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 void detail_super_imsm(struct supertype
*st
, char *homehost
,
2136 struct intel_super
*super
= st
->sb
;
2137 int temp_vol
= super
->current_vol
;
2140 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2142 getinfo_super_imsm(st
, &info
, NULL
);
2143 fname_from_uuid(st
, &info
, nbuf
, ':');
2144 printf("\n UUID : %s\n", nbuf
+ 5);
2146 super
->current_vol
= temp_vol
;
2149 static void brief_detail_super_imsm(struct supertype
*st
, char *subarray
)
2153 struct intel_super
*super
= st
->sb
;
2154 int temp_vol
= super
->current_vol
;
2157 super
->current_vol
= strtoul(subarray
, NULL
, 10);
2159 getinfo_super_imsm(st
, &info
, NULL
);
2160 fname_from_uuid(st
, &info
, nbuf
, ':');
2161 printf(" UUID=%s", nbuf
+ 5);
2163 super
->current_vol
= temp_vol
;
2166 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
,
2167 size_t serial_buf_len
);
2168 static void fd2devname(int fd
, char *name
);
2170 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2172 /* dump an unsorted list of devices attached to AHCI Intel storage
2173 * controller, as well as non-connected ports
2175 int hba_len
= strlen(hba_path
) + 1;
2180 unsigned long port_mask
= (1 << port_count
) - 1;
2182 if (port_count
> (int)sizeof(port_mask
) * 8) {
2184 pr_err("port_count %d out of range\n", port_count
);
2188 /* scroll through /sys/dev/block looking for devices attached to
2191 dir
= opendir("/sys/dev/block");
2195 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2206 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2208 path
= devt_to_devpath(makedev(major
, minor
));
2211 if (!path_attached_to_hba(path
, hba_path
)) {
2217 /* retrieve the scsi device type */
2218 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2220 pr_err("failed to allocate 'device'\n");
2224 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2225 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2227 pr_err("failed to read device type for %s\n",
2233 type
= strtoul(buf
, NULL
, 10);
2235 /* if it's not a disk print the vendor and model */
2236 if (!(type
== 0 || type
== 7 || type
== 14)) {
2239 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2240 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2241 strncpy(vendor
, buf
, sizeof(vendor
));
2242 vendor
[sizeof(vendor
) - 1] = '\0';
2243 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2244 while (isspace(*c
) || *c
== '\0')
2248 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2249 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2250 strncpy(model
, buf
, sizeof(model
));
2251 model
[sizeof(model
) - 1] = '\0';
2252 c
= (char *) &model
[sizeof(model
) - 1];
2253 while (isspace(*c
) || *c
== '\0')
2257 if (vendor
[0] && model
[0])
2258 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2260 switch (type
) { /* numbers from hald/linux/device.c */
2261 case 1: sprintf(buf
, "tape"); break;
2262 case 2: sprintf(buf
, "printer"); break;
2263 case 3: sprintf(buf
, "processor"); break;
2265 case 5: sprintf(buf
, "cdrom"); break;
2266 case 6: sprintf(buf
, "scanner"); break;
2267 case 8: sprintf(buf
, "media_changer"); break;
2268 case 9: sprintf(buf
, "comm"); break;
2269 case 12: sprintf(buf
, "raid"); break;
2270 default: sprintf(buf
, "unknown");
2276 /* chop device path to 'host%d' and calculate the port number */
2277 c
= strchr(&path
[hba_len
], '/');
2280 pr_err("%s - invalid path name\n", path
+ hba_len
);
2285 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2286 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2290 *c
= '/'; /* repair the full string */
2291 pr_err("failed to determine port number for %s\n",
2298 /* mark this port as used */
2299 port_mask
&= ~(1 << port
);
2301 /* print out the device information */
2303 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2307 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2309 printf(" Port%d : - disk info unavailable -\n", port
);
2311 fd2devname(fd
, buf
);
2312 printf(" Port%d : %s", port
, buf
);
2313 if (imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2315 printf(" (%s)\n", buf
);
2330 for (i
= 0; i
< port_count
; i
++)
2331 if (port_mask
& (1 << i
))
2332 printf(" Port%d : - no device attached -\n", i
);
2338 static int print_nvme_info(struct sys_dev
*hba
)
2346 dir
= opendir("/sys/block/");
2350 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2351 if (strstr(ent
->d_name
, "nvme")) {
2352 sprintf(buf
, "/sys/block/%s", ent
->d_name
);
2353 rp
= realpath(buf
, NULL
);
2356 if (path_attached_to_hba(rp
, hba
->path
)) {
2357 fd
= open_dev(ent
->d_name
);
2363 fd2devname(fd
, buf
);
2364 if (hba
->type
== SYS_DEV_VMD
)
2365 printf(" NVMe under VMD : %s", buf
);
2366 else if (hba
->type
== SYS_DEV_NVME
)
2367 printf(" NVMe Device : %s", buf
);
2368 if (!imsm_read_serial(fd
, NULL
, (__u8
*)buf
,
2370 printf(" (%s)\n", buf
);
2383 static void print_found_intel_controllers(struct sys_dev
*elem
)
2385 for (; elem
; elem
= elem
->next
) {
2386 pr_err("found Intel(R) ");
2387 if (elem
->type
== SYS_DEV_SATA
)
2388 fprintf(stderr
, "SATA ");
2389 else if (elem
->type
== SYS_DEV_SAS
)
2390 fprintf(stderr
, "SAS ");
2391 else if (elem
->type
== SYS_DEV_NVME
)
2392 fprintf(stderr
, "NVMe ");
2394 if (elem
->type
== SYS_DEV_VMD
)
2395 fprintf(stderr
, "VMD domain");
2397 fprintf(stderr
, "RAID controller");
2400 fprintf(stderr
, " at %s", elem
->pci_id
);
2401 fprintf(stderr
, ".\n");
2406 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2413 if ((dir
= opendir(hba_path
)) == NULL
)
2416 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2419 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2420 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2422 if (*port_count
== 0)
2424 else if (host
< host_base
)
2427 if (host
+ 1 > *port_count
+ host_base
)
2428 *port_count
= host
+ 1 - host_base
;
2434 static void print_imsm_capability(const struct imsm_orom
*orom
)
2436 printf(" Platform : Intel(R) ");
2437 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2438 printf("Matrix Storage Manager\n");
2439 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2440 printf("Virtual RAID on CPU\n");
2442 printf("Rapid Storage Technology%s\n",
2443 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2444 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2445 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2446 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2447 printf(" RAID Levels :%s%s%s%s%s\n",
2448 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2449 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2450 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2451 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2452 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2453 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2454 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2455 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2456 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2457 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2458 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2459 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2460 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2461 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2462 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2463 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2464 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2465 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2466 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2467 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2468 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2469 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2470 printf(" 2TB volumes :%s supported\n",
2471 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2472 printf(" 2TB disks :%s supported\n",
2473 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2474 printf(" Max Disks : %d\n", orom
->tds
);
2475 printf(" Max Volumes : %d per array, %d per %s\n",
2476 orom
->vpa
, orom
->vphba
,
2477 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2481 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2483 printf("MD_FIRMWARE_TYPE=imsm\n");
2484 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2485 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2486 orom
->hotfix_ver
, orom
->build
);
2487 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2488 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2489 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2490 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2491 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2492 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2493 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2494 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2495 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2496 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2497 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2498 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2499 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2500 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2501 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2502 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2503 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2504 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2505 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2506 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2507 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2508 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2509 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2510 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2511 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2512 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2513 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2514 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2517 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2519 /* There are two components to imsm platform support, the ahci SATA
2520 * controller and the option-rom. To find the SATA controller we
2521 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2522 * controller with the Intel vendor id is present. This approach
2523 * allows mdadm to leverage the kernel's ahci detection logic, with the
2524 * caveat that if ahci.ko is not loaded mdadm will not be able to
2525 * detect platform raid capabilities. The option-rom resides in a
2526 * platform "Adapter ROM". We scan for its signature to retrieve the
2527 * platform capabilities. If raid support is disabled in the BIOS the
2528 * option-rom capability structure will not be available.
2530 struct sys_dev
*list
, *hba
;
2535 if (enumerate_only
) {
2536 if (check_env("IMSM_NO_PLATFORM"))
2538 list
= find_intel_devices();
2541 for (hba
= list
; hba
; hba
= hba
->next
) {
2542 if (find_imsm_capability(hba
)) {
2552 list
= find_intel_devices();
2555 pr_err("no active Intel(R) RAID controller found.\n");
2557 } else if (verbose
> 0)
2558 print_found_intel_controllers(list
);
2560 for (hba
= list
; hba
; hba
= hba
->next
) {
2561 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2563 if (!find_imsm_capability(hba
)) {
2565 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2566 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2567 get_sys_dev_type(hba
->type
));
2573 if (controller_path
&& result
== 1) {
2574 pr_err("no active Intel(R) RAID controller found under %s\n",
2579 const struct orom_entry
*entry
;
2581 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2582 if (entry
->type
== SYS_DEV_VMD
) {
2583 print_imsm_capability(&entry
->orom
);
2584 printf(" 3rd party NVMe :%s supported\n",
2585 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2586 for (hba
= list
; hba
; hba
= hba
->next
) {
2587 if (hba
->type
== SYS_DEV_VMD
) {
2589 printf(" I/O Controller : %s (%s)\n",
2590 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2591 if (print_nvme_info(hba
)) {
2593 pr_err("failed to get devices attached to VMD domain.\n");
2602 print_imsm_capability(&entry
->orom
);
2603 if (entry
->type
== SYS_DEV_NVME
) {
2604 for (hba
= list
; hba
; hba
= hba
->next
) {
2605 if (hba
->type
== SYS_DEV_NVME
)
2606 print_nvme_info(hba
);
2612 struct devid_list
*devid
;
2613 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2614 hba
= device_by_id(devid
->devid
);
2618 printf(" I/O Controller : %s (%s)\n",
2619 hba
->path
, get_sys_dev_type(hba
->type
));
2620 if (hba
->type
== SYS_DEV_SATA
) {
2621 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2622 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2624 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2635 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2637 struct sys_dev
*list
, *hba
;
2640 list
= find_intel_devices();
2643 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2648 for (hba
= list
; hba
; hba
= hba
->next
) {
2649 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2651 if (!find_imsm_capability(hba
) && verbose
> 0) {
2653 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2654 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2660 const struct orom_entry
*entry
;
2662 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2663 if (entry
->type
== SYS_DEV_VMD
) {
2664 for (hba
= list
; hba
; hba
= hba
->next
)
2665 print_imsm_capability_export(&entry
->orom
);
2668 print_imsm_capability_export(&entry
->orom
);
2674 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2676 /* the imsm metadata format does not specify any host
2677 * identification information. We return -1 since we can never
2678 * confirm nor deny whether a given array is "meant" for this
2679 * host. We rely on compare_super and the 'family_num' fields to
2680 * exclude member disks that do not belong, and we rely on
2681 * mdadm.conf to specify the arrays that should be assembled.
2682 * Auto-assembly may still pick up "foreign" arrays.
2688 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2690 /* The uuid returned here is used for:
2691 * uuid to put into bitmap file (Create, Grow)
2692 * uuid for backup header when saving critical section (Grow)
2693 * comparing uuids when re-adding a device into an array
2694 * In these cases the uuid required is that of the data-array,
2695 * not the device-set.
2696 * uuid to recognise same set when adding a missing device back
2697 * to an array. This is a uuid for the device-set.
2699 * For each of these we can make do with a truncated
2700 * or hashed uuid rather than the original, as long as
2702 * In each case the uuid required is that of the data-array,
2703 * not the device-set.
2705 /* imsm does not track uuid's so we synthesis one using sha1 on
2706 * - The signature (Which is constant for all imsm array, but no matter)
2707 * - the orig_family_num of the container
2708 * - the index number of the volume
2709 * - the 'serial' number of the volume.
2710 * Hopefully these are all constant.
2712 struct intel_super
*super
= st
->sb
;
2715 struct sha1_ctx ctx
;
2716 struct imsm_dev
*dev
= NULL
;
2719 /* some mdadm versions failed to set ->orig_family_num, in which
2720 * case fall back to ->family_num. orig_family_num will be
2721 * fixed up with the first metadata update.
2723 family_num
= super
->anchor
->orig_family_num
;
2724 if (family_num
== 0)
2725 family_num
= super
->anchor
->family_num
;
2726 sha1_init_ctx(&ctx
);
2727 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2728 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2729 if (super
->current_vol
>= 0)
2730 dev
= get_imsm_dev(super
, super
->current_vol
);
2732 __u32 vol
= super
->current_vol
;
2733 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2734 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2736 sha1_finish_ctx(&ctx
, buf
);
2737 memcpy(uuid
, buf
, 4*4);
2742 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2744 __u8
*v
= get_imsm_version(mpb
);
2745 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2746 char major
[] = { 0, 0, 0 };
2747 char minor
[] = { 0 ,0, 0 };
2748 char patch
[] = { 0, 0, 0 };
2749 char *ver_parse
[] = { major
, minor
, patch
};
2753 while (*v
!= '\0' && v
< end
) {
2754 if (*v
!= '.' && j
< 2)
2755 ver_parse
[i
][j
++] = *v
;
2763 *m
= strtol(minor
, NULL
, 0);
2764 *p
= strtol(patch
, NULL
, 0);
2768 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2770 /* migr_strip_size when repairing or initializing parity */
2771 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2772 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2774 switch (get_imsm_raid_level(map
)) {
2779 return 128*1024 >> 9;
2783 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2785 /* migr_strip_size when rebuilding a degraded disk, no idea why
2786 * this is different than migr_strip_size_resync(), but it's good
2789 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2790 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2792 switch (get_imsm_raid_level(map
)) {
2795 if (map
->num_members
% map
->num_domains
== 0)
2796 return 128*1024 >> 9;
2800 return max((__u32
) 64*1024 >> 9, chunk
);
2802 return 128*1024 >> 9;
2806 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2808 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2809 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2810 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2811 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2813 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2816 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2818 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2819 int level
= get_imsm_raid_level(lo
);
2821 if (level
== 1 || level
== 10) {
2822 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2824 return hi
->num_domains
;
2826 return num_stripes_per_unit_resync(dev
);
2829 static __u8
imsm_num_data_members(struct imsm_map
*map
)
2831 /* named 'imsm_' because raid0, raid1 and raid10
2832 * counter-intuitively have the same number of data disks
2834 switch (get_imsm_raid_level(map
)) {
2836 return map
->num_members
;
2840 return map
->num_members
/2;
2842 return map
->num_members
- 1;
2844 dprintf("unsupported raid level\n");
2849 static unsigned long long calc_component_size(struct imsm_map
*map
,
2850 struct imsm_dev
*dev
)
2852 unsigned long long component_size
;
2853 unsigned long long dev_size
= imsm_dev_size(dev
);
2854 long long calc_dev_size
= 0;
2855 unsigned int member_disks
= imsm_num_data_members(map
);
2857 if (member_disks
== 0)
2860 component_size
= per_dev_array_size(map
);
2861 calc_dev_size
= component_size
* member_disks
;
2863 /* Component size is rounded to 1MB so difference between size from
2864 * metadata and size calculated from num_data_stripes equals up to
2865 * 2048 blocks per each device. If the difference is higher it means
2866 * that array size was expanded and num_data_stripes was not updated.
2868 if (llabs(calc_dev_size
- (long long)dev_size
) >
2869 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
2870 component_size
= dev_size
/ member_disks
;
2871 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
2872 component_size
/ map
->blocks_per_strip
,
2873 num_data_stripes(map
));
2876 return component_size
;
2879 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2881 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2882 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2884 switch(get_imsm_raid_level(map
)) {
2887 return chunk
* map
->num_domains
;
2889 return chunk
* map
->num_members
;
2895 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2897 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2898 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2899 __u32 strip
= block
/ chunk
;
2901 switch (get_imsm_raid_level(map
)) {
2904 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2905 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2907 return vol_stripe
* chunk
+ block
% chunk
;
2909 __u32 stripe
= strip
/ (map
->num_members
- 1);
2911 return stripe
* chunk
+ block
% chunk
;
2918 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2919 struct imsm_dev
*dev
)
2921 /* calculate the conversion factor between per member 'blocks'
2922 * (md/{resync,rebuild}_start) and imsm migration units, return
2923 * 0 for the 'not migrating' and 'unsupported migration' cases
2925 if (!dev
->vol
.migr_state
)
2928 switch (migr_type(dev
)) {
2929 case MIGR_GEN_MIGR
: {
2930 struct migr_record
*migr_rec
= super
->migr_rec
;
2931 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2936 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2937 __u32 stripes_per_unit
;
2938 __u32 blocks_per_unit
;
2947 /* yes, this is really the translation of migr_units to
2948 * per-member blocks in the 'resync' case
2950 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2951 migr_chunk
= migr_strip_blocks_resync(dev
);
2952 disks
= imsm_num_data_members(map
);
2953 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2954 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2955 segment
= blocks_per_unit
/ stripe
;
2956 block_rel
= blocks_per_unit
- segment
* stripe
;
2957 parity_depth
= parity_segment_depth(dev
);
2958 block_map
= map_migr_block(dev
, block_rel
);
2959 return block_map
+ parity_depth
* segment
;
2961 case MIGR_REBUILD
: {
2962 __u32 stripes_per_unit
;
2965 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2966 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2967 return migr_chunk
* stripes_per_unit
;
2969 case MIGR_STATE_CHANGE
:
2975 static int imsm_level_to_layout(int level
)
2983 return ALGORITHM_LEFT_ASYMMETRIC
;
2990 /*******************************************************************************
2991 * Function: read_imsm_migr_rec
2992 * Description: Function reads imsm migration record from last sector of disk
2994 * fd : disk descriptor
2995 * super : metadata info
2999 ******************************************************************************/
3000 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
3003 unsigned int sector_size
= super
->sector_size
;
3004 unsigned long long dsize
;
3006 get_dev_size(fd
, NULL
, &dsize
);
3007 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
3009 pr_err("Cannot seek to anchor block: %s\n",
3013 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
3014 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3015 MIGR_REC_BUF_SECTORS
*sector_size
) {
3016 pr_err("Cannot read migr record block: %s\n",
3021 if (sector_size
== 4096)
3022 convert_from_4k_imsm_migr_rec(super
);
3028 static struct imsm_dev
*imsm_get_device_during_migration(
3029 struct intel_super
*super
)
3032 struct intel_dev
*dv
;
3034 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
3035 if (is_gen_migration(dv
->dev
))
3041 /*******************************************************************************
3042 * Function: load_imsm_migr_rec
3043 * Description: Function reads imsm migration record (it is stored at the last
3046 * super : imsm internal array info
3047 * info : general array info
3051 * -2 : no migration in progress
3052 ******************************************************************************/
3053 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
3060 struct imsm_dev
*dev
;
3061 struct imsm_map
*map
;
3064 /* find map under migration */
3065 dev
= imsm_get_device_during_migration(super
);
3066 /* nothing to load,no migration in progress?
3072 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
3073 /* read only from one of the first two slots */
3074 if ((sd
->disk
.raid_disk
< 0) ||
3075 (sd
->disk
.raid_disk
> 1))
3078 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3079 fd
= dev_open(nm
, O_RDONLY
);
3085 map
= get_imsm_map(dev
, MAP_0
);
3086 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3087 /* skip spare and failed disks
3091 /* read only from one of the first two slots */
3093 slot
= get_imsm_disk_slot(map
, dl
->index
);
3094 if (map
== NULL
|| slot
> 1 || slot
< 0)
3096 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3097 fd
= dev_open(nm
, O_RDONLY
);
3104 retval
= read_imsm_migr_rec(fd
, super
);
3112 /*******************************************************************************
3113 * function: imsm_create_metadata_checkpoint_update
3114 * Description: It creates update for checkpoint change.
3116 * super : imsm internal array info
3117 * u : pointer to prepared update
3120 * If length is equal to 0, input pointer u contains no update
3121 ******************************************************************************/
3122 static int imsm_create_metadata_checkpoint_update(
3123 struct intel_super
*super
,
3124 struct imsm_update_general_migration_checkpoint
**u
)
3127 int update_memory_size
= 0;
3129 dprintf("(enter)\n");
3135 /* size of all update data without anchor */
3136 update_memory_size
=
3137 sizeof(struct imsm_update_general_migration_checkpoint
);
3139 *u
= xcalloc(1, update_memory_size
);
3141 dprintf("error: cannot get memory\n");
3144 (*u
)->type
= update_general_migration_checkpoint
;
3145 (*u
)->curr_migr_unit
= current_migr_unit(super
->migr_rec
);
3146 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3148 return update_memory_size
;
3151 static void imsm_update_metadata_locally(struct supertype
*st
,
3152 void *buf
, int len
);
3154 /*******************************************************************************
3155 * Function: write_imsm_migr_rec
3156 * Description: Function writes imsm migration record
3157 * (at the last sector of disk)
3159 * super : imsm internal array info
3163 ******************************************************************************/
3164 static int write_imsm_migr_rec(struct supertype
*st
)
3166 struct intel_super
*super
= st
->sb
;
3167 unsigned int sector_size
= super
->sector_size
;
3168 unsigned long long dsize
;
3174 struct imsm_update_general_migration_checkpoint
*u
;
3175 struct imsm_dev
*dev
;
3176 struct imsm_map
*map
;
3178 /* find map under migration */
3179 dev
= imsm_get_device_during_migration(super
);
3180 /* if no migration, write buffer anyway to clear migr_record
3181 * on disk based on first available device
3184 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3185 super
->current_vol
);
3187 map
= get_imsm_map(dev
, MAP_0
);
3189 if (sector_size
== 4096)
3190 convert_to_4k_imsm_migr_rec(super
);
3191 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3194 /* skip failed and spare devices */
3197 /* write to 2 first slots only */
3199 slot
= get_imsm_disk_slot(map
, sd
->index
);
3200 if (map
== NULL
|| slot
> 1 || slot
< 0)
3203 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3204 fd
= dev_open(nm
, O_RDWR
);
3207 get_dev_size(fd
, NULL
, &dsize
);
3208 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3210 pr_err("Cannot seek to anchor block: %s\n",
3214 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3215 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3216 MIGR_REC_BUF_SECTORS
*sector_size
) {
3217 pr_err("Cannot write migr record block: %s\n",
3224 if (sector_size
== 4096)
3225 convert_from_4k_imsm_migr_rec(super
);
3226 /* update checkpoint information in metadata */
3227 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3229 dprintf("imsm: Cannot prepare update\n");
3232 /* update metadata locally */
3233 imsm_update_metadata_locally(st
, u
, len
);
3234 /* and possibly remotely */
3235 if (st
->update_tail
) {
3236 append_metadata_update(st
, u
, len
);
3237 /* during reshape we do all work inside metadata handler
3238 * manage_reshape(), so metadata update has to be triggered
3241 flush_metadata_updates(st
);
3242 st
->update_tail
= &st
->updates
;
3253 /* spare/missing disks activations are not allowe when
3254 * array/container performs reshape operation, because
3255 * all arrays in container works on the same disks set
3257 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3260 struct intel_dev
*i_dev
;
3261 struct imsm_dev
*dev
;
3263 /* check whole container
3265 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3267 if (is_gen_migration(dev
)) {
3268 /* No repair during any migration in container
3276 static unsigned long long imsm_component_size_alignment_check(int level
,
3278 unsigned int sector_size
,
3279 unsigned long long component_size
)
3281 unsigned int component_size_alignment
;
3283 /* check component size alignment
3285 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3287 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3288 level
, chunk_size
, component_size
,
3289 component_size_alignment
);
3291 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3292 dprintf("imsm: reported component size aligned from %llu ",
3294 component_size
-= component_size_alignment
;
3295 dprintf_cont("to %llu (%i).\n",
3296 component_size
, component_size_alignment
);
3299 return component_size
;
3302 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3304 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3305 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3307 return pba_of_lba0(map
) +
3308 (num_data_stripes(map
) * map
->blocks_per_strip
);
3311 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3313 struct intel_super
*super
= st
->sb
;
3314 struct migr_record
*migr_rec
= super
->migr_rec
;
3315 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3316 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3317 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3318 struct imsm_map
*map_to_analyse
= map
;
3320 int map_disks
= info
->array
.raid_disks
;
3322 memset(info
, 0, sizeof(*info
));
3324 map_to_analyse
= prev_map
;
3326 dl
= super
->current_disk
;
3328 info
->container_member
= super
->current_vol
;
3329 info
->array
.raid_disks
= map
->num_members
;
3330 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3331 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3332 info
->array
.md_minor
= -1;
3333 info
->array
.ctime
= 0;
3334 info
->array
.utime
= 0;
3335 info
->array
.chunk_size
=
3336 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3337 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3338 info
->custom_array_size
= imsm_dev_size(dev
);
3339 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3341 if (is_gen_migration(dev
)) {
3342 info
->reshape_active
= 1;
3343 info
->new_level
= get_imsm_raid_level(map
);
3344 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3345 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3346 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3347 if (info
->delta_disks
) {
3348 /* this needs to be applied to every array
3351 info
->reshape_active
= CONTAINER_RESHAPE
;
3353 /* We shape information that we give to md might have to be
3354 * modify to cope with md's requirement for reshaping arrays.
3355 * For example, when reshaping a RAID0, md requires it to be
3356 * presented as a degraded RAID4.
3357 * Also if a RAID0 is migrating to a RAID5 we need to specify
3358 * the array as already being RAID5, but the 'before' layout
3359 * is a RAID4-like layout.
