).\n");
5087 while (super_list
) {
5088 struct intel_super
*s
= super_list
;
5090 super_list
= super_list
->next
;
5099 strcpy(st
->container_devnm
, fd2devnm(fd
));
5101 st
->container_devnm
[0] = 0;
5102 if (err
== 0 && st
->ss
== NULL
) {
5103 st
->ss
= &super_imsm
;
5104 st
->minor_version
= 0;
5105 st
->max_devs
= IMSM_MAX_DEVICES
;
5111 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5112 int *max
, int keep_fd
)
5114 struct md_list
*tmpdev
;
5118 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5119 if (tmpdev
->used
!= 1)
5121 if (tmpdev
->container
== 1) {
5123 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5125 pr_err("cannot open device %s: %s\n",
5126 tmpdev
->devname
, strerror(errno
));
5130 err
= get_sra_super_block(fd
, super_list
,
5131 tmpdev
->devname
, &lmax
,
5140 int major
= major(tmpdev
->st_rdev
);
5141 int minor
= minor(tmpdev
->st_rdev
);
5142 err
= get_super_block(super_list
,
5159 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5160 int major
, int minor
, int keep_fd
)
5162 struct intel_super
*s
;
5174 sprintf(nm
, "%d:%d", major
, minor
);
5175 dfd
= dev_open(nm
, O_RDWR
);
5181 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5182 find_intel_hba_capability(dfd
, s
, devname
);
5183 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5185 /* retry the load if we might have raced against mdmon */
5186 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5187 for (retry
= 0; retry
< 3; retry
++) {
5189 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5195 s
->next
= *super_list
;
5203 if (dfd
>= 0 && !keep_fd
)
5210 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5217 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5221 if (sra
->array
.major_version
!= -1 ||
5222 sra
->array
.minor_version
!= -2 ||
5223 strcmp(sra
->text_version
, "imsm") != 0) {
5228 devnm
= fd2devnm(fd
);
5229 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5230 if (get_super_block(super_list
, devnm
, devname
,
5231 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5242 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5244 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5247 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5249 struct intel_super
*super
;
5253 if (test_partition(fd
))
5254 /* IMSM not allowed on partitions */
5257 free_super_imsm(st
);
5259 super
= alloc_super();
5260 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5263 /* Load hba and capabilities if they exist.
5264 * But do not preclude loading metadata in case capabilities or hba are
5265 * non-compliant and ignore_hw_compat is set.
5267 rv
= find_intel_hba_capability(fd
, super
, devname
);
5268 /* no orom/efi or non-intel hba of the disk */
5269 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5271 pr_err("No OROM/EFI properties for %s\n", devname
);
5275 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5277 /* retry the load if we might have raced against mdmon */
5279 struct mdstat_ent
*mdstat
= NULL
;
5280 char *name
= fd2kname(fd
);
5283 mdstat
= mdstat_by_component(name
);
5285 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5286 for (retry
= 0; retry
< 3; retry
++) {
5288 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5294 free_mdstat(mdstat
);
5299 pr_err("Failed to load all information sections on %s\n", devname
);
5305 if (st
->ss
== NULL
) {
5306 st
->ss
= &super_imsm
;
5307 st
->minor_version
= 0;
5308 st
->max_devs
= IMSM_MAX_DEVICES
;
5311 /* load migration record */
5312 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5313 /* Check for unsupported migration features */
5314 if (check_mpb_migr_compatibility(super
) != 0) {
5315 pr_err("Unsupported migration detected");
5317 fprintf(stderr
, " on %s\n", devname
);
5319 fprintf(stderr
, " (IMSM).\n");
5327 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5329 if (info
->level
== 1)
5331 return info
->chunk_size
>> 9;
5334 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5335 unsigned long long size
)
5337 if (info
->level
== 1)
5340 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5343 static void imsm_update_version_info(struct intel_super
*super
)
5345 /* update the version and attributes */
5346 struct imsm_super
*mpb
= super
->anchor
;
5348 struct imsm_dev
*dev
;
5349 struct imsm_map
*map
;
5352 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5353 dev
= get_imsm_dev(super
, i
);
5354 map
= get_imsm_map(dev
, MAP_0
);
5355 if (__le32_to_cpu(dev
->size_high
) > 0)
5356 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5358 /* FIXME detect when an array spans a port multiplier */
5360 mpb
->attributes
|= MPB_ATTRIB_PM
;
5363 if (mpb
->num_raid_devs
> 1 ||
5364 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5365 version
= MPB_VERSION_ATTRIBS
;
5366 switch (get_imsm_raid_level(map
)) {
5367 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5368 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5369 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5370 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5373 if (map
->num_members
>= 5)
5374 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5375 else if (dev
->status
== DEV_CLONE_N_GO
)
5376 version
= MPB_VERSION_CNG
;
5377 else if (get_imsm_raid_level(map
) == 5)
5378 version
= MPB_VERSION_RAID5
;
5379 else if (map
->num_members
>= 3)
5380 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5381 else if (get_imsm_raid_level(map
) == 1)
5382 version
= MPB_VERSION_RAID1
;
5384 version
= MPB_VERSION_RAID0
;
5386 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5390 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5392 struct imsm_super
*mpb
= super
->anchor
;
5393 char *reason
= NULL
;
5395 size_t len
= strlen(name
);
5399 while (isspace(start
[len
- 1]))
5401 while (*start
&& isspace(*start
))
5403 memmove(name
, start
, len
+ 1);
5406 if (len
> MAX_RAID_SERIAL_LEN
)
5407 reason
= "must be 16 characters or less";
5409 reason
= "must be a non-empty string";
5411 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5412 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5414 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5415 reason
= "already exists";
5420 if (reason
&& !quiet
)
5421 pr_err("imsm volume name %s\n", reason
);
5426 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5427 struct shape
*s
, char *name
,
5428 char *homehost
, int *uuid
,
5429 long long data_offset
)
5431 /* We are creating a volume inside a pre-existing container.
5432 * so st->sb is already set.
5434 struct intel_super
*super
= st
->sb
;
5435 unsigned int sector_size
= super
->sector_size
;
5436 struct imsm_super
*mpb
= super
->anchor
;
5437 struct intel_dev
*dv
;
5438 struct imsm_dev
*dev
;
5439 struct imsm_vol
*vol
;
5440 struct imsm_map
*map
;
5441 int idx
= mpb
->num_raid_devs
;
5444 unsigned long long array_blocks
;
5445 size_t size_old
, size_new
;
5446 unsigned long long num_data_stripes
;
5447 unsigned int data_disks
;
5448 unsigned long long size_per_member
;
5450 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5451 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5455 /* ensure the mpb is large enough for the new data */
5456 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5457 size_new
= disks_to_mpb_size(info
->nr_disks
);
5458 if (size_new
> size_old
) {
5460 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5462 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5463 pr_err("could not allocate new mpb\n");
5466 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5467 MIGR_REC_BUF_SECTORS
*
5468 MAX_SECTOR_SIZE
) != 0) {
5469 pr_err("could not allocate migr_rec buffer\n");
5475 memcpy(mpb_new
, mpb
, size_old
);
5478 super
->anchor
= mpb_new
;
5479 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5480 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5481 super
->len
= size_round
;
5483 super
->current_vol
= idx
;
5485 /* handle 'failed_disks' by either:
5486 * a) create dummy disk entries in the table if this the first
5487 * volume in the array. We add them here as this is the only
5488 * opportunity to add them. add_to_super_imsm_volume()
5489 * handles the non-failed disks and continues incrementing
5491 * b) validate that 'failed_disks' matches the current number
5492 * of missing disks if the container is populated
5494 if (super
->current_vol
== 0) {
5496 for (i
= 0; i
< info
->failed_disks
; i
++) {
5497 struct imsm_disk
*disk
;
5500 disk
= __get_imsm_disk(mpb
, i
);
5501 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5502 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5503 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5504 "missing:%d", (__u8
)i
);
5506 find_missing(super
);
5511 for (d
= super
->missing
; d
; d
= d
->next
)
5513 if (info
->failed_disks
> missing
) {
5514 pr_err("unable to add 'missing' disk to container\n");
5519 if (!check_name(super
, name
, 0))
5521 dv
= xmalloc(sizeof(*dv
));
5522 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5524 * Explicitly allow truncating to not confuse gcc's
5525 * -Werror=stringop-truncation
5527 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5528 memcpy(dev
->volume
, name
, namelen
);
5529 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5530 info
->layout
, info
->chunk_size
,
5531 s
->size
* BLOCKS_PER_KB
);
5532 data_disks
= get_data_disks(info
->level
, info
->layout
,
5534 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5535 size_per_member
= array_blocks
/ data_disks
;
5537 set_imsm_dev_size(dev
, array_blocks
);
5538 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5540 vol
->migr_state
= 0;
5541 set_migr_type(dev
, MIGR_INIT
);
5542 vol
->dirty
= !info
->state
;
5543 vol
->curr_migr_unit
= 0;
5544 map
= get_imsm_map(dev
, MAP_0
);
5545 set_pba_of_lba0(map
, super
->create_offset
);
5546 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5547 map
->failed_disk_num
= ~0;
5548 if (info
->level
> 0)
5549 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5550 : IMSM_T_STATE_UNINITIALIZED
);
5552 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5553 IMSM_T_STATE_NORMAL
;
5556 if (info
->level
== 1 && info
->raid_disks
> 2) {
5559 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5563 map
->raid_level
= info
->level
;
5564 if (info
->level
== 10) {
5565 map
->raid_level
= 1;
5566 map
->num_domains
= info
->raid_disks
/ 2;
5567 } else if (info
->level
== 1)
5568 map
->num_domains
= info
->raid_disks
;
5570 map
->num_domains
= 1;
5572 /* info->size is only int so use the 'size' parameter instead */
5573 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5574 num_data_stripes
/= map
->num_domains
;
5575 set_num_data_stripes(map
, num_data_stripes
);
5577 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5578 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5582 map
->num_members
= info
->raid_disks
;
5583 for (i
= 0; i
< map
->num_members
; i
++) {
5584 /* initialized in add_to_super */
5585 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5587 mpb
->num_raid_devs
++;
5588 mpb
->num_raid_devs_created
++;
5589 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5591 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5592 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5593 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5594 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5598 pr_err("imsm does not support consistency policy %s\n",
5599 map_num(consistency_policies
, s
->consistency_policy
));
5604 dv
->index
= super
->current_vol
;
5605 dv
->next
= super
->devlist
;
5606 super
->devlist
= dv
;
5608 imsm_update_version_info(super
);
5613 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5614 struct shape
*s
, char *name
,
5615 char *homehost
, int *uuid
,
5616 unsigned long long data_offset
)
5618 /* This is primarily called by Create when creating a new array.
5619 * We will then get add_to_super called for each component, and then
5620 * write_init_super called to write it out to each device.
5621 * For IMSM, Create can create on fresh devices or on a pre-existing
5623 * To create on a pre-existing array a different method will be called.
5624 * This one is just for fresh drives.
5626 struct intel_super
*super
;
5627 struct imsm_super
*mpb
;
5631 if (data_offset
!= INVALID_SECTORS
) {
5632 pr_err("data-offset not supported by imsm\n");
5637 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5641 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5643 mpb_size
= MAX_SECTOR_SIZE
;
5645 super
= alloc_super();
5647 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5652 pr_err("could not allocate superblock\n");
5655 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5656 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5657 pr_err("could not allocate migr_rec buffer\n");
5662 memset(super
->buf
, 0, mpb_size
);
5664 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5668 /* zeroing superblock */
5672 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5674 version
= (char *) mpb
->sig
;
5675 strcpy(version
, MPB_SIGNATURE
);
5676 version
+= strlen(MPB_SIGNATURE
);
5677 strcpy(version
, MPB_VERSION_RAID0
);
5682 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5684 unsigned int member_sector_size
;
5687 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5691 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5693 if (member_sector_size
!= super
->sector_size
)
5698 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5699 int fd
, char *devname
)
5701 struct intel_super
*super
= st
->sb
;
5702 struct imsm_super
*mpb
= super
->anchor
;
5703 struct imsm_disk
*_disk
;
5704 struct imsm_dev
*dev
;
5705 struct imsm_map
*map
;
5709 dev
= get_imsm_dev(super
, super
->current_vol
);
5710 map
= get_imsm_map(dev
, MAP_0
);
5712 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5713 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5719 /* we're doing autolayout so grab the pre-marked (in
5720 * validate_geometry) raid_disk
5722 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5723 if (dl
->raiddisk
== dk
->raid_disk
)
5726 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5727 if (dl
->major
== dk
->major
&&
5728 dl
->minor
== dk
->minor
)
5733 pr_err("%s is not a member of the same container\n", devname
);
5737 if (mpb
->num_disks
== 0)
5738 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5739 &super
->sector_size
))
5742 if (!drive_validate_sector_size(super
, dl
)) {
5743 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5747 /* add a pristine spare to the metadata */
5748 if (dl
->index
< 0) {
5749 dl
->index
= super
->anchor
->num_disks
;
5750 super
->anchor
->num_disks
++;
5752 /* Check the device has not already been added */
5753 slot
= get_imsm_disk_slot(map
, dl
->index
);
5755 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5756 pr_err("%s has been included in this array twice\n",
5760 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5761 dl
->disk
.status
= CONFIGURED_DISK
;
5763 /* update size of 'missing' disks to be at least as large as the
5764 * largest acitve member (we only have dummy missing disks when
5765 * creating the first volume)
5767 if (super
->current_vol
== 0) {
5768 for (df
= super
->missing
; df
; df
= df
->next
) {
5769 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5770 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5771 _disk
= __get_imsm_disk(mpb
, df
->index
);
5776 /* refresh unset/failed slots to point to valid 'missing' entries */
5777 for (df
= super
->missing
; df
; df
= df
->next
)
5778 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5779 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5781 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5783 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5784 if (is_gen_migration(dev
)) {
5785 struct imsm_map
*map2
= get_imsm_map(dev
,
5787 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5788 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5789 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5792 if ((unsigned)df
->index
==
5794 set_imsm_ord_tbl_ent(map2
,
5800 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5804 /* if we are creating the first raid device update the family number */
5805 if (super
->current_vol
== 0) {
5807 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5809 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5810 if (!_dev
|| !_disk
) {
5811 pr_err("BUG mpb setup error\n");
5817 sum
+= __gen_imsm_checksum(mpb
);
5818 mpb
->family_num
= __cpu_to_le32(sum
);
5819 mpb
->orig_family_num
= mpb
->family_num
;
5821 super
->current_disk
= dl
;
5826 * Function marks disk as spare and restores disk serial
5827 * in case it was previously marked as failed by takeover operation
5829 * -1 : critical error
5830 * 0 : disk is marked as spare but serial is not set
5833 int mark_spare(struct dl
*disk
)
5835 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5842 if (!imsm_read_serial(disk
->fd
, NULL
, serial
, MAX_RAID_SERIAL_LEN
)) {
5843 /* Restore disk serial number, because takeover marks disk
5844 * as failed and adds to serial ':0' before it becomes
5847 serialcpy(disk
->serial
, serial
);
5848 serialcpy(disk
->disk
.serial
, serial
);
5851 disk
->disk
.status
= SPARE_DISK
;
5857 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5858 int fd
, char *devname
,
5859 unsigned long long data_offset
)
5861 struct intel_super
*super
= st
->sb
;
5863 unsigned long long size
;
5864 unsigned int member_sector_size
;
5869 /* If we are on an RAID enabled platform check that the disk is
5870 * attached to the raid controller.
5871 * We do not need to test disks attachment for container based additions,
5872 * they shall be already tested when container was created/assembled.
5874 rv
= find_intel_hba_capability(fd
, super
, devname
);
5875 /* no orom/efi or non-intel hba of the disk */
5877 dprintf("capability: %p fd: %d ret: %d\n",
5878 super
->orom
, fd
, rv
);
5882 if (super
->current_vol
>= 0)
5883 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5886 dd
= xcalloc(sizeof(*dd
), 1);
5887 dd
->major
= major(stb
.st_rdev
);
5888 dd
->minor
= minor(stb
.st_rdev
);
5889 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5892 dd
->action
= DISK_ADD
;
5893 rv
= imsm_read_serial(fd
, devname
, dd
->serial
, MAX_RAID_SERIAL_LEN
);
5895 pr_err("failed to retrieve scsi serial, aborting\n");
5901 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5902 (super
->hba
->type
== SYS_DEV_VMD
))) {
5904 char *devpath
= diskfd_to_devpath(fd
);
5905 char controller_path
[PATH_MAX
];
5908 pr_err("failed to get devpath, aborting\n");
5915 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5918 if (devpath_to_vendor(controller_path
) == 0x8086) {
5920 * If Intel's NVMe drive has serial ended with
5921 * "-A","-B","-1" or "-2" it means that this is "x8"
5922 * device (double drive on single PCIe card).
5923 * User should be warned about potential data loss.
