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 4096
92 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
93 #define SECT_PER_MB_SHIFT 11
94 #define MAX_SECTOR_SIZE 4096
95 #define MULTIPLE_PPL_AREA_SIZE_IMSM (1024 * 1024) /* Size of the whole
99 /* Disk configuration info. */
100 #define IMSM_MAX_DEVICES 255
102 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
103 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
104 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
105 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
106 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
107 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
108 #define JOURNAL_DISK __cpu_to_le32(0x2000000) /* Device marked as Journaling Drive */
109 __u32 status
; /* 0xF0 - 0xF3 */
110 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
111 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
112 #define IMSM_DISK_FILLERS 3
113 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
116 /* map selector for map managment
122 /* RAID map configuration infos. */
124 __u32 pba_of_lba0_lo
; /* start address of partition */
125 __u32 blocks_per_member_lo
;/* blocks per member */
126 __u32 num_data_stripes_lo
; /* number of data stripes */
127 __u16 blocks_per_strip
;
128 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
129 #define IMSM_T_STATE_NORMAL 0
130 #define IMSM_T_STATE_UNINITIALIZED 1
131 #define IMSM_T_STATE_DEGRADED 2
132 #define IMSM_T_STATE_FAILED 3
134 #define IMSM_T_RAID0 0
135 #define IMSM_T_RAID1 1
136 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
137 __u8 num_members
; /* number of member disks */
138 __u8 num_domains
; /* number of parity domains */
139 __u8 failed_disk_num
; /* valid only when state is degraded */
141 __u32 pba_of_lba0_hi
;
142 __u32 blocks_per_member_hi
;
143 __u32 num_data_stripes_hi
;
144 __u32 filler
[4]; /* expansion area */
145 #define IMSM_ORD_REBUILD (1 << 24)
146 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
147 * top byte contains some flags
149 } __attribute__ ((packed
));
152 __u32 curr_migr_unit
;
153 __u32 checkpoint_id
; /* id to access curr_migr_unit */
154 __u8 migr_state
; /* Normal or Migrating */
156 #define MIGR_REBUILD 1
157 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
158 #define MIGR_GEN_MIGR 3
159 #define MIGR_STATE_CHANGE 4
160 #define MIGR_REPAIR 5
161 __u8 migr_type
; /* Initializing, Rebuilding, ... */
162 #define RAIDVOL_CLEAN 0
163 #define RAIDVOL_DIRTY 1
164 #define RAIDVOL_DSRECORD_VALID 2
166 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
167 __u16 verify_errors
; /* number of mismatches */
168 __u16 bad_blocks
; /* number of bad blocks during verify */
170 struct imsm_map map
[1];
171 /* here comes another one if migr_state */
172 } __attribute__ ((packed
));
175 __u8 volume
[MAX_RAID_SERIAL_LEN
];
178 #define DEV_BOOTABLE __cpu_to_le32(0x01)
179 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
180 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
181 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
182 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
183 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
184 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
185 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
186 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
187 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
188 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
189 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
190 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
191 __u32 status
; /* Persistent RaidDev status */
192 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
196 __u8 cng_master_disk
;
200 __u16 my_vol_raid_dev_num
; /* Used in Unique volume Id for this RaidDev */
206 /* Unique Volume Id of the NvCache Volume associated with this volume */
207 __u32 nvc_vol_orig_family_num
;
208 __u16 nvc_vol_raid_dev_num
;
211 #define RWH_DISTRIBUTED 1
212 #define RWH_JOURNALING_DRIVE 2
213 #define RWH_MULTIPLE_DISTRIBUTED 3
214 #define RWH_MULTIPLE_PPLS_JOURNALING_DRIVE 4
215 #define RWH_MULTIPLE_OFF 5
216 __u8 rwh_policy
; /* Raid Write Hole Policy */
217 __u8 jd_serial
[MAX_RAID_SERIAL_LEN
]; /* Journal Drive serial number */
220 #define IMSM_DEV_FILLERS 3
221 __u32 filler
[IMSM_DEV_FILLERS
];
223 } __attribute__ ((packed
));
226 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
227 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