3361 switch (info
->array
.level
) {
3363 switch(info
->new_level
) {
3365 /* conversion is happening as RAID4 */
3366 info
->array
.level
= 4;
3367 info
->array
.raid_disks
+= 1;
3370 /* conversion is happening as RAID5 */
3371 info
->array
.level
= 5;
3372 info
->array
.layout
= ALGORITHM_PARITY_N
;
3373 info
->delta_disks
-= 1;
3376 /* FIXME error message */
3377 info
->array
.level
= UnSet
;
3383 info
->new_level
= UnSet
;
3384 info
->new_layout
= UnSet
;
3385 info
->new_chunk
= info
->array
.chunk_size
;
3386 info
->delta_disks
= 0;
3390 info
->disk
.major
= dl
->major
;
3391 info
->disk
.minor
= dl
->minor
;
3392 info
->disk
.number
= dl
->index
;
3393 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3397 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3398 info
->component_size
= calc_component_size(map
, dev
);
3399 info
->component_size
= imsm_component_size_alignment_check(
3401 info
->array
.chunk_size
,
3403 info
->component_size
);
3404 info
->bb
.supported
= 1;
3406 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3407 info
->recovery_start
= MaxSector
;
3409 if (info
->array
.level
== 5 &&
3410 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3411 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3412 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3413 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3414 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3415 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3417 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3419 } else if (info
->array
.level
<= 0) {
3420 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3422 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3425 info
->reshape_progress
= 0;
3426 info
->resync_start
= MaxSector
;
3427 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3428 !(info
->array
.state
& 1)) &&
3429 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3430 info
->resync_start
= 0;
3432 if (dev
->vol
.migr_state
) {
3433 switch (migr_type(dev
)) {
3436 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3438 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3440 info
->resync_start
= blocks_per_unit
* units
;
3443 case MIGR_GEN_MIGR
: {
3444 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3446 __u64 units
= current_migr_unit(migr_rec
);
3447 unsigned long long array_blocks
;
3450 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3452 (get_num_migr_units(migr_rec
)-1)) &&
3453 (super
->migr_rec
->rec_status
==
3454 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3457 info
->reshape_progress
= blocks_per_unit
* units
;
3459 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3460 (unsigned long long)units
,
3461 (unsigned long long)blocks_per_unit
,
3462 info
->reshape_progress
);
3464 used_disks
= imsm_num_data_members(prev_map
);
3465 if (used_disks
> 0) {
3466 array_blocks
= per_dev_array_size(map
) *
3468 info
->custom_array_size
=
3469 round_size_to_mb(array_blocks
,
3475 /* we could emulate the checkpointing of
3476 * 'sync_action=check' migrations, but for now
3477 * we just immediately complete them
3480 /* this is handled by container_content_imsm() */
3481 case MIGR_STATE_CHANGE
:
3482 /* FIXME handle other migrations */
3484 /* we are not dirty, so... */
3485 info
->resync_start
= MaxSector
;
3489 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3490 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3492 info
->array
.major_version
= -1;
3493 info
->array
.minor_version
= -2;
3494 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3495 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3496 uuid_from_super_imsm(st
, info
->uuid
);
3500 for (i
=0; i
<map_disks
; i
++) {
3502 if (i
< info
->array
.raid_disks
) {
3503 struct imsm_disk
*dsk
;
3504 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3505 dsk
= get_imsm_disk(super
, j
);
3506 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3513 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3514 int failed
, int look_in_map
);
3516 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3519 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3521 if (is_gen_migration(dev
)) {
3524 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3526 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3527 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3528 if (map2
->map_state
!= map_state
) {
3529 map2
->map_state
= map_state
;
3530 super
->updates_pending
++;
3535 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3539 for (d
= super
->missing
; d
; d
= d
->next
)
3540 if (d
->index
== index
)
3545 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3547 struct intel_super
*super
= st
->sb
;
3548 struct imsm_disk
*disk
;
3549 int map_disks
= info
->array
.raid_disks
;
3550 int max_enough
= -1;
3552 struct imsm_super
*mpb
;
3554 if (super
->current_vol
>= 0) {
3555 getinfo_super_imsm_volume(st
, info
, map
);
3558 memset(info
, 0, sizeof(*info
));
3560 /* Set raid_disks to zero so that Assemble will always pull in valid
3563 info
->array
.raid_disks
= 0;
3564 info
->array
.level
= LEVEL_CONTAINER
;
3565 info
->array
.layout
= 0;
3566 info
->array
.md_minor
= -1;
3567 info
->array
.ctime
= 0; /* N/A for imsm */
3568 info
->array
.utime
= 0;
3569 info
->array
.chunk_size
= 0;
3571 info
->disk
.major
= 0;
3572 info
->disk
.minor
= 0;
3573 info
->disk
.raid_disk
= -1;
3574 info
->reshape_active
= 0;
3575 info
->array
.major_version
= -1;
3576 info
->array
.minor_version
= -2;
3577 strcpy(info
->text_version
, "imsm");
3578 info
->safe_mode_delay
= 0;
3579 info
->disk
.number
= -1;
3580 info
->disk
.state
= 0;
3582 info
->recovery_start
= MaxSector
;
3583 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3584 info
->bb
.supported
= 1;
3586 /* do we have the all the insync disks that we expect? */
3587 mpb
= super
->anchor
;
3588 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3590 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3591 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3592 int failed
, enough
, j
, missing
= 0;
3593 struct imsm_map
*map
;
3596 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3597 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3598 map
= get_imsm_map(dev
, MAP_0
);
3600 /* any newly missing disks?
3601 * (catches single-degraded vs double-degraded)
3603 for (j
= 0; j
< map
->num_members
; j
++) {
3604 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3605 __u32 idx
= ord_to_idx(ord
);
3607 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3608 info
->disk
.raid_disk
= j
;
3610 if (!(ord
& IMSM_ORD_REBUILD
) &&
3611 get_imsm_missing(super
, idx
)) {
3617 if (state
== IMSM_T_STATE_FAILED
)
3619 else if (state
== IMSM_T_STATE_DEGRADED
&&
3620 (state
!= map
->map_state
|| missing
))
3622 else /* we're normal, or already degraded */
3624 if (is_gen_migration(dev
) && missing
) {
3625 /* during general migration we need all disks
3626 * that process is running on.
3627 * No new missing disk is allowed.
3631 /* no more checks necessary
3635 /* in the missing/failed disk case check to see
3636 * if at least one array is runnable
3638 max_enough
= max(max_enough
, enough
);
3640 dprintf("enough: %d\n", max_enough
);
3641 info
->container_enough
= max_enough
;
3644 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3646 disk
= &super
->disks
->disk
;
3647 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3648 info
->component_size
= reserved
;
3649 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3650 /* we don't change info->disk.raid_disk here because
3651 * this state will be finalized in mdmon after we have
3652 * found the 'most fresh' version of the metadata
3654 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3655 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3656 0 : (1 << MD_DISK_SYNC
);
3659 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3660 * ->compare_super may have updated the 'num_raid_devs' field for spares
3662 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3663 uuid_from_super_imsm(st
, info
->uuid
);
3665 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3667 /* I don't know how to compute 'map' on imsm, so use safe default */
3670 for (i
= 0; i
< map_disks
; i
++)
3676 /* allocates memory and fills disk in mdinfo structure
3677 * for each disk in array */
3678 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3680 struct mdinfo
*mddev
;
3681 struct intel_super
*super
= st
->sb
;
3682 struct imsm_disk
*disk
;
3685 if (!super
|| !super
->disks
)
3688 mddev
= xcalloc(1, sizeof(*mddev
));
3692 tmp
= xcalloc(1, sizeof(*tmp
));
3694 tmp
->next
= mddev
->devs
;
3696 tmp
->disk
.number
= count
++;
3697 tmp
->disk
.major
= dl
->major
;
3698 tmp
->disk
.minor
= dl
->minor
;
3699 tmp
->disk
.state
= is_configured(disk
) ?
3700 (1 << MD_DISK_ACTIVE
) : 0;
3701 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3702 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3703 tmp
->disk
.raid_disk
= -1;
3709 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3710 char *update
, char *devname
, int verbose
,
3711 int uuid_set
, char *homehost
)
3713 /* For 'assemble' and 'force' we need to return non-zero if any
3714 * change was made. For others, the return value is ignored.
3715 * Update options are:
3716 * force-one : This device looks a bit old but needs to be included,
3717 * update age info appropriately.
3718 * assemble: clear any 'faulty' flag to allow this device to
3720 * force-array: Array is degraded but being forced, mark it clean
3721 * if that will be needed to assemble it.
3723 * newdev: not used ????
3724 * grow: Array has gained a new device - this is currently for
3726 * resync: mark as dirty so a resync will happen.
3727 * name: update the name - preserving the homehost
3728 * uuid: Change the uuid of the array to match watch is given
3730 * Following are not relevant for this imsm:
3731 * sparc2.2 : update from old dodgey metadata
3732 * super-minor: change the preferred_minor number
3733 * summaries: update redundant counters.
3734 * homehost: update the recorded homehost
3735 * _reshape_progress: record new reshape_progress position.
3738 struct intel_super
*super
= st
->sb
;
3739 struct imsm_super
*mpb
;
3741 /* we can only update container info */
3742 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3745 mpb
= super
->anchor
;
3747 if (strcmp(update
, "uuid") == 0) {
3748 /* We take this to mean that the family_num should be updated.
3749 * However that is much smaller than the uuid so we cannot really
3750 * allow an explicit uuid to be given. And it is hard to reliably
3752 * So if !uuid_set we know the current uuid is random and just used
3753 * the first 'int' and copy it to the other 3 positions.
3754 * Otherwise we require the 4 'int's to be the same as would be the
3755 * case if we are using a random uuid. So an explicit uuid will be
3756 * accepted as long as all for ints are the same... which shouldn't hurt
3759 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3762 if (info
->uuid
[0] != info
->uuid
[1] ||
3763 info
->uuid
[1] != info
->uuid
[2] ||
3764 info
->uuid
[2] != info
->uuid
[3])
3770 mpb
->orig_family_num
= info
->uuid
[0];
3771 } else if (strcmp(update
, "assemble") == 0)
3776 /* successful update? recompute checksum */
3778 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3783 static size_t disks_to_mpb_size(int disks
)
3787 size
= sizeof(struct imsm_super
);
3788 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3789 size
+= 2 * sizeof(struct imsm_dev
);
3790 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3791 size
+= (4 - 2) * sizeof(struct imsm_map
);
3792 /* 4 possible disk_ord_tbl's */
3793 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3794 /* maximum bbm log */
3795 size
+= sizeof(struct bbm_log
);
3800 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3801 unsigned long long data_offset
)
3803 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3806 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3809 static void free_devlist(struct intel_super
*super
)
3811 struct intel_dev
*dv
;
3813 while (super
->devlist
) {
3814 dv
= super
->devlist
->next
;
3815 free(super
->devlist
->dev
);
3816 free(super
->devlist
);
3817 super
->devlist
= dv
;
3821 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3823 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3826 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3830 * 0 same, or first was empty, and second was copied
3831 * 1 second had wrong number
3833 * 3 wrong other info
3835 struct intel_super
*first
= st
->sb
;
3836 struct intel_super
*sec
= tst
->sb
;
3843 /* in platform dependent environment test if the disks
3844 * use the same Intel hba
3845 * If not on Intel hba at all, allow anything.
3847 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3848 if (first
->hba
->type
!= sec
->hba
->type
) {
3850 "HBAs of devices do not match %s != %s\n",
3851 get_sys_dev_type(first
->hba
->type
),
3852 get_sys_dev_type(sec
->hba
->type
));
3855 if (first
->orom
!= sec
->orom
) {
3857 "HBAs of devices do not match %s != %s\n",
3858 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3863 /* if an anchor does not have num_raid_devs set then it is a free
3866 if (first
->anchor
->num_raid_devs
> 0 &&
3867 sec
->anchor
->num_raid_devs
> 0) {
3868 /* Determine if these disks might ever have been
3869 * related. Further disambiguation can only take place
3870 * in load_super_imsm_all
3872 __u32 first_family
= first
->anchor
->orig_family_num
;
3873 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3875 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3876 MAX_SIGNATURE_LENGTH
) != 0)
3879 if (first_family
== 0)
3880 first_family
= first
->anchor
->family_num
;
3881 if (sec_family
== 0)
3882 sec_family
= sec
->anchor
->family_num
;
3884 if (first_family
!= sec_family
)
3889 /* if 'first' is a spare promote it to a populated mpb with sec's
3892 if (first
->anchor
->num_raid_devs
== 0 &&
3893 sec
->anchor
->num_raid_devs
> 0) {
3895 struct intel_dev
*dv
;
3896 struct imsm_dev
*dev
;
3898 /* we need to copy raid device info from sec if an allocation
3899 * fails here we don't associate the spare
3901 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3902 dv
= xmalloc(sizeof(*dv
));
3903 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3906 dv
->next
= first
->devlist
;
3907 first
->devlist
= dv
;
3909 if (i
< sec
->anchor
->num_raid_devs
) {
3910 /* allocation failure */
3911 free_devlist(first
);
3912 pr_err("imsm: failed to associate spare\n");
3915 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3916 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3917 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3918 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3919 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3920 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3926 static void fd2devname(int fd
, char *name
)
3930 char dname
[PATH_MAX
];
3935 if (fstat(fd
, &st
) != 0)
3937 sprintf(path
, "/sys/dev/block/%d:%d",
3938 major(st
.st_rdev
), minor(st
.st_rdev
));
3940 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3945 nm
= strrchr(dname
, '/');
3948 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3952 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3955 char *name
= fd2kname(fd
);
3960 if (strncmp(name
, "nvme", 4) != 0)
3963 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3965 return load_sys(path
, buf
, buf_len
);
3968 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3970 static int imsm_read_serial(int fd
, char *devname
,
3971 __u8
*serial
, size_t serial_buf_len
)
3980 memset(buf
, 0, sizeof(buf
));
3982 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3985 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3987 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3988 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3989 fd2devname(fd
, (char *) serial
);
3995 pr_err("Failed to retrieve serial for %s\n",
4000 /* trim all whitespace and non-printable characters and convert
4003 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
4006 /* ':' is reserved for use in placeholder serial
4007 * numbers for missing disks
4018 if (len
> serial_buf_len
) {
4019 /* truncate leading characters */
4020 dest
+= len
- serial_buf_len
;
4021 len
= serial_buf_len
;
4024 memset(serial
, 0, serial_buf_len
);
4025 memcpy(serial
, dest
, len
);
4030 static int serialcmp(__u8
*s1
, __u8
*s2
)
4032 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
4035 static void serialcpy(__u8
*dest
, __u8
*src
)
4037 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
4040 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
4044 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4045 if (serialcmp(dl
->serial
, serial
) == 0)
4051 static struct imsm_disk
*
4052 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4056 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4057 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4059 if (serialcmp(disk
->serial
, serial
) == 0) {
4070 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4072 struct imsm_disk
*disk
;
4077 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4079 rv
= imsm_read_serial(fd
, devname
, serial
, MAX_RAID_SERIAL_LEN
);
4084 dl
= xcalloc(1, sizeof(*dl
));
4087 dl
->major
= major(stb
.st_rdev
);
4088 dl
->minor
= minor(stb
.st_rdev
);
4089 dl
->next
= super
->disks
;
4090 dl
->fd
= keep_fd
? fd
: -1;
4091 assert(super
->disks
== NULL
);
4093 serialcpy(dl
->serial
, serial
);
4096 fd2devname(fd
, name
);
4098 dl
->devname
= xstrdup(devname
);
4100 dl
->devname
= xstrdup(name
);
4102 /* look up this disk's index in the current anchor */
4103 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4106 /* only set index on disks that are a member of a
4107 * populated contianer, i.e. one with raid_devs
4109 if (is_failed(&dl
->disk
))
4111 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4118 /* When migrating map0 contains the 'destination' state while map1
4119 * contains the current state. When not migrating map0 contains the
4120 * current state. This routine assumes that map[0].map_state is set to
4121 * the current array state before being called.
4123 * Migration is indicated by one of the following states
4124 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4125 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4126 * map1state=unitialized)
4127 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4129 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4130 * map1state=degraded)
4131 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4134 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4135 __u8 to_state
, int migr_type
)
4137 struct imsm_map
*dest
;
4138 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4140 dev
->vol
.migr_state
= 1;
4141 set_migr_type(dev
, migr_type
);
4142 dev
->vol
.curr_migr_unit
= 0;
4143 dest
= get_imsm_map(dev
, MAP_1
);
4145 /* duplicate and then set the target end state in map[0] */
4146 memcpy(dest
, src
, sizeof_imsm_map(src
));
4147 if (migr_type
== MIGR_GEN_MIGR
) {
4151 for (i
= 0; i
< src
->num_members
; i
++) {
4152 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4153 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4157 if (migr_type
== MIGR_GEN_MIGR
)
4158 /* Clear migration record */
4159 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4161 src
->map_state
= to_state
;
4164 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4167 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4168 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4172 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4173 * completed in the last migration.
4175 * FIXME add support for raid-level-migration
4177 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4178 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4179 /* when final map state is other than expected
4180 * merge maps (not for migration)
4184 for (i
= 0; i
< prev
->num_members
; i
++)
4185 for (j
= 0; j
< map
->num_members
; j
++)
4186 /* during online capacity expansion
4187 * disks position can be changed
4188 * if takeover is used
4190 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4191 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4192 map
->disk_ord_tbl
[j
] |=
4193 prev
->disk_ord_tbl
[i
];
4196 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4197 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4200 dev
->vol
.migr_state
= 0;
4201 set_migr_type(dev
, 0);
4202 dev
->vol
.curr_migr_unit
= 0;
4203 map
->map_state
= map_state
;
4206 static int parse_raid_devices(struct intel_super
*super
)
4209 struct imsm_dev
*dev_new
;
4210 size_t len
, len_migr
;
4212 size_t space_needed
= 0;
4213 struct imsm_super
*mpb
= super
->anchor
;
4215 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4216 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4217 struct intel_dev
*dv
;
4219 len
= sizeof_imsm_dev(dev_iter
, 0);
4220 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4222 space_needed
+= len_migr
- len
;
4224 dv
= xmalloc(sizeof(*dv
));
4225 if (max_len
< len_migr
)
4227 if (max_len
> len_migr
)
4228 space_needed
+= max_len
- len_migr
;
4229 dev_new
= xmalloc(max_len
);
4230 imsm_copy_dev(dev_new
, dev_iter
);
4233 dv
->next
= super
->devlist
;
4234 super
->devlist
= dv
;
4237 /* ensure that super->buf is large enough when all raid devices
4240 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4243 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4244 super
->sector_size
);
4245 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4248 memcpy(buf
, super
->buf
, super
->len
);
4249 memset(buf
+ super
->len
, 0, len
- super
->len
);
4255 super
->extra_space
+= space_needed
;
4260 /*******************************************************************************
4261 * Function: check_mpb_migr_compatibility
4262 * Description: Function checks for unsupported migration features:
4263 * - migration optimization area (pba_of_lba0)
4264 * - descending reshape (ascending_migr)
4266 * super : imsm metadata information
4268 * 0 : migration is compatible
4269 * -1 : migration is not compatible
4270 ******************************************************************************/
4271 int check_mpb_migr_compatibility(struct intel_super
*super
)
4273 struct imsm_map
*map0
, *map1
;
4274 struct migr_record
*migr_rec
= super
->migr_rec
;
4277 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4278 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4281 dev_iter
->vol
.migr_state
== 1 &&
4282 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4283 /* This device is migrating */
4284 map0
= get_imsm_map(dev_iter
, MAP_0
);
4285 map1
= get_imsm_map(dev_iter
, MAP_1
);
4286 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4287 /* migration optimization area was used */
4289 if (migr_rec
->ascending_migr
== 0 &&
4290 migr_rec
->dest_depth_per_unit
> 0)
4291 /* descending reshape not supported yet */
4298 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4300 /* load_imsm_mpb - read matrix metadata
4301 * allocates super->mpb to be freed by free_imsm
4303 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4305 unsigned long long dsize
;
4306 unsigned long long sectors
;
4307 unsigned int sector_size
= super
->sector_size
;
4309 struct imsm_super
*anchor
;
4312 get_dev_size(fd
, NULL
, &dsize
);
4313 if (dsize
< 2*sector_size
) {
4315 pr_err("%s: device to small for imsm\n",
4320 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4322 pr_err("Cannot seek to anchor block on %s: %s\n",
4323 devname
, strerror(errno
));
4327 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4329 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4332 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4334 pr_err("Cannot read anchor block on %s: %s\n",
4335 devname
, strerror(errno
));
4340 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4342 pr_err("no IMSM anchor on %s\n", devname
);
4347 __free_imsm(super
, 0);
4348 /* reload capability and hba */
4350 /* capability and hba must be updated with new super allocation */
4351 find_intel_hba_capability(fd
, super
, devname
);
4352 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4353 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4355 pr_err("unable to allocate %zu byte mpb buffer\n",
4360 memcpy(super
->buf
, anchor
, sector_size
);
4362 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4365 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4366 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4367 pr_err("could not allocate migr_rec buffer\n");
4371 super
->clean_migration_record_by_mdmon
= 0;
4374 check_sum
= __gen_imsm_checksum(super
->anchor
);
4375 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4377 pr_err("IMSM checksum %x != %x on %s\n",
4379 __le32_to_cpu(super
->anchor
->check_sum
),
4387 /* read the extended mpb */
4388 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4390 pr_err("Cannot seek to extended mpb on %s: %s\n",
4391 devname
, strerror(errno
));
4395 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4396 super
->len
- sector_size
) != super
->len
- sector_size
) {
4398 pr_err("Cannot read extended mpb on %s: %s\n",
4399 devname
, strerror(errno
));
4403 check_sum
= __gen_imsm_checksum(super
->anchor
);
4404 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4406 pr_err("IMSM checksum %x != %x on %s\n",
4407 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4415 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4417 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4418 static void clear_hi(struct intel_super
*super
)
4420 struct imsm_super
*mpb
= super
->anchor
;
4422 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4424 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4425 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4426 disk
->total_blocks_hi
= 0;
4428 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4429 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4432 for (n
= 0; n
< 2; ++n
) {
4433 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4436 map
->pba_of_lba0_hi
= 0;
4437 map
->blocks_per_member_hi
= 0;
4438 map
->num_data_stripes_hi
= 0;
4444 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4448 err
= load_imsm_mpb(fd
, super
, devname
);
4451 if (super
->sector_size
== 4096)
4452 convert_from_4k(super
);
4453 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4456 err
= parse_raid_devices(super
);
4459 err
= load_bbm_log(super
);
4464 static void __free_imsm_disk(struct dl
*d
)
4476 static void free_imsm_disks(struct intel_super
*super
)
4480 while (super
->disks
) {
4482 super
->disks
= d
->next
;
4483 __free_imsm_disk(d
);
4485 while (super
->disk_mgmt_list
) {
4486 d
= super
->disk_mgmt_list
;
4487 super
->disk_mgmt_list
= d
->next
;
4488 __free_imsm_disk(d
);
4490 while (super
->missing
) {
4492 super
->missing
= d
->next
;
4493 __free_imsm_disk(d
);
4498 /* free all the pieces hanging off of a super pointer */
4499 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4501 struct intel_hba
*elem
, *next
;
4507 /* unlink capability description */
4509 if (super
->migr_rec_buf
) {
4510 free(super
->migr_rec_buf
);
4511 super
->migr_rec_buf
= NULL
;
4514 free_imsm_disks(super
);
4515 free_devlist(super
);
4519 free((void *)elem
->path
);
4525 free(super
->bbm_log
);
4529 static void free_imsm(struct intel_super
*super
)
4531 __free_imsm(super
, 1);
4532 free(super
->bb
.entries
);
4536 static void free_super_imsm(struct supertype
*st
)
4538 struct intel_super
*super
= st
->sb
;
4547 static struct intel_super
*alloc_super(void)
4549 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4551 super
->current_vol
= -1;
4552 super
->create_offset
= ~((unsigned long long) 0);
4554 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4555 sizeof(struct md_bb_entry
));
4556 if (!super
->bb
.entries
) {
4565 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4567 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4569 struct sys_dev
*hba_name
;
4572 if (fd
>= 0 && test_partition(fd
)) {
4573 pr_err("imsm: %s is a partition, cannot be used in IMSM\n",
4577 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4582 hba_name
= find_disk_attached_hba(fd
, NULL
);
4585 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4589 rv
= attach_hba_to_super(super
, hba_name
);
4592 struct intel_hba
*hba
= super
->hba
;
4594 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4595 " but the container is assigned to Intel(R) %s %s (",
4597 get_sys_dev_type(hba_name
->type
),
4598 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4599 hba_name
->pci_id
? : "Err!",
4600 get_sys_dev_type(super
->hba
->type
),
4601 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4604 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4606 fprintf(stderr
, ", ");
4609 fprintf(stderr
, ").\n"
4610 " Mixing devices attached to different controllers is not allowed.\n");
4614 super
->orom
= find_imsm_capability(hba_name
);
4621 /* find_missing - helper routine for load_super_imsm_all that identifies
4622 * disks that have disappeared from the system. This routine relies on
4623 * the mpb being uptodate, which it is at load time.
4625 static int find_missing(struct intel_super
*super
)
4628 struct imsm_super
*mpb
= super
->anchor
;
4630 struct imsm_disk
*disk
;
4632 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4633 disk
= __get_imsm_disk(mpb
, i
);
4634 dl
= serial_to_dl(disk
->serial
, super
);
4638 dl
= xmalloc(sizeof(*dl
));
4642 dl
->devname
= xstrdup("missing");
4644 serialcpy(dl
->serial
, disk
->serial
);
4647 dl
->next
= super
->missing
;
4648 super
->missing
= dl
;
4654 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4656 struct intel_disk
*idisk
= disk_list
;
4659 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4661 idisk
= idisk
->next
;
4667 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4668 struct intel_super
*super
,
4669 struct intel_disk
**disk_list
)
4671 struct imsm_disk
*d
= &super
->disks
->disk
;
4672 struct imsm_super
*mpb
= super
->anchor
;
4675 for (i
= 0; i
< tbl_size
; i
++) {
4676 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4677 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4679 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4680 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4681 dprintf("mpb from %d:%d matches %d:%d\n",
4682 super
->disks
->major
,
4683 super
->disks
->minor
,
4684 table
[i
]->disks
->major
,
4685 table
[i
]->disks
->minor
);
4689 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4690 is_configured(d
) == is_configured(tbl_d
)) &&
4691 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4692 /* current version of the mpb is a
4693 * better candidate than the one in
4694 * super_table, but copy over "cross
4695 * generational" status
4697 struct intel_disk
*idisk
;
4699 dprintf("mpb from %d:%d replaces %d:%d\n",
4700 super
->disks
->major
,
4701 super
->disks
->minor
,
4702 table
[i
]->disks
->major
,
4703 table
[i
]->disks
->minor
);
4705 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4706 if (idisk
&& is_failed(&idisk
->disk
))
4707 tbl_d
->status
|= FAILED_DISK
;
4710 struct intel_disk
*idisk
;
4711 struct imsm_disk
*disk
;
4713 /* tbl_mpb is more up to date, but copy
4714 * over cross generational status before
4717 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4718 if (disk
&& is_failed(disk
))
4719 d
->status
|= FAILED_DISK
;
4721 idisk
= disk_list_get(d
->serial
, *disk_list
);
4724 if (disk
&& is_configured(disk
))
4725 idisk
->disk
.status
|= CONFIGURED_DISK
;
4728 dprintf("mpb from %d:%d prefer %d:%d\n",
4729 super
->disks
->major
,
4730 super
->disks
->minor
,
4731 table
[i
]->disks
->major
,
4732 table
[i
]->disks
->minor
);
4740 table
[tbl_size
++] = super
;
4744 /* update/extend the merged list of imsm_disk records */
4745 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4746 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4747 struct intel_disk
*idisk
;
4749 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4751 idisk
->disk
.status
|= disk
->status
;
4752 if (is_configured(&idisk
->disk
) ||
4753 is_failed(&idisk
->disk
))
4754 idisk
->disk
.status
&= ~(SPARE_DISK
);
4756 idisk
= xcalloc(1, sizeof(*idisk
));
4757 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4758 idisk
->disk
= *disk
;
4759 idisk
->next
= *disk_list
;
4763 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4770 static struct intel_super
*
4771 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4774 struct imsm_super
*mpb
= super
->anchor
;
4778 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4779 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4780 struct intel_disk
*idisk
;
4782 idisk
= disk_list_get(disk
->serial
, disk_list
);
4784 if (idisk
->owner
== owner
||
4785 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4788 dprintf("'%.16s' owner %d != %d\n",
4789 disk
->serial
, idisk
->owner
,
4792 dprintf("unknown disk %x [%d]: %.16s\n",
4793 __le32_to_cpu(mpb
->family_num
), i
,
4799 if (ok_count
== mpb
->num_disks
)
4804 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4806 struct intel_super
*s
;
4808 for (s
= super_list
; s
; s
= s
->next
) {
4809 if (family_num
!= s
->anchor
->family_num
)
4811 pr_err("Conflict, offlining family %#x on '%s'\n",
4812 __le32_to_cpu(family_num
), s
->disks
->devname
);
4816 static struct intel_super
*
4817 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4819 struct intel_super
*super_table
[len
];
4820 struct intel_disk
*disk_list
= NULL
;
4821 struct intel_super
*champion
, *spare
;
4822 struct intel_super
*s
, **del
;
4827 memset(super_table
, 0, sizeof(super_table
));
4828 for (s
= *super_list
; s
; s
= s
->next
)
4829 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4831 for (i
= 0; i
< tbl_size
; i
++) {
4832 struct imsm_disk
*d
;
4833 struct intel_disk
*idisk
;
4834 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4837 d
= &s
->disks
->disk
;
4839 /* 'd' must appear in merged disk list for its
4840 * configuration to be valid
4842 idisk
= disk_list_get(d
->serial
, disk_list
);
4843 if (idisk
&& idisk
->owner
== i
)
4844 s
= validate_members(s
, disk_list
, i
);
4849 dprintf("marking family: %#x from %d:%d offline\n",
4851 super_table
[i
]->disks
->major
,
4852 super_table
[i
]->disks
->minor
);
4856 /* This is where the mdadm implementation differs from the Windows
4857 * driver which has no strict concept of a container. We can only
4858 * assemble one family from a container, so when returning a prodigal
4859 * array member to this system the code will not be able to disambiguate
4860 * the container contents that should be assembled ("foreign" versus
4861 * "local"). It requires user intervention to set the orig_family_num
4862 * to a new value to establish a new container. The Windows driver in
4863 * this situation fixes up the volume name in place and manages the
4864 * foreign array as an independent entity.