5925 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5926 /* Skip empty character at the end */
5927 if (dd
->serial
[i
] == 0)
5930 if (((dd
->serial
[i
] == 'A') ||
5931 (dd
->serial
[i
] == 'B') ||
5932 (dd
->serial
[i
] == '1') ||
5933 (dd
->serial
[i
] == '2')) &&
5934 (dd
->serial
[i
-1] == '-'))
5935 pr_err("\tThe action you are about to take may put your data at risk.\n"
5936 "\tPlease note that x8 devices may consist of two separate x4 devices "
5937 "located on a single PCIe port.\n"
5938 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5941 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5942 !imsm_orom_has_tpv_support(super
->orom
)) {
5943 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5944 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
5951 get_dev_size(fd
, NULL
, &size
);
5952 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5954 if (super
->sector_size
== 0) {
5955 /* this a first device, so sector_size is not set yet */
5956 super
->sector_size
= member_sector_size
;
5959 /* clear migr_rec when adding disk to container */
5960 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5961 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5963 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5964 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5965 MIGR_REC_BUF_SECTORS
*member_sector_size
)
5966 perror("Write migr_rec failed");
5970 serialcpy(dd
->disk
.serial
, dd
->serial
);
5971 set_total_blocks(&dd
->disk
, size
);
5972 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5973 struct imsm_super
*mpb
= super
->anchor
;
5974 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5977 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5978 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5980 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5982 if (st
->update_tail
) {
5983 dd
->next
= super
->disk_mgmt_list
;
5984 super
->disk_mgmt_list
= dd
;
5986 dd
->next
= super
->disks
;
5988 super
->updates_pending
++;
5994 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5996 struct intel_super
*super
= st
->sb
;
5999 /* remove from super works only in mdmon - for communication
6000 * manager - monitor. Check if communication memory buffer
6003 if (!st
->update_tail
) {
6004 pr_err("shall be used in mdmon context only\n");
6007 dd
= xcalloc(1, sizeof(*dd
));
6008 dd
->major
= dk
->major
;
6009 dd
->minor
= dk
->minor
;
6012 dd
->action
= DISK_REMOVE
;
6014 dd
->next
= super
->disk_mgmt_list
;
6015 super
->disk_mgmt_list
= dd
;
6020 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
6023 char buf
[MAX_SECTOR_SIZE
];
6024 struct imsm_super anchor
;
6025 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
6027 /* spare records have their own family number and do not have any defined raid
6030 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
6032 struct imsm_super
*mpb
= super
->anchor
;
6033 struct imsm_super
*spare
= &spare_record
.anchor
;
6037 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
6038 spare
->generation_num
= __cpu_to_le32(1UL);
6039 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
6040 spare
->num_disks
= 1;
6041 spare
->num_raid_devs
= 0;
6042 spare
->cache_size
= mpb
->cache_size
;
6043 spare
->pwr_cycle_count
= __cpu_to_le32(1);
6045 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
6046 MPB_SIGNATURE MPB_VERSION_RAID0
);
6048 for (d
= super
->disks
; d
; d
= d
->next
) {
6052 spare
->disk
[0] = d
->disk
;
6053 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
6054 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6056 if (super
->sector_size
== 4096)
6057 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6059 sum
= __gen_imsm_checksum(spare
);
6060 spare
->family_num
= __cpu_to_le32(sum
);
6061 spare
->orig_family_num
= 0;
6062 sum
= __gen_imsm_checksum(spare
);
6063 spare
->check_sum
= __cpu_to_le32(sum
);
6065 if (store_imsm_mpb(d
->fd
, spare
)) {
6066 pr_err("failed for device %d:%d %s\n",
6067 d
->major
, d
->minor
, strerror(errno
));
6079 static int write_super_imsm(struct supertype
*st
, int doclose
)
6081 struct intel_super
*super
= st
->sb
;
6082 unsigned int sector_size
= super
->sector_size
;
6083 struct imsm_super
*mpb
= super
->anchor
;
6089 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6091 int clear_migration_record
= 1;
6094 /* 'generation' is incremented everytime the metadata is written */
6095 generation
= __le32_to_cpu(mpb
->generation_num
);
6097 mpb
->generation_num
= __cpu_to_le32(generation
);
6099 /* fix up cases where previous mdadm releases failed to set
6102 if (mpb
->orig_family_num
== 0)
6103 mpb
->orig_family_num
= mpb
->family_num
;
6105 for (d
= super
->disks
; d
; d
= d
->next
) {
6109 mpb
->disk
[d
->index
] = d
->disk
;
6113 for (d
= super
->missing
; d
; d
= d
->next
) {
6114 mpb
->disk
[d
->index
] = d
->disk
;
6117 mpb
->num_disks
= num_disks
;
6118 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6120 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6121 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6122 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6124 imsm_copy_dev(dev
, dev2
);
6125 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6127 if (is_gen_migration(dev2
))
6128 clear_migration_record
= 0;
6131 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6134 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6135 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6137 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6139 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6140 mpb_size
+= bbm_log_size
;
6141 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6144 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6147 /* recalculate checksum */
6148 sum
= __gen_imsm_checksum(mpb
);
6149 mpb
->check_sum
= __cpu_to_le32(sum
);
6151 if (super
->clean_migration_record_by_mdmon
) {
6152 clear_migration_record
= 1;
6153 super
->clean_migration_record_by_mdmon
= 0;
6155 if (clear_migration_record
)
6156 memset(super
->migr_rec_buf
, 0,
6157 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6159 if (sector_size
== 4096)
6160 convert_to_4k(super
);
6162 /* write the mpb for disks that compose raid devices */
6163 for (d
= super
->disks
; d
; d
= d
->next
) {
6164 if (d
->index
< 0 || is_failed(&d
->disk
))
6167 if (clear_migration_record
) {
6168 unsigned long long dsize
;
6170 get_dev_size(d
->fd
, NULL
, &dsize
);
6171 if (lseek64(d
->fd
, dsize
- sector_size
,
6173 if ((unsigned int)write(d
->fd
,
6174 super
->migr_rec_buf
,
6175 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6176 MIGR_REC_BUF_SECTORS
*sector_size
)
6177 perror("Write migr_rec failed");
6181 if (store_imsm_mpb(d
->fd
, mpb
))
6183 "failed for device %d:%d (fd: %d)%s\n",
6185 d
->fd
, strerror(errno
));
6194 return write_super_imsm_spares(super
, doclose
);
6199 static int create_array(struct supertype
*st
, int dev_idx
)
6202 struct imsm_update_create_array
*u
;
6203 struct intel_super
*super
= st
->sb
;
6204 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6205 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6206 struct disk_info
*inf
;
6207 struct imsm_disk
*disk
;
6210 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6211 sizeof(*inf
) * map
->num_members
;
6213 u
->type
= update_create_array
;
6214 u
->dev_idx
= dev_idx
;
6215 imsm_copy_dev(&u
->dev
, dev
);
6216 inf
= get_disk_info(u
);
6217 for (i
= 0; i
< map
->num_members
; i
++) {
6218 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6220 disk
= get_imsm_disk(super
, idx
);
6222 disk
= get_imsm_missing(super
, idx
);
6223 serialcpy(inf
[i
].serial
, disk
->serial
);
6225 append_metadata_update(st
, u
, len
);
6230 static int mgmt_disk(struct supertype
*st
)
6232 struct intel_super
*super
= st
->sb
;
6234 struct imsm_update_add_remove_disk
*u
;
6236 if (!super
->disk_mgmt_list
)
6241 u
->type
= update_add_remove_disk
;
6242 append_metadata_update(st
, u
, len
);
6247 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6249 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6251 struct ppl_header
*ppl_hdr
= buf
;
6254 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6256 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6258 perror("Failed to seek to PPL header location");
6262 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6264 perror("Write PPL header failed");
6273 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6275 struct intel_super
*super
= st
->sb
;
6277 struct ppl_header
*ppl_hdr
;
6280 /* first clear entire ppl space */
6281 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6285 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6287 pr_err("Failed to allocate PPL header buffer\n");
6291 memset(buf
, 0, PPL_HEADER_SIZE
);
6293 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6294 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6296 if (info
->mismatch_cnt
) {
6298 * We are overwriting an invalid ppl. Make one entry with wrong
6299 * checksum to prevent the kernel from skipping resync.
6301 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6302 ppl_hdr
->entries
[0].checksum
= ~0;
6305 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6311 static int is_rebuilding(struct imsm_dev
*dev
);
6313 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6314 struct mdinfo
*disk
)
6316 struct intel_super
*super
= st
->sb
;
6318 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6320 struct ppl_header
*ppl_hdr
= NULL
;
6322 struct imsm_dev
*dev
;
6325 unsigned long long ppl_offset
= 0;
6326 unsigned long long prev_gen_num
= 0;
6328 if (disk
->disk
.raid_disk
< 0)
6331 dev
= get_imsm_dev(super
, info
->container_member
);
6332 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6333 d
= get_imsm_dl_disk(super
, idx
);
6335 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6338 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6339 pr_err("Failed to allocate PPL header buffer\n");
6345 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6348 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6350 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6352 perror("Failed to seek to PPL header location");
6357 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6358 perror("Read PPL header failed");
6365 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6366 ppl_hdr
->checksum
= 0;
6368 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6369 dprintf("Wrong PPL header checksum on %s\n",
6374 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6375 /* previous was newest, it was already checked */
6379 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6380 super
->anchor
->orig_family_num
)) {
6381 dprintf("Wrong PPL header signature on %s\n",
6388 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6390 ppl_offset
+= PPL_HEADER_SIZE
;
6391 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6393 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6396 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6408 * Update metadata to use mutliple PPLs area (1MB).
6409 * This is done once for all RAID members
6411 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6412 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6414 struct mdinfo
*member_dev
;
6416 sprintf(subarray
, "%d", info
->container_member
);
6418 if (mdmon_running(st
->container_devnm
))
6419 st
->update_tail
= &st
->updates
;
6421 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6422 pr_err("Failed to update subarray %s\n",
6425 if (st
->update_tail
)
6426 flush_metadata_updates(st
);
6428 st
->ss
->sync_metadata(st
);
6429 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6430 for (member_dev
= info
->devs
; member_dev
;
6431 member_dev
= member_dev
->next
)
6432 member_dev
->ppl_size
=
6433 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6438 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6440 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6441 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6442 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6443 (is_rebuilding(dev
) &&
6444 dev
->vol
.curr_migr_unit
== 0 &&
6445 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6446 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6448 info
->mismatch_cnt
++;
6449 } else if (ret
== 0 &&
6450 ppl_hdr
->entries_count
== 0 &&
6451 is_rebuilding(dev
) &&
6452 info
->resync_start
== 0) {
6454 * The header has no entries - add a single empty entry and
6455 * rewrite the header to prevent the kernel from going into
6456 * resync after an interrupted rebuild.
6458 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6459 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6467 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6469 struct intel_super
*super
= st
->sb
;
6473 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6474 info
->array
.level
!= 5)
6477 for (d
= super
->disks
; d
; d
= d
->next
) {
6478 if (d
->index
< 0 || is_failed(&d
->disk
))
6481 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6489 static int write_init_super_imsm(struct supertype
*st
)
6491 struct intel_super
*super
= st
->sb
;
6492 int current_vol
= super
->current_vol
;
6496 getinfo_super_imsm(st
, &info
, NULL
);
6498 /* we are done with current_vol reset it to point st at the container */
6499 super
->current_vol
= -1;
6501 if (st
->update_tail
) {
6502 /* queue the recently created array / added disk
6503 * as a metadata update */
6505 /* determine if we are creating a volume or adding a disk */
6506 if (current_vol
< 0) {
6507 /* in the mgmt (add/remove) disk case we are running
6508 * in mdmon context, so don't close fd's
6512 rv
= write_init_ppl_imsm_all(st
, &info
);
6514 rv
= create_array(st
, current_vol
);
6518 for (d
= super
->disks
; d
; d
= d
->next
)
6519 Kill(d
->devname
, NULL
, 0, -1, 1);
6520 if (current_vol
>= 0)
6521 rv
= write_init_ppl_imsm_all(st
, &info
);
6523 rv
= write_super_imsm(st
, 1);
6529 static int store_super_imsm(struct supertype
*st
, int fd
)
6531 struct intel_super
*super
= st
->sb
;
6532 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6537 if (super
->sector_size
== 4096)
6538 convert_to_4k(super
);
6539 return store_imsm_mpb(fd
, mpb
);
6542 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6543 int layout
, int raiddisks
, int chunk
,
6544 unsigned long long size
,
6545 unsigned long long data_offset
,
6547 unsigned long long *freesize
,
6551 unsigned long long ldsize
;
6552 struct intel_super
*super
;
6555 if (level
!= LEVEL_CONTAINER
)
6560 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6563 pr_err("imsm: Cannot open %s: %s\n",
6564 dev
, strerror(errno
));
6567 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6572 /* capabilities retrieve could be possible
6573 * note that there is no fd for the disks in array.
6575 super
= alloc_super();
6580 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6586 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6590 fd2devname(fd
, str
);
6591 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6592 fd
, str
, super
->orom
, rv
, raiddisks
);
6594 /* no orom/efi or non-intel hba of the disk */
6601 if (raiddisks
> super
->orom
->tds
) {
6603 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6604 raiddisks
, super
->orom
->tds
);
6608 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6609 (ldsize
>> 9) >> 32 > 0) {
6611 pr_err("%s exceeds maximum platform supported size\n", dev
);
6617 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6623 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6625 const unsigned long long base_start
= e
[*idx
].start
;
6626 unsigned long long end
= base_start
+ e
[*idx
].size
;
6629 if (base_start
== end
)
6633 for (i
= *idx
; i
< num_extents
; i
++) {
6634 /* extend overlapping extents */
6635 if (e
[i
].start
>= base_start
&&
6636 e
[i
].start
<= end
) {
6639 if (e
[i
].start
+ e
[i
].size
> end
)
6640 end
= e
[i
].start
+ e
[i
].size
;
6641 } else if (e
[i
].start
> end
) {
6647 return end
- base_start
;
6650 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6652 /* build a composite disk with all known extents and generate a new
6653 * 'maxsize' given the "all disks in an array must share a common start
6654 * offset" constraint
6656 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6660 unsigned long long pos
;
6661 unsigned long long start
= 0;
6662 unsigned long long maxsize
;
6663 unsigned long reserve
;
6665 /* coalesce and sort all extents. also, check to see if we need to
6666 * reserve space between member arrays
6669 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6672 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6675 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6680 while (i
< sum_extents
) {
6681 e
[j
].start
= e
[i
].start
;
6682 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6684 if (e
[j
-1].size
== 0)
6693 unsigned long long esize
;
6695 esize
= e
[i
].start
- pos
;
6696 if (esize
>= maxsize
) {
6701 pos
= e
[i
].start
+ e
[i
].size
;
6703 } while (e
[i
-1].size
);
6709 /* FIXME assumes volume at offset 0 is the first volume in a
6712 if (start_extent
> 0)
6713 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6717 if (maxsize
< reserve
)
6720 super
->create_offset
= ~((unsigned long long) 0);
6721 if (start
+ reserve
> super
->create_offset
)
6722 return 0; /* start overflows create_offset */
6723 super
->create_offset
= start
+ reserve
;
6725 return maxsize
- reserve
;
6728 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6730 if (level
< 0 || level
== 6 || level
== 4)
6733 /* if we have an orom prevent invalid raid levels */
6736 case 0: return imsm_orom_has_raid0(orom
);
6739 return imsm_orom_has_raid1e(orom
);
6740 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6741 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6742 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6745 return 1; /* not on an Intel RAID platform so anything goes */
6751 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6752 int dpa
, int verbose
)
6754 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6755 struct mdstat_ent
*memb
;
6761 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6762 if (memb
->metadata_version
&&
6763 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6764 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6765 !is_subarray(memb
->metadata_version
+9) &&
6767 struct dev_member
*dev
= memb
->members
;
6769 while(dev
&& (fd
< 0)) {
6770 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6771 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6773 fd
= open(path
, O_RDONLY
, 0);
6774 if (num
<= 0 || fd
< 0) {
6775 pr_vrb("Cannot open %s: %s\n",
6776 dev
->name
, strerror(errno
));
6782 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6783 struct mdstat_ent
*vol
;
6784 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6785 if (vol
->active
> 0 &&
6786 vol
->metadata_version
&&
6787 is_container_member(vol
, memb
->devnm
)) {
6792 if (*devlist
&& (found
< dpa
)) {
6793 dv
= xcalloc(1, sizeof(*dv
));
6794 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6795 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6798 dv
->next
= *devlist
;
6806 free_mdstat(mdstat
);
6811 static struct md_list
*
6812 get_loop_devices(void)
6815 struct md_list
*devlist
= NULL
;
6818 for(i
= 0; i
< 12; i
++) {
6819 dv
= xcalloc(1, sizeof(*dv
));
6820 dv
->devname
= xmalloc(40);
6821 sprintf(dv
->devname
, "/dev/loop%d", i
);
6829 static struct md_list
*
6830 get_devices(const char *hba_path
)
6832 struct md_list
*devlist
= NULL
;
6839 devlist
= get_loop_devices();
6842 /* scroll through /sys/dev/block looking for devices attached to
6845 dir
= opendir("/sys/dev/block");
6846 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6851 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6853 path
= devt_to_devpath(makedev(major
, minor
));
6856 if (!path_attached_to_hba(path
, hba_path
)) {
6863 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6865 fd2devname(fd
, buf
);
6868 pr_err("cannot open device: %s\n",
6873 dv
= xcalloc(1, sizeof(*dv
));
6874 dv
->devname
= xstrdup(buf
);
6881 devlist
= devlist
->next
;
6891 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6892 int verbose
, int *found
)
6894 struct md_list
*tmpdev
;
6896 struct supertype
*st
;
6898 /* first walk the list of devices to find a consistent set
6899 * that match the criterea, if that is possible.
6900 * We flag the ones we like with 'used'.
6903 st
= match_metadata_desc_imsm("imsm");
6905 pr_vrb("cannot allocate memory for imsm supertype\n");
6909 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6910 char *devname
= tmpdev
->devname
;
6912 struct supertype
*tst
;
6914 if (tmpdev
->used
> 1)
6916 tst
= dup_super(st
);
6918 pr_vrb("cannot allocate memory for imsm supertype\n");
6921 tmpdev
->container
= 0;
6922 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6924 dprintf("cannot open device %s: %s\n",
6925 devname
, strerror(errno
));
6927 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
6929 } else if (must_be_container(dfd
)) {
6930 struct supertype
*cst
;
6931 cst
= super_by_fd(dfd
, NULL
);
6933 dprintf("cannot recognize container type %s\n",
6936 } else if (tst
->ss
!= st
->ss
) {
6937 dprintf("non-imsm container - ignore it: %s\n",
6940 } else if (!tst
->ss
->load_container
||
6941 tst
->ss
->load_container(tst
, dfd
, NULL
))
6944 tmpdev
->container
= 1;
6947 cst
->ss
->free_super(cst
);
6949 tmpdev
->st_rdev
= rdev
;
6950 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6951 dprintf("no RAID superblock on %s\n",
6954 } else if (tst
->ss
->compare_super
== NULL
) {
6955 dprintf("Cannot assemble %s metadata on %s\n",
6956 tst
->ss
->name
, devname
);
6962 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6963 /* Ignore unrecognised devices during auto-assembly */
6968 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6970 if (st
->minor_version
== -1)
6971 st
->minor_version
= tst
->minor_version
;
6973 if (memcmp(info
.uuid
, uuid_zero
,
6974 sizeof(int[4])) == 0) {
6975 /* this is a floating spare. It cannot define
6976 * an array unless there are no more arrays of
6977 * this type to be found. It can be included
6978 * in an array of this type though.
6984 if (st
->ss
!= tst
->ss
||
6985 st
->minor_version
!= tst
->minor_version
||
6986 st
->ss
->compare_super(st
, tst
) != 0) {
6987 /* Some mismatch. If exactly one array matches this host,
6988 * we can resolve on that one.
6989 * Or, if we are auto assembling, we just ignore the second
6992 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6998 dprintf("found: devname: %s\n", devname
);
7002 tst
->ss
->free_super(tst
);
7006 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
7007 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
7008 for (iter
= head
; iter
; iter
= iter
->next
) {
7009 dprintf("content->text_version: %s vol\n",
7010 iter
->text_version
);
7011 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
7012 /* do not assemble arrays with unsupported
7014 dprintf("Cannot activate member %s.\n",
7015 iter
->text_version
);
7022 dprintf("No valid super block on device list: err: %d %p\n",
7026 dprintf("no more devices to examine\n");
7029 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
7030 if (tmpdev
->used
== 1 && tmpdev
->found
) {
7032 if (count
< tmpdev
->found
)
7035 count
-= tmpdev
->found
;
7038 if (tmpdev
->used
== 1)
7043 st
->ss
->free_super(st
);
7047 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
7050 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
7052 const struct orom_entry
*entry
;
7053 struct devid_list
*dv
, *devid_list
;
7058 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7059 if (strstr(idev
->path
, hba_path
))
7063 if (!idev
|| !idev
->dev_id
)
7066 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7068 if (!entry
|| !entry
->devid_list
)
7071 devid_list
= entry
->devid_list
;
7072 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7073 struct md_list
*devlist
;
7074 struct sys_dev
*device
= NULL
;
7079 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7081 device
= device_by_id(dv
->devid
);
7084 hpath
= device
->path
;
7088 devlist
= get_devices(hpath
);
7089 /* if no intel devices return zero volumes */
7090 if (devlist
== NULL
)
7093 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7095 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7096 if (devlist
== NULL
)
7100 count
+= count_volumes_list(devlist
,
7104 dprintf("found %d count: %d\n", found
, count
);
7107 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7110 struct md_list
*dv
= devlist
;
7111 devlist
= devlist
->next
;
7119 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7123 if (hba
->type
== SYS_DEV_VMD
) {
7124 struct sys_dev
*dev
;
7127 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7128 if (dev
->type
== SYS_DEV_VMD
)
7129 count
+= __count_volumes(dev
->path
, dpa
,
7134 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7137 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7139 /* up to 512 if the plaform supports it, otherwise the platform max.