228 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
229 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
230 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
231 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
232 __u32 attributes
; /* 0x34 - 0x37 */
233 __u8 num_disks
; /* 0x38 Number of configured disks */
234 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
235 __u8 error_log_pos
; /* 0x3A */
236 __u8 fill
[1]; /* 0x3B */
237 __u32 cache_size
; /* 0x3c - 0x40 in mb */
238 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
239 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
240 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
241 __u16 num_raid_devs_created
; /* 0x4C - 0x4D Used for generating unique
242 * volume IDs for raid_dev created in this array
245 __u16 filler1
; /* 0x4E - 0x4F */
246 #define IMSM_FILLERS 34
247 __u32 filler
[IMSM_FILLERS
]; /* 0x50 - 0xD7 RAID_MPB_FILLERS */
248 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
249 /* here comes imsm_dev[num_raid_devs] */
250 /* here comes BBM logs */
251 } __attribute__ ((packed
));
253 #define BBM_LOG_MAX_ENTRIES 254
254 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
255 #define BBM_LOG_SIGNATURE 0xabadb10c
257 struct bbm_log_block_addr
{
260 } __attribute__ ((__packed__
));
262 struct bbm_log_entry
{
263 __u8 marked_count
; /* Number of blocks marked - 1 */
264 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
265 struct bbm_log_block_addr defective_block_start
;
266 } __attribute__ ((__packed__
));
269 __u32 signature
; /* 0xABADB10C */
271 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
272 } __attribute__ ((__packed__
));
274 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
276 #define BLOCKS_PER_KB (1024/512)
278 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
280 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
282 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
283 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
284 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
287 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
288 * be recovered using srcMap */
289 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
290 * already been migrated and must
291 * be recovered from checkpoint area */
293 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of single PPL, without the header */
296 __u32 rec_status
; /* Status used to determine how to restart
297 * migration in case it aborts
299 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
300 __u32 family_num
; /* Family number of MPB
301 * containing the RaidDev
302 * that is migrating */
303 __u32 ascending_migr
; /* True if migrating in increasing
305 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
306 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
308 * advances per unit-of-operation */
309 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
310 __u32 dest_1st_member_lba
; /* First member lba on first
311 * stripe of destination */
312 __u32 num_migr_units
; /* Total num migration units-of-op */
313 __u32 post_migr_vol_cap
; /* Size of volume after
314 * migration completes */
315 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
316 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
317 * migration ckpt record was read from
318 * (for recovered migrations) */
319 } __attribute__ ((__packed__
));
324 * 2: metadata does not match
332 struct md_list
*next
;
335 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
337 static __u8
migr_type(struct imsm_dev
*dev
)
339 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
340 dev
->status
& DEV_VERIFY_AND_FIX
)
343 return dev
->vol
.migr_type
;
346 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
348 /* for compatibility with older oroms convert MIGR_REPAIR, into
349 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
351 if (migr_type
== MIGR_REPAIR
) {
352 dev
->vol
.migr_type
= MIGR_VERIFY
;
353 dev
->status
|= DEV_VERIFY_AND_FIX
;
355 dev
->vol
.migr_type
= migr_type
;
356 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
360 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
362 return ROUND_UP(bytes
, sector_size
) / sector_size
;
365 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
366 unsigned int sector_size
)
368 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
372 struct imsm_dev
*dev
;
373 struct intel_dev
*next
;
378 enum sys_dev_type type
;
381 struct intel_hba
*next
;
388 /* internal representation of IMSM metadata */