4869 for (i
= 0; i
< tbl_size
; i
++) {
4870 struct intel_super
*tbl_ent
= super_table
[i
];
4876 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4881 if (s
&& !is_spare
) {
4882 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4884 } else if (!s
&& !is_spare
)
4897 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4898 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4900 /* collect all dl's onto 'champion', and update them to
4901 * champion's version of the status
4903 for (s
= *super_list
; s
; s
= s
->next
) {
4904 struct imsm_super
*mpb
= champion
->anchor
;
4905 struct dl
*dl
= s
->disks
;
4910 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4912 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4913 struct imsm_disk
*disk
;
4915 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4918 /* only set index on disks that are a member of
4919 * a populated contianer, i.e. one with
4922 if (is_failed(&dl
->disk
))
4924 else if (is_spare(&dl
->disk
))
4930 if (i
>= mpb
->num_disks
) {
4931 struct intel_disk
*idisk
;
4933 idisk
= disk_list_get(dl
->serial
, disk_list
);
4934 if (idisk
&& is_spare(&idisk
->disk
) &&
4935 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4943 dl
->next
= champion
->disks
;
4944 champion
->disks
= dl
;
4948 /* delete 'champion' from super_list */
4949 for (del
= super_list
; *del
; ) {
4950 if (*del
== champion
) {
4951 *del
= (*del
)->next
;
4954 del
= &(*del
)->next
;
4956 champion
->next
= NULL
;
4960 struct intel_disk
*idisk
= disk_list
;
4962 disk_list
= disk_list
->next
;
4970 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4971 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4972 int major
, int minor
, int keep_fd
);
4974 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4975 int *max
, int keep_fd
);
4977 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4978 char *devname
, struct md_list
*devlist
,
4981 struct intel_super
*super_list
= NULL
;
4982 struct intel_super
*super
= NULL
;
4987 /* 'fd' is an opened container */
4988 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4990 /* get super block from devlist devices */
4991 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4994 /* all mpbs enter, maybe one leaves */
4995 super
= imsm_thunderdome(&super_list
, i
);
5001 if (find_missing(super
) != 0) {
5007 /* load migration record */
5008 err
= load_imsm_migr_rec(super
, NULL
);
5010 /* migration is in progress,
5011 * but migr_rec cannot be loaded,
5017 /* Check migration compatibility */
5018 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
5019 pr_err("Unsupported migration detected");
5021 fprintf(stderr
, " on %s\n", devname
);
5023 fprintf(stderr
, " (IMSM).\n");
5032 while (super_list
) {
5033 struct intel_super
*s
= super_list
;
5035 super_list
= super_list
->next
;
5044 strcpy(st
->container_devnm
, fd2devnm(fd
));
5046 st
->container_devnm
[0] = 0;
5047 if (err
== 0 && st
->ss
== NULL
) {
5048 st
->ss
= &super_imsm
;
5049 st
->minor_version
= 0;
5050 st
->max_devs
= IMSM_MAX_DEVICES
;
5056 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5057 int *max
, int keep_fd
)
5059 struct md_list
*tmpdev
;
5063 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5064 if (tmpdev
->used
!= 1)
5066 if (tmpdev
->container
== 1) {
5068 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5070 pr_err("cannot open device %s: %s\n",
5071 tmpdev
->devname
, strerror(errno
));
5075 err
= get_sra_super_block(fd
, super_list
,
5076 tmpdev
->devname
, &lmax
,
5085 int major
= major(tmpdev
->st_rdev
);
5086 int minor
= minor(tmpdev
->st_rdev
);
5087 err
= get_super_block(super_list
,
5104 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5105 int major
, int minor
, int keep_fd
)
5107 struct intel_super
*s
;
5119 sprintf(nm
, "%d:%d", major
, minor
);
5120 dfd
= dev_open(nm
, O_RDWR
);
5126 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5127 find_intel_hba_capability(dfd
, s
, devname
);
5128 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5130 /* retry the load if we might have raced against mdmon */
5131 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5132 for (retry
= 0; retry
< 3; retry
++) {
5134 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5140 s
->next
= *super_list
;
5148 if (dfd
>= 0 && !keep_fd
)
5155 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5162 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5166 if (sra
->array
.major_version
!= -1 ||
5167 sra
->array
.minor_version
!= -2 ||
5168 strcmp(sra
->text_version
, "imsm") != 0) {
5173 devnm
= fd2devnm(fd
);
5174 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5175 if (get_super_block(super_list
, devnm
, devname
,
5176 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5187 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5189 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5192 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5194 struct intel_super
*super
;
5198 if (test_partition(fd
))
5199 /* IMSM not allowed on partitions */
5202 free_super_imsm(st
);
5204 super
= alloc_super();
5205 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5208 /* Load hba and capabilities if they exist.
5209 * But do not preclude loading metadata in case capabilities or hba are
5210 * non-compliant and ignore_hw_compat is set.
5212 rv
= find_intel_hba_capability(fd
, super
, devname
);
5213 /* no orom/efi or non-intel hba of the disk */
5214 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5216 pr_err("No OROM/EFI properties for %s\n", devname
);
5220 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5222 /* retry the load if we might have raced against mdmon */
5224 struct mdstat_ent
*mdstat
= NULL
;
5225 char *name
= fd2kname(fd
);
5228 mdstat
= mdstat_by_component(name
);
5230 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5231 for (retry
= 0; retry
< 3; retry
++) {
5233 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5239 free_mdstat(mdstat
);
5244 pr_err("Failed to load all information sections on %s\n", devname
);
5250 if (st
->ss
== NULL
) {
5251 st
->ss
= &super_imsm
;
5252 st
->minor_version
= 0;
5253 st
->max_devs
= IMSM_MAX_DEVICES
;
5256 /* load migration record */
5257 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5258 /* Check for unsupported migration features */
5259 if (check_mpb_migr_compatibility(super
) != 0) {
5260 pr_err("Unsupported migration detected");
5262 fprintf(stderr
, " on %s\n", devname
);
5264 fprintf(stderr
, " (IMSM).\n");
5272 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5274 if (info
->level
== 1)
5276 return info
->chunk_size
>> 9;
5279 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5280 unsigned long long size
)
5282 if (info
->level
== 1)
5285 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5288 static void imsm_update_version_info(struct intel_super
*super
)
5290 /* update the version and attributes */
5291 struct imsm_super
*mpb
= super
->anchor
;
5293 struct imsm_dev
*dev
;
5294 struct imsm_map
*map
;
5297 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5298 dev
= get_imsm_dev(super
, i
);
5299 map
= get_imsm_map(dev
, MAP_0
);
5300 if (__le32_to_cpu(dev
->size_high
) > 0)
5301 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5303 /* FIXME detect when an array spans a port multiplier */
5305 mpb
->attributes
|= MPB_ATTRIB_PM
;
5308 if (mpb
->num_raid_devs
> 1 ||
5309 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5310 version
= MPB_VERSION_ATTRIBS
;
5311 switch (get_imsm_raid_level(map
)) {
5312 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5313 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5314 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5315 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5318 if (map
->num_members
>= 5)
5319 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5320 else if (dev
->status
== DEV_CLONE_N_GO
)
5321 version
= MPB_VERSION_CNG
;
5322 else if (get_imsm_raid_level(map
) == 5)
5323 version
= MPB_VERSION_RAID5
;
5324 else if (map
->num_members
>= 3)
5325 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5326 else if (get_imsm_raid_level(map
) == 1)
5327 version
= MPB_VERSION_RAID1
;
5329 version
= MPB_VERSION_RAID0
;
5331 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5335 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5337 struct imsm_super
*mpb
= super
->anchor
;
5338 char *reason
= NULL
;
5340 size_t len
= strlen(name
);
5344 while (isspace(start
[len
- 1]))
5346 while (*start
&& isspace(*start
))
5348 memmove(name
, start
, len
+ 1);
5351 if (len
> MAX_RAID_SERIAL_LEN
)
5352 reason
= "must be 16 characters or less";
5354 reason
= "must be a non-empty string";
5356 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5357 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5359 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5360 reason
= "already exists";
5365 if (reason
&& !quiet
)
5366 pr_err("imsm volume name %s\n", reason
);
5371 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5372 struct shape
*s
, char *name
,
5373 char *homehost
, int *uuid
,
5374 long long data_offset
)
5376 /* We are creating a volume inside a pre-existing container.
5377 * so st->sb is already set.
5379 struct intel_super
*super
= st
->sb
;
5380 unsigned int sector_size
= super
->sector_size
;
5381 struct imsm_super
*mpb
= super
->anchor
;
5382 struct intel_dev
*dv
;
5383 struct imsm_dev
*dev
;
5384 struct imsm_vol
*vol
;
5385 struct imsm_map
*map
;
5386 int idx
= mpb
->num_raid_devs
;
5389 unsigned long long array_blocks
;
5390 size_t size_old
, size_new
;
5391 unsigned long long num_data_stripes
;
5392 unsigned int data_disks
;
5393 unsigned long long size_per_member
;
5395 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5396 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5400 /* ensure the mpb is large enough for the new data */
5401 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5402 size_new
= disks_to_mpb_size(info
->nr_disks
);
5403 if (size_new
> size_old
) {
5405 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5407 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5408 pr_err("could not allocate new mpb\n");
5411 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5412 MIGR_REC_BUF_SECTORS
*
5413 MAX_SECTOR_SIZE
) != 0) {
5414 pr_err("could not allocate migr_rec buffer\n");
5420 memcpy(mpb_new
, mpb
, size_old
);
5423 super
->anchor
= mpb_new
;
5424 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5425 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5426 super
->len
= size_round
;
5428 super
->current_vol
= idx
;
5430 /* handle 'failed_disks' by either:
5431 * a) create dummy disk entries in the table if this the first
5432 * volume in the array. We add them here as this is the only
5433 * opportunity to add them. add_to_super_imsm_volume()
5434 * handles the non-failed disks and continues incrementing
5436 * b) validate that 'failed_disks' matches the current number
5437 * of missing disks if the container is populated
5439 if (super
->current_vol
== 0) {
5441 for (i
= 0; i
< info
->failed_disks
; i
++) {
5442 struct imsm_disk
*disk
;
5445 disk
= __get_imsm_disk(mpb
, i
);
5446 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5447 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5448 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5449 "missing:%d", (__u8
)i
);
5451 find_missing(super
);
5456 for (d
= super
->missing
; d
; d
= d
->next
)
5458 if (info
->failed_disks
> missing
) {
5459 pr_err("unable to add 'missing' disk to container\n");
5464 if (!check_name(super
, name
, 0))
5466 dv
= xmalloc(sizeof(*dv
));
5467 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5469 * Explicitly allow truncating to not confuse gcc's
5470 * -Werror=stringop-truncation
5472 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5473 memcpy(dev
->volume
, name
, namelen
);
5474 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5475 info
->layout
, info
->chunk_size
,
5476 s
->size
* BLOCKS_PER_KB
);
5477 data_disks
= get_data_disks(info
->level
, info
->layout
,
5479 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5480 size_per_member
= array_blocks
/ data_disks
;
5482 set_imsm_dev_size(dev
, array_blocks
);
5483 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5485 vol
->migr_state
= 0;
5486 set_migr_type(dev
, MIGR_INIT
);
5487 vol
->dirty
= !info
->state
;
5488 vol
->curr_migr_unit
= 0;
5489 map
= get_imsm_map(dev
, MAP_0
);
5490 set_pba_of_lba0(map
, super
->create_offset
);
5491 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5492 map
->failed_disk_num
= ~0;
5493 if (info
->level
> 0)
5494 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5495 : IMSM_T_STATE_UNINITIALIZED
);
5497 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5498 IMSM_T_STATE_NORMAL
;
5501 if (info
->level
== 1 && info
->raid_disks
> 2) {
5504 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5508 map
->raid_level
= info
->level
;
5509 if (info
->level
== 10) {
5510 map
->raid_level
= 1;
5511 map
->num_domains
= info
->raid_disks
/ 2;
5512 } else if (info
->level
== 1)
5513 map
->num_domains
= info
->raid_disks
;
5515 map
->num_domains
= 1;
5517 /* info->size is only int so use the 'size' parameter instead */
5518 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5519 num_data_stripes
/= map
->num_domains
;
5520 set_num_data_stripes(map
, num_data_stripes
);
5522 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5523 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5527 map
->num_members
= info
->raid_disks
;
5528 for (i
= 0; i
< map
->num_members
; i
++) {
5529 /* initialized in add_to_super */
5530 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5532 mpb
->num_raid_devs
++;
5533 mpb
->num_raid_devs_created
++;
5534 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5536 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5537 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5538 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5539 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5543 pr_err("imsm does not support consistency policy %s\n",
5544 map_num(consistency_policies
, s
->consistency_policy
));
5549 dv
->index
= super
->current_vol
;
5550 dv
->next
= super
->devlist
;
5551 super
->devlist
= dv
;
5553 imsm_update_version_info(super
);
5558 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5559 struct shape
*s
, char *name
,
5560 char *homehost
, int *uuid
,
5561 unsigned long long data_offset
)
5563 /* This is primarily called by Create when creating a new array.
5564 * We will then get add_to_super called for each component, and then
5565 * write_init_super called to write it out to each device.
5566 * For IMSM, Create can create on fresh devices or on a pre-existing
5568 * To create on a pre-existing array a different method will be called.
5569 * This one is just for fresh drives.
5571 struct intel_super
*super
;
5572 struct imsm_super
*mpb
;
5576 if (data_offset
!= INVALID_SECTORS
) {
5577 pr_err("data-offset not supported by imsm\n");
5582 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5586 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5588 mpb_size
= MAX_SECTOR_SIZE
;
5590 super
= alloc_super();
5592 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5597 pr_err("could not allocate superblock\n");
5600 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5601 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5602 pr_err("could not allocate migr_rec buffer\n");
5607 memset(super
->buf
, 0, mpb_size
);
5609 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5613 /* zeroing superblock */
5617 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5619 version
= (char *) mpb
->sig
;
5620 strcpy(version
, MPB_SIGNATURE
);
5621 version
+= strlen(MPB_SIGNATURE
);
5622 strcpy(version
, MPB_VERSION_RAID0
);
5627 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5629 unsigned int member_sector_size
;
5632 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5636 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5638 if (member_sector_size
!= super
->sector_size
)
5643 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5644 int fd
, char *devname
)
5646 struct intel_super
*super
= st
->sb
;
5647 struct imsm_super
*mpb
= super
->anchor
;
5648 struct imsm_disk
*_disk
;
5649 struct imsm_dev
*dev
;
5650 struct imsm_map
*map
;
5654 dev
= get_imsm_dev(super
, super
->current_vol
);
5655 map
= get_imsm_map(dev
, MAP_0
);
5657 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5658 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5664 /* we're doing autolayout so grab the pre-marked (in
5665 * validate_geometry) raid_disk
5667 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5668 if (dl
->raiddisk
== dk
->raid_disk
)
5671 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5672 if (dl
->major
== dk
->major
&&
5673 dl
->minor
== dk
->minor
)
5678 pr_err("%s is not a member of the same container\n", devname
);
5682 if (mpb
->num_disks
== 0)
5683 if (!get_dev_sector_size(dl
->fd
, dl
->devname
,
5684 &super
->sector_size
))
5687 if (!drive_validate_sector_size(super
, dl
)) {
5688 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5692 /* add a pristine spare to the metadata */
5693 if (dl
->index
< 0) {
5694 dl
->index
= super
->anchor
->num_disks
;
5695 super
->anchor
->num_disks
++;
5697 /* Check the device has not already been added */
5698 slot
= get_imsm_disk_slot(map
, dl
->index
);
5700 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5701 pr_err("%s has been included in this array twice\n",
5705 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5706 dl
->disk
.status
= CONFIGURED_DISK
;
5708 /* update size of 'missing' disks to be at least as large as the
5709 * largest acitve member (we only have dummy missing disks when
5710 * creating the first volume)
5712 if (super
->current_vol
== 0) {
5713 for (df
= super
->missing
; df
; df
= df
->next
) {
5714 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5715 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5716 _disk
= __get_imsm_disk(mpb
, df
->index
);
5721 /* refresh unset/failed slots to point to valid 'missing' entries */
5722 for (df
= super
->missing
; df
; df
= df
->next
)
5723 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5724 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5726 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5728 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5729 if (is_gen_migration(dev
)) {
5730 struct imsm_map
*map2
= get_imsm_map(dev
,
5732 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5733 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5734 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5737 if ((unsigned)df
->index
==
5739 set_imsm_ord_tbl_ent(map2
,
5745 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5749 /* if we are creating the first raid device update the family number */
5750 if (super
->current_vol
== 0) {
5752 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5754 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5755 if (!_dev
|| !_disk
) {
5756 pr_err("BUG mpb setup error\n");
5762 sum
+= __gen_imsm_checksum(mpb
);
5763 mpb
->family_num
= __cpu_to_le32(sum
);
5764 mpb
->orig_family_num
= mpb
->family_num
;
5766 super
->current_disk
= dl
;
5771 * Function marks disk as spare and restores disk serial
5772 * in case it was previously marked as failed by takeover operation
5774 * -1 : critical error
5775 * 0 : disk is marked as spare but serial is not set
5778 int mark_spare(struct dl
*disk
)
5780 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5787 if (!imsm_read_serial(disk
->fd
, NULL
, serial
, MAX_RAID_SERIAL_LEN
)) {
5788 /* Restore disk serial number, because takeover marks disk
5789 * as failed and adds to serial ':0' before it becomes
5792 serialcpy(disk
->serial
, serial
);
5793 serialcpy(disk
->disk
.serial
, serial
);
5796 disk
->disk
.status
= SPARE_DISK
;
5802 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5803 int fd
, char *devname
,
5804 unsigned long long data_offset
)
5806 struct intel_super
*super
= st
->sb
;
5808 unsigned long long size
;
5809 unsigned int member_sector_size
;
5814 /* If we are on an RAID enabled platform check that the disk is
5815 * attached to the raid controller.
5816 * We do not need to test disks attachment for container based additions,
5817 * they shall be already tested when container was created/assembled.
5819 rv
= find_intel_hba_capability(fd
, super
, devname
);
5820 /* no orom/efi or non-intel hba of the disk */
5822 dprintf("capability: %p fd: %d ret: %d\n",
5823 super
->orom
, fd
, rv
);
5827 if (super
->current_vol
>= 0)
5828 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5831 dd
= xcalloc(sizeof(*dd
), 1);
5832 dd
->major
= major(stb
.st_rdev
);
5833 dd
->minor
= minor(stb
.st_rdev
);
5834 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5837 dd
->action
= DISK_ADD
;
5838 rv
= imsm_read_serial(fd
, devname
, dd
->serial
, MAX_RAID_SERIAL_LEN
);
5840 pr_err("failed to retrieve scsi serial, aborting\n");
5846 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5847 (super
->hba
->type
== SYS_DEV_VMD
))) {
5849 char *devpath
= diskfd_to_devpath(fd
);
5850 char controller_path
[PATH_MAX
];
5853 pr_err("failed to get devpath, aborting\n");
5860 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5863 if (devpath_to_vendor(controller_path
) == 0x8086) {
5865 * If Intel's NVMe drive has serial ended with
5866 * "-A","-B","-1" or "-2" it means that this is "x8"
5867 * device (double drive on single PCIe card).
5868 * User should be warned about potential data loss.
5870 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5871 /* Skip empty character at the end */
5872 if (dd
->serial
[i
] == 0)
5875 if (((dd
->serial
[i
] == 'A') ||
5876 (dd
->serial
[i
] == 'B') ||
5877 (dd
->serial
[i
] == '1') ||
5878 (dd
->serial
[i
] == '2')) &&
5879 (dd
->serial
[i
-1] == '-'))
5880 pr_err("\tThe action you are about to take may put your data at risk.\n"
5881 "\tPlease note that x8 devices may consist of two separate x4 devices "
5882 "located on a single PCIe port.\n"
5883 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5886 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5887 !imsm_orom_has_tpv_support(super
->orom
)) {
5888 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5889 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
5896 get_dev_size(fd
, NULL
, &size
);
5897 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5899 if (super
->sector_size
== 0) {
5900 /* this a first device, so sector_size is not set yet */
5901 super
->sector_size
= member_sector_size
;
5904 /* clear migr_rec when adding disk to container */
5905 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5906 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5908 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5909 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5910 MIGR_REC_BUF_SECTORS
*member_sector_size
)
5911 perror("Write migr_rec failed");
5915 serialcpy(dd
->disk
.serial
, dd
->serial
);
5916 set_total_blocks(&dd
->disk
, size
);
5917 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5918 struct imsm_super
*mpb
= super
->anchor
;
5919 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5922 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5923 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5925 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5927 if (st
->update_tail
) {
5928 dd
->next
= super
->disk_mgmt_list
;
5929 super
->disk_mgmt_list
= dd
;
5931 dd
->next
= super
->disks
;
5933 super
->updates_pending
++;
5939 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5941 struct intel_super
*super
= st
->sb
;
5944 /* remove from super works only in mdmon - for communication
5945 * manager - monitor. Check if communication memory buffer
5948 if (!st
->update_tail
) {
5949 pr_err("shall be used in mdmon context only\n");
5952 dd
= xcalloc(1, sizeof(*dd
));
5953 dd
->major
= dk
->major
;
5954 dd
->minor
= dk
->minor
;
5957 dd
->action
= DISK_REMOVE
;
5959 dd
->next
= super
->disk_mgmt_list
;
5960 super
->disk_mgmt_list
= dd
;
5965 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5968 char buf
[MAX_SECTOR_SIZE
];
5969 struct imsm_super anchor
;
5970 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5972 /* spare records have their own family number and do not have any defined raid
5975 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5977 struct imsm_super
*mpb
= super
->anchor
;
5978 struct imsm_super
*spare
= &spare_record
.anchor
;
5982 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5983 spare
->generation_num
= __cpu_to_le32(1UL);
5984 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5985 spare
->num_disks
= 1;
5986 spare
->num_raid_devs
= 0;
5987 spare
->cache_size
= mpb
->cache_size
;
5988 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5990 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5991 MPB_SIGNATURE MPB_VERSION_RAID0
);
5993 for (d
= super
->disks
; d
; d
= d
->next
) {
5997 spare
->disk
[0] = d
->disk
;
5998 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5999 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
6001 if (super
->sector_size
== 4096)
6002 convert_to_4k_imsm_disk(&spare
->disk
[0]);
6004 sum
= __gen_imsm_checksum(spare
);
6005 spare
->family_num
= __cpu_to_le32(sum
);
6006 spare
->orig_family_num
= 0;
6007 sum
= __gen_imsm_checksum(spare
);
6008 spare
->check_sum
= __cpu_to_le32(sum
);
6010 if (store_imsm_mpb(d
->fd
, spare
)) {
6011 pr_err("failed for device %d:%d %s\n",
6012 d
->major
, d
->minor
, strerror(errno
));
6024 static int write_super_imsm(struct supertype
*st
, int doclose
)
6026 struct intel_super
*super
= st
->sb
;
6027 unsigned int sector_size
= super
->sector_size
;
6028 struct imsm_super
*mpb
= super
->anchor
;
6034 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
6036 int clear_migration_record
= 1;
6039 /* 'generation' is incremented everytime the metadata is written */
6040 generation
= __le32_to_cpu(mpb
->generation_num
);
6042 mpb
->generation_num
= __cpu_to_le32(generation
);
6044 /* fix up cases where previous mdadm releases failed to set
6047 if (mpb
->orig_family_num
== 0)
6048 mpb
->orig_family_num
= mpb
->family_num
;
6050 for (d
= super
->disks
; d
; d
= d
->next
) {
6054 mpb
->disk
[d
->index
] = d
->disk
;
6058 for (d
= super
->missing
; d
; d
= d
->next
) {
6059 mpb
->disk
[d
->index
] = d
->disk
;
6062 mpb
->num_disks
= num_disks
;
6063 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
6065 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6066 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6067 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6069 imsm_copy_dev(dev
, dev2
);
6070 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6072 if (is_gen_migration(dev2
))
6073 clear_migration_record
= 0;
6076 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6079 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6080 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6082 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6084 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6085 mpb_size
+= bbm_log_size
;
6086 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6089 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6092 /* recalculate checksum */
6093 sum
= __gen_imsm_checksum(mpb
);
6094 mpb
->check_sum
= __cpu_to_le32(sum
);
6096 if (super
->clean_migration_record_by_mdmon
) {
6097 clear_migration_record
= 1;
6098 super
->clean_migration_record_by_mdmon
= 0;
6100 if (clear_migration_record
)
6101 memset(super
->migr_rec_buf
, 0,
6102 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6104 if (sector_size
== 4096)
6105 convert_to_4k(super
);
6107 /* write the mpb for disks that compose raid devices */
6108 for (d
= super
->disks
; d
; d
= d
->next
) {
6109 if (d
->index
< 0 || is_failed(&d
->disk
))
6112 if (clear_migration_record
) {
6113 unsigned long long dsize
;
6115 get_dev_size(d
->fd
, NULL
, &dsize
);
6116 if (lseek64(d
->fd
, dsize
- sector_size
,
6118 if ((unsigned int)write(d
->fd
,
6119 super
->migr_rec_buf
,
6120 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6121 MIGR_REC_BUF_SECTORS
*sector_size
)
6122 perror("Write migr_rec failed");
6126 if (store_imsm_mpb(d
->fd
, mpb
))
6128 "failed for device %d:%d (fd: %d)%s\n",
6130 d
->fd
, strerror(errno
));
6139 return write_super_imsm_spares(super
, doclose
);
6144 static int create_array(struct supertype
*st
, int dev_idx
)
6147 struct imsm_update_create_array
*u
;
6148 struct intel_super
*super
= st
->sb
;
6149 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6150 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6151 struct disk_info
*inf
;
6152 struct imsm_disk
*disk
;
6155 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6156 sizeof(*inf
) * map
->num_members
;
6158 u
->type
= update_create_array
;
6159 u
->dev_idx
= dev_idx
;
6160 imsm_copy_dev(&u
->dev
, dev
);
6161 inf
= get_disk_info(u
);
6162 for (i
= 0; i
< map
->num_members
; i
++) {
6163 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6165 disk
= get_imsm_disk(super
, idx
);
6167 disk
= get_imsm_missing(super
, idx
);
6168 serialcpy(inf
[i
].serial
, disk
->serial
);
6170 append_metadata_update(st
, u
, len
);
6175 static int mgmt_disk(struct supertype
*st
)
6177 struct intel_super
*super
= st
->sb
;
6179 struct imsm_update_add_remove_disk
*u
;
6181 if (!super
->disk_mgmt_list
)
6186 u
->type
= update_add_remove_disk
;
6187 append_metadata_update(st
, u
, len
);
6192 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6194 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6196 struct ppl_header
*ppl_hdr
= buf
;
6199 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6201 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6203 perror("Failed to seek to PPL header location");
6207 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6209 perror("Write PPL header failed");
6218 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6220 struct intel_super
*super
= st
->sb
;
6222 struct ppl_header
*ppl_hdr
;
6225 /* first clear entire ppl space */
6226 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6230 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6232 pr_err("Failed to allocate PPL header buffer\n");
6236 memset(buf
, 0, PPL_HEADER_SIZE
);
6238 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6239 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6241 if (info
->mismatch_cnt
) {
6243 * We are overwriting an invalid ppl. Make one entry with wrong
6244 * checksum to prevent the kernel from skipping resync.