7140 * 128 if no platform detected
7142 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7144 return min(512, (1 << fs
));
7148 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7149 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7151 /* check/set platform and metadata limits/defaults */
7152 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7153 pr_vrb("platform supports a maximum of %d disks per array\n",
7158 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7159 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7160 pr_vrb("platform does not support raid%d with %d disk%s\n",
7161 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7165 if (*chunk
== 0 || *chunk
== UnSet
)
7166 *chunk
= imsm_default_chunk(super
->orom
);
7168 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7169 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7173 if (layout
!= imsm_level_to_layout(level
)) {
7175 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7176 else if (level
== 10)
7177 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7179 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7184 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7185 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7186 pr_vrb("platform does not support a volume size over 2TB\n");
7193 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7194 * FIX ME add ahci details
7196 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7197 int layout
, int raiddisks
, int *chunk
,
7198 unsigned long long size
,
7199 unsigned long long data_offset
,
7201 unsigned long long *freesize
,
7205 struct intel_super
*super
= st
->sb
;
7206 struct imsm_super
*mpb
;
7208 unsigned long long pos
= 0;
7209 unsigned long long maxsize
;
7213 /* We must have the container info already read in. */
7217 mpb
= super
->anchor
;
7219 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7220 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7224 /* General test: make sure there is space for
7225 * 'raiddisks' device extents of size 'size' at a given
7228 unsigned long long minsize
= size
;
7229 unsigned long long start_offset
= MaxSector
;
7232 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7233 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7238 e
= get_extents(super
, dl
, 0);
7241 unsigned long long esize
;
7242 esize
= e
[i
].start
- pos
;
7243 if (esize
>= minsize
)
7245 if (found
&& start_offset
== MaxSector
) {
7248 } else if (found
&& pos
!= start_offset
) {
7252 pos
= e
[i
].start
+ e
[i
].size
;
7254 } while (e
[i
-1].size
);
7259 if (dcnt
< raiddisks
) {
7261 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7268 /* This device must be a member of the set */
7269 if (!stat_is_blkdev(dev
, &rdev
))
7271 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7272 if (dl
->major
== (int)major(rdev
) &&
7273 dl
->minor
== (int)minor(rdev
))
7278 pr_err("%s is not in the same imsm set\n", dev
);
7280 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7281 /* If a volume is present then the current creation attempt
7282 * cannot incorporate new spares because the orom may not
7283 * understand this configuration (all member disks must be
7284 * members of each array in the container).
7286 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7287 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7289 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7290 mpb
->num_disks
!= raiddisks
) {
7291 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7295 /* retrieve the largest free space block */
7296 e
= get_extents(super
, dl
, 0);
7301 unsigned long long esize
;
7303 esize
= e
[i
].start
- pos
;
7304 if (esize
>= maxsize
)
7306 pos
= e
[i
].start
+ e
[i
].size
;
7308 } while (e
[i
-1].size
);
7313 pr_err("unable to determine free space for: %s\n",
7317 if (maxsize
< size
) {
7319 pr_err("%s not enough space (%llu < %llu)\n",
7320 dev
, maxsize
, size
);
7324 /* count total number of extents for merge */
7326 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7328 i
+= dl
->extent_cnt
;
7330 maxsize
= merge_extents(super
, i
);
7332 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7333 pr_err("attempting to create a second volume with size less then remaining space.\n");
7335 if (maxsize
< size
|| maxsize
== 0) {
7338 pr_err("no free space left on device. Aborting...\n");
7340 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7346 *freesize
= maxsize
;
7349 int count
= count_volumes(super
->hba
,
7350 super
->orom
->dpa
, verbose
);
7351 if (super
->orom
->vphba
<= count
) {
7352 pr_vrb("platform does not support more than %d raid volumes.\n",
7353 super
->orom
->vphba
);
7360 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7361 unsigned long long size
, int chunk
,
7362 unsigned long long *freesize
)
7364 struct intel_super
*super
= st
->sb
;
7365 struct imsm_super
*mpb
= super
->anchor
;
7370 unsigned long long maxsize
;
7371 unsigned long long minsize
;
7375 /* find the largest common start free region of the possible disks */
7379 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7385 /* don't activate new spares if we are orom constrained
7386 * and there is already a volume active in the container
7388 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7391 e
= get_extents(super
, dl
, 0);
7394 for (i
= 1; e
[i
-1].size
; i
++)
7402 maxsize
= merge_extents(super
, extent_cnt
);
7406 minsize
= chunk
* 2;
7408 if (cnt
< raiddisks
||
7409 (super
->orom
&& used
&& used
!= raiddisks
) ||
7410 maxsize
< minsize
||
7412 pr_err("not enough devices with space to create array.\n");
7413 return 0; /* No enough free spaces large enough */
7424 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7425 pr_err("attempting to create a second volume with size less then remaining space.\n");
7427 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7429 dl
->raiddisk
= cnt
++;
7433 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7438 static int reserve_space(struct supertype
*st
, int raiddisks
,
7439 unsigned long long size
, int chunk
,
7440 unsigned long long *freesize
)
7442 struct intel_super
*super
= st
->sb
;
7447 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7450 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7452 dl
->raiddisk
= cnt
++;
7459 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7460 int raiddisks
, int *chunk
, unsigned long long size
,
7461 unsigned long long data_offset
,
7462 char *dev
, unsigned long long *freesize
,
7463 int consistency_policy
, int verbose
)
7470 * if given unused devices create a container
7471 * if given given devices in a container create a member volume
7473 if (level
== LEVEL_CONTAINER
) {
7474 /* Must be a fresh device to add to a container */
7475 return validate_geometry_imsm_container(st
, level
, layout
,
7483 if (size
&& (size
< 1024)) {
7484 pr_err("Given size must be greater than 1M.\n");
7485 /* Depends on algorithm in Create.c :
7486 * if container was given (dev == NULL) return -1,
7487 * if block device was given ( dev != NULL) return 0.
7489 return dev
? -1 : 0;
7494 struct intel_super
*super
= st
->sb
;
7495 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7496 raiddisks
, chunk
, size
,
7499 /* we are being asked to automatically layout a
7500 * new volume based on the current contents of
7501 * the container. If the the parameters can be
7502 * satisfied reserve_space will record the disks,
7503 * start offset, and size of the volume to be
7504 * created. add_to_super and getinfo_super
7505 * detect when autolayout is in progress.
7507 /* assuming that freesize is always given when array is
7509 if (super
->orom
&& freesize
) {
7511 count
= count_volumes(super
->hba
,
7512 super
->orom
->dpa
, verbose
);
7513 if (super
->orom
->vphba
<= count
) {
7514 pr_vrb("platform does not support more than %d raid volumes.\n",
7515 super
->orom
->vphba
);
7520 return reserve_space(st
, raiddisks
, size
,
7526 /* creating in a given container */
7527 return validate_geometry_imsm_volume(st
, level
, layout
,
7528 raiddisks
, chunk
, size
,
7530 dev
, freesize
, verbose
);
7533 /* This device needs to be a device in an 'imsm' container */
7534 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7537 pr_err("Cannot create this array on device %s\n",
7542 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7544 pr_err("Cannot open %s: %s\n",
7545 dev
, strerror(errno
));
7548 /* Well, it is in use by someone, maybe an 'imsm' container. */
7549 cfd
= open_container(fd
);
7553 pr_err("Cannot use %s: It is busy\n",
7557 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7558 if (sra
&& sra
->array
.major_version
== -1 &&
7559 strcmp(sra
->text_version
, "imsm") == 0)
7563 /* This is a member of a imsm container. Load the container
7564 * and try to create a volume
7566 struct intel_super
*super
;
7568 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7570 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7572 return validate_geometry_imsm_volume(st
, level
, layout
,
7574 size
, data_offset
, dev
,
7581 pr_err("failed container membership check\n");
7587 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7589 struct intel_super
*super
= st
->sb
;
7591 if (level
&& *level
== UnSet
)
7592 *level
= LEVEL_CONTAINER
;
7594 if (level
&& layout
&& *layout
== UnSet
)
7595 *layout
= imsm_level_to_layout(*level
);
7597 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7598 *chunk
= imsm_default_chunk(super
->orom
);
7601 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7603 static int kill_subarray_imsm(struct supertype
*st
, char *subarray_id
)
7605 /* remove the subarray currently referenced by subarray_id */
7607 struct intel_dev
**dp
;
7608 struct intel_super
*super
= st
->sb
;
7609 __u8 current_vol
= strtoul(subarray_id
, NULL
, 10);
7610 struct imsm_super
*mpb
= super
->anchor
;
7612 if (mpb
->num_raid_devs
== 0)
7615 /* block deletions that would change the uuid of active subarrays
7617 * FIXME when immutable ids are available, but note that we'll
7618 * also need to fixup the invalidated/active subarray indexes in
7621 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7624 if (i
< current_vol
)
7626 sprintf(subarray
, "%u", i
);
7627 if (is_subarray_active(subarray
, st
->devnm
)) {
7628 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7635 if (st
->update_tail
) {
7636 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7638 u
->type
= update_kill_array
;
7639 u
->dev_idx
= current_vol
;
7640 append_metadata_update(st
, u
, sizeof(*u
));
7645 for (dp
= &super
->devlist
; *dp
;)
7646 if ((*dp
)->index
== current_vol
) {
7649 handle_missing(super
, (*dp
)->dev
);
7650 if ((*dp
)->index
> current_vol
)
7655 /* no more raid devices, all active components are now spares,
7656 * but of course failed are still failed
7658 if (--mpb
->num_raid_devs
== 0) {
7661 for (d
= super
->disks
; d
; d
= d
->next
)
7666 super
->updates_pending
++;
7671 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7672 char *update
, struct mddev_ident
*ident
)
7674 /* update the subarray currently referenced by ->current_vol */
7675 struct intel_super
*super
= st
->sb
;
7676 struct imsm_super
*mpb
= super
->anchor
;
7678 if (strcmp(update
, "name") == 0) {
7679 char *name
= ident
->name
;
7683 if (is_subarray_active(subarray
, st
->devnm
)) {
7684 pr_err("Unable to update name of active subarray\n");
7688 if (!check_name(super
, name
, 0))
7691 vol
= strtoul(subarray
, &ep
, 10);
7692 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7695 if (st
->update_tail
) {
7696 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7698 u
->type
= update_rename_array
;
7700 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7701 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7702 append_metadata_update(st
, u
, sizeof(*u
));
7704 struct imsm_dev
*dev
;
7707 dev
= get_imsm_dev(super
, vol
);
7708 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7709 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7710 memcpy(dev
->volume
, name
, namelen
);
7711 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7712 dev
= get_imsm_dev(super
, i
);
7713 handle_missing(super
, dev
);
7715 super
->updates_pending
++;
7717 } else if (strcmp(update
, "ppl") == 0 ||
7718 strcmp(update
, "no-ppl") == 0) {
7721 int vol
= strtoul(subarray
, &ep
, 10);
7723 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7726 if (strcmp(update
, "ppl") == 0)
7727 new_policy
= RWH_MULTIPLE_DISTRIBUTED
;
7729 new_policy
= RWH_MULTIPLE_OFF
;
7731 if (st
->update_tail
) {
7732 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7734 u
->type
= update_rwh_policy
;
7736 u
->new_policy
= new_policy
;
7737 append_metadata_update(st
, u
, sizeof(*u
));
7739 struct imsm_dev
*dev
;
7741 dev
= get_imsm_dev(super
, vol
);
7742 dev
->rwh_policy
= new_policy
;
7743 super
->updates_pending
++;
7751 static int is_gen_migration(struct imsm_dev
*dev
)
7756 if (!dev
->vol
.migr_state
)
7759 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7765 static int is_rebuilding(struct imsm_dev
*dev
)
7767 struct imsm_map
*migr_map
;
7769 if (!dev
->vol
.migr_state
)
7772 if (migr_type(dev
) != MIGR_REBUILD
)
7775 migr_map
= get_imsm_map(dev
, MAP_1
);
7777 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7783 static int is_initializing(struct imsm_dev
*dev
)
7785 struct imsm_map
*migr_map
;
7787 if (!dev
->vol
.migr_state
)
7790 if (migr_type(dev
) != MIGR_INIT
)
7793 migr_map
= get_imsm_map(dev
, MAP_1
);
7795 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7801 static void update_recovery_start(struct intel_super
*super
,
7802 struct imsm_dev
*dev
,
7803 struct mdinfo
*array
)
7805 struct mdinfo
*rebuild
= NULL
;
7809 if (!is_rebuilding(dev
))
7812 /* Find the rebuild target, but punt on the dual rebuild case */
7813 for (d
= array
->devs
; d
; d
= d
->next
)
7814 if (d
->recovery_start
== 0) {
7821 /* (?) none of the disks are marked with
7822 * IMSM_ORD_REBUILD, so assume they are missing and the
7823 * disk_ord_tbl was not correctly updated
7825 dprintf("failed to locate out-of-sync disk\n");
7829 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7830 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7833 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7835 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7837 /* Given a container loaded by load_super_imsm_all,
7838 * extract information about all the arrays into
7840 * If 'subarray' is given, just extract info about that array.
7842 * For each imsm_dev create an mdinfo, fill it in,
7843 * then look for matching devices in super->disks
7844 * and create appropriate device mdinfo.
7846 struct intel_super
*super
= st
->sb
;
7847 struct imsm_super
*mpb
= super
->anchor
;
7848 struct mdinfo
*rest
= NULL
;
7852 int spare_disks
= 0;
7853 int current_vol
= super
->current_vol
;
7855 /* do not assemble arrays when not all attributes are supported */
7856 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7858 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7861 /* count spare devices, not used in maps
7863 for (d
= super
->disks
; d
; d
= d
->next
)
7867 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7868 struct imsm_dev
*dev
;
7869 struct imsm_map
*map
;
7870 struct imsm_map
*map2
;
7871 struct mdinfo
*this;
7878 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7881 dev
= get_imsm_dev(super
, i
);
7882 map
= get_imsm_map(dev
, MAP_0
);
7883 map2
= get_imsm_map(dev
, MAP_1
);
7884 level
= get_imsm_raid_level(map
);
7886 /* do not publish arrays that are in the middle of an
7887 * unsupported migration
7889 if (dev
->vol
.migr_state
&&
7890 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7891 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7895 /* do not publish arrays that are not support by controller's
7899 this = xmalloc(sizeof(*this));
7901 super
->current_vol
= i
;
7902 getinfo_super_imsm_volume(st
, this, NULL
);
7904 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7905 /* mdadm does not support all metadata features- set the bit in all arrays state */
7906 if (!validate_geometry_imsm_orom(super
,
7907 level
, /* RAID level */
7908 imsm_level_to_layout(level
),
7909 map
->num_members
, /* raid disks */
7910 &chunk
, imsm_dev_size(dev
),
7912 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7914 this->array
.state
|=
7915 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7916 (1<<MD_SB_BLOCK_VOLUME
);
7919 /* if array has bad blocks, set suitable bit in all arrays state */
7921 this->array
.state
|=
7922 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7923 (1<<MD_SB_BLOCK_VOLUME
);
7925 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7926 unsigned long long recovery_start
;
7927 struct mdinfo
*info_d
;
7935 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7936 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7937 for (d
= super
->disks
; d
; d
= d
->next
)
7938 if (d
->index
== idx
)
7941 recovery_start
= MaxSector
;
7944 if (d
&& is_failed(&d
->disk
))
7946 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
7948 if (!(ord
& IMSM_ORD_REBUILD
))
7949 this->array
.working_disks
++;
7951 * if we skip some disks the array will be assmebled degraded;
7952 * reset resync start to avoid a dirty-degraded
7953 * situation when performing the intial sync
7958 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
7959 if ((!able_to_resync(level
, missing
) ||
7960 recovery_start
== 0))
7961 this->resync_start
= MaxSector
;
7964 * FIXME handle dirty degraded
7971 info_d
= xcalloc(1, sizeof(*info_d
));
7972 info_d
->next
= this->devs
;
7973 this->devs
= info_d
;
7975 info_d
->disk
.number
= d
->index
;
7976 info_d
->disk
.major
= d
->major
;
7977 info_d
->disk
.minor
= d
->minor
;
7978 info_d
->disk
.raid_disk
= slot
;
7979 info_d
->recovery_start
= recovery_start
;
7981 if (slot
< map2
->num_members
)
7982 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7984 this->array
.spare_disks
++;
7986 if (slot
< map
->num_members
)
7987 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7989 this->array
.spare_disks
++;
7992 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7993 info_d
->data_offset
= pba_of_lba0(map
);
7994 info_d
->component_size
= calc_component_size(map
, dev
);
7996 if (map
->raid_level
== 5) {
7997 info_d
->ppl_sector
= this->ppl_sector
;
7998 info_d
->ppl_size
= this->ppl_size
;
7999 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
8000 recovery_start
== 0)
8001 this->resync_start
= 0;
8004 info_d
->bb
.supported
= 1;
8005 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
8006 info_d
->data_offset
,
8007 info_d
->component_size
,
8010 /* now that the disk list is up-to-date fixup recovery_start */
8011 update_recovery_start(super
, dev
, this);
8012 this->array
.spare_disks
+= spare_disks
;
8014 /* check for reshape */
8015 if (this->reshape_active
== 1)
8016 recover_backup_imsm(st
, this);
8020 super
->current_vol
= current_vol
;
8024 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
8025 int failed
, int look_in_map
)
8027 struct imsm_map
*map
;
8029 map
= get_imsm_map(dev
, look_in_map
);
8032 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
8033 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
8035 switch (get_imsm_raid_level(map
)) {
8037 return IMSM_T_STATE_FAILED
;
8040 if (failed
< map
->num_members
)
8041 return IMSM_T_STATE_DEGRADED
;
8043 return IMSM_T_STATE_FAILED
;
8048 * check to see if any mirrors have failed, otherwise we
8049 * are degraded. Even numbered slots are mirrored on
8053 /* gcc -Os complains that this is unused */
8054 int insync
= insync
;
8056 for (i
= 0; i
< map
->num_members
; i
++) {
8057 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8058 int idx
= ord_to_idx(ord
);
8059 struct imsm_disk
*disk
;
8061 /* reset the potential in-sync count on even-numbered
8062 * slots. num_copies is always 2 for imsm raid10
8067 disk
= get_imsm_disk(super
, idx
);
8068 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8071 /* no in-sync disks left in this mirror the
8075 return IMSM_T_STATE_FAILED
;
8078 return IMSM_T_STATE_DEGRADED
;
8082 return IMSM_T_STATE_DEGRADED
;
8084 return IMSM_T_STATE_FAILED
;
8090 return map
->map_state
;
8093 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8098 struct imsm_disk
*disk
;
8099 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8100 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8101 struct imsm_map
*map_for_loop
;
8106 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8107 * disks that are being rebuilt. New failures are recorded to
8108 * map[0]. So we look through all the disks we started with and
8109 * see if any failures are still present, or if any new ones
8113 if (prev
&& (map
->num_members
< prev
->num_members
))
8114 map_for_loop
= prev
;
8116 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8118 /* when MAP_X is passed both maps failures are counted
8121 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8122 i
< prev
->num_members
) {
8123 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8124 idx_1
= ord_to_idx(ord
);
8126 disk
= get_imsm_disk(super
, idx_1
);
8127 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8130 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8131 i
< map
->num_members
) {
8132 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8133 idx
= ord_to_idx(ord
);
8136 disk
= get_imsm_disk(super
, idx
);
8137 if (!disk
|| is_failed(disk
) ||
8138 ord
& IMSM_ORD_REBUILD
)
8147 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8150 struct intel_super
*super
= c
->sb
;
8151 struct imsm_super
*mpb
= super
->anchor
;
8152 struct imsm_update_prealloc_bb_mem u
;
8154 if (atoi(inst
) >= mpb
->num_raid_devs
) {
8155 pr_err("subarry index %d, out of range\n", atoi(inst
));
8159 dprintf("imsm: open_new %s\n", inst
);
8160 a
->info
.container_member
= atoi(inst
);
8162 u
.type
= update_prealloc_badblocks_mem
;
8163 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8168 static int is_resyncing(struct imsm_dev
*dev
)
8170 struct imsm_map
*migr_map
;
8172 if (!dev
->vol
.migr_state
)
8175 if (migr_type(dev
) == MIGR_INIT
||
8176 migr_type(dev
) == MIGR_REPAIR
)
8179 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8182 migr_map
= get_imsm_map(dev
, MAP_1
);
8184 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8185 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8191 /* return true if we recorded new information */
8192 static int mark_failure(struct intel_super
*super
,
8193 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8197 struct imsm_map
*map
;
8198 char buf
[MAX_RAID_SERIAL_LEN
+3];
8199 unsigned int len
, shift
= 0;
8201 /* new failures are always set in map[0] */
8202 map
= get_imsm_map(dev
, MAP_0
);
8204 slot
= get_imsm_disk_slot(map
, idx
);
8208 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8209 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8212 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8213 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8215 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8216 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8217 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8219 disk
->status
|= FAILED_DISK
;
8220 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8221 /* mark failures in second map if second map exists and this disk
8223 * This is valid for migration, initialization and rebuild
8225 if (dev
->vol
.migr_state
) {
8226 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8227 int slot2
= get_imsm_disk_slot(map2
, idx
);
8229 if (slot2
< map2
->num_members
&& slot2
>= 0)
8230 set_imsm_ord_tbl_ent(map2
, slot2
,
8231 idx
| IMSM_ORD_REBUILD
);
8233 if (map
->failed_disk_num
== 0xff ||
8234 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8235 map
->failed_disk_num
= slot
;
8237 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8242 static void mark_missing(struct intel_super
*super
,
8243 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8245 mark_failure(super
, dev
, disk
, idx
);
8247 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8250 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8251 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8254 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8258 if (!super
->missing
)
8261 /* When orom adds replacement for missing disk it does
8262 * not remove entry of missing disk, but just updates map with
8263 * new added disk. So it is not enough just to test if there is
8264 * any missing disk, we have to look if there are any failed disks
8265 * in map to stop migration */
8267 dprintf("imsm: mark missing\n");
8268 /* end process for initialization and rebuild only
8270 if (is_gen_migration(dev
) == 0) {
8271 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8275 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8276 struct imsm_map
*map1
;
8277 int i
, ord
, ord_map1
;
8280 for (i
= 0; i
< map
->num_members
; i
++) {
8281 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8282 if (!(ord
& IMSM_ORD_REBUILD
))
8285 map1
= get_imsm_map(dev
, MAP_1
);
8289 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8290 if (ord_map1
& IMSM_ORD_REBUILD
)
8295 map_state
= imsm_check_degraded(super
, dev
,
8297 end_migration(dev
, super
, map_state
);
8301 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8302 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8303 super
->updates_pending
++;
8306 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8309 unsigned long long array_blocks
;
8310 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8311 int used_disks
= imsm_num_data_members(map
);
8313 if (used_disks
== 0) {
8314 /* when problems occures
8315 * return current array_blocks value
8317 array_blocks
= imsm_dev_size(dev
);
8319 return array_blocks
;
8322 /* set array size in metadata
8325 /* OLCE size change is caused by added disks
8327 array_blocks
= per_dev_array_size(map
) * used_disks
;
8329 /* Online Volume Size Change
8330 * Using available free space
8332 array_blocks
= new_size
;
8334 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8335 set_imsm_dev_size(dev
, array_blocks
);
8337 return array_blocks
;
8340 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8342 static void imsm_progress_container_reshape(struct intel_super
*super
)
8344 /* if no device has a migr_state, but some device has a
8345 * different number of members than the previous device, start
8346 * changing the number of devices in this device to match
8349 struct imsm_super
*mpb
= super
->anchor
;
8350 int prev_disks
= -1;
8354 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8355 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8356 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8357 struct imsm_map
*map2
;
8358 int prev_num_members
;
8360 if (dev
->vol
.migr_state
)
8363 if (prev_disks
== -1)
8364 prev_disks
= map
->num_members
;
8365 if (prev_disks
== map
->num_members
)
8368 /* OK, this array needs to enter reshape mode.