391 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
392 struct imsm_super
*anchor
; /* immovable parameters */
395 void *migr_rec_buf
; /* buffer for I/O operations */
396 struct migr_record
*migr_rec
; /* migration record */
398 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
399 array, it indicates that mdmon is allowed to clean migration
401 size_t len
; /* size of the 'buf' allocation */
402 size_t extra_space
; /* extra space in 'buf' that is not used yet */
403 void *next_buf
; /* for realloc'ing buf from the manager */
405 int updates_pending
; /* count of pending updates for mdmon */
406 int current_vol
; /* index of raid device undergoing creation */
407 unsigned long long create_offset
; /* common start for 'current_vol' */
408 __u32 random
; /* random data for seeding new family numbers */
409 struct intel_dev
*devlist
;
410 unsigned int sector_size
; /* sector size of used member drives */
414 __u8 serial
[MAX_RAID_SERIAL_LEN
];
417 struct imsm_disk disk
;
420 struct extent
*e
; /* for determining freespace @ create */
421 int raiddisk
; /* slot to fill in autolayout */
423 } *disks
, *current_disk
;
424 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
426 struct dl
*missing
; /* disks removed while we weren't looking */
427 struct bbm_log
*bbm_log
;
428 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
429 const struct imsm_orom
*orom
; /* platform firmware support */
430 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
431 struct md_bb bb
; /* memory for get_bad_blocks call */
435 struct imsm_disk disk
;
436 #define IMSM_UNKNOWN_OWNER (-1)
438 struct intel_disk
*next
;
442 unsigned long long start
, size
;
445 /* definitions of reshape process types */
446 enum imsm_reshape_type
{
452 /* definition of messages passed to imsm_process_update */
453 enum imsm_update_type
{
454 update_activate_spare
,
458 update_add_remove_disk
,
459 update_reshape_container_disks
,
460 update_reshape_migration
,
462 update_general_migration_checkpoint
,
464 update_prealloc_badblocks_mem
,
468 struct imsm_update_activate_spare
{
469 enum imsm_update_type type
;
473 struct imsm_update_activate_spare
*next
;
479 unsigned long long size
;
486 enum takeover_direction
{
490 struct imsm_update_takeover
{
491 enum imsm_update_type type
;
493 enum takeover_direction direction
;
496 struct imsm_update_reshape
{
497 enum imsm_update_type type
;
501 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
504 struct imsm_update_reshape_migration
{
505 enum imsm_update_type type
;
508 /* fields for array migration changes
515 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
518 struct imsm_update_size_change
{
519 enum imsm_update_type type
;
524 struct imsm_update_general_migration_checkpoint
{
525 enum imsm_update_type type
;
526 __u32 curr_migr_unit
;
530 __u8 serial
[MAX_RAID_SERIAL_LEN
];
533 struct imsm_update_create_array
{
534 enum imsm_update_type type
;
539 struct imsm_update_kill_array
{
540 enum imsm_update_type type
;
544 struct imsm_update_rename_array
{
545 enum imsm_update_type type
;
546 __u8 name
[MAX_RAID_SERIAL_LEN
];
550 struct imsm_update_add_remove_disk
{
551 enum imsm_update_type type
;
554 struct imsm_update_prealloc_bb_mem
{
555 enum imsm_update_type type
;
558 struct imsm_update_rwh_policy
{
559 enum imsm_update_type type
;
564 static const char *_sys_dev_type
[] = {
565 [SYS_DEV_UNKNOWN
] = "Unknown",
566 [SYS_DEV_SAS
] = "SAS",
567 [SYS_DEV_SATA
] = "SATA",
568 [SYS_DEV_NVME
] = "NVMe",
569 [SYS_DEV_VMD
] = "VMD"
572 const char *get_sys_dev_type(enum sys_dev_type type
)
574 if (type
>= SYS_DEV_MAX
)
575 type
= SYS_DEV_UNKNOWN
;
577 return _sys_dev_type
[type
];
580 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
582 struct intel_hba
*result
= xmalloc(sizeof(*result
));
584 result
->type
= device
->type
;
585 result
->path
= xstrdup(device
->path
);
587 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
593 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
595 struct intel_hba
*result
;
597 for (result
= hba
; result
; result
= result
->next
) {
598 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
604 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
606 struct intel_hba
*hba
;
608 /* check if disk attached to Intel HBA */
609 hba
= find_intel_hba(super
->hba
, device
);
612 /* Check if HBA is already attached to super */
613 if (super
->hba
== NULL
) {
614 super
->hba
= alloc_intel_hba(device
);
619 /* Intel metadata allows for all disks attached to the same type HBA.