6246 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6247 ppl_hdr
->entries
[0].checksum
= ~0;
6250 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6256 static int is_rebuilding(struct imsm_dev
*dev
);
6258 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6259 struct mdinfo
*disk
)
6261 struct intel_super
*super
= st
->sb
;
6263 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6265 struct ppl_header
*ppl_hdr
= NULL
;
6267 struct imsm_dev
*dev
;
6270 unsigned long long ppl_offset
= 0;
6271 unsigned long long prev_gen_num
= 0;
6273 if (disk
->disk
.raid_disk
< 0)
6276 dev
= get_imsm_dev(super
, info
->container_member
);
6277 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6278 d
= get_imsm_dl_disk(super
, idx
);
6280 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6283 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6284 pr_err("Failed to allocate PPL header buffer\n");
6290 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6293 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6295 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6297 perror("Failed to seek to PPL header location");
6302 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6303 perror("Read PPL header failed");
6310 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6311 ppl_hdr
->checksum
= 0;
6313 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6314 dprintf("Wrong PPL header checksum on %s\n",
6319 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6320 /* previous was newest, it was already checked */
6324 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6325 super
->anchor
->orig_family_num
)) {
6326 dprintf("Wrong PPL header signature on %s\n",
6333 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6335 ppl_offset
+= PPL_HEADER_SIZE
;
6336 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6338 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6341 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6353 * Update metadata to use mutliple PPLs area (1MB).
6354 * This is done once for all RAID members
6356 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6357 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6359 struct mdinfo
*member_dev
;
6361 sprintf(subarray
, "%d", info
->container_member
);
6363 if (mdmon_running(st
->container_devnm
))
6364 st
->update_tail
= &st
->updates
;
6366 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6367 pr_err("Failed to update subarray %s\n",
6370 if (st
->update_tail
)
6371 flush_metadata_updates(st
);
6373 st
->ss
->sync_metadata(st
);
6374 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6375 for (member_dev
= info
->devs
; member_dev
;
6376 member_dev
= member_dev
->next
)
6377 member_dev
->ppl_size
=
6378 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6383 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6385 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6386 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6387 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6388 (is_rebuilding(dev
) &&
6389 dev
->vol
.curr_migr_unit
== 0 &&
6390 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6391 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6393 info
->mismatch_cnt
++;
6394 } else if (ret
== 0 &&
6395 ppl_hdr
->entries_count
== 0 &&
6396 is_rebuilding(dev
) &&
6397 info
->resync_start
== 0) {
6399 * The header has no entries - add a single empty entry and
6400 * rewrite the header to prevent the kernel from going into
6401 * resync after an interrupted rebuild.
6403 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6404 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6412 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6414 struct intel_super
*super
= st
->sb
;
6418 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6419 info
->array
.level
!= 5)
6422 for (d
= super
->disks
; d
; d
= d
->next
) {
6423 if (d
->index
< 0 || is_failed(&d
->disk
))
6426 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6434 static int write_init_super_imsm(struct supertype
*st
)
6436 struct intel_super
*super
= st
->sb
;
6437 int current_vol
= super
->current_vol
;
6441 getinfo_super_imsm(st
, &info
, NULL
);
6443 /* we are done with current_vol reset it to point st at the container */
6444 super
->current_vol
= -1;
6446 if (st
->update_tail
) {
6447 /* queue the recently created array / added disk
6448 * as a metadata update */
6450 /* determine if we are creating a volume or adding a disk */
6451 if (current_vol
< 0) {
6452 /* in the mgmt (add/remove) disk case we are running
6453 * in mdmon context, so don't close fd's
6457 rv
= write_init_ppl_imsm_all(st
, &info
);
6459 rv
= create_array(st
, current_vol
);
6463 for (d
= super
->disks
; d
; d
= d
->next
)
6464 Kill(d
->devname
, NULL
, 0, -1, 1);
6465 if (current_vol
>= 0)
6466 rv
= write_init_ppl_imsm_all(st
, &info
);
6468 rv
= write_super_imsm(st
, 1);
6474 static int store_super_imsm(struct supertype
*st
, int fd
)
6476 struct intel_super
*super
= st
->sb
;
6477 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6482 if (super
->sector_size
== 4096)
6483 convert_to_4k(super
);
6484 return store_imsm_mpb(fd
, mpb
);
6487 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6488 int layout
, int raiddisks
, int chunk
,
6489 unsigned long long size
,
6490 unsigned long long data_offset
,
6492 unsigned long long *freesize
,
6496 unsigned long long ldsize
;
6497 struct intel_super
*super
;
6500 if (level
!= LEVEL_CONTAINER
)
6505 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6508 pr_err("imsm: Cannot open %s: %s\n",
6509 dev
, strerror(errno
));
6512 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6517 /* capabilities retrieve could be possible
6518 * note that there is no fd for the disks in array.
6520 super
= alloc_super();
6525 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6531 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6535 fd2devname(fd
, str
);
6536 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6537 fd
, str
, super
->orom
, rv
, raiddisks
);
6539 /* no orom/efi or non-intel hba of the disk */
6546 if (raiddisks
> super
->orom
->tds
) {
6548 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6549 raiddisks
, super
->orom
->tds
);
6553 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6554 (ldsize
>> 9) >> 32 > 0) {
6556 pr_err("%s exceeds maximum platform supported size\n", dev
);
6562 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6568 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6570 const unsigned long long base_start
= e
[*idx
].start
;
6571 unsigned long long end
= base_start
+ e
[*idx
].size
;
6574 if (base_start
== end
)
6578 for (i
= *idx
; i
< num_extents
; i
++) {
6579 /* extend overlapping extents */
6580 if (e
[i
].start
>= base_start
&&
6581 e
[i
].start
<= end
) {
6584 if (e
[i
].start
+ e
[i
].size
> end
)
6585 end
= e
[i
].start
+ e
[i
].size
;
6586 } else if (e
[i
].start
> end
) {
6592 return end
- base_start
;
6595 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6597 /* build a composite disk with all known extents and generate a new
6598 * 'maxsize' given the "all disks in an array must share a common start
6599 * offset" constraint
6601 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6605 unsigned long long pos
;
6606 unsigned long long start
= 0;
6607 unsigned long long maxsize
;
6608 unsigned long reserve
;
6610 /* coalesce and sort all extents. also, check to see if we need to
6611 * reserve space between member arrays
6614 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6617 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6620 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6625 while (i
< sum_extents
) {
6626 e
[j
].start
= e
[i
].start
;
6627 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6629 if (e
[j
-1].size
== 0)
6638 unsigned long long esize
;
6640 esize
= e
[i
].start
- pos
;
6641 if (esize
>= maxsize
) {
6646 pos
= e
[i
].start
+ e
[i
].size
;
6648 } while (e
[i
-1].size
);
6654 /* FIXME assumes volume at offset 0 is the first volume in a
6657 if (start_extent
> 0)
6658 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6662 if (maxsize
< reserve
)
6665 super
->create_offset
= ~((unsigned long long) 0);
6666 if (start
+ reserve
> super
->create_offset
)
6667 return 0; /* start overflows create_offset */
6668 super
->create_offset
= start
+ reserve
;
6670 return maxsize
- reserve
;
6673 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6675 if (level
< 0 || level
== 6 || level
== 4)
6678 /* if we have an orom prevent invalid raid levels */
6681 case 0: return imsm_orom_has_raid0(orom
);
6684 return imsm_orom_has_raid1e(orom
);
6685 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6686 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6687 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6690 return 1; /* not on an Intel RAID platform so anything goes */
6696 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6697 int dpa
, int verbose
)
6699 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6700 struct mdstat_ent
*memb
;
6706 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6707 if (memb
->metadata_version
&&
6708 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6709 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6710 !is_subarray(memb
->metadata_version
+9) &&
6712 struct dev_member
*dev
= memb
->members
;
6714 while(dev
&& (fd
< 0)) {
6715 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6716 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6718 fd
= open(path
, O_RDONLY
, 0);
6719 if (num
<= 0 || fd
< 0) {
6720 pr_vrb("Cannot open %s: %s\n",
6721 dev
->name
, strerror(errno
));
6727 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6728 struct mdstat_ent
*vol
;
6729 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6730 if (vol
->active
> 0 &&
6731 vol
->metadata_version
&&
6732 is_container_member(vol
, memb
->devnm
)) {
6737 if (*devlist
&& (found
< dpa
)) {
6738 dv
= xcalloc(1, sizeof(*dv
));
6739 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6740 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6743 dv
->next
= *devlist
;
6751 free_mdstat(mdstat
);
6756 static struct md_list
*
6757 get_loop_devices(void)
6760 struct md_list
*devlist
= NULL
;
6763 for(i
= 0; i
< 12; i
++) {
6764 dv
= xcalloc(1, sizeof(*dv
));
6765 dv
->devname
= xmalloc(40);
6766 sprintf(dv
->devname
, "/dev/loop%d", i
);
6774 static struct md_list
*
6775 get_devices(const char *hba_path
)
6777 struct md_list
*devlist
= NULL
;
6784 devlist
= get_loop_devices();
6787 /* scroll through /sys/dev/block looking for devices attached to
6790 dir
= opendir("/sys/dev/block");
6791 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6796 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6798 path
= devt_to_devpath(makedev(major
, minor
));
6801 if (!path_attached_to_hba(path
, hba_path
)) {
6808 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6810 fd2devname(fd
, buf
);
6813 pr_err("cannot open device: %s\n",
6818 dv
= xcalloc(1, sizeof(*dv
));
6819 dv
->devname
= xstrdup(buf
);
6826 devlist
= devlist
->next
;
6836 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6837 int verbose
, int *found
)
6839 struct md_list
*tmpdev
;
6841 struct supertype
*st
;
6843 /* first walk the list of devices to find a consistent set
6844 * that match the criterea, if that is possible.
6845 * We flag the ones we like with 'used'.
6848 st
= match_metadata_desc_imsm("imsm");
6850 pr_vrb("cannot allocate memory for imsm supertype\n");
6854 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6855 char *devname
= tmpdev
->devname
;
6857 struct supertype
*tst
;
6859 if (tmpdev
->used
> 1)
6861 tst
= dup_super(st
);
6863 pr_vrb("cannot allocate memory for imsm supertype\n");
6866 tmpdev
->container
= 0;
6867 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6869 dprintf("cannot open device %s: %s\n",
6870 devname
, strerror(errno
));
6872 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
6874 } else if (must_be_container(dfd
)) {
6875 struct supertype
*cst
;
6876 cst
= super_by_fd(dfd
, NULL
);
6878 dprintf("cannot recognize container type %s\n",
6881 } else if (tst
->ss
!= st
->ss
) {
6882 dprintf("non-imsm container - ignore it: %s\n",
6885 } else if (!tst
->ss
->load_container
||
6886 tst
->ss
->load_container(tst
, dfd
, NULL
))
6889 tmpdev
->container
= 1;
6892 cst
->ss
->free_super(cst
);
6894 tmpdev
->st_rdev
= rdev
;
6895 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6896 dprintf("no RAID superblock on %s\n",
6899 } else if (tst
->ss
->compare_super
== NULL
) {
6900 dprintf("Cannot assemble %s metadata on %s\n",
6901 tst
->ss
->name
, devname
);
6907 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6908 /* Ignore unrecognised devices during auto-assembly */
6913 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6915 if (st
->minor_version
== -1)
6916 st
->minor_version
= tst
->minor_version
;
6918 if (memcmp(info
.uuid
, uuid_zero
,
6919 sizeof(int[4])) == 0) {
6920 /* this is a floating spare. It cannot define
6921 * an array unless there are no more arrays of
6922 * this type to be found. It can be included
6923 * in an array of this type though.
6929 if (st
->ss
!= tst
->ss
||
6930 st
->minor_version
!= tst
->minor_version
||
6931 st
->ss
->compare_super(st
, tst
) != 0) {
6932 /* Some mismatch. If exactly one array matches this host,
6933 * we can resolve on that one.
6934 * Or, if we are auto assembling, we just ignore the second
6937 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6943 dprintf("found: devname: %s\n", devname
);
6947 tst
->ss
->free_super(tst
);
6951 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6952 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6953 for (iter
= head
; iter
; iter
= iter
->next
) {
6954 dprintf("content->text_version: %s vol\n",
6955 iter
->text_version
);
6956 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6957 /* do not assemble arrays with unsupported
6959 dprintf("Cannot activate member %s.\n",
6960 iter
->text_version
);
6967 dprintf("No valid super block on device list: err: %d %p\n",
6971 dprintf("no more devices to examine\n");
6974 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6975 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6977 if (count
< tmpdev
->found
)
6980 count
-= tmpdev
->found
;
6983 if (tmpdev
->used
== 1)
6988 st
->ss
->free_super(st
);
6992 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
6995 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6997 const struct orom_entry
*entry
;
6998 struct devid_list
*dv
, *devid_list
;
7003 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
7004 if (strstr(idev
->path
, hba_path
))
7008 if (!idev
|| !idev
->dev_id
)
7011 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
7013 if (!entry
|| !entry
->devid_list
)
7016 devid_list
= entry
->devid_list
;
7017 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
7018 struct md_list
*devlist
;
7019 struct sys_dev
*device
= NULL
;
7024 device
= device_by_id_and_path(dv
->devid
, hba_path
);
7026 device
= device_by_id(dv
->devid
);
7029 hpath
= device
->path
;
7033 devlist
= get_devices(hpath
);
7034 /* if no intel devices return zero volumes */
7035 if (devlist
== NULL
)
7038 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
7040 dprintf("path: %s active arrays: %d\n", hpath
, count
);
7041 if (devlist
== NULL
)
7045 count
+= count_volumes_list(devlist
,
7049 dprintf("found %d count: %d\n", found
, count
);
7052 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
7055 struct md_list
*dv
= devlist
;
7056 devlist
= devlist
->next
;
7064 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7068 if (hba
->type
== SYS_DEV_VMD
) {
7069 struct sys_dev
*dev
;
7072 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7073 if (dev
->type
== SYS_DEV_VMD
)
7074 count
+= __count_volumes(dev
->path
, dpa
,
7079 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7082 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7084 /* up to 512 if the plaform supports it, otherwise the platform max.
7085 * 128 if no platform detected
7087 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7089 return min(512, (1 << fs
));
7093 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7094 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7096 /* check/set platform and metadata limits/defaults */
7097 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7098 pr_vrb("platform supports a maximum of %d disks per array\n",
7103 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7104 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7105 pr_vrb("platform does not support raid%d with %d disk%s\n",
7106 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7110 if (*chunk
== 0 || *chunk
== UnSet
)
7111 *chunk
= imsm_default_chunk(super
->orom
);
7113 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7114 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7118 if (layout
!= imsm_level_to_layout(level
)) {
7120 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7121 else if (level
== 10)
7122 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7124 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7129 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7130 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7131 pr_vrb("platform does not support a volume size over 2TB\n");
7138 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7139 * FIX ME add ahci details
7141 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7142 int layout
, int raiddisks
, int *chunk
,
7143 unsigned long long size
,
7144 unsigned long long data_offset
,
7146 unsigned long long *freesize
,
7150 struct intel_super
*super
= st
->sb
;
7151 struct imsm_super
*mpb
;
7153 unsigned long long pos
= 0;
7154 unsigned long long maxsize
;
7158 /* We must have the container info already read in. */
7162 mpb
= super
->anchor
;
7164 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7165 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7169 /* General test: make sure there is space for
7170 * 'raiddisks' device extents of size 'size' at a given
7173 unsigned long long minsize
= size
;
7174 unsigned long long start_offset
= MaxSector
;
7177 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7178 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7183 e
= get_extents(super
, dl
, 0);
7186 unsigned long long esize
;
7187 esize
= e
[i
].start
- pos
;
7188 if (esize
>= minsize
)
7190 if (found
&& start_offset
== MaxSector
) {
7193 } else if (found
&& pos
!= start_offset
) {
7197 pos
= e
[i
].start
+ e
[i
].size
;
7199 } while (e
[i
-1].size
);
7204 if (dcnt
< raiddisks
) {
7206 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7213 /* This device must be a member of the set */
7214 if (!stat_is_blkdev(dev
, &rdev
))
7216 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7217 if (dl
->major
== (int)major(rdev
) &&
7218 dl
->minor
== (int)minor(rdev
))
7223 pr_err("%s is not in the same imsm set\n", dev
);
7225 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7226 /* If a volume is present then the current creation attempt
7227 * cannot incorporate new spares because the orom may not
7228 * understand this configuration (all member disks must be
7229 * members of each array in the container).
7231 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7232 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7234 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7235 mpb
->num_disks
!= raiddisks
) {
7236 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7240 /* retrieve the largest free space block */
7241 e
= get_extents(super
, dl
, 0);
7246 unsigned long long esize
;
7248 esize
= e
[i
].start
- pos
;
7249 if (esize
>= maxsize
)
7251 pos
= e
[i
].start
+ e
[i
].size
;
7253 } while (e
[i
-1].size
);
7258 pr_err("unable to determine free space for: %s\n",
7262 if (maxsize
< size
) {
7264 pr_err("%s not enough space (%llu < %llu)\n",
7265 dev
, maxsize
, size
);
7269 /* count total number of extents for merge */
7271 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7273 i
+= dl
->extent_cnt
;
7275 maxsize
= merge_extents(super
, i
);
7277 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7278 pr_err("attempting to create a second volume with size less then remaining space.\n");
7280 if (maxsize
< size
|| maxsize
== 0) {
7283 pr_err("no free space left on device. Aborting...\n");
7285 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7291 *freesize
= maxsize
;
7294 int count
= count_volumes(super
->hba
,
7295 super
->orom
->dpa
, verbose
);
7296 if (super
->orom
->vphba
<= count
) {
7297 pr_vrb("platform does not support more than %d raid volumes.\n",
7298 super
->orom
->vphba
);
7305 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7306 unsigned long long size
, int chunk
,
7307 unsigned long long *freesize
)
7309 struct intel_super
*super
= st
->sb
;
7310 struct imsm_super
*mpb
= super
->anchor
;
7315 unsigned long long maxsize
;
7316 unsigned long long minsize
;
7320 /* find the largest common start free region of the possible disks */
7324 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7330 /* don't activate new spares if we are orom constrained
7331 * and there is already a volume active in the container
7333 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7336 e
= get_extents(super
, dl
, 0);
7339 for (i
= 1; e
[i
-1].size
; i
++)
7347 maxsize
= merge_extents(super
, extent_cnt
);
7351 minsize
= chunk
* 2;
7353 if (cnt
< raiddisks
||
7354 (super
->orom
&& used
&& used
!= raiddisks
) ||
7355 maxsize
< minsize
||
7357 pr_err("not enough devices with space to create array.\n");
7358 return 0; /* No enough free spaces large enough */
7369 if (mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
)
7370 pr_err("attempting to create a second volume with size less then remaining space.\n");
7372 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7374 dl
->raiddisk
= cnt
++;
7378 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7383 static int reserve_space(struct supertype
*st
, int raiddisks
,
7384 unsigned long long size
, int chunk
,
7385 unsigned long long *freesize
)
7387 struct intel_super
*super
= st
->sb
;
7392 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7395 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7397 dl
->raiddisk
= cnt
++;
7404 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7405 int raiddisks
, int *chunk
, unsigned long long size
,
7406 unsigned long long data_offset
,
7407 char *dev
, unsigned long long *freesize
,
7408 int consistency_policy
, int verbose
)
7415 * if given unused devices create a container
7416 * if given given devices in a container create a member volume
7418 if (level
== LEVEL_CONTAINER
) {
7419 /* Must be a fresh device to add to a container */
7420 return validate_geometry_imsm_container(st
, level
, layout
,
7428 if (size
&& (size
< 1024)) {
7429 pr_err("Given size must be greater than 1M.\n");
7430 /* Depends on algorithm in Create.c :
7431 * if container was given (dev == NULL) return -1,
7432 * if block device was given ( dev != NULL) return 0.
7434 return dev
? -1 : 0;
7439 struct intel_super
*super
= st
->sb
;
7440 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7441 raiddisks
, chunk
, size
,
7444 /* we are being asked to automatically layout a
7445 * new volume based on the current contents of
7446 * the container. If the the parameters can be
7447 * satisfied reserve_space will record the disks,
7448 * start offset, and size of the volume to be
7449 * created. add_to_super and getinfo_super
7450 * detect when autolayout is in progress.