8369 * i.e it needs a migr_state
8372 copy_map_size
= sizeof_imsm_map(map
);
8373 prev_num_members
= map
->num_members
;
8374 map
->num_members
= prev_disks
;
8375 dev
->vol
.migr_state
= 1;
8376 dev
->vol
.curr_migr_unit
= 0;
8377 set_migr_type(dev
, MIGR_GEN_MIGR
);
8378 for (i
= prev_num_members
;
8379 i
< map
->num_members
; i
++)
8380 set_imsm_ord_tbl_ent(map
, i
, i
);
8381 map2
= get_imsm_map(dev
, MAP_1
);
8382 /* Copy the current map */
8383 memcpy(map2
, map
, copy_map_size
);
8384 map2
->num_members
= prev_num_members
;
8386 imsm_set_array_size(dev
, -1);
8387 super
->clean_migration_record_by_mdmon
= 1;
8388 super
->updates_pending
++;
8392 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8393 * states are handled in imsm_set_disk() with one exception, when a
8394 * resync is stopped due to a new failure this routine will set the
8395 * 'degraded' state for the array.
8397 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8399 int inst
= a
->info
.container_member
;
8400 struct intel_super
*super
= a
->container
->sb
;
8401 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8402 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8403 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8404 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8405 __u32 blocks_per_unit
;
8407 if (dev
->vol
.migr_state
&&
8408 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8409 /* array state change is blocked due to reshape action
8411 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8412 * - finish the reshape (if last_checkpoint is big and action != reshape)
8413 * - update curr_migr_unit
8415 if (a
->curr_action
== reshape
) {
8416 /* still reshaping, maybe update curr_migr_unit */
8417 goto mark_checkpoint
;
8419 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8420 /* for some reason we aborted the reshape.
8422 * disable automatic metadata rollback
8423 * user action is required to recover process
8426 struct imsm_map
*map2
=
8427 get_imsm_map(dev
, MAP_1
);
8428 dev
->vol
.migr_state
= 0;
8429 set_migr_type(dev
, 0);
8430 dev
->vol
.curr_migr_unit
= 0;
8432 sizeof_imsm_map(map2
));
8433 super
->updates_pending
++;
8436 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8437 unsigned long long array_blocks
;
8441 used_disks
= imsm_num_data_members(map
);
8442 if (used_disks
> 0) {
8444 per_dev_array_size(map
) *
8447 round_size_to_mb(array_blocks
,
8449 a
->info
.custom_array_size
= array_blocks
;
8450 /* encourage manager to update array
8454 a
->check_reshape
= 1;
8456 /* finalize online capacity expansion/reshape */
8457 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8459 mdi
->disk
.raid_disk
,
8462 imsm_progress_container_reshape(super
);
8467 /* before we activate this array handle any missing disks */
8468 if (consistent
== 2)
8469 handle_missing(super
, dev
);
8471 if (consistent
== 2 &&
8472 (!is_resync_complete(&a
->info
) ||
8473 map_state
!= IMSM_T_STATE_NORMAL
||
8474 dev
->vol
.migr_state
))
8477 if (is_resync_complete(&a
->info
)) {
8478 /* complete intialization / resync,
8479 * recovery and interrupted recovery is completed in
8482 if (is_resyncing(dev
)) {
8483 dprintf("imsm: mark resync done\n");
8484 end_migration(dev
, super
, map_state
);
8485 super
->updates_pending
++;
8486 a
->last_checkpoint
= 0;
8488 } else if ((!is_resyncing(dev
) && !failed
) &&
8489 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8490 /* mark the start of the init process if nothing is failed */
8491 dprintf("imsm: mark resync start\n");
8492 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8493 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8495 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8496 super
->updates_pending
++;
8500 /* skip checkpointing for general migration,
8501 * it is controlled in mdadm
8503 if (is_gen_migration(dev
))
8504 goto skip_mark_checkpoint
;
8506 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8507 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8508 if (blocks_per_unit
) {
8512 units
= a
->last_checkpoint
/ blocks_per_unit
;
8515 /* check that we did not overflow 32-bits, and that
8516 * curr_migr_unit needs updating
8518 if (units32
== units
&&
8520 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8521 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8522 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8523 super
->updates_pending
++;
8527 skip_mark_checkpoint
:
8528 /* mark dirty / clean */
8529 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8530 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8531 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8533 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8535 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8536 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8537 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8538 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8540 super
->updates_pending
++;
8546 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8548 int inst
= a
->info
.container_member
;
8549 struct intel_super
*super
= a
->container
->sb
;
8550 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8551 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8553 if (slot
> map
->num_members
) {
8554 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8555 slot
, map
->num_members
- 1);
8562 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8565 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8567 int inst
= a
->info
.container_member
;
8568 struct intel_super
*super
= a
->container
->sb
;
8569 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8570 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8571 struct imsm_disk
*disk
;
8573 int recovery_not_finished
= 0;
8577 int rebuild_done
= 0;
8580 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8584 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8585 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8587 /* check for new failures */
8588 if (disk
&& (state
& DS_FAULTY
)) {
8589 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8590 super
->updates_pending
++;
8593 /* check if in_sync */
8594 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8595 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8597 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8599 super
->updates_pending
++;
8602 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8603 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8605 /* check if recovery complete, newly degraded, or failed */
8606 dprintf("imsm: Detected transition to state ");
8607 switch (map_state
) {
8608 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8609 dprintf("normal: ");
8610 if (is_rebuilding(dev
)) {
8611 dprintf_cont("while rebuilding");
8612 /* check if recovery is really finished */
8613 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8614 if (mdi
->recovery_start
!= MaxSector
) {
8615 recovery_not_finished
= 1;
8618 if (recovery_not_finished
) {
8620 dprintf("Rebuild has not finished yet, state not changed");
8621 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8622 a
->last_checkpoint
= mdi
->recovery_start
;
8623 super
->updates_pending
++;
8627 end_migration(dev
, super
, map_state
);
8628 map
= get_imsm_map(dev
, MAP_0
);
8629 map
->failed_disk_num
= ~0;
8630 super
->updates_pending
++;
8631 a
->last_checkpoint
= 0;
8634 if (is_gen_migration(dev
)) {
8635 dprintf_cont("while general migration");
8636 if (a
->last_checkpoint
>= a
->info
.component_size
)
8637 end_migration(dev
, super
, map_state
);
8639 map
->map_state
= map_state
;
8640 map
= get_imsm_map(dev
, MAP_0
);
8641 map
->failed_disk_num
= ~0;
8642 super
->updates_pending
++;
8646 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8647 dprintf_cont("degraded: ");
8648 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8649 dprintf_cont("mark degraded");
8650 map
->map_state
= map_state
;
8651 super
->updates_pending
++;
8652 a
->last_checkpoint
= 0;
8655 if (is_rebuilding(dev
)) {
8656 dprintf_cont("while rebuilding ");
8657 if (state
& DS_FAULTY
) {
8658 dprintf_cont("removing failed drive ");
8659 if (n
== map
->failed_disk_num
) {
8660 dprintf_cont("end migration");
8661 end_migration(dev
, super
, map_state
);
8662 a
->last_checkpoint
= 0;
8664 dprintf_cont("fail detected during rebuild, changing map state");
8665 map
->map_state
= map_state
;
8667 super
->updates_pending
++;
8673 /* check if recovery is really finished */
8674 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8675 if (mdi
->recovery_start
!= MaxSector
) {
8676 recovery_not_finished
= 1;
8679 if (recovery_not_finished
) {
8681 dprintf_cont("Rebuild has not finished yet");
8682 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8683 a
->last_checkpoint
=
8684 mdi
->recovery_start
;
8685 super
->updates_pending
++;
8690 dprintf_cont(" Rebuild done, still degraded");
8691 end_migration(dev
, super
, map_state
);
8692 a
->last_checkpoint
= 0;
8693 super
->updates_pending
++;
8695 for (i
= 0; i
< map
->num_members
; i
++) {
8696 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8698 if (idx
& IMSM_ORD_REBUILD
)
8699 map
->failed_disk_num
= i
;
8701 super
->updates_pending
++;
8704 if (is_gen_migration(dev
)) {
8705 dprintf_cont("while general migration");
8706 if (a
->last_checkpoint
>= a
->info
.component_size
)
8707 end_migration(dev
, super
, map_state
);
8709 map
->map_state
= map_state
;
8710 manage_second_map(super
, dev
);
8712 super
->updates_pending
++;
8715 if (is_initializing(dev
)) {
8716 dprintf_cont("while initialization.");
8717 map
->map_state
= map_state
;
8718 super
->updates_pending
++;
8722 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8723 dprintf_cont("failed: ");
8724 if (is_gen_migration(dev
)) {
8725 dprintf_cont("while general migration");
8726 map
->map_state
= map_state
;
8727 super
->updates_pending
++;
8730 if (map
->map_state
!= map_state
) {
8731 dprintf_cont("mark failed");
8732 end_migration(dev
, super
, map_state
);
8733 super
->updates_pending
++;
8734 a
->last_checkpoint
= 0;
8739 dprintf_cont("state %i\n", map_state
);
8744 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8747 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8748 unsigned long long dsize
;
8749 unsigned long long sectors
;
8750 unsigned int sector_size
;
8752 get_dev_sector_size(fd
, NULL
, §or_size
);
8753 get_dev_size(fd
, NULL
, &dsize
);
8755 if (mpb_size
> sector_size
) {
8756 /* -1 to account for anchor */
8757 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8759 /* write the extended mpb to the sectors preceeding the anchor */
8760 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8764 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8765 sector_size
* sectors
) != sector_size
* sectors
)
8769 /* first block is stored on second to last sector of the disk */
8770 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8773 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8779 static void imsm_sync_metadata(struct supertype
*container
)
8781 struct intel_super
*super
= container
->sb
;
8783 dprintf("sync metadata: %d\n", super
->updates_pending
);
8784 if (!super
->updates_pending
)
8787 write_super_imsm(container
, 0);
8789 super
->updates_pending
= 0;
8792 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8794 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8795 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8798 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8802 if (dl
&& is_failed(&dl
->disk
))
8806 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8811 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8812 struct active_array
*a
, int activate_new
,
8813 struct mdinfo
*additional_test_list
)
8815 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8816 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8817 struct imsm_super
*mpb
= super
->anchor
;
8818 struct imsm_map
*map
;
8819 unsigned long long pos
;
8824 __u32 array_start
= 0;
8825 __u32 array_end
= 0;
8827 struct mdinfo
*test_list
;
8829 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8830 /* If in this array, skip */
8831 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8832 if (d
->state_fd
>= 0 &&
8833 d
->disk
.major
== dl
->major
&&
8834 d
->disk
.minor
== dl
->minor
) {
8835 dprintf("%x:%x already in array\n",
8836 dl
->major
, dl
->minor
);
8841 test_list
= additional_test_list
;
8843 if (test_list
->disk
.major
== dl
->major
&&
8844 test_list
->disk
.minor
== dl
->minor
) {
8845 dprintf("%x:%x already in additional test list\n",
8846 dl
->major
, dl
->minor
);
8849 test_list
= test_list
->next
;
8854 /* skip in use or failed drives */
8855 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8857 dprintf("%x:%x status (failed: %d index: %d)\n",
8858 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8862 /* skip pure spares when we are looking for partially
8863 * assimilated drives
8865 if (dl
->index
== -1 && !activate_new
)
8868 if (!drive_validate_sector_size(super
, dl
))
8871 /* Does this unused device have the requisite free space?
8872 * It needs to be able to cover all member volumes
8874 ex
= get_extents(super
, dl
, 1);
8876 dprintf("cannot get extents\n");
8879 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8880 dev
= get_imsm_dev(super
, i
);
8881 map
= get_imsm_map(dev
, MAP_0
);
8883 /* check if this disk is already a member of
8886 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8892 array_start
= pba_of_lba0(map
);
8893 array_end
= array_start
+
8894 per_dev_array_size(map
) - 1;
8897 /* check that we can start at pba_of_lba0 with
8898 * num_data_stripes*blocks_per_stripe of space
8900 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8904 pos
= ex
[j
].start
+ ex
[j
].size
;
8906 } while (ex
[j
-1].size
);
8913 if (i
< mpb
->num_raid_devs
) {
8914 dprintf("%x:%x does not have %u to %u available\n",
8915 dl
->major
, dl
->minor
, array_start
, array_end
);
8925 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8927 struct imsm_dev
*dev2
;
8928 struct imsm_map
*map
;
8934 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8936 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8937 if (state
== IMSM_T_STATE_FAILED
) {
8938 map
= get_imsm_map(dev2
, MAP_0
);
8941 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8943 * Check if failed disks are deleted from intel
8944 * disk list or are marked to be deleted
8946 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8947 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8949 * Do not rebuild the array if failed disks
8950 * from failed sub-array are not removed from
8954 is_failed(&idisk
->disk
) &&
8955 (idisk
->action
!= DISK_REMOVE
))
8963 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8964 struct metadata_update
**updates
)
8967 * Find a device with unused free space and use it to replace a
8968 * failed/vacant region in an array. We replace failed regions one a
8969 * array at a time. The result is that a new spare disk will be added
8970 * to the first failed array and after the monitor has finished
8971 * propagating failures the remainder will be consumed.
8973 * FIXME add a capability for mdmon to request spares from another
8977 struct intel_super
*super
= a
->container
->sb
;
8978 int inst
= a
->info
.container_member
;
8979 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8980 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8981 int failed
= a
->info
.array
.raid_disks
;
8982 struct mdinfo
*rv
= NULL
;
8985 struct metadata_update
*mu
;
8987 struct imsm_update_activate_spare
*u
;
8992 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8993 if ((d
->curr_state
& DS_FAULTY
) &&
8995 /* wait for Removal to happen */
8997 if (d
->state_fd
>= 0)
9001 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
9002 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
9004 if (imsm_reshape_blocks_arrays_changes(super
))
9007 /* Cannot activate another spare if rebuild is in progress already
9009 if (is_rebuilding(dev
)) {
9010 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
9014 if (a
->info
.array
.level
== 4)
9015 /* No repair for takeovered array
9016 * imsm doesn't support raid4
9020 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
9021 IMSM_T_STATE_DEGRADED
)
9024 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
9025 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
9030 * If there are any failed disks check state of the other volume.
9031 * Block rebuild if the another one is failed until failed disks
9032 * are removed from container.