620 * Do not support HBA types mixing
622 if (device
->type
!= hba
->type
)
625 /* Multiple same type HBAs can be used if they share the same OROM */
626 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
628 if (device_orom
!= super
->orom
)
634 hba
->next
= alloc_intel_hba(device
);
638 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
640 struct sys_dev
*list
, *elem
;
643 if ((list
= find_intel_devices()) == NULL
)
647 disk_path
= (char *) devname
;
649 disk_path
= diskfd_to_devpath(fd
);
654 for (elem
= list
; elem
; elem
= elem
->next
)
655 if (path_attached_to_hba(disk_path
, elem
->path
))
658 if (disk_path
!= devname
)
664 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
667 static struct supertype
*match_metadata_desc_imsm(char *arg
)
669 struct supertype
*st
;
671 if (strcmp(arg
, "imsm") != 0 &&
672 strcmp(arg
, "default") != 0
676 st
= xcalloc(1, sizeof(*st
));
677 st
->ss
= &super_imsm
;
678 st
->max_devs
= IMSM_MAX_DEVICES
;
679 st
->minor_version
= 0;
684 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
686 return &mpb
->sig
[MPB_SIG_LEN
];
689 /* retrieve a disk directly from the anchor when the anchor is known to be
690 * up-to-date, currently only at load time
692 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
694 if (index
>= mpb
->num_disks
)
696 return &mpb
->disk
[index
];
699 /* retrieve the disk description based on a index of the disk
702 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
706 for (d
= super
->disks
; d
; d
= d
->next
)
707 if (d
->index
== index
)
712 /* retrieve a disk from the parsed metadata */
713 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
717 dl
= get_imsm_dl_disk(super
, index
);
724 /* generate a checksum directly from the anchor when the anchor is known to be
725 * up-to-date, currently only at load or write_super after coalescing
727 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
729 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
730 __u32
*p
= (__u32
*) mpb
;
734 sum
+= __le32_to_cpu(*p
);
738 return sum
- __le32_to_cpu(mpb
->check_sum
);
741 static size_t sizeof_imsm_map(struct imsm_map
*map
)
743 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
746 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
748 /* A device can have 2 maps if it is in the middle of a migration.
750 * MAP_0 - we return the first map
751 * MAP_1 - we return the second map if it exists, else NULL
752 * MAP_X - we return the second map if it exists, else the first
754 struct imsm_map
*map
= &dev
->vol
.map
[0];
755 struct imsm_map
*map2
= NULL
;
757 if (dev
->vol
.migr_state
)
758 map2
= (void *)map
+ sizeof_imsm_map(map
);
760 switch (second_map
) {
777 /* return the size of the device.
778 * migr_state increases the returned size if map[0] were to be duplicated
780 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
782 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
783 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
785 /* migrating means an additional map */
786 if (dev
->vol
.migr_state
)
787 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
789 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
794 /* retrieve disk serial number list from a metadata update */
795 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
798 struct disk_info
*inf
;
800 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
801 sizeof_imsm_dev(&update
->dev
, 0);
806 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
812 if (index
>= mpb
->num_raid_devs
)
815 /* devices start after all disks */
816 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
818 for (i
= 0; i
<= index
; i
++)
820 return _mpb
+ offset
;
822 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
827 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
829 struct intel_dev
*dv
;
831 if (index
>= super
->anchor
->num_raid_devs
)
833 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
834 if (dv
->index
== index
)
839 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
842 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
843 __le16_to_cpu(addr
->w1
));
846 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
848 struct bbm_log_block_addr addr
;
850 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
851 addr
.dw1
= __cpu_to_le32((__u32
)(sec
>> 16) & 0xffffffff);
855 /* get size of the bbm log */
856 static __u32
get_imsm_bbm_log_size(struct bbm_log
*log
)
858 if (!log
|| log
->entry_count
== 0)
861 return sizeof(log
->signature
) +
862 sizeof(log
->entry_count
) +
863 log
->entry_count
* sizeof(struct bbm_log_entry
);
866 /* check if bad block is not partially stored in bbm log */
867 static int is_stored_in_bbm(struct bbm_log
*log
, const __u8 idx
, const unsigned
868 long long sector
, const int length
, __u32
*pos
)
872 for (i
= *pos
; i
< log
->entry_count
; i
++) {
873 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
874 unsigned long long bb_start
;
875 unsigned long long bb_end
;
877 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
878 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
880 if ((entry
->disk_ordinal
== idx
) && (bb_start
>= sector
) &&
881 (bb_end