7452 /* assuming that freesize is always given when array is
7454 if (super
->orom
&& freesize
) {
7456 count
= count_volumes(super
->hba
,
7457 super
->orom
->dpa
, verbose
);
7458 if (super
->orom
->vphba
<= count
) {
7459 pr_vrb("platform does not support more than %d raid volumes.\n",
7460 super
->orom
->vphba
);
7465 return reserve_space(st
, raiddisks
, size
,
7471 /* creating in a given container */
7472 return validate_geometry_imsm_volume(st
, level
, layout
,
7473 raiddisks
, chunk
, size
,
7475 dev
, freesize
, verbose
);
7478 /* This device needs to be a device in an 'imsm' container */
7479 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7482 pr_err("Cannot create this array on device %s\n",
7487 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7489 pr_err("Cannot open %s: %s\n",
7490 dev
, strerror(errno
));
7493 /* Well, it is in use by someone, maybe an 'imsm' container. */
7494 cfd
= open_container(fd
);
7498 pr_err("Cannot use %s: It is busy\n",
7502 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7503 if (sra
&& sra
->array
.major_version
== -1 &&
7504 strcmp(sra
->text_version
, "imsm") == 0)
7508 /* This is a member of a imsm container. Load the container
7509 * and try to create a volume
7511 struct intel_super
*super
;
7513 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7515 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7517 return validate_geometry_imsm_volume(st
, level
, layout
,
7519 size
, data_offset
, dev
,
7526 pr_err("failed container membership check\n");
7532 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7534 struct intel_super
*super
= st
->sb
;
7536 if (level
&& *level
== UnSet
)
7537 *level
= LEVEL_CONTAINER
;
7539 if (level
&& layout
&& *layout
== UnSet
)
7540 *layout
= imsm_level_to_layout(*level
);
7542 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7543 *chunk
= imsm_default_chunk(super
->orom
);
7546 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7548 static int kill_subarray_imsm(struct supertype
*st
, char *subarray_id
)
7550 /* remove the subarray currently referenced by subarray_id */
7552 struct intel_dev
**dp
;
7553 struct intel_super
*super
= st
->sb
;
7554 __u8 current_vol
= strtoul(subarray_id
, NULL
, 10);
7555 struct imsm_super
*mpb
= super
->anchor
;
7557 if (mpb
->num_raid_devs
== 0)
7560 /* block deletions that would change the uuid of active subarrays
7562 * FIXME when immutable ids are available, but note that we'll
7563 * also need to fixup the invalidated/active subarray indexes in
7566 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7569 if (i
< current_vol
)
7571 sprintf(subarray
, "%u", i
);
7572 if (is_subarray_active(subarray
, st
->devnm
)) {
7573 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7580 if (st
->update_tail
) {
7581 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7583 u
->type
= update_kill_array
;
7584 u
->dev_idx
= current_vol
;
7585 append_metadata_update(st
, u
, sizeof(*u
));
7590 for (dp
= &super
->devlist
; *dp
;)
7591 if ((*dp
)->index
== current_vol
) {
7594 handle_missing(super
, (*dp
)->dev
);
7595 if ((*dp
)->index
> current_vol
)
7600 /* no more raid devices, all active components are now spares,
7601 * but of course failed are still failed
7603 if (--mpb
->num_raid_devs
== 0) {
7606 for (d
= super
->disks
; d
; d
= d
->next
)
7611 super
->updates_pending
++;
7616 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7617 char *update
, struct mddev_ident
*ident
)
7619 /* update the subarray currently referenced by ->current_vol */
7620 struct intel_super
*super
= st
->sb
;
7621 struct imsm_super
*mpb
= super
->anchor
;
7623 if (strcmp(update
, "name") == 0) {
7624 char *name
= ident
->name
;
7628 if (is_subarray_active(subarray
, st
->devnm
)) {
7629 pr_err("Unable to update name of active subarray\n");
7633 if (!check_name(super
, name
, 0))
7636 vol
= strtoul(subarray
, &ep
, 10);
7637 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7640 if (st
->update_tail
) {
7641 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7643 u
->type
= update_rename_array
;
7645 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7646 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7647 append_metadata_update(st
, u
, sizeof(*u
));
7649 struct imsm_dev
*dev
;
7652 dev
= get_imsm_dev(super
, vol
);
7653 memset(dev
->volume
, '\0', MAX_RAID_SERIAL_LEN
);
7654 namelen
= min((int)strlen(name
), MAX_RAID_SERIAL_LEN
);
7655 memcpy(dev
->volume
, name
, namelen
);
7656 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7657 dev
= get_imsm_dev(super
, i
);
7658 handle_missing(super
, dev
);
7660 super
->updates_pending
++;
7662 } else if (strcmp(update
, "ppl") == 0 ||
7663 strcmp(update
, "no-ppl") == 0) {
7666 int vol
= strtoul(subarray
, &ep
, 10);
7668 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7671 if (strcmp(update
, "ppl") == 0)
7672 new_policy
= RWH_MULTIPLE_DISTRIBUTED
;
7674 new_policy
= RWH_MULTIPLE_OFF
;
7676 if (st
->update_tail
) {
7677 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7679 u
->type
= update_rwh_policy
;
7681 u
->new_policy
= new_policy
;
7682 append_metadata_update(st
, u
, sizeof(*u
));
7684 struct imsm_dev
*dev
;
7686 dev
= get_imsm_dev(super
, vol
);
7687 dev
->rwh_policy
= new_policy
;
7688 super
->updates_pending
++;
7696 static int is_gen_migration(struct imsm_dev
*dev
)
7701 if (!dev
->vol
.migr_state
)
7704 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7710 static int is_rebuilding(struct imsm_dev
*dev
)
7712 struct imsm_map
*migr_map
;
7714 if (!dev
->vol
.migr_state
)
7717 if (migr_type(dev
) != MIGR_REBUILD
)
7720 migr_map
= get_imsm_map(dev
, MAP_1
);
7722 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7728 static int is_initializing(struct imsm_dev
*dev
)
7730 struct imsm_map
*migr_map
;
7732 if (!dev
->vol
.migr_state
)
7735 if (migr_type(dev
) != MIGR_INIT
)
7738 migr_map
= get_imsm_map(dev
, MAP_1
);
7740 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7746 static void update_recovery_start(struct intel_super
*super
,
7747 struct imsm_dev
*dev
,
7748 struct mdinfo
*array
)
7750 struct mdinfo
*rebuild
= NULL
;
7754 if (!is_rebuilding(dev
))
7757 /* Find the rebuild target, but punt on the dual rebuild case */
7758 for (d
= array
->devs
; d
; d
= d
->next
)
7759 if (d
->recovery_start
== 0) {
7766 /* (?) none of the disks are marked with
7767 * IMSM_ORD_REBUILD, so assume they are missing and the
7768 * disk_ord_tbl was not correctly updated
7770 dprintf("failed to locate out-of-sync disk\n");
7774 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7775 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7778 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7780 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7782 /* Given a container loaded by load_super_imsm_all,
7783 * extract information about all the arrays into
7785 * If 'subarray' is given, just extract info about that array.
7787 * For each imsm_dev create an mdinfo, fill it in,
7788 * then look for matching devices in super->disks
7789 * and create appropriate device mdinfo.
7791 struct intel_super
*super
= st
->sb
;
7792 struct imsm_super
*mpb
= super
->anchor
;
7793 struct mdinfo
*rest
= NULL
;
7797 int spare_disks
= 0;
7798 int current_vol
= super
->current_vol
;
7800 /* do not assemble arrays when not all attributes are supported */
7801 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7803 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7806 /* count spare devices, not used in maps
7808 for (d
= super
->disks
; d
; d
= d
->next
)
7812 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7813 struct imsm_dev
*dev
;
7814 struct imsm_map
*map
;
7815 struct imsm_map
*map2
;
7816 struct mdinfo
*this;
7823 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7826 dev
= get_imsm_dev(super
, i
);
7827 map
= get_imsm_map(dev
, MAP_0
);
7828 map2
= get_imsm_map(dev
, MAP_1
);
7829 level
= get_imsm_raid_level(map
);
7831 /* do not publish arrays that are in the middle of an
7832 * unsupported migration
7834 if (dev
->vol
.migr_state
&&
7835 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7836 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7840 /* do not publish arrays that are not support by controller's
7844 this = xmalloc(sizeof(*this));
7846 super
->current_vol
= i
;
7847 getinfo_super_imsm_volume(st
, this, NULL
);
7849 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7850 /* mdadm does not support all metadata features- set the bit in all arrays state */
7851 if (!validate_geometry_imsm_orom(super
,
7852 level
, /* RAID level */
7853 imsm_level_to_layout(level
),
7854 map
->num_members
, /* raid disks */
7855 &chunk
, imsm_dev_size(dev
),
7857 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7859 this->array
.state
|=
7860 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7861 (1<<MD_SB_BLOCK_VOLUME
);
7864 /* if array has bad blocks, set suitable bit in all arrays state */
7866 this->array
.state
|=
7867 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7868 (1<<MD_SB_BLOCK_VOLUME
);
7870 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7871 unsigned long long recovery_start
;
7872 struct mdinfo
*info_d
;
7880 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7881 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7882 for (d
= super
->disks
; d
; d
= d
->next
)
7883 if (d
->index
== idx
)
7886 recovery_start
= MaxSector
;
7889 if (d
&& is_failed(&d
->disk
))
7891 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
7893 if (!(ord
& IMSM_ORD_REBUILD
))
7894 this->array
.working_disks
++;
7896 * if we skip some disks the array will be assmebled degraded;
7897 * reset resync start to avoid a dirty-degraded
7898 * situation when performing the intial sync
7903 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
7904 if ((!able_to_resync(level
, missing
) ||
7905 recovery_start
== 0))
7906 this->resync_start
= MaxSector
;
7909 * FIXME handle dirty degraded
7916 info_d
= xcalloc(1, sizeof(*info_d
));
7917 info_d
->next
= this->devs
;
7918 this->devs
= info_d
;
7920 info_d
->disk
.number
= d
->index
;
7921 info_d
->disk
.major
= d
->major
;
7922 info_d
->disk
.minor
= d
->minor
;
7923 info_d
->disk
.raid_disk
= slot
;
7924 info_d
->recovery_start
= recovery_start
;
7926 if (slot
< map2
->num_members
)
7927 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7929 this->array
.spare_disks
++;
7931 if (slot
< map
->num_members
)
7932 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7934 this->array
.spare_disks
++;
7937 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7938 info_d
->data_offset
= pba_of_lba0(map
);
7939 info_d
->component_size
= calc_component_size(map
, dev
);
7941 if (map
->raid_level
== 5) {
7942 info_d
->ppl_sector
= this->ppl_sector
;
7943 info_d
->ppl_size
= this->ppl_size
;
7944 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
7945 recovery_start
== 0)
7946 this->resync_start
= 0;
7949 info_d
->bb
.supported
= 1;
7950 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7951 info_d
->data_offset
,
7952 info_d
->component_size
,
7955 /* now that the disk list is up-to-date fixup recovery_start */
7956 update_recovery_start(super
, dev
, this);
7957 this->array
.spare_disks
+= spare_disks
;
7959 /* check for reshape */
7960 if (this->reshape_active
== 1)
7961 recover_backup_imsm(st
, this);
7965 super
->current_vol
= current_vol
;
7969 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7970 int failed
, int look_in_map
)
7972 struct imsm_map
*map
;
7974 map
= get_imsm_map(dev
, look_in_map
);
7977 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
7978 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
7980 switch (get_imsm_raid_level(map
)) {
7982 return IMSM_T_STATE_FAILED
;
7985 if (failed
< map
->num_members
)
7986 return IMSM_T_STATE_DEGRADED
;
7988 return IMSM_T_STATE_FAILED
;
7993 * check to see if any mirrors have failed, otherwise we
7994 * are degraded. Even numbered slots are mirrored on
7998 /* gcc -Os complains that this is unused */
7999 int insync
= insync
;
8001 for (i
= 0; i
< map
->num_members
; i
++) {
8002 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
8003 int idx
= ord_to_idx(ord
);
8004 struct imsm_disk
*disk
;
8006 /* reset the potential in-sync count on even-numbered
8007 * slots. num_copies is always 2 for imsm raid10
8012 disk
= get_imsm_disk(super
, idx
);
8013 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8016 /* no in-sync disks left in this mirror the
8020 return IMSM_T_STATE_FAILED
;
8023 return IMSM_T_STATE_DEGRADED
;
8027 return IMSM_T_STATE_DEGRADED
;
8029 return IMSM_T_STATE_FAILED
;
8035 return map
->map_state
;
8038 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
8043 struct imsm_disk
*disk
;
8044 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8045 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
8046 struct imsm_map
*map_for_loop
;
8051 /* at the beginning of migration we set IMSM_ORD_REBUILD on
8052 * disks that are being rebuilt. New failures are recorded to
8053 * map[0]. So we look through all the disks we started with and
8054 * see if any failures are still present, or if any new ones
8058 if (prev
&& (map
->num_members
< prev
->num_members
))
8059 map_for_loop
= prev
;
8061 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8063 /* when MAP_X is passed both maps failures are counted
8066 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8067 i
< prev
->num_members
) {
8068 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8069 idx_1
= ord_to_idx(ord
);
8071 disk
= get_imsm_disk(super
, idx_1
);
8072 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8075 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8076 i
< map
->num_members
) {
8077 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8078 idx
= ord_to_idx(ord
);
8081 disk
= get_imsm_disk(super
, idx
);
8082 if (!disk
|| is_failed(disk
) ||
8083 ord
& IMSM_ORD_REBUILD
)
8092 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8095 struct intel_super
*super
= c
->sb
;
8096 struct imsm_super
*mpb
= super
->anchor
;
8097 struct imsm_update_prealloc_bb_mem u
;
8099 if (atoi(inst
) >= mpb
->num_raid_devs
) {
8100 pr_err("subarry index %d, out of range\n", atoi(inst
));
8104 dprintf("imsm: open_new %s\n", inst
);
8105 a
->info
.container_member
= atoi(inst
);
8107 u
.type
= update_prealloc_badblocks_mem
;
8108 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8113 static int is_resyncing(struct imsm_dev
*dev
)
8115 struct imsm_map
*migr_map
;
8117 if (!dev
->vol
.migr_state
)
8120 if (migr_type(dev
) == MIGR_INIT
||
8121 migr_type(dev
) == MIGR_REPAIR
)
8124 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8127 migr_map
= get_imsm_map(dev
, MAP_1
);
8129 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8130 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8136 /* return true if we recorded new information */
8137 static int mark_failure(struct intel_super
*super
,
8138 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8142 struct imsm_map
*map
;
8143 char buf
[MAX_RAID_SERIAL_LEN
+3];
8144 unsigned int len
, shift
= 0;
8146 /* new failures are always set in map[0] */
8147 map
= get_imsm_map(dev
, MAP_0
);
8149 slot
= get_imsm_disk_slot(map
, idx
);
8153 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8154 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8157 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8158 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8160 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8161 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8162 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8164 disk
->status
|= FAILED_DISK
;
8165 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8166 /* mark failures in second map if second map exists and this disk
8168 * This is valid for migration, initialization and rebuild
8170 if (dev
->vol
.migr_state
) {
8171 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8172 int slot2
= get_imsm_disk_slot(map2
, idx
);
8174 if (slot2
< map2
->num_members
&& slot2
>= 0)
8175 set_imsm_ord_tbl_ent(map2
, slot2
,
8176 idx
| IMSM_ORD_REBUILD
);
8178 if (map
->failed_disk_num
== 0xff ||
8179 (!is_rebuilding(dev
) && map
->failed_disk_num
> slot
))
8180 map
->failed_disk_num
= slot
;
8182 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8187 static void mark_missing(struct intel_super
*super
,
8188 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8190 mark_failure(super
, dev
, disk
, idx
);
8192 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8195 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8196 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8199 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8203 if (!super
->missing
)
8206 /* When orom adds replacement for missing disk it does
8207 * not remove entry of missing disk, but just updates map with
8208 * new added disk. So it is not enough just to test if there is
8209 * any missing disk, we have to look if there are any failed disks
8210 * in map to stop migration */
8212 dprintf("imsm: mark missing\n");
8213 /* end process for initialization and rebuild only
8215 if (is_gen_migration(dev
) == 0) {
8216 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8220 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8221 struct imsm_map
*map1
;
8222 int i
, ord
, ord_map1
;
8225 for (i
= 0; i
< map
->num_members
; i
++) {
8226 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8227 if (!(ord
& IMSM_ORD_REBUILD
))
8230 map1
= get_imsm_map(dev
, MAP_1
);
8234 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8235 if (ord_map1
& IMSM_ORD_REBUILD
)
8240 map_state
= imsm_check_degraded(super
, dev
,
8242 end_migration(dev
, super
, map_state
);
8246 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8247 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8248 super
->updates_pending
++;
8251 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8254 unsigned long long array_blocks
;
8255 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8256 int used_disks
= imsm_num_data_members(map
);
8258 if (used_disks
== 0) {
8259 /* when problems occures
8260 * return current array_blocks value
8262 array_blocks
= imsm_dev_size(dev
);
8264 return array_blocks
;
8267 /* set array size in metadata
8270 /* OLCE size change is caused by added disks
8272 array_blocks
= per_dev_array_size(map
) * used_disks
;
8274 /* Online Volume Size Change
8275 * Using available free space
8277 array_blocks
= new_size
;
8279 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8280 set_imsm_dev_size(dev
, array_blocks
);
8282 return array_blocks
;
8285 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8287 static void imsm_progress_container_reshape(struct intel_super
*super
)
8289 /* if no device has a migr_state, but some device has a
8290 * different number of members than the previous device, start
8291 * changing the number of devices in this device to match
8294 struct imsm_super
*mpb
= super
->anchor
;
8295 int prev_disks
= -1;
8299 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8300 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8301 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8302 struct imsm_map
*map2
;
8303 int prev_num_members
;
8305 if (dev
->vol
.migr_state
)
8308 if (prev_disks
== -1)
8309 prev_disks
= map
->num_members
;
8310 if (prev_disks
== map
->num_members
)
8313 /* OK, this array needs to enter reshape mode.
8314 * i.e it needs a migr_state
8317 copy_map_size
= sizeof_imsm_map(map
);
8318 prev_num_members
= map
->num_members
;
8319 map
->num_members
= prev_disks
;
8320 dev
->vol
.migr_state
= 1;
8321 dev
->vol
.curr_migr_unit
= 0;
8322 set_migr_type(dev
, MIGR_GEN_MIGR
);
8323 for (i
= prev_num_members
;
8324 i
< map
->num_members
; i
++)
8325 set_imsm_ord_tbl_ent(map
, i
, i
);
8326 map2
= get_imsm_map(dev
, MAP_1
);
8327 /* Copy the current map */
8328 memcpy(map2
, map
, copy_map_size
);
8329 map2
->num_members
= prev_num_members
;
8331 imsm_set_array_size(dev
, -1);
8332 super
->clean_migration_record_by_mdmon
= 1;
8333 super
->updates_pending
++;
8337 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8338 * states are handled in imsm_set_disk() with one exception, when a
8339 * resync is stopped due to a new failure this routine will set the
8340 * 'degraded' state for the array.
8342 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8344 int inst
= a
->info
.container_member
;
8345 struct intel_super
*super
= a
->container
->sb
;
8346 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8347 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8348 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8349 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8350 __u32 blocks_per_unit
;
8352 if (dev
->vol
.migr_state
&&
8353 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8354 /* array state change is blocked due to reshape action
8356 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8357 * - finish the reshape (if last_checkpoint is big and action != reshape)
8358 * - update curr_migr_unit
8360 if (a
->curr_action
== reshape
) {
8361 /* still reshaping, maybe update curr_migr_unit */
8362 goto mark_checkpoint
;
8364 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8365 /* for some reason we aborted the reshape.
8367 * disable automatic metadata rollback
8368 * user action is required to recover process
8371 struct imsm_map
*map2
=
8372 get_imsm_map(dev
, MAP_1
);
8373 dev
->vol
.migr_state
= 0;
8374 set_migr_type(dev
, 0);
8375 dev
->vol
.curr_migr_unit
= 0;
8377 sizeof_imsm_map(map2
));
8378 super
->updates_pending
++;
8381 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8382 unsigned long long array_blocks
;
8386 used_disks
= imsm_num_data_members(map
);
8387 if (used_disks
> 0) {
8389 per_dev_array_size(map
) *
8392 round_size_to_mb(array_blocks
,
8394 a
->info
.custom_array_size
= array_blocks
;
8395 /* encourage manager to update array
8399 a
->check_reshape
= 1;
8401 /* finalize online capacity expansion/reshape */
8402 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8404 mdi
->disk
.raid_disk
,
8407 imsm_progress_container_reshape(super
);
8412 /* before we activate this array handle any missing disks */
8413 if (consistent
== 2)
8414 handle_missing(super
, dev
);
8416 if (consistent
== 2 &&
8417 (!is_resync_complete(&a
->info
) ||
8418 map_state
!= IMSM_T_STATE_NORMAL
||
8419 dev
->vol
.migr_state
))
8422 if (is_resync_complete(&a
->info
)) {
8423 /* complete intialization / resync,
8424 * recovery and interrupted recovery is completed in
8427 if (is_resyncing(dev
)) {
8428 dprintf("imsm: mark resync done\n");
8429 end_migration(dev
, super
, map_state
);
8430 super
->updates_pending
++;
8431 a
->last_checkpoint
= 0;
8433 } else if ((!is_resyncing(dev
) && !failed
) &&
8434 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8435 /* mark the start of the init process if nothing is failed */
8436 dprintf("imsm: mark resync start\n");
8437 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8438 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8440 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8441 super
->updates_pending
++;
8445 /* skip checkpointing for general migration,
8446 * it is controlled in mdadm
8448 if (is_gen_migration(dev
))
8449 goto skip_mark_checkpoint
;
8451 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8452 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8453 if (blocks_per_unit
) {
8457 units
= a
->last_checkpoint
/ blocks_per_unit
;
8460 /* check that we did not overflow 32-bits, and that
8461 * curr_migr_unit needs updating
8463 if (units32
== units
&&
8465 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8466 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8467 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8468 super
->updates_pending
++;
8472 skip_mark_checkpoint
:
8473 /* mark dirty / clean */
8474 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8475 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8476 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8478 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8480 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8481 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8482 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8483 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8485 super
->updates_pending
++;
8491 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8493 int inst
= a
->info
.container_member
;
8494 struct intel_super
*super
= a
->container
->sb
;
8495 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8496 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8498 if (slot
> map
->num_members
) {
8499 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8500 slot
, map
->num_members
- 1);
8507 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8510 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8512 int inst
= a
->info
.container_member
;
8513 struct intel_super
*super
= a
->container
->sb
;
8514 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8515 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8516 struct imsm_disk
*disk
;
8518 int recovery_not_finished
= 0;
8522 int rebuild_done
= 0;
8525 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8529 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8530 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8532 /* check for new failures */
8533 if (disk
&& (state
& DS_FAULTY
)) {
8534 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8535 super
->updates_pending
++;
8538 /* check if in_sync */
8539 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8540 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8542 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8544 super
->updates_pending
++;
8547 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8548 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8550 /* check if recovery complete, newly degraded, or failed */
8551 dprintf("imsm: Detected transition to state ");
8552 switch (map_state
) {
8553 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8554 dprintf("normal: ");
8555 if (is_rebuilding(dev
)) {
8556 dprintf_cont("while rebuilding");
8557 /* check if recovery is really finished */
8558 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8559 if (mdi
->recovery_start
!= MaxSector
) {
8560 recovery_not_finished
= 1;
8563 if (recovery_not_finished
) {
8565 dprintf("Rebuild has not finished yet, state not changed");
8566 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8567 a
->last_checkpoint
= mdi
->recovery_start
;
8568 super
->updates_pending
++;
8572 end_migration(dev
, super
, map_state
);
8573 map
= get_imsm_map(dev
, MAP_0
);
8574 map
->failed_disk_num
= ~0;
8575 super
->updates_pending
++;
8576 a
->last_checkpoint
= 0;
8579 if (is_gen_migration(dev
)) {
8580 dprintf_cont("while general migration");
8581 if (a
->last_checkpoint
>= a
->info
.component_size
)
8582 end_migration(dev
, super
, map_state
);
8584 map
->map_state
= map_state
;
8585 map
= get_imsm_map(dev
, MAP_0
);
8586 map
->failed_disk_num
= ~0;
8587 super
->updates_pending
++;
8591 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8592 dprintf_cont("degraded: ");
8593 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8594 dprintf_cont("mark degraded");
8595 map
->map_state
= map_state
;
8596 super
->updates_pending
++;
8597 a
->last_checkpoint
= 0;
8600 if (is_rebuilding(dev
)) {
8601 dprintf_cont("while rebuilding ");
8602 if (state
& DS_FAULTY
) {
8603 dprintf_cont("removing failed drive ");
8604 if (n
== map
->failed_disk_num
) {
8605 dprintf_cont("end migration");
8606 end_migration(dev
, super
, map_state
);
8607 a
->last_checkpoint
= 0;
8609 dprintf_cont("fail detected during rebuild, changing map state");
8610 map
->map_state
= map_state
;
8612 super
->updates_pending
++;
8618 /* check if recovery is really finished */
8619 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8620 if (mdi
->recovery_start
!= MaxSector
) {
8621 recovery_not_finished
= 1;
8624 if (recovery_not_finished
) {
8626 dprintf_cont("Rebuild has not finished yet");
8627 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8628 a
->last_checkpoint
=
8629 mdi
->recovery_start
;
8630 super
->updates_pending
++;
8635 dprintf_cont(" Rebuild done, still degraded");
8636 end_migration(dev
, super
, map_state
);
8637 a
->last_checkpoint
= 0;
8638 super
->updates_pending
++;
8640 for (i
= 0; i
< map
->num_members
; i
++) {
8641 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8643 if (idx
& IMSM_ORD_REBUILD
)
8644 map
->failed_disk_num
= i
;
8646 super
->updates_pending
++;
8649 if (is_gen_migration(dev
)) {
8650 dprintf_cont("while general migration");
8651 if (a
->last_checkpoint
>= a
->info
.component_size
)
8652 end_migration(dev
, super
, map_state
);
8654 map
->map_state
= map_state
;
8655 manage_second_map(super
, dev
);
8657 super
->updates_pending
++;
8660 if (is_initializing(dev
)) {
8661 dprintf_cont("while initialization.");
8662 map
->map_state
= map_state
;
8663 super
->updates_pending
++;
8667 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8668 dprintf_cont("failed: ");
8669 if (is_gen_migration(dev
)) {
8670 dprintf_cont("while general migration");
8671 map
->map_state
= map_state
;
8672 super
->updates_pending
++;
8675 if (map
->map_state
!= map_state
) {
8676 dprintf_cont("mark failed");
8677 end_migration(dev
, super
, map_state
);
8678 super
->updates_pending
++;
8679 a
->last_checkpoint
= 0;
8684 dprintf_cont("state %i\n", map_state
);
8689 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8692 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8693 unsigned long long dsize
;
8694 unsigned long long sectors
;
8695 unsigned int sector_size
;
8697 get_dev_sector_size(fd
, NULL
, §or_size
);
8698 get_dev_size(fd
, NULL
, &dsize
);
8700 if (mpb_size
> sector_size
) {
8701 /* -1 to account for anchor */
8702 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8704 /* write the extended mpb to the sectors preceeding the anchor */
8705 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8709 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8710 sector_size
* sectors
) != sector_size
* sectors
)
8714 /* first block is stored on second to last sector of the disk */
8715 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8718 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8724 static void imsm_sync_metadata(struct supertype
*container
)
8726 struct intel_super
*super
= container
->sb
;
8728 dprintf("sync metadata: %d\n", super
->updates_pending
);
8729 if (!super
->updates_pending
)
8732 write_super_imsm(container
, 0);
8734 super
->updates_pending
= 0;
8737 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8739 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8740 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8743 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8747 if (dl
&& is_failed(&dl
->disk
))
8751 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8756 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8757 struct active_array
*a
, int activate_new
,
8758 struct mdinfo
*additional_test_list
)
8760 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8761 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8762 struct imsm_super
*mpb
= super
->anchor
;
8763 struct imsm_map
*map
;
8764 unsigned long long pos
;
8769 __u32 array_start
= 0;
8770 __u32 array_end
= 0;
8772 struct mdinfo
*test_list
;
8774 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8775 /* If in this array, skip */
8776 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8777 if (d
->state_fd
>= 0 &&
8778 d
->disk
.major
== dl
->major
&&
8779 d
->disk
.minor
== dl
->minor
) {
8780 dprintf("%x:%x already in array\n",
8781 dl
->major
, dl
->minor
);
8786 test_list
= additional_test_list
;
8788 if (test_list
->disk
.major
== dl
->major
&&
8789 test_list
->disk
.minor
== dl
->minor
) {
8790 dprintf("%x:%x already in additional test list\n",
8791 dl
->major
, dl
->minor
);
8794 test_list
= test_list
->next
;
8799 /* skip in use or failed drives */
8800 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8802 dprintf("%x:%x status (failed: %d index: %d)\n",
8803 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8807 /* skip pure spares when we are looking for partially
8808 * assimilated drives
8810 if (dl
->index
== -1 && !activate_new
)
8813 if (!drive_validate_sector_size(super
, dl
))
8816 /* Does this unused device have the requisite free space?