9035 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
9036 MAX_RAID_SERIAL_LEN
, dev
->volume
);
9037 /* check if states of the other volumes allow for rebuild */
9038 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
9040 allowed
= imsm_rebuild_allowed(a
->container
,
9048 /* For each slot, if it is not working, find a spare */
9049 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
9050 for (d
= a
->info
.devs
; d
; d
= d
->next
)
9051 if (d
->disk
.raid_disk
== i
)
9053 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
9054 if (d
&& (d
->state_fd
>= 0))
9058 * OK, this device needs recovery. Try to re-add the
9059 * previous occupant of this slot, if this fails see if
9060 * we can continue the assimilation of a spare that was
9061 * partially assimilated, finally try to activate a new
9064 dl
= imsm_readd(super
, i
, a
);
9066 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9068 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9072 /* found a usable disk with enough space */
9073 di
= xcalloc(1, sizeof(*di
));
9075 /* dl->index will be -1 in the case we are activating a
9076 * pristine spare. imsm_process_update() will create a
9077 * new index in this case. Once a disk is found to be
9078 * failed in all member arrays it is kicked from the
9081 di
->disk
.number
= dl
->index
;
9083 /* (ab)use di->devs to store a pointer to the device
9086 di
->devs
= (struct mdinfo
*) dl
;
9088 di
->disk
.raid_disk
= i
;
9089 di
->disk
.major
= dl
->major
;
9090 di
->disk
.minor
= dl
->minor
;
9092 di
->recovery_start
= 0;
9093 di
->data_offset
= pba_of_lba0(map
);
9094 di
->component_size
= a
->info
.component_size
;
9095 di
->container_member
= inst
;
9096 di
->bb
.supported
= 1;
9097 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9098 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9099 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9101 super
->random
= random32();
9105 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9106 i
, di
->data_offset
);
9110 /* No spares found */
9112 /* Now 'rv' has a list of devices to return.
9113 * Create a metadata_update record to update the
9114 * disk_ord_tbl for the array
9116 mu
= xmalloc(sizeof(*mu
));
9117 mu
->buf
= xcalloc(num_spares
,
9118 sizeof(struct imsm_update_activate_spare
));
9120 mu
->space_list
= NULL
;
9121 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9122 mu
->next
= *updates
;
9123 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9125 for (di
= rv
; di
; di
= di
->next
) {
9126 u
->type
= update_activate_spare
;
9127 u
->dl
= (struct dl
*) di
->devs
;
9129 u
->slot
= di
->disk
.raid_disk
;
9140 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9142 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9143 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9144 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9145 struct disk_info
*inf
= get_disk_info(u
);
9146 struct imsm_disk
*disk
;
9150 for (i
= 0; i
< map
->num_members
; i
++) {
9151 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9152 for (j
= 0; j
< new_map
->num_members
; j
++)
9153 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9160 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9164 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9165 if (dl
->major
== major
&& dl
->minor
== minor
)
9170 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9176 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9177 if (dl
->major
== major
&& dl
->minor
== minor
) {
9180 prev
->next
= dl
->next
;
9182 super
->disks
= dl
->next
;
9184 __free_imsm_disk(dl
);
9185 dprintf("removed %x:%x\n", major
, minor
);
9193 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9195 static int add_remove_disk_update(struct intel_super
*super
)
9197 int check_degraded
= 0;
9200 /* add/remove some spares to/from the metadata/contrainer */
9201 while (super
->disk_mgmt_list
) {
9202 struct dl
*disk_cfg
;
9204 disk_cfg
= super
->disk_mgmt_list
;
9205 super
->disk_mgmt_list
= disk_cfg
->next
;
9206 disk_cfg
->next
= NULL
;
9208 if (disk_cfg
->action
== DISK_ADD
) {
9209 disk_cfg
->next
= super
->disks
;
9210 super
->disks
= disk_cfg
;
9212 dprintf("added %x:%x\n",
9213 disk_cfg
->major
, disk_cfg
->minor
);
9214 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9215 dprintf("Disk remove action processed: %x.%x\n",
9216 disk_cfg
->major
, disk_cfg
->minor
);
9217 disk
= get_disk_super(super
,
9221 /* store action status */
9222 disk
->action
= DISK_REMOVE
;
9223 /* remove spare disks only */
9224 if (disk
->index
== -1) {
9225 remove_disk_super(super
,
9229 disk_cfg
->fd
= disk
->fd
;
9233 /* release allocate disk structure */
9234 __free_imsm_disk(disk_cfg
);
9237 return check_degraded
;
9240 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9241 struct intel_super
*super
,
9244 struct intel_dev
*id
;
9245 void **tofree
= NULL
;
9248 dprintf("(enter)\n");
9249 if (u
->subdev
< 0 || u
->subdev
> 1) {
9250 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9253 if (space_list
== NULL
|| *space_list
== NULL
) {
9254 dprintf("imsm: Error: Memory is not allocated\n");
9258 for (id
= super
->devlist
; id
; id
= id
->next
) {
9259 if (id
->index
== (unsigned)u
->subdev
) {
9260 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9261 struct imsm_map
*map
;
9262 struct imsm_dev
*new_dev
=
9263 (struct imsm_dev
*)*space_list
;
9264 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9266 struct dl
*new_disk
;
9268 if (new_dev
== NULL
)
9270 *space_list
= **space_list
;
9271 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9272 map
= get_imsm_map(new_dev
, MAP_0
);
9274 dprintf("imsm: Error: migration in progress");
9278 to_state
= map
->map_state
;
9279 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9281 /* this should not happen */
9282 if (u
->new_disks
[0] < 0) {
9283 map
->failed_disk_num
=
9284 map
->num_members
- 1;
9285 to_state
= IMSM_T_STATE_DEGRADED
;
9287 to_state
= IMSM_T_STATE_NORMAL
;
9289 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9290 if (u
->new_level
> -1)
9291 map
->raid_level
= u
->new_level
;
9292 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9293 if ((u
->new_level
== 5) &&
9294 (migr_map
->raid_level
== 0)) {
9295 int ord
= map
->num_members
- 1;
9296 migr_map
->num_members
--;
9297 if (u
->new_disks
[0] < 0)
9298 ord
|= IMSM_ORD_REBUILD
;
9299 set_imsm_ord_tbl_ent(map
,
9300 map
->num_members
- 1,
9304 tofree
= (void **)dev
;
9306 /* update chunk size
9308 if (u
->new_chunksize
> 0) {
9309 unsigned long long num_data_stripes
;
9310 struct imsm_map
*dest_map
=
9311 get_imsm_map(dev
, MAP_0
);
9313 imsm_num_data_members(dest_map
);
9315 if (used_disks
== 0)
9318 map
->blocks_per_strip
=
9319 __cpu_to_le16(u
->new_chunksize
* 2);
9321 imsm_dev_size(dev
) / used_disks
;
9322 num_data_stripes
/= map
->blocks_per_strip
;
9323 num_data_stripes
/= map
->num_domains
;
9324 set_num_data_stripes(map
, num_data_stripes
);
9327 /* ensure blocks_per_member has valid value
9329 set_blocks_per_member(map
,
9330 per_dev_array_size(map
) +
9331 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9335 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9336 migr_map
->raid_level
== map
->raid_level
)
9339 if (u
->new_disks
[0] >= 0) {
9342 new_disk
= get_disk_super(super
,
9343 major(u
->new_disks
[0]),
9344 minor(u
->new_disks
[0]));
9345 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9346 major(u
->new_disks
[0]),
9347 minor(u
->new_disks
[0]),
9348 new_disk
, new_disk
->index
);
9349 if (new_disk
== NULL
)
9350 goto error_disk_add
;
9352 new_disk
->index
= map
->num_members
- 1;
9353 /* slot to fill in autolayout
9355 new_disk
->raiddisk
= new_disk
->index
;
9356 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9357 new_disk
->disk
.status
&= ~SPARE_DISK
;
9359 goto error_disk_add
;
9362 *tofree
= *space_list
;
9363 /* calculate new size
9365 imsm_set_array_size(new_dev
, -1);
9372 *space_list
= tofree
;
9376 dprintf("Error: imsm: Cannot find disk.\n");
9380 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9381 struct intel_super
*super
)
9383 struct intel_dev
*id
;
9386 dprintf("(enter)\n");
9387 if (u
->subdev
< 0 || u
->subdev
> 1) {
9388 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9392 for (id
= super
->devlist
; id
; id
= id
->next
) {
9393 if (id
->index
== (unsigned)u
->subdev
) {
9394 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9395 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9396 int used_disks
= imsm_num_data_members(map
);
9397 unsigned long long blocks_per_member
;
9398 unsigned long long num_data_stripes
;
9399 unsigned long long new_size_per_disk
;
9401 if (used_disks
== 0)
9404 /* calculate new size
9406 new_size_per_disk
= u
->new_size
/ used_disks
;
9407 blocks_per_member
= new_size_per_disk
+
9408 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9409 num_data_stripes
= new_size_per_disk
/
9410 map
->blocks_per_strip
;
9411 num_data_stripes
/= map
->num_domains
;
9412 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9413 u
->new_size
, new_size_per_disk
,
9415 set_blocks_per_member(map
, blocks_per_member
);
9416 set_num_data_stripes(map
, num_data_stripes
);
9417 imsm_set_array_size(dev
, u
->new_size
);
9427 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9428 struct intel_super
*super
,
9429 struct active_array
*active_array
)
9431 struct imsm_super
*mpb
= super
->anchor
;
9432 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9433 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9434 struct imsm_map
*migr_map
;
9435 struct active_array
*a
;
9436 struct imsm_disk
*disk
;
9443 int second_map_created
= 0;
9445 for (; u
; u
= u
->next
) {
9446 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9451 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9456 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9461 /* count failures (excluding rebuilds and the victim)
9462 * to determine map[0] state
9465 for (i
= 0; i
< map
->num_members
; i
++) {
9468 disk
= get_imsm_disk(super
,
9469 get_imsm_disk_idx(dev
, i
, MAP_X
));
9470 if (!disk
|| is_failed(disk
))
9474 /* adding a pristine spare, assign a new index */
9475 if (dl
->index
< 0) {
9476 dl
->index
= super
->anchor
->num_disks
;
9477 super
->anchor
->num_disks
++;
9480 disk
->status
|= CONFIGURED_DISK
;
9481 disk
->status
&= ~SPARE_DISK
;
9484 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9485 if (!second_map_created
) {
9486 second_map_created
= 1;
9487 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9488 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9490 map
->map_state
= to_state
;
9491 migr_map
= get_imsm_map(dev
, MAP_1
);
9492 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9493 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9494 dl
->index
| IMSM_ORD_REBUILD
);
9496 /* update the family_num to mark a new container
9497 * generation, being careful to record the existing
9498 * family_num in orig_family_num to clean up after
9499 * earlier mdadm versions that neglected to set it.
9501 if (mpb
->orig_family_num
== 0)
9502 mpb
->orig_family_num
= mpb
->family_num
;
9503 mpb
->family_num
+= super
->random
;
9505 /* count arrays using the victim in the metadata */
9507 for (a
= active_array
; a
; a
= a
->next
) {
9508 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9509 map
= get_imsm_map(dev
, MAP_0
);
9511 if (get_imsm_disk_slot(map
, victim
) >= 0)
9515 /* delete the victim if it is no longer being
9521 /* We know that 'manager' isn't touching anything,
9522 * so it is safe to delete
9524 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9525 if ((*dlp
)->index
== victim
)
9528 /* victim may be on the missing list */
9530 for (dlp
= &super
->missing
; *dlp
;
9531 dlp
= &(*dlp
)->next
)
9532 if ((*dlp
)->index
== victim
)
9534 imsm_delete(super
, dlp
, victim
);
9541 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9542 struct intel_super
*super
,
9545 struct dl
*new_disk
;
9546 struct intel_dev
*id
;
9548 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9549 int disk_count
= u
->old_raid_disks
;
9550 void **tofree
= NULL
;
9551 int devices_to_reshape
= 1;
9552 struct imsm_super
*mpb
= super
->anchor
;
9554 unsigned int dev_id
;
9556 dprintf("(enter)\n");
9558 /* enable spares to use in array */
9559 for (i
= 0; i
< delta_disks
; i
++) {
9560 new_disk
= get_disk_super(super
,
9561 major(u
->new_disks
[i
]),
9562 minor(u
->new_disks
[i
]));
9563 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9564 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9565 new_disk
, new_disk
->index
);
9566 if (new_disk
== NULL
||
9567 (new_disk
->index
>= 0 &&
9568 new_disk
->index
< u
->old_raid_disks
))
9569 goto update_reshape_exit
;
9570 new_disk
->index
= disk_count
++;
9571 /* slot to fill in autolayout
9573 new_disk
->raiddisk
= new_disk
->index
;
9574 new_disk
->disk
.status
|=
9576 new_disk
->disk
.status
&= ~SPARE_DISK
;
9579 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9580 mpb
->num_raid_devs
);
9581 /* manage changes in volume
9583 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9584 void **sp
= *space_list
;
9585 struct imsm_dev
*newdev
;
9586 struct imsm_map
*newmap
, *oldmap
;
9588 for (id
= super
->devlist
; id
; id
= id
->next
) {
9589 if (id
->index
== dev_id
)
9598 /* Copy the dev, but not (all of) the map */
9599 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9600 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9601 newmap
= get_imsm_map(newdev
, MAP_0
);
9602 /* Copy the current map */
9603 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9604 /* update one device only
9606 if (devices_to_reshape
) {
9607 dprintf("imsm: modifying subdev: %i\n",
9609 devices_to_reshape
--;
9610 newdev
->vol
.migr_state
= 1;
9611 newdev
->vol
.curr_migr_unit
= 0;
9612 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9613 newmap
->num_members
= u
->new_raid_disks
;
9614 for (i
= 0; i
< delta_disks
; i
++) {
9615 set_imsm_ord_tbl_ent(newmap
,
9616 u
->old_raid_disks
+ i
,
9617 u
->old_raid_disks
+ i
);
9619 /* New map is correct, now need to save old map
9621 newmap
= get_imsm_map(newdev
, MAP_1
);
9622 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9624 imsm_set_array_size(newdev
, -1);
9627 sp
= (void **)id
->dev
;
9632 /* Clear migration record */
9633 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9636 *space_list
= tofree
;
9639 update_reshape_exit
:
9644 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9645 struct intel_super
*super
,
9648 struct imsm_dev
*dev
= NULL
;
9649 struct intel_dev
*dv
;
9650 struct imsm_dev
*dev_new
;
9651 struct imsm_map
*map
;
9655 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9656 if (dv
->index
== (unsigned int)u
->subarray
) {
9664 map
= get_imsm_map(dev
, MAP_0
);
9666 if (u
->direction
== R10_TO_R0
) {
9667 unsigned long long num_data_stripes
;
9669 /* Number of failed disks must be half of initial disk number */
9670 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9671 (map
->num_members
/ 2))
9674 /* iterate through devices to mark removed disks as spare */
9675 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9676 if (dm
->disk
.status
& FAILED_DISK
) {
9677 int idx
= dm
->index
;
9678 /* update indexes on the disk list */
9679 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9680 the index values will end up being correct.... NB */
9681 for (du
= super
->disks
; du
; du
= du
->next
)
9682 if (du
->index
> idx
)
9684 /* mark as spare disk */
9689 map
->num_members
= map
->num_members
/ 2;
9690 map
->map_state
= IMSM_T_STATE_NORMAL
;
9691 map
->num_domains
= 1;
9692 map
->raid_level
= 0;
9693 map
->failed_disk_num
= -1;
9694 num_data_stripes
= imsm_dev_size(dev
) / 2;
9695 num_data_stripes
/= map
->blocks_per_strip
;
9696 set_num_data_stripes(map
, num_data_stripes
);
9699 if (u
->direction
== R0_TO_R10
) {
9701 unsigned long long num_data_stripes
;
9703 /* update slots in current disk list */
9704 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9708 /* create new *missing* disks */
9709 for (i
= 0; i
< map
->num_members
; i
++) {
9710 space
= *space_list
;
9713 *space_list
= *space
;
9715 memcpy(du
, super
->disks
, sizeof(*du
));
9719 du
->index
= (i
* 2) + 1;
9720 sprintf((char *)du
->disk
.serial
,
9721 " MISSING_%d", du
->index
);
9722 sprintf((char *)du
->serial
,
9723 "MISSING_%d", du
->index
);
9724 du
->next
= super
->missing
;
9725 super
->missing
= du
;
9727 /* create new dev and map */
9728 space
= *space_list
;
9731 *space_list
= *space
;
9732 dev_new
= (void *)space
;
9733 memcpy(dev_new
, dev
, sizeof(*dev
));
9734 /* update new map */
9735 map
= get_imsm_map(dev_new
, MAP_0
);
9736 map
->num_members
= map
->num_members
* 2;
9737 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9738 map
->num_domains
= 2;
9739 map
->raid_level
= 1;
9740 num_data_stripes
= imsm_dev_size(dev
) / 2;
9741 num_data_stripes
/= map
->blocks_per_strip
;
9742 num_data_stripes
/= map
->num_domains
;
9743 set_num_data_stripes(map
, num_data_stripes
);
9745 /* replace dev<->dev_new */
9748 /* update disk order table */
9749 for (du
= super
->disks
; du
; du
= du
->next
)
9751 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9752 for (du
= super
->missing
; du
; du
= du
->next
)
9753 if (du
->index
>= 0) {
9754 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9755 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9761 static void imsm_process_update(struct supertype
*st
,
9762 struct metadata_update
*update
)
9765 * crack open the metadata_update envelope to find the update record
9766 * update can be one of:
9767 * update_reshape_container_disks - all the arrays in the container
9768 * are being reshaped to have more devices. We need to mark
9769 * the arrays for general migration and convert selected spares
9770 * into active devices.
9771 * update_activate_spare - a spare device has replaced a failed
9772 * device in an array, update the disk_ord_tbl. If this disk is
9773 * present in all member arrays then also clear the SPARE_DISK
9775 * update_create_array
9777 * update_rename_array
9778 * update_add_remove_disk
9780 struct intel_super
*super
= st
->sb
;
9781 struct imsm_super
*mpb
;
9782 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9784 /* update requires a larger buf but the allocation failed */
9785 if (super
->next_len
&& !super
->next_buf
) {
9786 super
->next_len
= 0;
9790 if (super
->next_buf
) {
9791 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9793 super
->len
= super
->next_len
;
9794 super
->buf
= super
->next_buf
;
9796 super
->next_len
= 0;
9797 super
->next_buf
= NULL
;
9800 mpb
= super
->anchor
;
9803 case update_general_migration_checkpoint
: {
9804 struct intel_dev
*id
;
9805 struct imsm_update_general_migration_checkpoint
*u
=
9806 (void *)update
->buf
;
9808 dprintf("called for update_general_migration_checkpoint\n");
9810 /* find device under general migration */
9811 for (id
= super
->devlist
; id
; id
= id
->next
) {
9812 if (is_gen_migration(id
->dev
)) {
9813 id
->dev
->vol
.curr_migr_unit
=
9814 __cpu_to_le32(u
->curr_migr_unit
);
9815 super
->updates_pending
++;
9820 case update_takeover
: {
9821 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9822 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9823 imsm_update_version_info(super
);
9824 super
->updates_pending
++;
9829 case update_reshape_container_disks
: {
9830 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9831 if (apply_reshape_container_disks_update(
9832 u
, super
, &update
->space_list
))
9833 super
->updates_pending
++;
9836 case update_reshape_migration
: {
9837 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9838 if (apply_reshape_migration_update(
9839 u
, super
, &update
->space_list
))
9840 super
->updates_pending
++;
9843 case update_size_change
: {
9844 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9845 if (apply_size_change_update(u
, super
))
9846 super
->updates_pending
++;
9849 case update_activate_spare
: {
9850 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9851 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9852 super
->updates_pending
++;
9855 case update_create_array
: {
9856 /* someone wants to create a new array, we need to be aware of
9857 * a few races/collisions:
9858 * 1/ 'Create' called by two separate instances of mdadm
9859 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9860 * devices that have since been assimilated via
9862 * In the event this update can not be carried out mdadm will
9863 * (FIX ME) notice that its update did not take hold.