<= sector
+ length
)) {
889 /* record new bad block in bbm log */
890 static int record_new_badblock(struct bbm_log
*log
, const __u8 idx
, unsigned
891 long long sector
, int length
)
895 struct bbm_log_entry
*entry
= NULL
;
897 while (is_stored_in_bbm(log
, idx
, sector
, length
, &pos
)) {
898 struct bbm_log_entry
*e
= &log
->marked_block_entries
[pos
];
900 if ((e
->marked_count
+ 1 == BBM_LOG_MAX_LBA_ENTRY_VAL
) &&
901 (__le48_to_cpu(&e
->defective_block_start
) == sector
)) {
902 sector
+= BBM_LOG_MAX_LBA_ENTRY_VAL
;
903 length
-= BBM_LOG_MAX_LBA_ENTRY_VAL
;
912 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
913 BBM_LOG_MAX_LBA_ENTRY_VAL
;
914 entry
->defective_block_start
= __cpu_to_le48(sector
);
915 entry
->marked_count
= cnt
- 1;
922 new_bb
= ROUND_UP(length
, BBM_LOG_MAX_LBA_ENTRY_VAL
) /
923 BBM_LOG_MAX_LBA_ENTRY_VAL
;
924 if (log
->entry_count
+ new_bb
> BBM_LOG_MAX_ENTRIES
)
928 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
929 BBM_LOG_MAX_LBA_ENTRY_VAL
;
930 struct bbm_log_entry
*entry
=
931 &log
->marked_block_entries
[log
->entry_count
];
933 entry
->defective_block_start
= __cpu_to_le48(sector
);
934 entry
->marked_count
= cnt
- 1;
935 entry
->disk_ordinal
= idx
;
946 /* clear all bad blocks for given disk */
947 static void clear_disk_badblocks(struct bbm_log
*log
, const __u8 idx
)
951 while (i
< log
->entry_count
) {
952 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
954 if (entries
[i
].disk_ordinal
== idx
) {
955 if (i
< log
->entry_count
- 1)
956 entries
[i
] = entries
[log
->entry_count
- 1];
964 /* clear given bad block */
965 static int clear_badblock(struct bbm_log
*log
, const __u8 idx
, const unsigned
966 long long sector
, const int length
) {
969 while (i
< log
->entry_count
) {
970 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
972 if ((entries
[i
].disk_ordinal
== idx
) &&
973 (__le48_to_cpu(&entries
[i
].defective_block_start
) ==
974 sector
) && (entries
[i
].marked_count
+ 1 == length
)) {
975 if (i
< log
->entry_count
- 1)
976 entries
[i
] = entries
[log
->entry_count
- 1];
986 /* allocate and load BBM log from metadata */
987 static int load_bbm_log(struct intel_super
*super
)
989 struct imsm_super
*mpb
= super
->anchor
;
990 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
992 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
997 struct bbm_log
*log
= (void *)mpb
+
998 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1002 if (bbm_log_size
< sizeof(log
->signature
) +
1003 sizeof(log
->entry_count
))
1006 entry_count
= __le32_to_cpu(log
->entry_count
);
1007 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
1008 (entry_count
> BBM_LOG_MAX_ENTRIES
))
1012 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
1013 entry_count
* sizeof(struct bbm_log_entry
))
1016 memcpy(super
->bbm_log
, log
, bbm_log_size
);
1018 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
1019 super
->bbm_log
->entry_count
= 0;
1025 /* checks if bad block is within volume boundaries */
1026 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
1027 const unsigned long long start_sector
,
1028 const unsigned long long size
)
1030 unsigned long long bb_start
;
1031 unsigned long long bb_end
;
1033 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1034 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1036 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1037 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1043 /* get list of bad blocks on a drive for a volume */
1044 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1045 const unsigned long long start_sector
,
1046 const unsigned long long size
,
1052 for (i
= 0; i
< log
->entry_count
; i
++) {
1053 const struct bbm_log_entry
*ent
=
1054 &log
->marked_block_entries
[i
];
1055 struct md_bb_entry
*bb
;
1057 if ((ent
->disk_ordinal
== idx
) &&
1058 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1060 if (!bbs
->entries
) {
1061 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1067 bb
= &bbs
->entries
[count
++];
1068 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1069 bb
->length
= ent
->marked_count
+ 1;
1077 * == MAP_0 get first map
1078 * == MAP_1 get second map
1079 * == MAP_X than get map according to the current migr_state
1081 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1085 struct imsm_map
*map
;
1087 map
= get_imsm_map(dev
, second_map
);
1089 /* top byte identifies disk under rebuild */
1090 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1093 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1094 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1096 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1098 return ord_to_idx(ord
);
1101 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1103 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1106 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
1111 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1112 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1113 if (ord_to_idx(ord
) == idx
)
1120 static int get_imsm_raid_level(struct imsm_map
*map
)
1122 if (map
->raid_level
== 1) {
1123 if (map
->num_members
== 2)
1129 return map