8817 * It needs to be able to cover all member volumes
8819 ex
= get_extents(super
, dl
, 1);
8821 dprintf("cannot get extents\n");
8824 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8825 dev
= get_imsm_dev(super
, i
);
8826 map
= get_imsm_map(dev
, MAP_0
);
8828 /* check if this disk is already a member of
8831 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8837 array_start
= pba_of_lba0(map
);
8838 array_end
= array_start
+
8839 per_dev_array_size(map
) - 1;
8842 /* check that we can start at pba_of_lba0 with
8843 * num_data_stripes*blocks_per_stripe of space
8845 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8849 pos
= ex
[j
].start
+ ex
[j
].size
;
8851 } while (ex
[j
-1].size
);
8858 if (i
< mpb
->num_raid_devs
) {
8859 dprintf("%x:%x does not have %u to %u available\n",
8860 dl
->major
, dl
->minor
, array_start
, array_end
);
8870 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8872 struct imsm_dev
*dev2
;
8873 struct imsm_map
*map
;
8879 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8881 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8882 if (state
== IMSM_T_STATE_FAILED
) {
8883 map
= get_imsm_map(dev2
, MAP_0
);
8886 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8888 * Check if failed disks are deleted from intel
8889 * disk list or are marked to be deleted
8891 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8892 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8894 * Do not rebuild the array if failed disks
8895 * from failed sub-array are not removed from
8899 is_failed(&idisk
->disk
) &&
8900 (idisk
->action
!= DISK_REMOVE
))
8908 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8909 struct metadata_update
**updates
)
8912 * Find a device with unused free space and use it to replace a
8913 * failed/vacant region in an array. We replace failed regions one a
8914 * array at a time. The result is that a new spare disk will be added
8915 * to the first failed array and after the monitor has finished
8916 * propagating failures the remainder will be consumed.
8918 * FIXME add a capability for mdmon to request spares from another
8922 struct intel_super
*super
= a
->container
->sb
;
8923 int inst
= a
->info
.container_member
;
8924 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8925 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8926 int failed
= a
->info
.array
.raid_disks
;
8927 struct mdinfo
*rv
= NULL
;
8930 struct metadata_update
*mu
;
8932 struct imsm_update_activate_spare
*u
;
8937 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8938 if ((d
->curr_state
& DS_FAULTY
) &&
8940 /* wait for Removal to happen */
8942 if (d
->state_fd
>= 0)
8946 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8947 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8949 if (imsm_reshape_blocks_arrays_changes(super
))
8952 /* Cannot activate another spare if rebuild is in progress already
8954 if (is_rebuilding(dev
)) {
8955 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8959 if (a
->info
.array
.level
== 4)
8960 /* No repair for takeovered array
8961 * imsm doesn't support raid4
8965 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8966 IMSM_T_STATE_DEGRADED
)
8969 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8970 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8975 * If there are any failed disks check state of the other volume.
8976 * Block rebuild if the another one is failed until failed disks
8977 * are removed from container.
8980 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8981 MAX_RAID_SERIAL_LEN
, dev
->volume
);
8982 /* check if states of the other volumes allow for rebuild */
8983 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
8985 allowed
= imsm_rebuild_allowed(a
->container
,
8993 /* For each slot, if it is not working, find a spare */
8994 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
8995 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8996 if (d
->disk
.raid_disk
== i
)
8998 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
8999 if (d
&& (d
->state_fd
>= 0))
9003 * OK, this device needs recovery. Try to re-add the
9004 * previous occupant of this slot, if this fails see if
9005 * we can continue the assimilation of a spare that was
9006 * partially assimilated, finally try to activate a new
9009 dl
= imsm_readd(super
, i
, a
);
9011 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
9013 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
9017 /* found a usable disk with enough space */
9018 di
= xcalloc(1, sizeof(*di
));
9020 /* dl->index will be -1 in the case we are activating a
9021 * pristine spare. imsm_process_update() will create a
9022 * new index in this case. Once a disk is found to be
9023 * failed in all member arrays it is kicked from the
9026 di
->disk
.number
= dl
->index
;
9028 /* (ab)use di->devs to store a pointer to the device
9031 di
->devs
= (struct mdinfo
*) dl
;
9033 di
->disk
.raid_disk
= i
;
9034 di
->disk
.major
= dl
->major
;
9035 di
->disk
.minor
= dl
->minor
;
9037 di
->recovery_start
= 0;
9038 di
->data_offset
= pba_of_lba0(map
);
9039 di
->component_size
= a
->info
.component_size
;
9040 di
->container_member
= inst
;
9041 di
->bb
.supported
= 1;
9042 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
9043 di
->ppl_sector
= get_ppl_sector(super
, inst
);
9044 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
9046 super
->random
= random32();
9050 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
9051 i
, di
->data_offset
);
9055 /* No spares found */
9057 /* Now 'rv' has a list of devices to return.
9058 * Create a metadata_update record to update the
9059 * disk_ord_tbl for the array
9061 mu
= xmalloc(sizeof(*mu
));
9062 mu
->buf
= xcalloc(num_spares
,
9063 sizeof(struct imsm_update_activate_spare
));
9065 mu
->space_list
= NULL
;
9066 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
9067 mu
->next
= *updates
;
9068 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9070 for (di
= rv
; di
; di
= di
->next
) {
9071 u
->type
= update_activate_spare
;
9072 u
->dl
= (struct dl
*) di
->devs
;
9074 u
->slot
= di
->disk
.raid_disk
;
9085 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9087 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9088 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9089 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9090 struct disk_info
*inf
= get_disk_info(u
);
9091 struct imsm_disk
*disk
;
9095 for (i
= 0; i
< map
->num_members
; i
++) {
9096 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9097 for (j
= 0; j
< new_map
->num_members
; j
++)
9098 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9105 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9109 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9110 if (dl
->major
== major
&& dl
->minor
== minor
)
9115 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9121 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9122 if (dl
->major
== major
&& dl
->minor
== minor
) {
9125 prev
->next
= dl
->next
;
9127 super
->disks
= dl
->next
;
9129 __free_imsm_disk(dl
);
9130 dprintf("removed %x:%x\n", major
, minor
);
9138 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9140 static int add_remove_disk_update(struct intel_super
*super
)
9142 int check_degraded
= 0;
9145 /* add/remove some spares to/from the metadata/contrainer */
9146 while (super
->disk_mgmt_list
) {
9147 struct dl
*disk_cfg
;
9149 disk_cfg
= super
->disk_mgmt_list
;
9150 super
->disk_mgmt_list
= disk_cfg
->next
;
9151 disk_cfg
->next
= NULL
;
9153 if (disk_cfg
->action
== DISK_ADD
) {
9154 disk_cfg
->next
= super
->disks
;
9155 super
->disks
= disk_cfg
;
9157 dprintf("added %x:%x\n",
9158 disk_cfg
->major
, disk_cfg
->minor
);
9159 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9160 dprintf("Disk remove action processed: %x.%x\n",
9161 disk_cfg
->major
, disk_cfg
->minor
);
9162 disk
= get_disk_super(super
,
9166 /* store action status */
9167 disk
->action
= DISK_REMOVE
;
9168 /* remove spare disks only */
9169 if (disk
->index
== -1) {
9170 remove_disk_super(super
,
9174 disk_cfg
->fd
= disk
->fd
;
9178 /* release allocate disk structure */
9179 __free_imsm_disk(disk_cfg
);
9182 return check_degraded
;
9185 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9186 struct intel_super
*super
,
9189 struct intel_dev
*id
;
9190 void **tofree
= NULL
;
9193 dprintf("(enter)\n");
9194 if (u
->subdev
< 0 || u
->subdev
> 1) {
9195 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9198 if (space_list
== NULL
|| *space_list
== NULL
) {
9199 dprintf("imsm: Error: Memory is not allocated\n");
9203 for (id
= super
->devlist
; id
; id
= id
->next
) {
9204 if (id
->index
== (unsigned)u
->subdev
) {
9205 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9206 struct imsm_map
*map
;
9207 struct imsm_dev
*new_dev
=
9208 (struct imsm_dev
*)*space_list
;
9209 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9211 struct dl
*new_disk
;
9213 if (new_dev
== NULL
)
9215 *space_list
= **space_list
;
9216 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9217 map
= get_imsm_map(new_dev
, MAP_0
);
9219 dprintf("imsm: Error: migration in progress");
9223 to_state
= map
->map_state
;
9224 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9226 /* this should not happen */
9227 if (u
->new_disks
[0] < 0) {
9228 map
->failed_disk_num
=
9229 map
->num_members
- 1;
9230 to_state
= IMSM_T_STATE_DEGRADED
;
9232 to_state
= IMSM_T_STATE_NORMAL
;
9234 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9235 if (u
->new_level
> -1)
9236 map
->raid_level
= u
->new_level
;
9237 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9238 if ((u
->new_level
== 5) &&
9239 (migr_map
->raid_level
== 0)) {
9240 int ord
= map
->num_members
- 1;
9241 migr_map
->num_members
--;
9242 if (u
->new_disks
[0] < 0)
9243 ord
|= IMSM_ORD_REBUILD
;
9244 set_imsm_ord_tbl_ent(map
,
9245 map
->num_members
- 1,
9249 tofree
= (void **)dev
;
9251 /* update chunk size
9253 if (u
->new_chunksize
> 0) {
9254 unsigned long long num_data_stripes
;
9255 struct imsm_map
*dest_map
=
9256 get_imsm_map(dev
, MAP_0
);
9258 imsm_num_data_members(dest_map
);
9260 if (used_disks
== 0)
9263 map
->blocks_per_strip
=
9264 __cpu_to_le16(u
->new_chunksize
* 2);
9266 imsm_dev_size(dev
) / used_disks
;
9267 num_data_stripes
/= map
->blocks_per_strip
;
9268 num_data_stripes
/= map
->num_domains
;
9269 set_num_data_stripes(map
, num_data_stripes
);
9272 /* ensure blocks_per_member has valid value
9274 set_blocks_per_member(map
,
9275 per_dev_array_size(map
) +
9276 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9280 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9281 migr_map
->raid_level
== map
->raid_level
)
9284 if (u
->new_disks
[0] >= 0) {
9287 new_disk
= get_disk_super(super
,
9288 major(u
->new_disks
[0]),
9289 minor(u
->new_disks
[0]));
9290 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9291 major(u
->new_disks
[0]),
9292 minor(u
->new_disks
[0]),
9293 new_disk
, new_disk
->index
);
9294 if (new_disk
== NULL
)
9295 goto error_disk_add
;
9297 new_disk
->index
= map
->num_members
- 1;
9298 /* slot to fill in autolayout
9300 new_disk
->raiddisk
= new_disk
->index
;
9301 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9302 new_disk
->disk
.status
&= ~SPARE_DISK
;
9304 goto error_disk_add
;
9307 *tofree
= *space_list
;
9308 /* calculate new size
9310 imsm_set_array_size(new_dev
, -1);
9317 *space_list
= tofree
;
9321 dprintf("Error: imsm: Cannot find disk.\n");
9325 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9326 struct intel_super
*super
)
9328 struct intel_dev
*id
;
9331 dprintf("(enter)\n");
9332 if (u
->subdev
< 0 || u
->subdev
> 1) {
9333 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9337 for (id
= super
->devlist
; id
; id
= id
->next
) {
9338 if (id
->index
== (unsigned)u
->subdev
) {
9339 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9340 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9341 int used_disks
= imsm_num_data_members(map
);
9342 unsigned long long blocks_per_member
;
9343 unsigned long long num_data_stripes
;
9344 unsigned long long new_size_per_disk
;
9346 if (used_disks
== 0)
9349 /* calculate new size
9351 new_size_per_disk
= u
->new_size
/ used_disks
;
9352 blocks_per_member
= new_size_per_disk
+
9353 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9354 num_data_stripes
= new_size_per_disk
/
9355 map
->blocks_per_strip
;
9356 num_data_stripes
/= map
->num_domains
;
9357 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9358 u
->new_size
, new_size_per_disk
,
9360 set_blocks_per_member(map
, blocks_per_member
);
9361 set_num_data_stripes(map
, num_data_stripes
);
9362 imsm_set_array_size(dev
, u
->new_size
);
9372 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9373 struct intel_super
*super
,
9374 struct active_array
*active_array
)
9376 struct imsm_super
*mpb
= super
->anchor
;
9377 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9378 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9379 struct imsm_map
*migr_map
;
9380 struct active_array
*a
;
9381 struct imsm_disk
*disk
;
9388 int second_map_created
= 0;
9390 for (; u
; u
= u
->next
) {
9391 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9396 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9401 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9406 /* count failures (excluding rebuilds and the victim)
9407 * to determine map[0] state
9410 for (i
= 0; i
< map
->num_members
; i
++) {
9413 disk
= get_imsm_disk(super
,
9414 get_imsm_disk_idx(dev
, i
, MAP_X
));
9415 if (!disk
|| is_failed(disk
))
9419 /* adding a pristine spare, assign a new index */
9420 if (dl
->index
< 0) {
9421 dl
->index
= super
->anchor
->num_disks
;
9422 super
->anchor
->num_disks
++;
9425 disk
->status
|= CONFIGURED_DISK
;
9426 disk
->status
&= ~SPARE_DISK
;
9429 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9430 if (!second_map_created
) {
9431 second_map_created
= 1;
9432 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9433 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9435 map
->map_state
= to_state
;
9436 migr_map
= get_imsm_map(dev
, MAP_1
);
9437 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9438 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9439 dl
->index
| IMSM_ORD_REBUILD
);
9441 /* update the family_num to mark a new container
9442 * generation, being careful to record the existing
9443 * family_num in orig_family_num to clean up after
9444 * earlier mdadm versions that neglected to set it.
9446 if (mpb
->orig_family_num
== 0)
9447 mpb
->orig_family_num
= mpb
->family_num
;
9448 mpb
->family_num
+= super
->random
;
9450 /* count arrays using the victim in the metadata */
9452 for (a
= active_array
; a
; a
= a
->next
) {
9453 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9454 map
= get_imsm_map(dev
, MAP_0
);
9456 if (get_imsm_disk_slot(map
, victim
) >= 0)
9460 /* delete the victim if it is no longer being
9466 /* We know that 'manager' isn't touching anything,
9467 * so it is safe to delete
9469 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9470 if ((*dlp
)->index
== victim
)
9473 /* victim may be on the missing list */
9475 for (dlp
= &super
->missing
; *dlp
;
9476 dlp
= &(*dlp
)->next
)
9477 if ((*dlp
)->index
== victim
)
9479 imsm_delete(super
, dlp
, victim
);
9486 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9487 struct intel_super
*super
,
9490 struct dl
*new_disk
;
9491 struct intel_dev
*id
;
9493 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9494 int disk_count
= u
->old_raid_disks
;
9495 void **tofree
= NULL
;
9496 int devices_to_reshape
= 1;
9497 struct imsm_super
*mpb
= super
->anchor
;
9499 unsigned int dev_id
;
9501 dprintf("(enter)\n");
9503 /* enable spares to use in array */
9504 for (i
= 0; i
< delta_disks
; i
++) {
9505 new_disk
= get_disk_super(super
,
9506 major(u
->new_disks
[i
]),
9507 minor(u
->new_disks
[i
]));
9508 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9509 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9510 new_disk
, new_disk
->index
);
9511 if (new_disk
== NULL
||
9512 (new_disk
->index
>= 0 &&
9513 new_disk
->index
< u
->old_raid_disks
))
9514 goto update_reshape_exit
;
9515 new_disk
->index
= disk_count
++;
9516 /* slot to fill in autolayout
9518 new_disk
->raiddisk
= new_disk
->index
;
9519 new_disk
->disk
.status
|=
9521 new_disk
->disk
.status
&= ~SPARE_DISK
;
9524 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9525 mpb
->num_raid_devs
);
9526 /* manage changes in volume
9528 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9529 void **sp
= *space_list
;
9530 struct imsm_dev
*newdev
;
9531 struct imsm_map
*newmap
, *oldmap
;
9533 for (id
= super
->devlist
; id
; id
= id
->next
) {
9534 if (id
->index
== dev_id
)
9543 /* Copy the dev, but not (all of) the map */
9544 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9545 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9546 newmap
= get_imsm_map(newdev
, MAP_0
);
9547 /* Copy the current map */
9548 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9549 /* update one device only
9551 if (devices_to_reshape
) {
9552 dprintf("imsm: modifying subdev: %i\n",
9554 devices_to_reshape
--;
9555 newdev
->vol
.migr_state
= 1;
9556 newdev
->vol
.curr_migr_unit
= 0;
9557 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9558 newmap
->num_members
= u
->new_raid_disks
;
9559 for (i
= 0; i
< delta_disks
; i
++) {
9560 set_imsm_ord_tbl_ent(newmap
,
9561 u
->old_raid_disks
+ i
,
9562 u
->old_raid_disks
+ i
);
9564 /* New map is correct, now need to save old map
9566 newmap
= get_imsm_map(newdev
, MAP_1
);
9567 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9569 imsm_set_array_size(newdev
, -1);
9572 sp
= (void **)id
->dev
;
9577 /* Clear migration record */
9578 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9581 *space_list
= tofree
;
9584 update_reshape_exit
:
9589 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9590 struct intel_super
*super
,
9593 struct imsm_dev
*dev
= NULL
;
9594 struct intel_dev
*dv
;
9595 struct imsm_dev
*dev_new
;
9596 struct imsm_map
*map
;
9600 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9601 if (dv
->index
== (unsigned int)u
->subarray
) {
9609 map
= get_imsm_map(dev
, MAP_0
);
9611 if (u
->direction
== R10_TO_R0
) {
9612 unsigned long long num_data_stripes
;
9614 /* Number of failed disks must be half of initial disk number */
9615 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9616 (map
->num_members
/ 2))
9619 /* iterate through devices to mark removed disks as spare */
9620 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9621 if (dm
->disk
.status
& FAILED_DISK
) {
9622 int idx
= dm
->index
;
9623 /* update indexes on the disk list */
9624 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9625 the index values will end up being correct.... NB */
9626 for (du
= super
->disks
; du
; du
= du
->next
)
9627 if (du
->index
> idx
)
9629 /* mark as spare disk */
9634 map
->num_members
= map
->num_members
/ 2;
9635 map
->map_state
= IMSM_T_STATE_NORMAL
;
9636 map
->num_domains
= 1;
9637 map
->raid_level
= 0;
9638 map
->failed_disk_num
= -1;
9639 num_data_stripes
= imsm_dev_size(dev
) / 2;
9640 num_data_stripes
/= map
->blocks_per_strip
;
9641 set_num_data_stripes(map
, num_data_stripes
);
9644 if (u
->direction
== R0_TO_R10
) {
9646 unsigned long long num_data_stripes
;
9648 /* update slots in current disk list */
9649 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9653 /* create new *missing* disks */
9654 for (i
= 0; i
< map
->num_members
; i
++) {
9655 space
= *space_list
;
9658 *space_list
= *space
;
9660 memcpy(du
, super
->disks
, sizeof(*du
));
9664 du
->index
= (i
* 2) + 1;
9665 sprintf((char *)du
->disk
.serial
,
9666 " MISSING_%d", du
->index
);
9667 sprintf((char *)du
->serial
,
9668 "MISSING_%d", du
->index
);
9669 du
->next
= super
->missing
;
9670 super
->missing
= du
;
9672 /* create new dev and map */
9673 space
= *space_list
;
9676 *space_list
= *space
;
9677 dev_new
= (void *)space
;
9678 memcpy(dev_new
, dev
, sizeof(*dev
));
9679 /* update new map */
9680 map
= get_imsm_map(dev_new
, MAP_0
);
9681 map
->num_members
= map
->num_members
* 2;
9682 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9683 map
->num_domains
= 2;
9684 map
->raid_level
= 1;
9685 num_data_stripes
= imsm_dev_size(dev
) / 2;
9686 num_data_stripes
/= map
->blocks_per_strip
;
9687 num_data_stripes
/= map
->num_domains
;
9688 set_num_data_stripes(map
, num_data_stripes
);
9690 /* replace dev<->dev_new */
9693 /* update disk order table */
9694 for (du
= super
->disks
; du
; du
= du
->next
)
9696 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9697 for (du
= super
->missing
; du
; du
= du
->next
)
9698 if (du
->index
>= 0) {
9699 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9700 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9706 static void imsm_process_update(struct supertype
*st
,
9707 struct metadata_update
*update
)
9710 * crack open the metadata_update envelope to find the update record
9711 * update can be one of:
9712 * update_reshape_container_disks - all the arrays in the container
9713 * are being reshaped to have more devices. We need to mark
9714 * the arrays for general migration and convert selected spares
9715 * into active devices.
9716 * update_activate_spare - a spare device has replaced a failed
9717 * device in an array, update the disk_ord_tbl. If this disk is
9718 * present in all member arrays then also clear the SPARE_DISK
9720 * update_create_array
9722 * update_rename_array
9723 * update_add_remove_disk
9725 struct intel_super
*super
= st
->sb
;
9726 struct imsm_super
*mpb
;
9727 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9729 /* update requires a larger buf but the allocation failed */
9730 if (super
->next_len
&& !super
->next_buf
) {
9731 super
->next_len
= 0;
9735 if (super
->next_buf
) {
9736 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9738 super
->len
= super
->next_len
;
9739 super
->buf
= super
->next_buf
;
9741 super
->next_len
= 0;
9742 super
->next_buf
= NULL
;
9745 mpb
= super
->anchor
;
9748 case update_general_migration_checkpoint
: {
9749 struct intel_dev
*id
;
9750 struct imsm_update_general_migration_checkpoint
*u
=
9751 (void *)update
->buf
;
9753 dprintf("called for update_general_migration_checkpoint\n");
9755 /* find device under general migration */
9756 for (id
= super
->devlist
; id
; id
= id
->next
) {
9757 if (is_gen_migration(id
->dev
)) {
9758 id
->dev
->vol
.curr_migr_unit
=
9759 __cpu_to_le32(u
->curr_migr_unit
);
9760 super
->updates_pending
++;
9765 case update_takeover
: {
9766 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9767 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9768 imsm_update_version_info(super
);
9769 super
->updates_pending
++;
9774 case update_reshape_container_disks
: {
9775 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9776 if (apply_reshape_container_disks_update(
9777 u
, super
, &update
->space_list
))
9778 super
->updates_pending
++;
9781 case update_reshape_migration
: {
9782 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9783 if (apply_reshape_migration_update(
9784 u
, super
, &update
->space_list
))
9785 super
->updates_pending
++;
9788 case update_size_change
: {
9789 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9790 if (apply_size_change_update(u
, super
))
9791 super
->updates_pending
++;
9794 case update_activate_spare
: {
9795 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9796 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9797 super
->updates_pending
++;
9800 case update_create_array
: {
9801 /* someone wants to create a new array, we need to be aware of
9802 * a few races/collisions:
9803 * 1/ 'Create' called by two separate instances of mdadm
9804 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9805 * devices that have since been assimilated via
9807 * In the event this update can not be carried out mdadm will
9808 * (FIX ME) notice that its update did not take hold.
9810 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9811 struct intel_dev
*dv
;
9812 struct imsm_dev
*dev
;
9813 struct imsm_map
*map
, *new_map
;
9814 unsigned long long start
, end
;
9815 unsigned long long new_start
, new_end
;
9817 struct disk_info
*inf
;
9820 /* handle racing creates: first come first serve */
9821 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9822 dprintf("subarray %d already defined\n", u
->dev_idx
);
9826 /* check update is next in sequence */
9827 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9828 dprintf("can not create array %d expected index %d\n",
9829 u
->dev_idx
, mpb
->num_raid_devs
);
9833 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9834 new_start
= pba_of_lba0(new_map
);
9835 new_end
= new_start
+ per_dev_array_size(new_map
);
9836 inf
= get_disk_info(u
);
9838 /* handle activate_spare versus create race:
9839 * check to make sure that overlapping arrays do not include
9842 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9843 dev
= get_imsm_dev(super
, i
);
9844 map
= get_imsm_map(dev
, MAP_0
);
9845 start
= pba_of_lba0(map
);
9846 end
= start
+ per_dev_array_size(map
);
9847 if ((new_start
>= start
&& new_start
<= end
) ||
9848 (start
>= new_start
&& start
<= new_end
))
9853 if (disks_overlap(super
, i
, u
)) {
9854 dprintf("arrays overlap\n");
9859 /* check that prepare update was successful */
9860 if (!update
->space
) {
9861 dprintf("prepare update failed\n");
9865 /* check that all disks are still active before committing
9866 * changes. FIXME: could we instead handle this by creating a
9867 * degraded array? That's probably not what the user expects,
9868 * so better to drop this update on the floor.
9870 for (i
= 0; i
< new_map
->num_members
; i
++) {
9871 dl
= serial_to_dl(inf
[i
].serial
, super
);
9873 dprintf("disk disappeared\n");
9878 super
->updates_pending
++;
9880 /* convert spares to members and fixup ord_tbl */
9881 for (i
= 0; i
< new_map
->num_members
; i
++) {
9882 dl
= serial_to_dl(inf
[i
].serial
, super
);
9883 if (dl
->index
== -1) {
9884 dl
->index
= mpb
->num_disks
;
9886 dl
->disk
.status
|= CONFIGURED_DISK
;
9887 dl
->disk
.status
&= ~SPARE_DISK
;
9889 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9894 update
->space
= NULL
;
9895 imsm_copy_dev(dev
, &u
->dev
);
9896 dv
->index
= u
->dev_idx
;
9897 dv
->next
= super
->devlist
;
9898 super
->devlist
= dv
;
9899 mpb
->num_raid_devs
++;
9901 imsm_update_version_info(super
);
9904 /* mdmon knows how to release update->space, but not
9905 * ((struct intel_dev *) update->space)->dev
9907 if (update
->space
) {
9913 case update_kill_array
: {
9914 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9915 int victim
= u
->dev_idx
;
9916 struct active_array
*a
;
9917 struct intel_dev
**dp
;
9918 struct imsm_dev
*dev
;
9920 /* sanity check that we are not affecting the uuid of
9921 * active arrays, or deleting an active array
9923 * FIXME when immutable ids are available, but note that
9924 * we'll also need to fixup the invalidated/active
9925 * subarray indexes in mdstat
9927 for (a
= st
->arrays
; a
; a
= a
->next
)
9928 if (a
->info
.container_member
>= victim
)
9930 /* by definition if mdmon is running at least one array
9931 * is active in the container, so checking
9932 * mpb->num_raid_devs is just extra paranoia
9934 dev
= get_imsm_dev(super
, victim
);
9935 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9936 dprintf("failed to delete subarray-%d\n", victim
);
9940 for (dp
= &super
->devlist
; *dp
;)
9941 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9944 if ((*dp
)->index
> (unsigned)victim
)
9948 mpb
->num_raid_devs
--;
9949 super
->updates_pending
++;
9952 case update_rename_array
: {
9953 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9954 char name
[MAX_RAID_SERIAL_LEN
+1];
9955 int target
= u
->dev_idx
;
9956 struct active_array
*a
;
9957 struct imsm_dev
*dev
;
9959 /* sanity check that we are not affecting the uuid of
9962 memset(name
, 0, sizeof(name
));
9963 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9964 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9965 for (a
= st
->arrays
; a
; a
= a
->next
)
9966 if (a
->info
.container_member
== target
)
9968 dev
= get_imsm_dev(super
, u
->dev_idx
);
9969 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9970 dprintf("failed to rename subarray-%d\n", target
);
9974 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
9975 super
->updates_pending
++;
9978 case update_add_remove_disk
: {
9979 /* we may be able to repair some arrays if disks are
9980 * being added, check the status of add_remove_disk
9981 * if discs has been added.