9865 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9866 struct intel_dev
*dv
;
9867 struct imsm_dev
*dev
;
9868 struct imsm_map
*map
, *new_map
;
9869 unsigned long long start
, end
;
9870 unsigned long long new_start
, new_end
;
9872 struct disk_info
*inf
;
9875 /* handle racing creates: first come first serve */
9876 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9877 dprintf("subarray %d already defined\n", u
->dev_idx
);
9881 /* check update is next in sequence */
9882 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9883 dprintf("can not create array %d expected index %d\n",
9884 u
->dev_idx
, mpb
->num_raid_devs
);
9888 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9889 new_start
= pba_of_lba0(new_map
);
9890 new_end
= new_start
+ per_dev_array_size(new_map
);
9891 inf
= get_disk_info(u
);
9893 /* handle activate_spare versus create race:
9894 * check to make sure that overlapping arrays do not include
9897 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9898 dev
= get_imsm_dev(super
, i
);
9899 map
= get_imsm_map(dev
, MAP_0
);
9900 start
= pba_of_lba0(map
);
9901 end
= start
+ per_dev_array_size(map
);
9902 if ((new_start
>= start
&& new_start
<= end
) ||
9903 (start
>= new_start
&& start
<= new_end
))
9908 if (disks_overlap(super
, i
, u
)) {
9909 dprintf("arrays overlap\n");
9914 /* check that prepare update was successful */
9915 if (!update
->space
) {
9916 dprintf("prepare update failed\n");
9920 /* check that all disks are still active before committing
9921 * changes. FIXME: could we instead handle this by creating a
9922 * degraded array? That's probably not what the user expects,
9923 * so better to drop this update on the floor.
9925 for (i
= 0; i
< new_map
->num_members
; i
++) {
9926 dl
= serial_to_dl(inf
[i
].serial
, super
);
9928 dprintf("disk disappeared\n");
9933 super
->updates_pending
++;
9935 /* convert spares to members and fixup ord_tbl */
9936 for (i
= 0; i
< new_map
->num_members
; i
++) {
9937 dl
= serial_to_dl(inf
[i
].serial
, super
);
9938 if (dl
->index
== -1) {
9939 dl
->index
= mpb
->num_disks
;
9941 dl
->disk
.status
|= CONFIGURED_DISK
;
9942 dl
->disk
.status
&= ~SPARE_DISK
;
9944 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9949 update
->space
= NULL
;
9950 imsm_copy_dev(dev
, &u
->dev
);
9951 dv
->index
= u
->dev_idx
;
9952 dv
->next
= super
->devlist
;
9953 super
->devlist
= dv
;
9954 mpb
->num_raid_devs
++;
9956 imsm_update_version_info(super
);
9959 /* mdmon knows how to release update->space, but not
9960 * ((struct intel_dev *) update->space)->dev
9962 if (update
->space
) {
9968 case update_kill_array
: {
9969 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9970 int victim
= u
->dev_idx
;
9971 struct active_array
*a
;
9972 struct intel_dev
**dp
;
9973 struct imsm_dev
*dev
;
9975 /* sanity check that we are not affecting the uuid of
9976 * active arrays, or deleting an active array
9978 * FIXME when immutable ids are available, but note that
9979 * we'll also need to fixup the invalidated/active
9980 * subarray indexes in mdstat
9982 for (a
= st
->arrays
; a
; a
= a
->next
)
9983 if (a
->info
.container_member
>= victim
)
9985 /* by definition if mdmon is running at least one array
9986 * is active in the container, so checking
9987 * mpb->num_raid_devs is just extra paranoia
9989 dev
= get_imsm_dev(super
, victim
);
9990 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9991 dprintf("failed to delete subarray-%d\n", victim
);
9995 for (dp
= &super
->devlist
; *dp
;)
9996 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9999 if ((*dp
)->index
> (unsigned)victim
)
10003 mpb
->num_raid_devs
--;
10004 super
->updates_pending
++;
10007 case update_rename_array
: {
10008 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
10009 char name
[MAX_RAID_SERIAL_LEN
+1];
10010 int target
= u
->dev_idx
;
10011 struct active_array
*a
;
10012 struct imsm_dev
*dev
;
10014 /* sanity check that we are not affecting the uuid of
10017 memset(name
, 0, sizeof(name
));
10018 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
10019 name
[MAX_RAID_SERIAL_LEN
] = '\0';
10020 for (a
= st
->arrays
; a
; a
= a
->next
)
10021 if (a
->info
.container_member
== target
)
10023 dev
= get_imsm_dev(super
, u
->dev_idx
);
10024 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
10025 dprintf("failed to rename subarray-%d\n", target
);
10029 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
10030 super
->updates_pending
++;
10033 case update_add_remove_disk
: {
10034 /* we may be able to repair some arrays if disks are
10035 * being added, check the status of add_remove_disk
10036 * if discs has been added.
10038 if (add_remove_disk_update(super
)) {
10039 struct active_array
*a
;
10041 super
->updates_pending
++;
10042 for (a
= st
->arrays
; a
; a
= a
->next
)
10043 a
->check_degraded
= 1;
10047 case update_prealloc_badblocks_mem
:
10049 case update_rwh_policy
: {
10050 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
10051 int target
= u
->dev_idx
;
10052 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
10054 dprintf("could not find subarray-%d\n", target
);
10058 if (dev
->rwh_policy
!= u
->new_policy
) {
10059 dev
->rwh_policy
= u
->new_policy
;
10060 super
->updates_pending
++;
10065 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10069 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10071 static int imsm_prepare_update(struct supertype
*st
,
10072 struct metadata_update
*update
)
10075 * Allocate space to hold new disk entries, raid-device entries or a new
10076 * mpb if necessary. The manager synchronously waits for updates to
10077 * complete in the monitor, so new mpb buffers allocated here can be
10078 * integrated by the monitor thread without worrying about live pointers
10079 * in the manager thread.
10081 enum imsm_update_type type
;
10082 struct intel_super
*super
= st
->sb
;
10083 unsigned int sector_size
= super
->sector_size
;
10084 struct imsm_super
*mpb
= super
->anchor
;
10088 if (update
->len
< (int)sizeof(type
))
10091 type
= *(enum imsm_update_type
*) update
->buf
;
10094 case update_general_migration_checkpoint
:
10095 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10097 dprintf("called for update_general_migration_checkpoint\n");
10099 case update_takeover
: {
10100 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10101 if (update
->len
< (int)sizeof(*u
))
10103 if (u
->direction
== R0_TO_R10
) {
10104 void **tail
= (void **)&update
->space_list
;
10105 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10106 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10107 int num_members
= map
->num_members
;
10110 /* allocate memory for added disks */
10111 for (i
= 0; i
< num_members
; i
++) {
10112 size
= sizeof(struct dl
);
10113 space
= xmalloc(size
);
10118 /* allocate memory for new device */
10119 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10120 (num_members
* sizeof(__u32
));
10121 space
= xmalloc(size
);
10125 len
= disks_to_mpb_size(num_members
* 2);
10130 case update_reshape_container_disks
: {
10131 /* Every raid device in the container is about to
10132 * gain some more devices, and we will enter a
10134 * So each 'imsm_map' will be bigger, and the imsm_vol
10135 * will now hold 2 of them.
10136 * Thus we need new 'struct imsm_dev' allocations sized
10137 * as sizeof_imsm_dev but with more devices in both maps.
10139 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10140 struct intel_dev
*dl
;
10141 void **space_tail
= (void**)&update
->space_list
;
10143 if (update
->len
< (int)sizeof(*u
))
10146 dprintf("for update_reshape\n");
10148 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10149 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10151 if (u
->new_raid_disks
> u
->old_raid_disks
)
10152 size
+= sizeof(__u32
)*2*
10153 (u
->new_raid_disks
- u
->old_raid_disks
);
10157 *space_tail
= NULL
;
10160 len
= disks_to_mpb_size(u
->new_raid_disks
);
10161 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10164 case update_reshape_migration
: {
10165 /* for migration level 0->5 we need to add disks
10166 * so the same as for container operation we will copy
10167 * device to the bigger location.
10168 * in memory prepared device and new disk area are prepared
10169 * for usage in process update
10171 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10172 struct intel_dev
*id
;
10173 void **space_tail
= (void **)&update
->space_list
;
10176 int current_level
= -1;
10178 if (update
->len
< (int)sizeof(*u
))
10181 dprintf("for update_reshape\n");
10183 /* add space for bigger array in update
10185 for (id
= super
->devlist
; id
; id
= id
->next
) {
10186 if (id
->index
== (unsigned)u
->subdev
) {
10187 size
= sizeof_imsm_dev(id
->dev
, 1);
10188 if (u
->new_raid_disks
> u
->old_raid_disks
)
10189 size
+= sizeof(__u32
)*2*
10190 (u
->new_raid_disks
- u
->old_raid_disks
);
10194 *space_tail
= NULL
;
10198 if (update
->space_list
== NULL
)
10201 /* add space for disk in update
10203 size
= sizeof(struct dl
);
10207 *space_tail
= NULL
;
10209 /* add spare device to update
10211 for (id
= super
->devlist
; id
; id
= id
->next
)
10212 if (id
->index
== (unsigned)u
->subdev
) {
10213 struct imsm_dev
*dev
;
10214 struct imsm_map
*map
;
10216 dev
= get_imsm_dev(super
, u
->subdev
);
10217 map
= get_imsm_map(dev
, MAP_0
);
10218 current_level
= map
->raid_level
;
10221 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10222 struct mdinfo
*spares
;
10224 spares
= get_spares_for_grow(st
);
10227 struct mdinfo
*dev
;
10229 dev
= spares
->devs
;
10232 makedev(dev
->disk
.major
,
10234 dl
= get_disk_super(super
,
10237 dl
->index
= u
->old_raid_disks
;
10240 sysfs_free(spares
);
10243 len
= disks_to_mpb_size(u
->new_raid_disks
);
10244 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10247 case update_size_change
: {
10248 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10252 case update_activate_spare
: {
10253 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10257 case update_create_array
: {
10258 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10259 struct intel_dev
*dv
;
10260 struct imsm_dev
*dev
= &u
->dev
;
10261 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10263 struct disk_info
*inf
;
10267 if (update
->len
< (int)sizeof(*u
))
10270 inf
= get_disk_info(u
);
10271 len
= sizeof_imsm_dev(dev
, 1);
10272 /* allocate a new super->devlist entry */
10273 dv
= xmalloc(sizeof(*dv
));
10274 dv
->dev
= xmalloc(len
);
10275 update
->space
= dv
;
10277 /* count how many spares will be converted to members */
10278 for (i
= 0; i
< map
->num_members
; i
++) {
10279 dl
= serial_to_dl(inf
[i
].serial
, super
);
10281 /* hmm maybe it failed?, nothing we can do about
10286 if (count_memberships(dl
, super
) == 0)
10289 len
+= activate
* sizeof(struct imsm_disk
);
10292 case update_kill_array
: {
10293 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10297 case update_rename_array
: {
10298 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10302 case update_add_remove_disk
:
10303 /* no update->len needed */
10305 case update_prealloc_badblocks_mem
:
10306 super
->extra_space
+= sizeof(struct bbm_log
) -
10307 get_imsm_bbm_log_size(super
->bbm_log
);
10309 case update_rwh_policy
: {
10310 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10318 /* check if we need a larger metadata buffer */
10319 if (super
->next_buf
)
10320 buf_len
= super
->next_len
;
10322 buf_len
= super
->len
;
10324 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10325 /* ok we need a larger buf than what is currently allocated
10326 * if this allocation fails process_update will notice that
10327 * ->next_len is set and ->next_buf is NULL
10329 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10330 super
->extra_space
+ len
, sector_size
);
10331 if (super
->next_buf
)
10332 free(super
->next_buf
);
10334 super
->next_len
= buf_len
;
10335 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10336 memset(super
->next_buf
, 0, buf_len
);
10338 super
->next_buf
= NULL
;
10343 /* must be called while manager is quiesced */
10344 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10346 struct imsm_super
*mpb
= super
->anchor
;
10348 struct imsm_dev
*dev
;
10349 struct imsm_map
*map
;
10350 unsigned int i
, j
, num_members
;
10351 __u32 ord
, ord_map0
;
10352 struct bbm_log
*log
= super
->bbm_log
;
10354 dprintf("deleting device[%d] from imsm_super\n", index
);
10356 /* shift all indexes down one */
10357 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10358 if (iter
->index
> (int)index
)
10360 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10361 if (iter
->index
> (int)index
)
10364 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10365 dev
= get_imsm_dev(super
, i
);
10366 map
= get_imsm_map(dev
, MAP_0
);
10367 num_members
= map
->num_members
;
10368 for (j
= 0; j
< num_members
; j
++) {
10369 /* update ord entries being careful not to propagate
10370 * ord-flags to the first map
10372 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10373 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10375 if (ord_to_idx(ord
) <= index
)
10378 map
= get_imsm_map(dev
, MAP_0
);
10379 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10380 map
= get_imsm_map(dev
, MAP_1
);
10382 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10386 for (i
= 0; i
< log
->entry_count
; i
++) {
10387 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10389 if (entry
->disk_ordinal
<= index
)
10391 entry
->disk_ordinal
--;
10395 super
->updates_pending
++;
10397 struct dl
*dl
= *dlp
;
10399 *dlp
= (*dlp
)->next
;
10400 __free_imsm_disk(dl
);
10404 static void close_targets(int *targets
, int new_disks
)
10411 for (i
= 0; i
< new_disks
; i
++) {
10412 if (targets
[i
] >= 0) {
10419 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10420 struct intel_super
*super
,
10421 struct imsm_dev
*dev
)
10427 struct imsm_map
*map
;
10430 ret_val
= raid_disks
/2;
10431 /* check map if all disks pairs not failed
10434 map
= get_imsm_map(dev
, MAP_0
);
10435 for (i
= 0; i
< ret_val
; i
++) {
10436 int degradation
= 0;
10437 if (get_imsm_disk(super
, i
) == NULL
)
10439 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10441 if (degradation
== 2)
10444 map
= get_imsm_map(dev
, MAP_1
);
10445 /* if there is no second map
10446 * result can be returned
10450 /* check degradation in second map
10452 for (i
= 0; i
< ret_val
; i
++) {
10453 int degradation
= 0;
10454 if (get_imsm_disk(super
, i
) == NULL
)
10456 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10458 if (degradation
== 2)
10472 /*******************************************************************************
10473 * Function: open_backup_targets
10474 * Description: Function opens file descriptors for all devices given in
10477 * info : general array info
10478 * raid_disks : number of disks
10479 * raid_fds : table of device's file descriptors
10480 * super : intel super for raid10 degradation check
10481 * dev : intel device for raid10 degradation check
10485 ******************************************************************************/
10486 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10487 struct intel_super
*super
, struct imsm_dev
*dev
)
10493 for (i
= 0; i
< raid_disks
; i
++)
10496 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10499 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10500 dprintf("disk is faulty!!\n");
10504 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10507 dn
= map_dev(sd
->disk
.major
,
10508 sd
->disk
.minor
, 1);
10509 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10510 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10511 pr_err("cannot open component\n");
10516 /* check if maximum array degradation level is not exceeded
10518 if ((raid_disks
- opened
) >
10519 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10521 pr_err("Not enough disks can be opened.\n");
10522 close_targets(raid_fds
, raid_disks
);
10528 /*******************************************************************************
10529 * Function: validate_container_imsm
10530 * Description: This routine validates container after assemble,
10531 * eg. if devices in container are under the same controller.
10534 * info : linked list with info about devices used in array
10538 ******************************************************************************/
10539 int validate_container_imsm(struct mdinfo
*info
)
10541 if (check_env("IMSM_NO_PLATFORM"))
10544 struct sys_dev
*idev
;
10545 struct sys_dev
*hba
= NULL
;
10546 struct sys_dev
*intel_devices
= find_intel_devices();
10547 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10548 info
->disk
.minor
));
10550 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10551 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10560 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10561 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10565 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10566 struct mdinfo
*dev
;
10568 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10569 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10571 struct sys_dev
*hba2
= NULL
;
10572 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10573 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10581 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10582 get_orom_by_device_id(hba2
->dev_id
);
10584 if (hba2
&& hba
->type
!= hba2
->type
) {
10585 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10586 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10590 if (orom
!= orom2
) {
10591 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10592 " This operation is not supported and can lead to data loss.\n");
10597 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10598 " This operation is not supported and can lead to data loss.\n");
10606 /*******************************************************************************
10607 * Function: imsm_record_badblock
10608 * Description: This routine stores new bad block record in BBM log
10611 * a : array containing a bad block
10612 * slot : disk number containing a bad block
10613 * sector : bad block sector
10614 * length : bad block sectors range
10618 ******************************************************************************/
10619 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10620 unsigned long long sector
, int length
)
10622 struct intel_super
*super
= a
->container
->sb
;
10626 ord
= imsm_disk_slot_to_ord(a
, slot
);
10630 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10633 super
->updates_pending
++;
10637 /*******************************************************************************
10638 * Function: imsm_clear_badblock
10639 * Description: This routine clears bad block record from BBM log
10642 * a : array containing a bad block
10643 * slot : disk number containing a bad block
10644 * sector : bad block sector
10645 * length : bad block sectors range
10649 ******************************************************************************/
10650 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10651 unsigned long long sector
, int length
)
10653 struct intel_super
*super
= a
->container
->sb
;
10657 ord
= imsm_disk_slot_to_ord(a
, slot
);
10661 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10663 super
->updates_pending
++;
10667 /*******************************************************************************
10668 * Function: imsm_get_badblocks
10669 * Description: This routine get list of bad blocks for an array
10673 * slot : disk number
10675 * bb : structure containing bad blocks
10677 ******************************************************************************/
10678 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10680 int inst
= a
->info
.container_member
;
10681 struct intel_super
*super
= a
->container
->sb
;
10682 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10683 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10686 ord
= imsm_disk_slot_to_ord(a
, slot
);
10690 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10691 per_dev_array_size(map
), &super
->bb
);
10695 /*******************************************************************************
10696 * Function: examine_badblocks_imsm
10697 * Description: Prints list of bad blocks on a disk to the standard output
10700 * st : metadata handler
10701 * fd : open file descriptor for device
10702 * devname : device name
10706 ******************************************************************************/
10707 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10709 struct intel_super
*super
= st
->sb
;
10710 struct bbm_log
*log
= super
->bbm_log
;
10711 struct dl
*d
= NULL
;
10714 for (d
= super
->disks
; d
; d
= d
->next
) {
10715 if (strcmp(d
->devname
, devname
) == 0)
10719 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10720 pr_err("%s doesn't appear to be part of a raid array\n",
10727 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10729 for (i
= 0; i
< log
->entry_count
; i
++) {
10730 if (entry
[i
].disk_ordinal
== d
->index
) {
10731 unsigned long long sector
= __le48_to_cpu(
10732 &entry
[i
].defective_block_start
);
10733 int cnt
= entry
[i
].marked_count
+ 1;
10736 printf("Bad-blocks on %s:\n", devname
);
10740 printf("%20llu for %d sectors\n", sector
, cnt
);
10746 printf("No bad-blocks list configured on %s\n", devname
);
10750 /*******************************************************************************
10751 * Function: init_migr_record_imsm
10752 * Description: Function inits imsm migration record
10754 * super : imsm internal array info
10755 * dev : device under migration
10756 * info : general array info to find the smallest device
10759 ******************************************************************************/
10760 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10761 struct mdinfo
*info
)
10763 struct intel_super
*super
= st
->sb
;
10764 struct migr_record
*migr_rec
= super
->migr_rec
;
10765 int new_data_disks
;
10766 unsigned long long dsize
, dev_sectors
;
10767 long long unsigned min_dev_sectors
= -1LLU;
10771 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10772 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10773 unsigned long long num_migr_units
;
10774 unsigned long long array_blocks
;
10776 memset(migr_rec
, 0, sizeof(struct migr_record
));
10777 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10779 /* only ascending reshape supported now */
10780 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10782 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10783 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10784 migr_rec
->dest_depth_per_unit
*=
10785 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10786 new_data_disks
= imsm_num_data_members(map_dest
);
10787 migr_rec
->blocks_per_unit
=
10788 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10789 migr_rec
->dest_depth_per_unit
=
10790 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10791 array_blocks
= info
->component_size
* new_data_disks
;
10793 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10795 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10797 set_num_migr_units(migr_rec
, num_migr_units
);
10799 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10800 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10802 /* Find the smallest dev */
10803 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10804 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10805 fd
= dev_open(nm
, O_RDONLY
);
10808 get_dev_size(fd
, NULL
, &dsize
);
10809 dev_sectors
= dsize
/ 512;
10810 if (dev_sectors
< min_dev_sectors
)
10811 min_dev_sectors
= dev_sectors
;
10814 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
10815 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10817 write_imsm_migr_rec(st
);
10822 /*******************************************************************************
10823 * Function: save_backup_imsm
10824 * Description: Function saves critical data stripes to Migration Copy Area
10825 * and updates the current migration unit status.