9983 if (add_remove_disk_update(super
)) {
9984 struct active_array
*a
;
9986 super
->updates_pending
++;
9987 for (a
= st
->arrays
; a
; a
= a
->next
)
9988 a
->check_degraded
= 1;
9992 case update_prealloc_badblocks_mem
:
9994 case update_rwh_policy
: {
9995 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
9996 int target
= u
->dev_idx
;
9997 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
9999 dprintf("could not find subarray-%d\n", target
);
10003 if (dev
->rwh_policy
!= u
->new_policy
) {
10004 dev
->rwh_policy
= u
->new_policy
;
10005 super
->updates_pending
++;
10010 pr_err("error: unsupported process update type:(type: %d)\n", type
);
10014 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
10016 static int imsm_prepare_update(struct supertype
*st
,
10017 struct metadata_update
*update
)
10020 * Allocate space to hold new disk entries, raid-device entries or a new
10021 * mpb if necessary. The manager synchronously waits for updates to
10022 * complete in the monitor, so new mpb buffers allocated here can be
10023 * integrated by the monitor thread without worrying about live pointers
10024 * in the manager thread.
10026 enum imsm_update_type type
;
10027 struct intel_super
*super
= st
->sb
;
10028 unsigned int sector_size
= super
->sector_size
;
10029 struct imsm_super
*mpb
= super
->anchor
;
10033 if (update
->len
< (int)sizeof(type
))
10036 type
= *(enum imsm_update_type
*) update
->buf
;
10039 case update_general_migration_checkpoint
:
10040 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
10042 dprintf("called for update_general_migration_checkpoint\n");
10044 case update_takeover
: {
10045 struct imsm_update_takeover
*u
= (void *)update
->buf
;
10046 if (update
->len
< (int)sizeof(*u
))
10048 if (u
->direction
== R0_TO_R10
) {
10049 void **tail
= (void **)&update
->space_list
;
10050 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
10051 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10052 int num_members
= map
->num_members
;
10055 /* allocate memory for added disks */
10056 for (i
= 0; i
< num_members
; i
++) {
10057 size
= sizeof(struct dl
);
10058 space
= xmalloc(size
);
10063 /* allocate memory for new device */
10064 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
10065 (num_members
* sizeof(__u32
));
10066 space
= xmalloc(size
);
10070 len
= disks_to_mpb_size(num_members
* 2);
10075 case update_reshape_container_disks
: {
10076 /* Every raid device in the container is about to
10077 * gain some more devices, and we will enter a
10079 * So each 'imsm_map' will be bigger, and the imsm_vol
10080 * will now hold 2 of them.
10081 * Thus we need new 'struct imsm_dev' allocations sized
10082 * as sizeof_imsm_dev but with more devices in both maps.
10084 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10085 struct intel_dev
*dl
;
10086 void **space_tail
= (void**)&update
->space_list
;
10088 if (update
->len
< (int)sizeof(*u
))
10091 dprintf("for update_reshape\n");
10093 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10094 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10096 if (u
->new_raid_disks
> u
->old_raid_disks
)
10097 size
+= sizeof(__u32
)*2*
10098 (u
->new_raid_disks
- u
->old_raid_disks
);
10102 *space_tail
= NULL
;
10105 len
= disks_to_mpb_size(u
->new_raid_disks
);
10106 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10109 case update_reshape_migration
: {
10110 /* for migration level 0->5 we need to add disks
10111 * so the same as for container operation we will copy
10112 * device to the bigger location.
10113 * in memory prepared device and new disk area are prepared
10114 * for usage in process update
10116 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10117 struct intel_dev
*id
;
10118 void **space_tail
= (void **)&update
->space_list
;
10121 int current_level
= -1;
10123 if (update
->len
< (int)sizeof(*u
))
10126 dprintf("for update_reshape\n");
10128 /* add space for bigger array in update
10130 for (id
= super
->devlist
; id
; id
= id
->next
) {
10131 if (id
->index
== (unsigned)u
->subdev
) {
10132 size
= sizeof_imsm_dev(id
->dev
, 1);
10133 if (u
->new_raid_disks
> u
->old_raid_disks
)
10134 size
+= sizeof(__u32
)*2*
10135 (u
->new_raid_disks
- u
->old_raid_disks
);
10139 *space_tail
= NULL
;
10143 if (update
->space_list
== NULL
)
10146 /* add space for disk in update
10148 size
= sizeof(struct dl
);
10152 *space_tail
= NULL
;
10154 /* add spare device to update
10156 for (id
= super
->devlist
; id
; id
= id
->next
)
10157 if (id
->index
== (unsigned)u
->subdev
) {
10158 struct imsm_dev
*dev
;
10159 struct imsm_map
*map
;
10161 dev
= get_imsm_dev(super
, u
->subdev
);
10162 map
= get_imsm_map(dev
, MAP_0
);
10163 current_level
= map
->raid_level
;
10166 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10167 struct mdinfo
*spares
;
10169 spares
= get_spares_for_grow(st
);
10172 struct mdinfo
*dev
;
10174 dev
= spares
->devs
;
10177 makedev(dev
->disk
.major
,
10179 dl
= get_disk_super(super
,
10182 dl
->index
= u
->old_raid_disks
;
10185 sysfs_free(spares
);
10188 len
= disks_to_mpb_size(u
->new_raid_disks
);
10189 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10192 case update_size_change
: {
10193 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10197 case update_activate_spare
: {
10198 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10202 case update_create_array
: {
10203 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10204 struct intel_dev
*dv
;
10205 struct imsm_dev
*dev
= &u
->dev
;
10206 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10208 struct disk_info
*inf
;
10212 if (update
->len
< (int)sizeof(*u
))
10215 inf
= get_disk_info(u
);
10216 len
= sizeof_imsm_dev(dev
, 1);
10217 /* allocate a new super->devlist entry */
10218 dv
= xmalloc(sizeof(*dv
));
10219 dv
->dev
= xmalloc(len
);
10220 update
->space
= dv
;
10222 /* count how many spares will be converted to members */
10223 for (i
= 0; i
< map
->num_members
; i
++) {
10224 dl
= serial_to_dl(inf
[i
].serial
, super
);
10226 /* hmm maybe it failed?, nothing we can do about
10231 if (count_memberships(dl
, super
) == 0)
10234 len
+= activate
* sizeof(struct imsm_disk
);
10237 case update_kill_array
: {
10238 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10242 case update_rename_array
: {
10243 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10247 case update_add_remove_disk
:
10248 /* no update->len needed */
10250 case update_prealloc_badblocks_mem
:
10251 super
->extra_space
+= sizeof(struct bbm_log
) -
10252 get_imsm_bbm_log_size(super
->bbm_log
);
10254 case update_rwh_policy
: {
10255 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10263 /* check if we need a larger metadata buffer */
10264 if (super
->next_buf
)
10265 buf_len
= super
->next_len
;
10267 buf_len
= super
->len
;
10269 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10270 /* ok we need a larger buf than what is currently allocated
10271 * if this allocation fails process_update will notice that
10272 * ->next_len is set and ->next_buf is NULL
10274 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10275 super
->extra_space
+ len
, sector_size
);
10276 if (super
->next_buf
)
10277 free(super
->next_buf
);
10279 super
->next_len
= buf_len
;
10280 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10281 memset(super
->next_buf
, 0, buf_len
);
10283 super
->next_buf
= NULL
;
10288 /* must be called while manager is quiesced */
10289 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10291 struct imsm_super
*mpb
= super
->anchor
;
10293 struct imsm_dev
*dev
;
10294 struct imsm_map
*map
;
10295 unsigned int i
, j
, num_members
;
10296 __u32 ord
, ord_map0
;
10297 struct bbm_log
*log
= super
->bbm_log
;
10299 dprintf("deleting device[%d] from imsm_super\n", index
);
10301 /* shift all indexes down one */
10302 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10303 if (iter
->index
> (int)index
)
10305 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10306 if (iter
->index
> (int)index
)
10309 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10310 dev
= get_imsm_dev(super
, i
);
10311 map
= get_imsm_map(dev
, MAP_0
);
10312 num_members
= map
->num_members
;
10313 for (j
= 0; j
< num_members
; j
++) {
10314 /* update ord entries being careful not to propagate
10315 * ord-flags to the first map
10317 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10318 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10320 if (ord_to_idx(ord
) <= index
)
10323 map
= get_imsm_map(dev
, MAP_0
);
10324 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10325 map
= get_imsm_map(dev
, MAP_1
);
10327 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10331 for (i
= 0; i
< log
->entry_count
; i
++) {
10332 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10334 if (entry
->disk_ordinal
<= index
)
10336 entry
->disk_ordinal
--;
10340 super
->updates_pending
++;
10342 struct dl
*dl
= *dlp
;
10344 *dlp
= (*dlp
)->next
;
10345 __free_imsm_disk(dl
);
10349 static void close_targets(int *targets
, int new_disks
)
10356 for (i
= 0; i
< new_disks
; i
++) {
10357 if (targets
[i
] >= 0) {
10364 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10365 struct intel_super
*super
,
10366 struct imsm_dev
*dev
)
10372 struct imsm_map
*map
;
10375 ret_val
= raid_disks
/2;
10376 /* check map if all disks pairs not failed
10379 map
= get_imsm_map(dev
, MAP_0
);
10380 for (i
= 0; i
< ret_val
; i
++) {
10381 int degradation
= 0;
10382 if (get_imsm_disk(super
, i
) == NULL
)
10384 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10386 if (degradation
== 2)
10389 map
= get_imsm_map(dev
, MAP_1
);
10390 /* if there is no second map
10391 * result can be returned
10395 /* check degradation in second map
10397 for (i
= 0; i
< ret_val
; i
++) {
10398 int degradation
= 0;
10399 if (get_imsm_disk(super
, i
) == NULL
)
10401 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10403 if (degradation
== 2)
10417 /*******************************************************************************
10418 * Function: open_backup_targets
10419 * Description: Function opens file descriptors for all devices given in
10422 * info : general array info
10423 * raid_disks : number of disks
10424 * raid_fds : table of device's file descriptors
10425 * super : intel super for raid10 degradation check
10426 * dev : intel device for raid10 degradation check
10430 ******************************************************************************/
10431 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10432 struct intel_super
*super
, struct imsm_dev
*dev
)
10438 for (i
= 0; i
< raid_disks
; i
++)
10441 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10444 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10445 dprintf("disk is faulty!!\n");
10449 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10452 dn
= map_dev(sd
->disk
.major
,
10453 sd
->disk
.minor
, 1);
10454 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10455 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10456 pr_err("cannot open component\n");
10461 /* check if maximum array degradation level is not exceeded
10463 if ((raid_disks
- opened
) >
10464 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10466 pr_err("Not enough disks can be opened.\n");
10467 close_targets(raid_fds
, raid_disks
);
10473 /*******************************************************************************
10474 * Function: validate_container_imsm
10475 * Description: This routine validates container after assemble,
10476 * eg. if devices in container are under the same controller.
10479 * info : linked list with info about devices used in array
10483 ******************************************************************************/
10484 int validate_container_imsm(struct mdinfo
*info
)
10486 if (check_env("IMSM_NO_PLATFORM"))
10489 struct sys_dev
*idev
;
10490 struct sys_dev
*hba
= NULL
;
10491 struct sys_dev
*intel_devices
= find_intel_devices();
10492 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10493 info
->disk
.minor
));
10495 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10496 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10505 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10506 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10510 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10511 struct mdinfo
*dev
;
10513 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10514 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10516 struct sys_dev
*hba2
= NULL
;
10517 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10518 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10526 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10527 get_orom_by_device_id(hba2
->dev_id
);
10529 if (hba2
&& hba
->type
!= hba2
->type
) {
10530 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10531 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10535 if (orom
!= orom2
) {
10536 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10537 " This operation is not supported and can lead to data loss.\n");
10542 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10543 " This operation is not supported and can lead to data loss.\n");
10551 /*******************************************************************************
10552 * Function: imsm_record_badblock
10553 * Description: This routine stores new bad block record in BBM log
10556 * a : array containing a bad block
10557 * slot : disk number containing a bad block
10558 * sector : bad block sector
10559 * length : bad block sectors range
10563 ******************************************************************************/
10564 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10565 unsigned long long sector
, int length
)
10567 struct intel_super
*super
= a
->container
->sb
;
10571 ord
= imsm_disk_slot_to_ord(a
, slot
);
10575 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10578 super
->updates_pending
++;
10582 /*******************************************************************************
10583 * Function: imsm_clear_badblock
10584 * Description: This routine clears bad block record from BBM log
10587 * a : array containing a bad block
10588 * slot : disk number containing a bad block
10589 * sector : bad block sector
10590 * length : bad block sectors range
10594 ******************************************************************************/
10595 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10596 unsigned long long sector
, int length
)
10598 struct intel_super
*super
= a
->container
->sb
;
10602 ord
= imsm_disk_slot_to_ord(a
, slot
);
10606 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10608 super
->updates_pending
++;
10612 /*******************************************************************************
10613 * Function: imsm_get_badblocks
10614 * Description: This routine get list of bad blocks for an array
10618 * slot : disk number
10620 * bb : structure containing bad blocks
10622 ******************************************************************************/
10623 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10625 int inst
= a
->info
.container_member
;
10626 struct intel_super
*super
= a
->container
->sb
;
10627 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10628 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10631 ord
= imsm_disk_slot_to_ord(a
, slot
);
10635 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10636 per_dev_array_size(map
), &super
->bb
);
10640 /*******************************************************************************
10641 * Function: examine_badblocks_imsm
10642 * Description: Prints list of bad blocks on a disk to the standard output
10645 * st : metadata handler
10646 * fd : open file descriptor for device
10647 * devname : device name
10651 ******************************************************************************/
10652 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10654 struct intel_super
*super
= st
->sb
;
10655 struct bbm_log
*log
= super
->bbm_log
;
10656 struct dl
*d
= NULL
;
10659 for (d
= super
->disks
; d
; d
= d
->next
) {
10660 if (strcmp(d
->devname
, devname
) == 0)
10664 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10665 pr_err("%s doesn't appear to be part of a raid array\n",
10672 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10674 for (i
= 0; i
< log
->entry_count
; i
++) {
10675 if (entry
[i
].disk_ordinal
== d
->index
) {
10676 unsigned long long sector
= __le48_to_cpu(
10677 &entry
[i
].defective_block_start
);
10678 int cnt
= entry
[i
].marked_count
+ 1;
10681 printf("Bad-blocks on %s:\n", devname
);
10685 printf("%20llu for %d sectors\n", sector
, cnt
);
10691 printf("No bad-blocks list configured on %s\n", devname
);
10695 /*******************************************************************************
10696 * Function: init_migr_record_imsm
10697 * Description: Function inits imsm migration record
10699 * super : imsm internal array info
10700 * dev : device under migration
10701 * info : general array info to find the smallest device
10704 ******************************************************************************/
10705 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10706 struct mdinfo
*info
)
10708 struct intel_super
*super
= st
->sb
;
10709 struct migr_record
*migr_rec
= super
->migr_rec
;
10710 int new_data_disks
;
10711 unsigned long long dsize
, dev_sectors
;
10712 long long unsigned min_dev_sectors
= -1LLU;
10716 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10717 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10718 unsigned long long num_migr_units
;
10719 unsigned long long array_blocks
;
10721 memset(migr_rec
, 0, sizeof(struct migr_record
));
10722 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10724 /* only ascending reshape supported now */
10725 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10727 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10728 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10729 migr_rec
->dest_depth_per_unit
*=
10730 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10731 new_data_disks
= imsm_num_data_members(map_dest
);
10732 migr_rec
->blocks_per_unit
=
10733 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10734 migr_rec
->dest_depth_per_unit
=
10735 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10736 array_blocks
= info
->component_size
* new_data_disks
;
10738 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10740 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10742 set_num_migr_units(migr_rec
, num_migr_units
);
10744 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10745 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10747 /* Find the smallest dev */
10748 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10749 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10750 fd
= dev_open(nm
, O_RDONLY
);
10753 get_dev_size(fd
, NULL
, &dsize
);
10754 dev_sectors
= dsize
/ 512;
10755 if (dev_sectors
< min_dev_sectors
)
10756 min_dev_sectors
= dev_sectors
;
10759 set_migr_chkp_area_pba(migr_rec
, min_dev_sectors
-
10760 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10762 write_imsm_migr_rec(st
);
10767 /*******************************************************************************
10768 * Function: save_backup_imsm
10769 * Description: Function saves critical data stripes to Migration Copy Area
10770 * and updates the current migration unit status.
10771 * Use restore_stripes() to form a destination stripe,
10772 * and to write it to the Copy Area.
10774 * st : supertype information
10775 * dev : imsm device that backup is saved for
10776 * info : general array info
10777 * buf : input buffer
10778 * length : length of data to backup (blocks_per_unit)
10782 ******************************************************************************/
10783 int save_backup_imsm(struct supertype
*st
,
10784 struct imsm_dev
*dev
,
10785 struct mdinfo
*info
,
10790 struct intel_super
*super
= st
->sb
;
10791 unsigned long long *target_offsets
;
10794 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10795 int new_disks
= map_dest
->num_members
;
10796 int dest_layout
= 0;
10798 unsigned long long start
;
10799 int data_disks
= imsm_num_data_members(map_dest
);
10801 targets
= xmalloc(new_disks
* sizeof(int));
10803 for (i
= 0; i
< new_disks
; i
++)
10806 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10808 start
= info
->reshape_progress
* 512;
10809 for (i
= 0; i
< new_disks
; i
++) {
10810 target_offsets
[i
] = migr_chkp_area_pba(super
->migr_rec
) * 512;
10811 /* move back copy area adderss, it will be moved forward
10812 * in restore_stripes() using start input variable
10814 target_offsets
[i
] -= start
/data_disks
;
10817 if (open_backup_targets(info
, new_disks
, targets
,
10821 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10822 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10824 if (restore_stripes(targets
, /* list of dest devices */
10825 target_offsets
, /* migration record offsets */
10828 map_dest
->raid_level
,
10830 -1, /* source backup file descriptor */
10831 0, /* input buf offset
10832 * always 0 buf is already offseted */
10836 pr_err("Error restoring stripes\n");
10844 close_targets(targets
, new_disks
);
10847 free(target_offsets
);
10852 /*******************************************************************************
10853 * Function: save_checkpoint_imsm
10854 * Description: Function called for current unit status update
10855 * in the migration record. It writes it to disk.
10857 * super : imsm internal array info
10858 * info : general array info
10862 * 2: failure, means no valid migration record
10863 * / no general migration in progress /
10864 ******************************************************************************/
10865 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10867 struct intel_super
*super
= st
->sb
;
10868 unsigned long long blocks_per_unit
;
10869 unsigned long long curr_migr_unit
;
10871 if (load_imsm_migr_rec(super
, info
) != 0) {
10872 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10876 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10877 if (blocks_per_unit
== 0) {
10878 dprintf("imsm: no migration in progress.\n");
10881 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10882 /* check if array is alligned to copy area
10883 * if it is not alligned, add one to current migration unit value
10884 * this can happend on array reshape finish only
10886 if (info
->reshape_progress
% blocks_per_unit
)
10889 set_current_migr_unit(super
->migr_rec
, curr_migr_unit
);
10890 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10891 set_migr_dest_1st_member_lba(super
->migr_rec
,
10892 super
->migr_rec
->dest_depth_per_unit
* curr_migr_unit
);
10894 if (write_imsm_migr_rec(st
) < 0) {
10895 dprintf("imsm: Cannot write migration record outside backup area\n");
10902 /*******************************************************************************
10903 * Function: recover_backup_imsm
10904 * Description: Function recovers critical data from the Migration Copy Area
10905 * while assembling an array.
10907 * super : imsm internal array info
10908 * info : general array info
10910 * 0 : success (or there is no data to recover)
10912 ******************************************************************************/
10913 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10915 struct intel_super
*super
= st
->sb
;
10916 struct migr_record
*migr_rec
= super
->migr_rec
;
10917 struct imsm_map
*map_dest
;
10918 struct intel_dev
*id
= NULL
;
10919 unsigned long long read_offset
;
10920 unsigned long long write_offset
;
10922 int *targets
= NULL
;
10923 int new_disks
, i
, err
;
10926 unsigned int sector_size
= super
->sector_size
;
10927 unsigned long curr_migr_unit
= current_migr_unit(migr_rec
);
10928 unsigned long num_migr_units
= get_num_migr_units(migr_rec
);
10930 int skipped_disks
= 0;
10932 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10936 /* recover data only during assemblation */
10937 if (strncmp(buffer
, "inactive", 8) != 0)
10939 /* no data to recover */
10940 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10942 if (curr_migr_unit
>= num_migr_units
)
10945 /* find device during reshape */
10946 for (id
= super
->devlist
; id
; id
= id
->next
)
10947 if (is_gen_migration(id
->dev
))
10952 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10953 new_disks
= map_dest
->num_members
;
10955 read_offset
= migr_chkp_area_pba(migr_rec
) * 512;
10957 write_offset
= (migr_dest_1st_member_lba(migr_rec
) +
10958 pba_of_lba0(map_dest
)) * 512;
10960 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10961 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10963 targets
= xcalloc(new_disks
, sizeof(int));
10965 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10966 pr_err("Cannot open some devices belonging to array.\n");
10970 for (i
= 0; i
< new_disks
; i
++) {
10971 if (targets
[i
] < 0) {
10975 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
10976 pr_err("Cannot seek to block: %s\n",
10981 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
10982 pr_err("Cannot read copy area block: %s\n",
10987 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
10988 pr_err("Cannot seek to block: %s\n",
10993 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
10994 pr_err("Cannot restore block: %s\n",
11001 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
11005 pr_err("Cannot restore data from backup. Too many failed disks\n");
11009 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
11010 /* ignore error == 2, this can mean end of reshape here
11012 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
11018 for (i
= 0; i
< new_disks
; i
++)
11027 static char disk_by_path
[] = "/dev/disk/by-path/";
11029 static const char *imsm_get_disk_controller_domain(const char *path
)
11031 char disk_path
[PATH_MAX
];
11035 strcpy(disk_path
, disk_by_path
);
11036 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
11037 if (stat(disk_path
, &st
) == 0) {
11038 struct sys_dev
* hba
;
11041 path
= devt_to_devpath(st
.st_rdev
);
11044 hba
= find_disk_attached_hba(-1, path
);
11045 if (hba
&& hba
->type
== SYS_DEV_SAS
)
11047 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
11049 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
11051 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
11055 dprintf("path: %s hba: %s attached: %s\n",
11056 path
, (hba
) ? hba
->path
: "NULL", drv
);
11062 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
11064 static char devnm
[32];
11065 char subdev_name
[20];
11066 struct mdstat_ent
*mdstat
;
11068 sprintf(subdev_name
, "%d", subdev
);
11069 mdstat
= mdstat_by_subdev(subdev_name
, container
);
11073 strcpy(devnm
, mdstat
->devnm
);
11074 free_mdstat(mdstat
);
11078 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11079 struct geo_params
*geo
,
11080 int *old_raid_disks
,
11083 /* currently we only support increasing the number of devices
11084 * for a container. This increases the number of device for each
11085 * member array. They must all be RAID0 or RAID5.
11088 struct mdinfo
*info
, *member
;
11089 int devices_that_can_grow
= 0;
11091 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11093 if (geo
->size
> 0 ||
11094 geo
->level
!= UnSet
||
11095 geo
->layout
!= UnSet
||
11096 geo
->chunksize
!= 0 ||
11097 geo
->raid_disks
== UnSet
) {
11098 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11102 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11103 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11107 info
= container_content_imsm(st
, NULL
);
11108 for (member
= info
; member
; member
= member
->next
) {
11111 dprintf("imsm: checking device_num: %i\n",
11112 member
->container_member
);
11114 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11115 /* we work on container for Online Capacity Expansion
11116 * only so raid_disks has to grow
11118 dprintf("imsm: for container operation raid disks increase is required\n");
11122 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11123 /* we cannot use this container with other raid level
11125 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11126 info
->array
.level
);
11129 /* check for platform support
11130 * for this raid level configuration
11132 struct intel_super
*super
= st
->sb
;
11133 if (!is_raid_level_supported(super
->orom
,
11134 member
->array
.level
,
11135 geo
->raid_disks
)) {
11136 dprintf("platform does not support raid%d with %d disk%s\n",
11139 geo
->raid_disks
> 1 ? "s" : "");
11142 /* check if component size is aligned to chunk size
11144 if (info
->component_size
%
11145 (info
->array
.chunk_size
/512)) {
11146 dprintf("Component size is not aligned to chunk size\n");
11151 if (*old_raid_disks
&&
11152 info
->array
.raid_disks
!= *old_raid_disks
)
11154 *old_raid_disks
= info
->array
.raid_disks
;
11156 /* All raid5 and raid0 volumes in container
11157 * have to be ready for Online Capacity Expansion
11158 * so they need to be assembled. We have already
11159 * checked that no recovery etc is happening.
11161 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11162 st
->container_devnm
);
11163 if (result
== NULL
) {
11164 dprintf("imsm: cannot find array\n");
11167 devices_that_can_grow
++;
11170 if (!member
&& devices_that_can_grow
)
11174 dprintf("Container operation allowed\n");
11176 dprintf("Error: %i\n", ret_val
);
11181 /* Function: get_spares_for_grow
11182 * Description: Allocates memory and creates list of spare devices
11183 * avaliable in container. Checks if spare drive size is acceptable.
11184 * Parameters: Pointer to the supertype structure
11185 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11188 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11190 struct spare_criteria sc
;
11192 get_spare_criteria_imsm(st
, &sc
);
11193 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11196 /******************************************************************************
11197 * function: imsm_create_metadata_update_for_reshape
11198 * Function creates update for whole IMSM container.
11200 ******************************************************************************/
11201 static int imsm_create_metadata_update_for_reshape(
11202 struct supertype
*st
,
11203 struct geo_params
*geo
,
11204 int old_raid_disks
,
11205 struct imsm_update_reshape
**updatep
)
11207 struct intel_super
*super
= st
->sb
;
11208 struct imsm_super
*mpb
= super
->anchor
;
11209 int update_memory_size
;
11210 struct imsm_update_reshape
*u
;
11211 struct mdinfo
*spares
;
11214 struct mdinfo
*dev
;
11216 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11218 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11220 /* size of all update data without anchor */
11221 update_memory_size
= sizeof(struct imsm_update_reshape
);
11223 /* now add space for spare disks that we need to add. */
11224 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11226 u
= xcalloc(1, update_memory_size
);
11227 u
->type
= update_reshape_container_disks
;
11228 u
->old_raid_disks
= old_raid_disks
;
11229 u
->new_raid_disks
= geo
->raid_disks
;
11231 /* now get spare disks list
11233 spares
= get_spares_for_grow(st
);
11235 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11236 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11241 /* we have got spares
11242 * update disk list in imsm_disk list table in anchor
11244 dprintf("imsm: %i spares are available.\n\n",
11245 spares
->array
.spare_disks
);
11247 dev
= spares
->devs
;
11248 for (i
= 0; i
< delta_disks
; i
++) {
11253 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11255 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11256 dl
->index
= mpb
->num_disks
;
11264 sysfs_free(spares
);
11266 dprintf("imsm: reshape update preparation :");
11267 if (i
== delta_disks
) {
11268 dprintf_cont(" OK\n");
11270 return update_memory_size
;
11273 dprintf_cont(" Error\n");
11278 /******************************************************************************
11279 * function: imsm_create_metadata_update_for_size_change()
11280 * Creates update for IMSM array for array size change.
11282 ******************************************************************************/
11283 static int imsm_create_metadata_update_for_size_change(
11284 struct supertype
*st
,
11285 struct geo_params
*geo
,
11286 struct imsm_update_size_change
**updatep
)
11288 struct intel_super
*super
= st
->sb
;
11289 int update_memory_size
;
11290 struct imsm_update_size_change
*u
;
11292 dprintf("(enter) New size = %llu\n", geo
->size
);
11294 /* size of all update data without anchor */
11295 update_memory_size
= sizeof(struct imsm_update_size_change
);
11297 u
= xcalloc(1, update_memory_size
);
11298 u
->type
= update_size_change
;
11299 u
->subdev
= super
->current_vol
;
11300 u
->new_size
= geo
->size
;
11302 dprintf("imsm: reshape update preparation : OK\n");
11305 return update_memory_size
;
11308 /******************************************************************************
11309 * function: imsm_create_metadata_update_for_migration()
11310 * Creates update for IMSM array.
11312 ******************************************************************************/
11313 static int imsm_create_metadata_update_for_migration(
11314 struct supertype
*st
,
11315 struct geo_params
*geo
,
11316 struct imsm_update_reshape_migration
**updatep
)
11318 struct intel_super
*super
= st
->sb
;
11319 int update_memory_size
;
11320 struct imsm_update_reshape_migration
*u
;
11321 struct imsm_dev
*dev
;
11322 int previous_level
= -1;
11324 dprintf("(enter) New Level = %i\n", geo
->level
);
11326 /* size of all update data without anchor */
11327 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11329 u
= xcalloc(1, update_memory_size
);
11330 u
->type
= update_reshape_migration
;
11331 u
->subdev
= super
->current_vol
;
11332 u
->new_level
= geo
->level
;
11333 u
->new_layout
= geo
->layout
;
11334 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11335 u
->new_disks
[0] = -1;
11336 u
->new_chunksize
= -1;
11338 dev
= get_imsm_dev(super
, u
->subdev
);
11340 struct imsm_map
*map
;
11342 map
= get_imsm_map(dev
, MAP_0
);
11344 int current_chunk_size
=
11345 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11347 if (geo
->chunksize
!= current_chunk_size
) {
11348 u
->new_chunksize
= geo
->chunksize
/ 1024;
11349 dprintf("imsm: chunk size change from %i to %i\n",
11350 current_chunk_size
, u
->new_chunksize
);
11352 previous_level
= map
->raid_level
;
11355 if (geo
->level
== 5 && previous_level
== 0) {
11356 struct mdinfo
*spares
= NULL
;
11358 u
->new_raid_disks
++;
11359 spares
= get_spares_for_grow(st
);
11360 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11362 sysfs_free(spares
);
11363 update_memory_size
= 0;
11364 pr_err("cannot get spare device for requested migration\n");
11367 sysfs_free(spares
);
11369 dprintf("imsm: reshape update preparation : OK\n");
11372 return update_memory_size
;
11375 static void imsm_update_metadata_locally(struct supertype
*st
,
11376 void *buf
, int len
)
11378 struct metadata_update mu
;
11383 mu
.space_list
= NULL
;
11385 if (imsm_prepare_update(st
, &mu
))
11386 imsm_process_update(st
, &mu
);
11388 while (mu
.space_list
) {
11389 void **space
= mu
.space_list
;
11390 mu
.space_list
= *space
;
11395 /***************************************************************************
11396 * Function: imsm_analyze_change
11397 * Description: Function analyze change for single volume
11398 * and validate if transition is supported
11399 * Parameters: Geometry parameters, supertype structure,
11400 * metadata change direction (apply/rollback)
11401 * Returns: Operation type code on success, -1 if fail
11402 ****************************************************************************/
11403 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11404 struct geo_params
*geo
,
11407 struct mdinfo info
;
11409 int check_devs
= 0;
11411 /* number of added/removed disks in operation result */
11412 int devNumChange
= 0;
11413 /* imsm compatible layout value for array geometry verification */
11414 int imsm_layout
= -1;
11416 struct imsm_dev
*dev
;
11417 struct imsm_map
*map
;
11418 struct intel_super
*super
;
11419 unsigned long long current_size
;
11420 unsigned long long free_size
;
11421 unsigned long long max_size
;
11424 getinfo_super_imsm_volume(st
, &info
, NULL
);
11425 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11426 geo
->level
!= UnSet
) {
11427 switch (info
.array
.level
) {
11429 if (geo
->level
== 5) {
11430 change
= CH_MIGRATION
;
11431 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11432 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11434 goto analyse_change_exit
;
11436 imsm_layout
= geo
->layout
;
11438 devNumChange
= 1; /* parity disk added */
11439 } else if (geo
->level
== 10) {
11440 change
= CH_TAKEOVER
;
11442 devNumChange
= 2; /* two mirrors added */
11443 imsm_layout
= 0x102; /* imsm supported layout */
11448 if (geo
->level
== 0) {
11449 change
= CH_TAKEOVER
;
11451 devNumChange
= -(geo
->raid_disks
/2);
11452 imsm_layout
= 0; /* imsm raid0 layout */
11456 if (change
== -1) {
11457 pr_err("Error. Level Migration from %d to %d not supported!\n",
11458 info
.array
.level
, geo
->level
);
11459 goto analyse_change_exit
;
11462 geo
->level
= info
.array
.level
;
11464 if (geo
->layout
!= info
.array
.layout
&&
11465 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11466 change
= CH_MIGRATION
;
11467 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11468 geo
->layout
== 5) {
11469 /* reshape 5 -> 4 */
11470 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11471 geo
->layout
== 0) {
11472 /* reshape 4 -> 5 */
11476 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11477 info
.array
.layout
, geo
->layout
);
11479 goto analyse_change_exit
;
11482 geo
->layout
= info
.array
.layout
;
11483 if (imsm_layout
== -1)
11484 imsm_layout
= info
.array
.layout
;
11487 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11488 geo
->chunksize
!= info
.array
.chunk_size
) {
11489 if (info
.array
.level
== 10) {
11490 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11492 goto analyse_change_exit
;
11493 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11494 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11495 geo
->chunksize
/1024, info
.component_size
/2);
11497 goto analyse_change_exit
;
11499 change
= CH_MIGRATION
;
11501 geo
->chunksize
= info
.array
.chunk_size
;
11504 chunk
= geo
->chunksize
/ 1024;
11507 dev
= get_imsm_dev(super
, super
->current_vol
);
11508 map
= get_imsm_map(dev
, MAP_0
);
11509 data_disks
= imsm_num_data_members(map
);
11510 /* compute current size per disk member
11512 current_size
= info
.custom_array_size
/ data_disks
;
11514 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11515 /* align component size
11517 geo
->size
= imsm_component_size_alignment_check(
11518 get_imsm_raid_level(dev
->vol
.map
),
11519 chunk
* 1024, super
->sector_size
,
11521 if (geo
->size
== 0) {
11522 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11524 goto analyse_change_exit
;
11528 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11529 if (change
!= -1) {
11530 pr_err("Error. Size change should be the only one at a time.\n");
11532 goto analyse_change_exit
;
11534 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11535 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11536 super
->current_vol
, st
->devnm
);
11537 goto analyse_change_exit
;
11539 /* check the maximum available size
11541 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11542 0, chunk
, &free_size
);
11544 /* Cannot find maximum available space
11548 max_size
= free_size
+ current_size
;
11549 /* align component size
11551 max_size
= imsm_component_size_alignment_check(
11552 get_imsm_raid_level(dev
->vol
.map
),
11553 chunk
* 1024, super
->sector_size
,
11556 if (geo
->size
== MAX_SIZE
) {
11557 /* requested size change to the maximum available size
11559 if (max_size
== 0) {
11560 pr_err("Error. Cannot find maximum available space.\n");
11562 goto analyse_change_exit
;
11564 geo
->size
= max_size
;
11567 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11568 /* accept size for rollback only
11571 /* round size due to metadata compatibility
11573 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11574 << SECT_PER_MB_SHIFT
;
11575 dprintf("Prepare update for size change to %llu\n",
11577 if (current_size
>= geo
->size
) {
11578 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11579 current_size
, geo
->size
);
11580 goto analyse_change_exit
;
11582 if (max_size
&& geo
->size
> max_size
) {
11583 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11584 max_size
, geo
->size
);
11585 goto analyse_change_exit
;
11588 geo
->size
*= data_disks
;
11589 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11590 change
= CH_ARRAY_SIZE
;
11592 if (!validate_geometry_imsm(st
,
11595 geo
->raid_disks
+ devNumChange
,
11597 geo
->size
, INVALID_SECTORS
,
11598 0, 0, info
.consistency_policy
, 1))
11602 struct intel_super
*super
= st
->sb
;
11603 struct imsm_super
*mpb
= super
->anchor
;
11605 if (mpb
->num_raid_devs
> 1) {
11606 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11612 analyse_change_exit
:
11613 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11614 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11615 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11621 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11623 struct intel_super
*super
= st
->sb
;
11624 struct imsm_update_takeover
*u
;
11626 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11628 u
->type
= update_takeover
;
11629 u
->subarray
= super
->current_vol
;
11631 /* 10->0 transition */
11632 if (geo
->level
== 0)
11633 u
->direction
= R10_TO_R0
;
11635 /* 0->10 transition */
11636 if (geo
->level
== 10)
11637 u
->direction
= R0_TO_R10
;
11639 /* update metadata locally */
11640 imsm_update_metadata_locally(st
, u
,
11641 sizeof(struct imsm_update_takeover
));
11642 /* and possibly remotely */
11643 if (st
->update_tail
)
11644 append_metadata_update(st
, u
,
11645 sizeof(struct imsm_update_takeover
));
11652 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11654 int layout
, int chunksize
, int raid_disks
,
11655 int delta_disks
, char *backup
, char *dev
,
11656 int direction
, int verbose
)
11659 struct geo_params geo
;
11661 dprintf("(enter)\n");
11663 memset(&geo
, 0, sizeof(struct geo_params
));
11665 geo
.dev_name
= dev
;
11666 strcpy(geo
.devnm
, st
->devnm
);
11669 geo
.layout
= layout
;
11670 geo
.chunksize
= chunksize
;
11671 geo
.raid_disks
= raid_disks
;
11672 if (delta_disks
!= UnSet
)
11673 geo
.raid_disks
+= delta_disks
;
11675 dprintf("for level : %i\n", geo
.level
);
11676 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11678 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11679 /* On container level we can only increase number of devices. */
11680 dprintf("imsm: info: Container operation\n");
11681 int old_raid_disks
= 0;
11683 if (imsm_reshape_is_allowed_on_container(
11684 st
, &geo
, &old_raid_disks
, direction
)) {
11685 struct imsm_update_reshape
*u
= NULL
;
11688 len
= imsm_create_metadata_update_for_reshape(
11689 st
, &geo
, old_raid_disks
, &u
);
11692 dprintf("imsm: Cannot prepare update\n");
11693 goto exit_imsm_reshape_super
;
11697 /* update metadata locally */
11698 imsm_update_metadata_locally(st
, u
, len
);
11699 /* and possibly remotely */
11700 if (st
->update_tail
)
11701 append_metadata_update(st
, u
, len
);
11706 pr_err("(imsm) Operation is not allowed on this container\n");
11709 /* On volume level we support following operations
11710 * - takeover: raid10 -> raid0; raid0 -> raid10
11711 * - chunk size migration
11712 * - migration: raid5 -> raid0; raid0 -> raid5
11714 struct intel_super
*super
= st
->sb
;
11715 struct intel_dev
*dev
= super
->devlist
;
11717 dprintf("imsm: info: Volume operation\n");
11718 /* find requested device */
11721 imsm_find_array_devnm_by_subdev(
11722 dev
->index
, st
->container_devnm
);
11723 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11728 pr_err("Cannot find %s (%s) subarray\n",
11729 geo
.dev_name
, geo
.devnm
);
11730 goto exit_imsm_reshape_super
;
11732 super
->current_vol
= dev
->index
;
11733 change
= imsm_analyze_change(st
, &geo
, direction
);
11736 ret_val
= imsm_takeover(st
, &geo
);
11738 case CH_MIGRATION
: {
11739 struct imsm_update_reshape_migration
*u
= NULL
;
11741 imsm_create_metadata_update_for_migration(
11744 dprintf("imsm: Cannot prepare update\n");
11748 /* update metadata locally */
11749 imsm_update_metadata_locally(st
, u
, len
);
11750 /* and possibly remotely */
11751 if (st
->update_tail
)
11752 append_metadata_update(st
, u
, len
);
11757 case CH_ARRAY_SIZE
: {
11758 struct imsm_update_size_change
*u
= NULL
;
11760 imsm_create_metadata_update_for_size_change(
11763 dprintf("imsm: Cannot prepare update\n");
11767 /* update metadata locally */
11768 imsm_update_metadata_locally(st
, u
, len
);
11769 /* and possibly remotely */
11770 if (st
->update_tail
)
11771 append_metadata_update(st
, u
, len
);
11781 exit_imsm_reshape_super
:
11782 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11786 #define COMPLETED_OK 0
11787 #define COMPLETED_NONE 1
11788 #define COMPLETED_DELAYED 2
11790 static int read_completed(int fd
, unsigned long long *val
)
11795 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11799 ret
= COMPLETED_OK
;
11800 if (strncmp(buf
, "none", 4) == 0) {
11801 ret
= COMPLETED_NONE
;
11802 } else if (strncmp(buf
, "delayed", 7) == 0) {
11803 ret
= COMPLETED_DELAYED
;
11806 *val
= strtoull(buf
, &ep
, 0);
11807 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11813 /*******************************************************************************
11814 * Function: wait_for_reshape_imsm
11815 * Description: Function writes new sync_max value and waits until
11816 * reshape process reach new position
11818 * sra : general array info
11819 * ndata : number of disks in new array's layout
11822 * 1 : there is no reshape in progress,
11824 ******************************************************************************/
11825 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11827 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11829 unsigned long long completed
;
11830 /* to_complete : new sync_max position */
11831 unsigned long long to_complete
= sra
->reshape_progress
;
11832 unsigned long long position_to_set
= to_complete
/ ndata
;
11835 dprintf("cannot open reshape_position\n");
11840 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11842 dprintf("cannot read reshape_position (no reshape in progres)\n");
11851 if (completed
> position_to_set
) {
11852 dprintf("wrong next position to set %llu (%llu)\n",
11853 to_complete
, position_to_set
);
11857 dprintf("Position set: %llu\n", position_to_set
);
11858 if (sysfs_set_num(sra
, NULL
, "sync_max",
11859 position_to_set
) != 0) {
11860 dprintf("cannot set reshape position to %llu\n",
11869 int timeout
= 3000;
11871 sysfs_wait(fd
, &timeout
);
11872 if (sysfs_get_str(sra
, NULL
, "sync_action",
11874 strncmp(action
, "reshape", 7) != 0) {
11875 if (strncmp(action
, "idle", 4) == 0)
11881 rc
= read_completed(fd
, &completed
);
11883 dprintf("cannot read reshape_position (in loop)\n");
11886 } else if (rc
== COMPLETED_NONE
)
11888 } while (completed
< position_to_set
);
11894 /*******************************************************************************
11895 * Function: check_degradation_change
11896 * Description: Check that array hasn't become failed.
11898 * info : for sysfs access
11899 * sources : source disks descriptors
11900 * degraded: previous degradation level
11902 * degradation level
11903 ******************************************************************************/
11904 int check_degradation_change(struct mdinfo
*info
,
11908 unsigned long long new_degraded
;
11911 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11912 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11913 /* check each device to ensure it is still working */
11916 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11917 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11919 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11922 if (sysfs_get_str(info
,
11923 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11924 strstr(sbuf
, "faulty") ||
11925 strstr(sbuf
, "in_sync") == NULL
) {
11926 /* this device is dead */
11927 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11928 if (sd
->disk
.raid_disk
>= 0 &&
11929 sources
[sd
->disk
.raid_disk
] >= 0) {
11931 sd
->disk
.raid_disk
]);
11932 sources
[sd
->disk
.raid_disk
] =
11941 return new_degraded
;
11944 /*******************************************************************************
11945 * Function: imsm_manage_reshape
11946 * Description: Function finds array under reshape and it manages reshape
11947 * process. It creates stripes backups (if required) and sets
11950 * afd : Backup handle (nattive) - not used
11951 * sra : general array info
11952 * reshape : reshape parameters - not used
11953 * st : supertype structure
11954 * blocks : size of critical section [blocks]
11955 * fds : table of source device descriptor
11956 * offsets : start of array (offest per devices)
11958 * destfd : table of destination device descriptor
11959 * destoffsets : table of destination offsets (per device)
11961 * 1 : success, reshape is done
11963 ******************************************************************************/
11964 static int imsm_manage_reshape(
11965 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
11966 struct supertype
*st
, unsigned long backup_blocks
,
11967 int *fds
, unsigned long long *offsets
,
11968 int dests
, int *destfd
, unsigned long long *destoffsets
)
11971 struct intel_super
*super
= st
->sb
;
11972 struct intel_dev
*dv
;
11973 unsigned int sector_size
= super
->sector_size
;
11974 struct imsm_dev
*dev
= NULL
;
11975 struct imsm_map
*map_src
, *map_dest
;
11976 int migr_vol_qan
= 0;
11977 int ndata
, odata
; /* [bytes] */
11978 int chunk
; /* [bytes] */
11979 struct migr_record
*migr_rec
;
11981 unsigned int buf_size
; /* [bytes] */
11982 unsigned long long max_position
; /* array size [bytes] */
11983 unsigned long long next_step
; /* [blocks]/[bytes] */
11984 unsigned long long old_data_stripe_length
;
11985 unsigned long long start_src
; /* [bytes] */
11986 unsigned long long start
; /* [bytes] */
11987 unsigned long long start_buf_shift
; /* [bytes] */
11989 int source_layout
= 0;
11994 if (!fds
|| !offsets
)
11997 /* Find volume during the reshape */
11998 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
11999 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
12000 dv
->dev
->vol
.migr_state
== 1) {
12005 /* Only one volume can migrate at the same time */
12006 if (migr_vol_qan
!= 1) {
12007 pr_err("%s", migr_vol_qan
?
12008 "Number of migrating volumes greater than 1\n" :
12009 "There is no volume during migrationg\n");
12013 map_dest
= get_imsm_map(dev
, MAP_0
);
12014 map_src
= get_imsm_map(dev
, MAP_1
);
12015 if (map_src
== NULL
)
12018 ndata
= imsm_num_data_members(map_dest
);
12019 odata
= imsm_num_data_members(map_src
);
12021 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
12022 old_data_stripe_length
= odata
* chunk
;
12024 migr_rec
= super
->migr_rec
;
12026 /* initialize migration record for start condition */
12027 if (sra
->reshape_progress
== 0)
12028 init_migr_record_imsm(st
, dev
, sra
);
12030 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
12031 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
12034 /* Save checkpoint to update migration record for current
12035 * reshape position (in md). It can be farther than current
12036 * reshape position in metadata.
12038 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12039 /* ignore error == 2, this can mean end of reshape here
12041 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
12046 /* size for data */
12047 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
12048 /* extend buffer size for parity disk */
12049 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
12050 /* add space for stripe alignment */
12051 buf_size
+= old_data_stripe_length
;
12052 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
12053 dprintf("imsm: Cannot allocate checkpoint buffer\n");
12057 max_position
= sra
->component_size
* ndata
;
12058 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
12060 while (current_migr_unit(migr_rec
) <
12061 get_num_migr_units(migr_rec
)) {
12062 /* current reshape position [blocks] */
12063 unsigned long long current_position
=
12064 __le32_to_cpu(migr_rec
->blocks_per_unit
)
12065 * current_migr_unit(migr_rec
);
12066 unsigned long long border
;
12068 /* Check that array hasn't become failed.
12070 degraded
= check_degradation_change(sra
, fds
, degraded
);
12071 if (degraded
> 1) {
12072 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12076 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12078 if ((current_position
+ next_step
) > max_position
)
12079 next_step
= max_position
- current_position
;
12081 start
= current_position
* 512;
12083 /* align reading start to old geometry */
12084 start_buf_shift
= start
% old_data_stripe_length
;
12085 start_src
= start
- start_buf_shift
;
12087 border
= (start_src
/ odata
) - (start
/ ndata
);
12089 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12090 /* save critical stripes to buf
12091 * start - start address of current unit
12092 * to backup [bytes]
12093 * start_src - start address of current unit
12094 * to backup alligned to source array
12097 unsigned long long next_step_filler
;
12098 unsigned long long copy_length
= next_step
* 512;
12100 /* allign copy area length to stripe in old geometry */
12101 next_step_filler
= ((copy_length
+ start_buf_shift
)
12102 % old_data_stripe_length
);
12103 if (next_step_filler
)
12104 next_step_filler
= (old_data_stripe_length
12105 - next_step_filler
);
12106 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12107 start
, start_src
, copy_length
,
12108 start_buf_shift
, next_step_filler
);
12110 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12111 chunk
, map_src
->raid_level
,
12112 source_layout
, 0, NULL
, start_src
,
12114 next_step_filler
+ start_buf_shift
,
12116 dprintf("imsm: Cannot save stripes to buffer\n");
12119 /* Convert data to destination format and store it
12120 * in backup general migration area
12122 if (save_backup_imsm(st
, dev
, sra
,
12123 buf
+ start_buf_shift
, copy_length
)) {
12124 dprintf("imsm: Cannot save stripes to target devices\n");
12127 if (save_checkpoint_imsm(st
, sra
,
12128 UNIT_SRC_IN_CP_AREA
)) {
12129 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12133 /* set next step to use whole border area */
12134 border
/= next_step
;
12136 next_step
*= border
;
12138 /* When data backed up, checkpoint stored,
12139 * kick the kernel to reshape unit of data
12141 next_step
= next_step
+ sra
->reshape_progress
;
12142 /* limit next step to array max position */
12143 if (next_step
> max_position
)
12144 next_step
= max_position
;
12145 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12146 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12147 sra
->reshape_progress
= next_step
;
12149 /* wait until reshape finish */
12150 if (wait_for_reshape_imsm(sra
, ndata
)) {
12151 dprintf("wait_for_reshape_imsm returned error!\n");
12157 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12158 /* ignore error == 2, this can mean end of reshape here
12160 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12166 /* clear migr_rec on disks after successful migration */
12169 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12170 for (d
= super
->disks
; d
; d
= d
->next
) {
12171 if (d
->index
< 0 || is_failed(&d
->disk
))
12173 unsigned long long dsize
;
12175 get_dev_size(d
->fd
, NULL
, &dsize
);
12176 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12178 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12179 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12180 MIGR_REC_BUF_SECTORS
*sector_size
)
12181 perror("Write migr_rec failed");
12185 /* return '1' if done */
12189 /* See Grow.c: abort_reshape() for further explanation */
12190 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12191 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12192 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12197 struct superswitch super_imsm
= {
12198 .examine_super
= examine_super_imsm
,
12199 .brief_examine_super
= brief_examine_super_imsm
,
12200 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12201 .export_examine_super
= export_examine_super_imsm
,
12202 .detail_super
= detail_super_imsm
,
12203 .brief_detail_super
= brief_detail_super_imsm
,
12204 .write_init_super
= write_init_super_imsm
,
12205 .validate_geometry
= validate_geometry_imsm
,
12206 .add_to_super
= add_to_super_imsm
,
12207 .remove_from_super
= remove_from_super_imsm
,
12208 .detail_platform
= detail_platform_imsm
,
12209 .export_detail_platform
= export_detail_platform_imsm
,
12210 .kill_subarray
= kill_subarray_imsm
,
12211 .update_subarray
= update_subarray_imsm
,
12212 .load_container
= load_container_imsm
,
12213 .default_geometry
= default_geometry_imsm
,
12214 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12215 .reshape_super
= imsm_reshape_super
,
12216 .manage_reshape
= imsm_manage_reshape
,
12217 .recover_backup
= recover_backup_imsm
,
12218 .examine_badblocks
= examine_badblocks_imsm
,
12219 .match_home
= match_home_imsm
,
12220 .uuid_from_super
= uuid_from_super_imsm
,
12221 .getinfo_super
= getinfo_super_imsm
,
12222 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12223 .update_super
= update_super_imsm
,
12225 .avail_size
= avail_size_imsm
,
12226 .get_spare_criteria
= get_spare_criteria_imsm
,
12228 .compare_super
= compare_super_imsm
,
12230 .load_super
= load_super_imsm
,
12231 .init_super
= init_super_imsm
,
12232 .store_super
= store_super_imsm
,
12233 .free_super
= free_super_imsm
,
12234 .match_metadata_desc
= match_metadata_desc_imsm
,
12235 .container_content
= container_content_imsm
,
12236 .validate_container
= validate_container_imsm
,
12238 .write_init_ppl
= write_init_ppl_imsm
,
12239 .validate_ppl
= validate_ppl_imsm
,
12245 .open_new
= imsm_open_new
,
12246 .set_array_state
= imsm_set_array_state
,
12247 .set_disk
= imsm_set_disk
,
12248 .sync_metadata
= imsm_sync_metadata
,
12249 .activate_spare
= imsm_activate_spare
,
12250 .process_update
= imsm_process_update
,
12251 .prepare_update
= imsm_prepare_update
,
12252 .record_bad_block
= imsm_record_badblock
,
12253 .clear_bad_block
= imsm_clear_badblock
,
12254 .get_bad_blocks
= imsm_get_badblocks
,