10826 * Use restore_stripes() to form a destination stripe,
10827 * and to write it to the Copy Area.
10829 * st : supertype information
10830 * dev : imsm device that backup is saved for
10831 * info : general array info
10832 * buf : input buffer
10833 * length : length of data to backup (blocks_per_unit)
10837 ******************************************************************************/
10838 int save_backup_imsm(struct supertype
*st
,
10839 struct imsm_dev
*dev
,
10840 struct mdinfo
*info
,
10845 struct intel_super
*super
= st
->sb
;
10846 unsigned long long *target_offsets
;
10849 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10850 int new_disks
= map_dest
->num_members
;
10851 int dest_layout
= 0;
10853 unsigned long long start
;
10854 int data_disks
= imsm_num_data_members(map_dest
);
10856 targets
= xmalloc(new_disks
* sizeof(int));
10858 for (i
= 0; i
< new_disks
; i
++)
10861 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10863 start
= info
->reshape_progress
* 512;
10864 for (i
= 0; i
< new_disks
; i
++) {
10865 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
10866 /* move back copy area adderss, it will be moved forward
10867 * in restore_stripes() using start input variable
10869 target_offsets
[i
] -= start
/data_disks
;
10872 if (open_backup_targets(info
, new_disks
, targets
,
10876 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10877 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10879 if (restore_stripes(targets
, /* list of dest devices */
10880 target_offsets
, /* migration record offsets */
10883 map_dest
->raid_level
,
10885 -1, /* source backup file descriptor */
10886 0, /* input buf offset
10887 * always 0 buf is already offseted */
10891 pr_err("Error restoring stripes\n");
10899 close_targets(targets
, new_disks
);
10902 free(target_offsets
);
10907 /*******************************************************************************
10908 * Function: save_checkpoint_imsm
10909 * Description: Function called for current unit status update
10910 * in the migration record. It writes it to disk.
10912 * super : imsm internal array info
10913 * info : general array info
10917 * 2: failure, means no valid migration record
10918 * / no general migration in progress /
10919 ******************************************************************************/
10920 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10922 struct intel_super
*super
= st
->sb
;
10923 unsigned long long blocks_per_unit
;
10924 unsigned long long curr_migr_unit
;
10926 if (load_imsm_migr_rec(super
, info
) != 0) {
10927 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10931 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10932 if (blocks_per_unit
== 0) {
10933 dprintf("imsm: no migration in progress.\n");
10936 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10937 /* check if array is alligned to copy area
10938 * if it is not alligned, add one to current migration unit value
10939 * this can happend on array reshape finish only
10941 if (info
->reshape_progress
% blocks_per_unit
)
10944 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
10945 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10946 set_migr_dest_1st_member_lba(super
->migr_rec
,
10947 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
10949 if (write_imsm_migr_rec(st
) < 0) {
10950 dprintf("imsm: Cannot write migration record outside backup area\n");
10957 /*******************************************************************************
10958 * Function: recover_backup_imsm
10959 * Description: Function recovers critical data from the Migration Copy Area
10960 * while assembling an array.
10962 * super : imsm internal array info
10963 * info : general array info
10965 * 0 : success (or there is no data to recover)
10967 ******************************************************************************/
10968 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10970 struct intel_super
*super
= st
->sb
;
10971 struct migr_record
*migr_rec
= super
->migr_rec
;
10972 struct imsm_map
*map_dest
;
10973 struct intel_dev
*id
= NULL
;
10974 unsigned long long read_offset
;
10975 unsigned long long write_offset
;
10977 int *targets
= NULL
;
10978 int new_disks
, i
, err
;
10981 unsigned int sector_size
= super
->sector_size
;
10982 unsigned long curr_migr_unit
= current_migr_unit(migr_rec
);
10983 unsigned long num_migr_units
= get_num_migr_units(migr_rec
);
10985 int skipped_disks
= 0;
10987 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10991 /* recover data only during assemblation */
10992 if (strncmp(buffer
, "inactive", 8) != 0)
10994 /* no data to recover */
10995 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10997 if (curr_migr_unit
>= num_migr_units
)
11000 /* find device during reshape */
11001 for (id
= super
->devlist
; id
; id
= id
->next
)
11002 if (is_gen_migration(id
->dev
))
11007 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
11008 new_disks
= map_dest
->num_members
;
11010 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
11012 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
11013 pba_of_lba0(map_dest
)) * 512;
11015 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11016 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
11018 targets
= xcalloc(new_disks
, sizeof(int));
11020 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
11021 pr_err("Cannot open some devices belonging to array.\n");
11025 for (i
= 0; i
< new_disks
; i
++) {
11026 if (targets
[i
] < 0) {
11030 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
11031 pr_err("Cannot seek to block: %s\n",
11036 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
11037 pr_err("Cannot read copy area block: %s\n",
11042 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
11043 pr_err("Cannot seek to block: %s\n",
11048 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
11049 pr_err("Cannot restore block: %s\n",
11056 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11060 pr_err("Cannot restore data from backup. Too many failed disks\n");
11064 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11065 /* ignore error == 2, this can mean end of reshape here
11067 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11073 for (i
= 0; i
< new_disks
; i
++)
11082 static char disk_by_path
[] = "/dev/disk/by-path/";
11084 static const char *imsm_get_disk_controller_domain(const char *path
)
11086 char disk_path
[PATH_MAX
];
11090 strcpy(disk_path
, disk_by_path
);
11091 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
11092 if (stat(disk_path
, &st
) == 0) {
11093 struct sys_dev
* hba
;
11096 path
= devt_to_devpath(st
.st_rdev
);
11099 hba
= find_disk_attached_hba(-1, path
);
11100 if (hba
&& hba
->type
== SYS_DEV_SAS
)
11102 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
11104 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
11106 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
11110 dprintf("path: %s hba: %s attached: %s\n",
11111 path
, (hba
) ? hba
->path
: "NULL", drv
);
11117 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11119 static char devnm
[32];
11120 char subdev_name
[20];
11121 struct mdstat_ent
*mdstat
;
11123 sprintf(subdev_name
, "%d", subdev
);
11124 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11128 strcpy(devnm
, mdstat
->devnm
);
11129 free_mdstat(mdstat
);
11133 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11134 struct geo_params
*geo
,
11135 int *old_raid_disks
,
11138 /* currently we only support increasing the number of devices
11139 * for a container. This increases the number of device for each
11140 * member array. They must all be RAID0 or RAID5.
11143 struct mdinfo
*info
, *member
;
11144 int devices_that_can_grow
= 0;
11146 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11148 if (geo
->size
> 0 ||
11149 geo
->level
!= UnSet
||
11150 geo
->layout
!= UnSet
||
11151 geo
->chunksize
!= 0 ||
11152 geo
->raid_disks
== UnSet
) {
11153 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11157 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11158 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11162 info
= container_content_imsm(st
, NULL
);
11163 for (member
= info
; member
; member
= member
->next
) {
11166 dprintf("imsm: checking device_num: %i\n",
11167 member
->container_member
);
11169 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11170 /* we work on container for Online Capacity Expansion
11171 * only so raid_disks has to grow
11173 dprintf("imsm: for container operation raid disks increase is required\n");
11177 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11178 /* we cannot use this container with other raid level
11180 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11181 info
->array
.level
);
11184 /* check for platform support
11185 * for this raid level configuration
11187 struct intel_super
*super
= st
->sb
;
11188 if (!is_raid_level_supported(super
->orom
,
11189 member
->array
.level
,
11190 geo
->raid_disks
)) {
11191 dprintf("platform does not support raid%d with %d disk%s\n",
11194 geo
->raid_disks
> 1 ? "s" : "");
11197 /* check if component size is aligned to chunk size
11199 if (info
->component_size
%
11200 (info
->array
.chunk_size
/512)) {
11201 dprintf("Component size is not aligned to chunk size\n");
11206 if (*old_raid_disks
&&
11207 info
->array
.raid_disks
!= *old_raid_disks
)
11209 *old_raid_disks
= info
->array
.raid_disks
;
11211 /* All raid5 and raid0 volumes in container
11212 * have to be ready for Online Capacity Expansion
11213 * so they need to be assembled. We have already
11214 * checked that no recovery etc is happening.
11216 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11217 st
->container_devnm
);
11218 if (result
== NULL
) {
11219 dprintf("imsm: cannot find array\n");
11222 devices_that_can_grow
++;
11225 if (!member
&& devices_that_can_grow
)
11229 dprintf("Container operation allowed\n");
11231 dprintf("Error: %i\n", ret_val
);
11236 /* Function: get_spares_for_grow
11237 * Description: Allocates memory and creates list of spare devices
11238 * avaliable in container. Checks if spare drive size is acceptable.
11239 * Parameters: Pointer to the supertype structure
11240 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11243 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11245 struct spare_criteria sc
;
11247 get_spare_criteria_imsm(st
, &sc
);
11248 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11251 /******************************************************************************
11252 * function: imsm_create_metadata_update_for_reshape
11253 * Function creates update for whole IMSM container.
11255 ******************************************************************************/
11256 static int imsm_create_metadata_update_for_reshape(
11257 struct supertype
*st
,
11258 struct geo_params
*geo
,
11259 int old_raid_disks
,
11260 struct imsm_update_reshape
**updatep
)
11262 struct intel_super
*super
= st
->sb
;
11263 struct imsm_super
*mpb
= super
->anchor
;
11264 int update_memory_size
;
11265 struct imsm_update_reshape
*u
;
11266 struct mdinfo
*spares
;
11269 struct mdinfo
*dev
;
11271 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11273 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11275 /* size of all update data without anchor */
11276 update_memory_size
= sizeof(struct imsm_update_reshape
);
11278 /* now add space for spare disks that we need to add. */
11279 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11281 u
= xcalloc(1, update_memory_size
);
11282 u
->type
= update_reshape_container_disks
;
11283 u
->old_raid_disks
= old_raid_disks
;
11284 u
->new_raid_disks
= geo
->raid_disks
;
11286 /* now get spare disks list
11288 spares
= get_spares_for_grow(st
);
11290 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11291 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11296 /* we have got spares
11297 * update disk list in imsm_disk list table in anchor
11299 dprintf("imsm: %i spares are available.\n\n",
11300 spares
->array
.spare_disks
);
11302 dev
= spares
->devs
;
11303 for (i
= 0; i
< delta_disks
; i
++) {
11308 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11310 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11311 dl
->index
= mpb
->num_disks
;
11319 sysfs_free(spares
);
11321 dprintf("imsm: reshape update preparation :");
11322 if (i
== delta_disks
) {
11323 dprintf_cont(" OK\n");
11325 return update_memory_size
;
11328 dprintf_cont(" Error\n");
11333 /******************************************************************************
11334 * function: imsm_create_metadata_update_for_size_change()
11335 * Creates update for IMSM array for array size change.
11337 ******************************************************************************/
11338 static int imsm_create_metadata_update_for_size_change(
11339 struct supertype
*st
,
11340 struct geo_params
*geo
,
11341 struct imsm_update_size_change
**updatep
)
11343 struct intel_super
*super
= st
->sb
;
11344 int update_memory_size
;
11345 struct imsm_update_size_change
*u
;
11347 dprintf("(enter) New size = %llu\n", geo
->size
);
11349 /* size of all update data without anchor */
11350 update_memory_size
= sizeof(struct imsm_update_size_change
);
11352 u
= xcalloc(1, update_memory_size
);
11353 u
->type
= update_size_change
;
11354 u
->subdev
= super
->current_vol
;
11355 u
->new_size
= geo
->size
;
11357 dprintf("imsm: reshape update preparation : OK\n");
11360 return update_memory_size
;
11363 /******************************************************************************
11364 * function: imsm_create_metadata_update_for_migration()
11365 * Creates update for IMSM array.
11367 ******************************************************************************/
11368 static int imsm_create_metadata_update_for_migration(
11369 struct supertype
*st
,
11370 struct geo_params
*geo
,
11371 struct imsm_update_reshape_migration
**updatep
)
11373 struct intel_super
*super
= st
->sb
;
11374 int update_memory_size
;
11375 struct imsm_update_reshape_migration
*u
;
11376 struct imsm_dev
*dev
;
11377 int previous_level
= -1;
11379 dprintf("(enter) New Level = %i\n", geo
->level
);
11381 /* size of all update data without anchor */
11382 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11384 u
= xcalloc(1, update_memory_size
);
11385 u
->type
= update_reshape_migration
;
11386 u
->subdev
= super
->current_vol
;
11387 u
->new_level
= geo
->level
;
11388 u
->new_layout
= geo
->layout
;
11389 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11390 u
->new_disks
[0] = -1;
11391 u
->new_chunksize
= -1;
11393 dev
= get_imsm_dev(super
, u
->subdev
);
11395 struct imsm_map
*map
;
11397 map
= get_imsm_map(dev
, MAP_0
);
11399 int current_chunk_size
=
11400 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11402 if (geo
->chunksize
!= current_chunk_size
) {
11403 u
->new_chunksize
= geo
->chunksize
/ 1024;
11404 dprintf("imsm: chunk size change from %i to %i\n",
11405 current_chunk_size
, u
->new_chunksize
);
11407 previous_level
= map
->raid_level
;
11410 if (geo
->level
== 5 && previous_level
== 0) {
11411 struct mdinfo
*spares
= NULL
;
11413 u
->new_raid_disks
++;
11414 spares
= get_spares_for_grow(st
);
11415 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11417 sysfs_free(spares
);
11418 update_memory_size
= 0;
11419 pr_err("cannot get spare device for requested migration\n");
11422 sysfs_free(spares
);
11424 dprintf("imsm: reshape update preparation : OK\n");
11427 return update_memory_size
;
11430 static void imsm_update_metadata_locally(struct supertype
*st
,
11431 void *buf
, int len
)
11433 struct metadata_update mu
;
11438 mu
.space_list
= NULL
;
11440 if (imsm_prepare_update(st
, &mu
))
11441 imsm_process_update(st
, &mu
);
11443 while (mu
.space_list
) {
11444 void **space
= mu
.space_list
;
11445 mu
.space_list
= *space
;
11450 /***************************************************************************
11451 * Function: imsm_analyze_change
11452 * Description: Function analyze change for single volume
11453 * and validate if transition is supported
11454 * Parameters: Geometry parameters, supertype structure,
11455 * metadata change direction (apply/rollback)
11456 * Returns: Operation type code on success, -1 if fail
11457 ****************************************************************************/
11458 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11459 struct geo_params
*geo
,
11462 struct mdinfo info
;
11464 int check_devs
= 0;
11466 /* number of added/removed disks in operation result */
11467 int devNumChange
= 0;
11468 /* imsm compatible layout value for array geometry verification */
11469 int imsm_layout
= -1;
11471 struct imsm_dev
*dev
;
11472 struct imsm_map
*map
;
11473 struct intel_super
*super
;
11474 unsigned long long current_size
;
11475 unsigned long long free_size
;
11476 unsigned long long max_size
;
11479 getinfo_super_imsm_volume(st
, &info
, NULL
);
11480 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11481 geo
->level
!= UnSet
) {
11482 switch (info
.array
.level
) {
11484 if (geo
->level
== 5) {
11485 change
= CH_MIGRATION
;
11486 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11487 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11489 goto analyse_change_exit
;
11491 imsm_layout
= geo
->layout
;
11493 devNumChange
= 1; /* parity disk added */
11494 } else if (geo
->level
== 10) {
11495 change
= CH_TAKEOVER
;
11497 devNumChange
= 2; /* two mirrors added */
11498 imsm_layout
= 0x102; /* imsm supported layout */
11503 if (geo
->level
== 0) {
11504 change
= CH_TAKEOVER
;
11506 devNumChange
= -(geo
->raid_disks
/2);
11507 imsm_layout
= 0; /* imsm raid0 layout */
11511 if (change
== -1) {
11512 pr_err("Error. Level Migration from %d to %d not supported!\n",
11513 info
.array
.level
, geo
->level
);
11514 goto analyse_change_exit
;
11517 geo
->level
= info
.array
.level
;
11519 if (geo
->layout
!= info
.array
.layout
&&
11520 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11521 change
= CH_MIGRATION
;
11522 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11523 geo
->layout
== 5) {
11524 /* reshape 5 -> 4 */
11525 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11526 geo
->layout
== 0) {
11527 /* reshape 4 -> 5 */
11531 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11532 info
.array
.layout
, geo
->layout
);
11534 goto analyse_change_exit
;
11537 geo
->layout
= info
.array
.layout
;
11538 if (imsm_layout
== -1)
11539 imsm_layout
= info
.array
.layout
;
11542 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11543 geo
->chunksize
!= info
.array
.chunk_size
) {
11544 if (info
.array
.level
== 10) {
11545 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11547 goto analyse_change_exit
;
11548 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11549 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11550 geo
->chunksize
/1024, info
.component_size
/2);
11552 goto analyse_change_exit
;
11554 change
= CH_MIGRATION
;
11556 geo
->chunksize
= info
.array
.chunk_size
;
11559 chunk
= geo
->chunksize
/ 1024;
11562 dev
= get_imsm_dev(super
, super
->current_vol
);
11563 map
= get_imsm_map(dev
, MAP_0
);
11564 data_disks
= imsm_num_data_members(map
);
11565 /* compute current size per disk member
11567 current_size
= info
.custom_array_size
/ data_disks
;
11569 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11570 /* align component size
11572 geo
->size
= imsm_component_size_alignment_check(
11573 get_imsm_raid_level(dev
->vol
.map
),
11574 chunk
* 1024, super
->sector_size
,
11576 if (geo
->size
== 0) {
11577 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11579 goto analyse_change_exit
;
11583 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11584 if (change
!= -1) {
11585 pr_err("Error. Size change should be the only one at a time.\n");
11587 goto analyse_change_exit
;
11589 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11590 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11591 super
->current_vol
, st
->devnm
);
11592 goto analyse_change_exit
;
11594 /* check the maximum available size
11596 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11597 0, chunk
, &free_size
);
11599 /* Cannot find maximum available space
11603 max_size
= free_size
+ current_size
;
11604 /* align component size
11606 max_size
= imsm_component_size_alignment_check(
11607 get_imsm_raid_level(dev
->vol
.map
),
11608 chunk
* 1024, super
->sector_size
,
11611 if (geo
->size
== MAX_SIZE
) {
11612 /* requested size change to the maximum available size
11614 if (max_size
== 0) {
11615 pr_err("Error. Cannot find maximum available space.\n");
11617 goto analyse_change_exit
;
11619 geo
->size
= max_size
;
11622 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11623 /* accept size for rollback only
11626 /* round size due to metadata compatibility
11628 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11629 << SECT_PER_MB_SHIFT
;
11630 dprintf("Prepare update for size change to %llu\n",
11632 if (current_size
>= geo
->size
) {
11633 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11634 current_size
, geo
->size
);
11635 goto analyse_change_exit
;
11637 if (max_size
&& geo
->size
> max_size
) {
11638 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11639 max_size
, geo
->size
);
11640 goto analyse_change_exit
;
11643 geo
->size
*= data_disks
;
11644 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11645 change
= CH_ARRAY_SIZE
;
11647 if (!validate_geometry_imsm(st
,
11650 geo
->raid_disks
+ devNumChange
,
11652 geo
->size
, INVALID_SECTORS
,
11653 0, 0, info
.consistency_policy
, 1))
11657 struct intel_super
*super
= st
->sb
;
11658 struct imsm_super
*mpb
= super
->anchor
;
11660 if (mpb
->num_raid_devs
> 1) {
11661 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11667 analyse_change_exit
:
11668 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11669 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11670 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11676 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11678 struct intel_super
*super
= st
->sb
;
11679 struct imsm_update_takeover
*u
;
11681 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11683 u
->type
= update_takeover
;
11684 u
->subarray
= super
->current_vol
;
11686 /* 10->0 transition */
11687 if (geo
->level
== 0)
11688 u
->direction
= R10_TO_R0
;
11690 /* 0->10 transition */
11691 if (geo
->level
== 10)
11692 u
->direction
= R0_TO_R10
;
11694 /* update metadata locally */
11695 imsm_update_metadata_locally(st
, u
,
11696 sizeof(struct imsm_update_takeover
));
11697 /* and possibly remotely */
11698 if (st
->update_tail
)
11699 append_metadata_update(st
, u
,
11700 sizeof(struct imsm_update_takeover
));
11707 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11709 int layout
, int chunksize
, int raid_disks
,
11710 int delta_disks
, char *backup
, char *dev
,
11711 int direction
, int verbose
)
11714 struct geo_params geo
;
11716 dprintf("(enter)\n");
11718 memset(&geo
, 0, sizeof(struct geo_params
));
11720 geo
.dev_name
= dev
;
11721 strcpy(geo
.devnm
, st
->devnm
);
11724 geo
.layout
= layout
;
11725 geo
.chunksize
= chunksize
;
11726 geo
.raid_disks
= raid_disks
;
11727 if (delta_disks
!= UnSet
)
11728 geo
.raid_disks
+= delta_disks
;
11730 dprintf("for level : %i\n", geo
.level
);
11731 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11733 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11734 /* On container level we can only increase number of devices. */
11735 dprintf("imsm: info: Container operation\n");
11736 int old_raid_disks
= 0;
11738 if (imsm_reshape_is_allowed_on_container(
11739 st
, &geo
, &old_raid_disks
, direction
)) {
11740 struct imsm_update_reshape
*u
= NULL
;
11743 len
= imsm_create_metadata_update_for_reshape(
11744 st
, &geo
, old_raid_disks
, &u
);
11747 dprintf("imsm: Cannot prepare update\n");
11748 goto exit_imsm_reshape_super
;
11752 /* update metadata locally */
11753 imsm_update_metadata_locally(st
, u
, len
);
11754 /* and possibly remotely */
11755 if (st
->update_tail
)
11756 append_metadata_update(st
, u
, len
);
11761 pr_err("(imsm) Operation is not allowed on this container\n");
11764 /* On volume level we support following operations
11765 * - takeover: raid10 -> raid0; raid0 -> raid10
11766 * - chunk size migration
11767 * - migration: raid5 -> raid0; raid0 -> raid5
11769 struct intel_super
*super
= st
->sb
;
11770 struct intel_dev
*dev
= super
->devlist
;
11772 dprintf("imsm: info: Volume operation\n");
11773 /* find requested device */
11776 imsm_find_array_devnm_by_subdev(
11777 dev
->index
, st
->container_devnm
);
11778 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11783 pr_err("Cannot find %s (%s) subarray\n",
11784 geo
.dev_name
, geo
.devnm
);
11785 goto exit_imsm_reshape_super
;
11787 super
->current_vol
= dev
->index
;
11788 change
= imsm_analyze_change(st
, &geo
, direction
);
11791 ret_val
= imsm_takeover(st
, &geo
);
11793 case CH_MIGRATION
: {
11794 struct imsm_update_reshape_migration
*u
= NULL
;
11796 imsm_create_metadata_update_for_migration(
11799 dprintf("imsm: Cannot prepare update\n");
11803 /* update metadata locally */
11804 imsm_update_metadata_locally(st
, u
, len
);
11805 /* and possibly remotely */
11806 if (st
->update_tail
)
11807 append_metadata_update(st
, u
, len
);
11812 case CH_ARRAY_SIZE
: {
11813 struct imsm_update_size_change
*u
= NULL
;
11815 imsm_create_metadata_update_for_size_change(
11818 dprintf("imsm: Cannot prepare update\n");
11822 /* update metadata locally */
11823 imsm_update_metadata_locally(st
, u
, len
);
11824 /* and possibly remotely */
11825 if (st
->update_tail
)
11826 append_metadata_update(st
, u
, len
);
11836 exit_imsm_reshape_super
:
11837 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11841 #define COMPLETED_OK 0
11842 #define COMPLETED_NONE 1
11843 #define COMPLETED_DELAYED 2
11845 static int read_completed(int fd
, unsigned long long *val
)
11850 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11854 ret
= COMPLETED_OK
;
11855 if (strncmp(buf
, "none", 4) == 0) {
11856 ret
= COMPLETED_NONE
;
11857 } else if (strncmp(buf
, "delayed", 7) == 0) {
11858 ret
= COMPLETED_DELAYED
;
11861 *val
= strtoull(buf
, &ep
, 0);
11862 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11868 /*******************************************************************************
11869 * Function: wait_for_reshape_imsm
11870 * Description: Function writes new sync_max value and waits until
11871 * reshape process reach new position
11873 * sra : general array info
11874 * ndata : number of disks in new array's layout
11877 * 1 : there is no reshape in progress,
11879 ******************************************************************************/
11880 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11882 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11884 unsigned long long completed
;
11885 /* to_complete : new sync_max position */
11886 unsigned long long to_complete
= sra
->reshape_progress
;
11887 unsigned long long position_to_set
= to_complete
/ ndata
;
11890 dprintf("cannot open reshape_position\n");
11895 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11897 dprintf("cannot read reshape_position (no reshape in progres)\n");
11906 if (completed
> position_to_set
) {
11907 dprintf("wrong next position to set %llu (%llu)\n",
11908 to_complete
, position_to_set
);
11912 dprintf("Position set: %llu\n", position_to_set
);
11913 if (sysfs_set_num(sra
, NULL
, "sync_max",
11914 position_to_set
) != 0) {
11915 dprintf("cannot set reshape position to %llu\n",
11924 int timeout
= 3000;
11926 sysfs_wait(fd
, &timeout
);
11927 if (sysfs_get_str(sra
, NULL
, "sync_action",
11929 strncmp(action
, "reshape", 7) != 0) {
11930 if (strncmp(action
, "idle", 4) == 0)
11936 rc
= read_completed(fd
, &completed
);
11938 dprintf("cannot read reshape_position (in loop)\n");
11941 } else if (rc
== COMPLETED_NONE
)
11943 } while (completed
< position_to_set
);
11949 /*******************************************************************************
11950 * Function: check_degradation_change
11951 * Description: Check that array hasn't become failed.
11953 * info : for sysfs access
11954 * sources : source disks descriptors
11955 * degraded: previous degradation level
11957 * degradation level
11958 ******************************************************************************/
11959 int check_degradation_change(struct mdinfo
*info
,
11963 unsigned long long new_degraded
;
11966 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11967 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11968 /* check each device to ensure it is still working */
11971 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11972 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11974 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11977 if (sysfs_get_str(info
,
11978 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11979 strstr(sbuf
, "faulty") ||
11980 strstr(sbuf
, "in_sync") == NULL
) {
11981 /* this device is dead */
11982 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11983 if (sd
->disk
.raid_disk
>= 0 &&
11984 sources
[sd
->disk
.raid_disk
] >= 0) {
11986 sd
->disk
.raid_disk
]);
11987 sources
[sd
->disk
.raid_disk
] =
11996 return new_degraded
;
11999 /*******************************************************************************
12000 * Function: imsm_manage_reshape
12001 * Description: Function finds array under reshape and it manages reshape
12002 * process. It creates stripes backups (if required) and sets
12005 * afd : Backup handle (nattive) - not used
12006 * sra : general array info
12007 * reshape : reshape parameters - not used
12008 * st : supertype structure
12009 * blocks : size of critical section [blocks]
12010 * fds : table of source device descriptor
12011 * offsets : start of array (offest per devices)
12013 * destfd : table of destination device descriptor
12014 * destoffsets : table of destination offsets (per device)
12016 * 1 : success, reshape is done
12018 ******************************************************************************/
12019 static int imsm_manage_reshape(
12020 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
12021 struct supertype
*st
, unsigned long backup_blocks
,
12022 int *fds
, unsigned long long *offsets
,
12023 int dests
, int *destfd
, unsigned long long *destoffsets
)
12026 struct intel_super
*super
= st
->sb
;
12027 struct intel_dev
*dv
;
12028 unsigned int sector_size
= super
->sector_size
;
12029 struct imsm_dev
*dev
= NULL
;
12030 struct imsm_map
*map_src
, *map_dest
;
12031 int migr_vol_qan
= 0;
12032 int ndata
, odata
; /* [bytes] */
12033 int chunk
; /* [bytes] */
12034 struct migr_record
*migr_rec
;
12036 unsigned int buf_size
; /* [bytes] */
12037 unsigned long long max_position
; /* array size [bytes] */
12038 unsigned long long next_step
; /* [blocks]/[bytes] */
12039 unsigned long long old_data_stripe_length
;
12040 unsigned long long start_src
; /* [bytes] */
12041 unsigned long long start
; /* [bytes] */
12042 unsigned long long start_buf_shift
; /* [bytes] */
12044 int source_layout
= 0;
12049 if (!fds
|| !offsets
)
12052 /* Find volume during the reshape */
12053 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
12054 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12055 dv
->dev
->vol
.migr_state
== 1) {
12060 /* Only one volume can migrate at the same time */
12061 if (migr_vol_qan
!= 1) {
12062 pr_err("%s", migr_vol_qan
?
12063 "Number of migrating volumes greater than 1\n" :
12064 "There is no volume during migrationg\n");
12068 map_dest
= get_imsm_map(dev
, MAP_0
);
12069 map_src
= get_imsm_map(dev
, MAP_1
);
12070 if (map_src
== NULL
)
12073 ndata
= imsm_num_data_members(map_dest
);
12074 odata
= imsm_num_data_members(map_src
);
12076 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12077 old_data_stripe_length
= odata
* chunk
;
12079 migr_rec
= super
->migr_rec
;
12081 /* initialize migration record for start condition */
12082 if (sra
->reshape_progress
== 0)
12083 init_migr_record_imsm(st
, dev
, sra
);
12085 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12086 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12089 /* Save checkpoint to update migration record for current
12090 * reshape position (in md). It can be farther than current
12091 * reshape position in metadata.
12093 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12094 /* ignore error == 2, this can mean end of reshape here
12096 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12101 /* size for data */
12102 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12103 /* extend buffer size for parity disk */
12104 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12105 /* add space for stripe alignment */
12106 buf_size
+= old_data_stripe_length
;
12107 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12108 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12112 max_position
= sra
->component_size
* ndata
;
12113 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12115 while (current_migr_unit(migr_rec
) <
12116 get_num_migr_units(migr_rec
)) {
12117 /* current reshape position [blocks] */
12118 unsigned long long current_position
=
12119 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12120 * current_migr_unit(migr_rec
);
12121 unsigned long long border
;
12123 /* Check that array hasn't become failed.
12125 degraded
= check_degradation_change(sra
, fds
, degraded
);
12126 if (degraded
> 1) {
12127 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12131 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12133 if ((current_position
+ next_step
) > max_position
)
12134 next_step
= max_position
- current_position
;
12136 start
= current_position
* 512;
12138 /* align reading start to old geometry */
12139 start_buf_shift
= start
% old_data_stripe_length
;
12140 start_src
= start
- start_buf_shift
;
12142 border
= (start_src
/ odata
) - (start
/ ndata
);
12144 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12145 /* save critical stripes to buf
12146 * start - start address of current unit
12147 * to backup [bytes]
12148 * start_src - start address of current unit
12149 * to backup alligned to source array
12152 unsigned long long next_step_filler
;
12153 unsigned long long copy_length
= next_step
* 512;
12155 /* allign copy area length to stripe in old geometry */
12156 next_step_filler
= ((copy_length
+ start_buf_shift
)
12157 % old_data_stripe_length
);
12158 if (next_step_filler
)
12159 next_step_filler
= (old_data_stripe_length
12160 - next_step_filler
);
12161 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12162 start
, start_src
, copy_length
,
12163 start_buf_shift
, next_step_filler
);
12165 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12166 chunk
, map_src
->raid_level
,
12167 source_layout
, 0, NULL
, start_src
,
12169 next_step_filler
+ start_buf_shift
,
12171 dprintf("imsm: Cannot save stripes to buffer\n");
12174 /* Convert data to destination format and store it
12175 * in backup general migration area
12177 if (save_backup_imsm(st
, dev
, sra
,
12178 buf
+ start_buf_shift
, copy_length
)) {
12179 dprintf("imsm: Cannot save stripes to target devices\n");
12182 if (save_checkpoint_imsm(st
, sra
,
12183 UNIT_SRC_IN_CP_AREA
)) {
12184 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12188 /* set next step to use whole border area */
12189 border
/= next_step
;
12191 next_step
*= border
;
12193 /* When data backed up, checkpoint stored,
12194 * kick the kernel to reshape unit of data
12196 next_step
= next_step
+ sra
->reshape_progress
;
12197 /* limit next step to array max position */
12198 if (next_step
> max_position
)
12199 next_step
= max_position
;
12200 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12201 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12202 sra
->reshape_progress
= next_step
;
12204 /* wait until reshape finish */
12205 if (wait_for_reshape_imsm(sra
, ndata
)) {
12206 dprintf("wait_for_reshape_imsm returned error!\n");
12212 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12213 /* ignore error == 2, this can mean end of reshape here
12215 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12221 /* clear migr_rec on disks after successful migration */
12224 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12225 for (d
= super
->disks
; d
; d
= d
->next
) {
12226 if (d
->index
< 0 || is_failed(&d
->disk
))
12228 unsigned long long dsize
;
12230 get_dev_size(d
->fd
, NULL
, &dsize
);
12231 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12233 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12234 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12235 MIGR_REC_BUF_SECTORS
*sector_size
)
12236 perror("Write migr_rec failed");
12240 /* return '1' if done */
12244 /* See Grow.c: abort_reshape() for further explanation */
12245 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12246 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12247 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12252 struct superswitch super_imsm
= {
12253 .examine_super
= examine_super_imsm
,
12254 .brief_examine_super
= brief_examine_super_imsm
,
12255 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12256 .export_examine_super
= export_examine_super_imsm
,
12257 .detail_super
= detail_super_imsm
,
12258 .brief_detail_super
= brief_detail_super_imsm
,
12259 .write_init_super
= write_init_super_imsm
,
12260 .validate_geometry
= validate_geometry_imsm
,
12261 .add_to_super
= add_to_super_imsm
,
12262 .remove_from_super
= remove_from_super_imsm
,
12263 .detail_platform
= detail_platform_imsm
,
12264 .export_detail_platform
= export_detail_platform_imsm
,
12265 .kill_subarray
= kill_subarray_imsm
,
12266 .update_subarray
= update_subarray_imsm
,
12267 .load_container
= load_container_imsm
,
12268 .default_geometry
= default_geometry_imsm
,
12269 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12270 .reshape_super
= imsm_reshape_super
,
12271 .manage_reshape
= imsm_manage_reshape
,
12272 .recover_backup
= recover_backup_imsm
,
12273 .copy_metadata
= copy_metadata_imsm
,
12274 .examine_badblocks
= examine_badblocks_imsm
,
12275 .match_home
= match_home_imsm
,
12276 .uuid_from_super
= uuid_from_super_imsm
,
12277 .getinfo_super
= getinfo_super_imsm
,
12278 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12279 .update_super
= update_super_imsm
,
12281 .avail_size
= avail_size_imsm
,
12282 .get_spare_criteria
= get_spare_criteria_imsm
,
12284 .compare_super
= compare_super_imsm
,
12286 .load_super
= load_super_imsm
,
12287 .init_super
= init_super_imsm
,
12288 .store_super
= store_super_imsm
,
12289 .free_super
= free_super_imsm
,
12290 .match_metadata_desc
= match_metadata_desc_imsm
,
12291 .container_content
= container_content_imsm
,
12292 .validate_container
= validate_container_imsm
,
12294 .write_init_ppl
= write_init_ppl_imsm
,
12295 .validate_ppl
= validate_ppl_imsm
,
12301 .open_new
= imsm_open_new
,
12302 .set_array_state
= imsm_set_array_state
,
12303 .set_disk
= imsm_set_disk
,
12304 .sync_metadata
= imsm_sync_metadata
,
12305 .activate_spare
= imsm_activate_spare
,
12306 .process_update
= imsm_process_update
,
12307 .prepare_update
= imsm_prepare_update
,
12308 .record_bad_block
= imsm_record_badblock
,
12309 .clear_bad_block
= imsm_clear_badblock
,
12310 .get_bad_blocks
= imsm_get_badblocks
,