->raid_level
;
1132 static int cmp_extent(const void *av
, const void *bv
)
1134 const struct extent
*a
= av
;
1135 const struct extent
*b
= bv
;
1136 if (a
->start
< b
->start
)
1138 if (a
->start
> b
->start
)
1143 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1145 int memberships
= 0;
1148 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1149 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1150 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1152 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
1159 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1161 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
1163 if (lo
== 0 || hi
== 0)
1165 *lo
= __le32_to_cpu((unsigned)n
);
1166 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
1170 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1172 return (unsigned long long)__le32_to_cpu(lo
) |
1173 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1176 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1180 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1183 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1187 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1190 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1194 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1197 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1201 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1204 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1206 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1209 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1211 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1214 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1216 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1219 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1221 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1224 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
1226 /* find a list of used extents on the given physical device */
1227 struct extent
*rv
, *e
;
1229 int memberships
= count_memberships(dl
, super
);
1232 /* trim the reserved area for spares, so they can join any array
1233 * regardless of whether the OROM has assigned sectors from the
1234 * IMSM_RESERVED_SECTORS region
1236 if (dl
->index
== -1)
1237 reservation
= imsm_min_reserved_sectors(super
);
1239 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1241 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
1244 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1245 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1246 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1248 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1249 e
->start
= pba_of_lba0(map
);
1250 e
->size
= blocks_per_member(map
);
1254 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1256 /* determine the start of the metadata
1257 * when no raid devices are defined use the default
1258 * ...otherwise allow the metadata to truncate the value
1259 * as is the case with older versions of imsm
1262 struct extent
*last
= &rv
[memberships
- 1];
1263 unsigned long long remainder
;
1265 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1266 /* round down to 1k block to satisfy precision of the kernel
1270 /* make sure remainder is still sane */
1271 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1272 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1273 if (reservation
> remainder
)
1274 reservation
= remainder
;
1276 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1281 /* try to determine how much space is reserved for metadata from
1282 * the last get_extents() entry, otherwise fallback to the
1285 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1291 /* for spares just return a minimal reservation which will grow
1292 * once the spare is picked up by an array
1294 if (dl
->index
== -1)
1295 return MPB_SECTOR_CNT
;
1297 e
= get_extents(super
, dl
);
1299 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1301 /* scroll to last entry */
1302 for (i
= 0; e
[i
].size
; i
++)
1305 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1312 static int is_spare(struct imsm_disk
*disk
)
1314 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1317 static int is_configured(struct imsm_disk
*disk
)
1319 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1322 static int is_failed(struct imsm_disk
*disk
)
1324 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1327 static int is_journal(struct imsm_disk
*disk
)
1329 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1332 /* round array size down to closest MB and ensure it splits evenly
1335 static unsigned long long round_size_to_mb(unsigned long long size
, unsigned int
1339 size
= (size
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
1345 static int able_to_resync(int raid_level
, int missing_disks
)
1347 int max_missing_disks
= 0;
1349 switch (raid_level
) {
1351 max_missing_disks
= 1;
1354 max_missing_disks
= 0;
1356 return missing_disks
<= max_missing_disks
;
1359 /* try to determine how much space is reserved for metadata from
1360 * the last get_extents() entry on the smallest active disk,
1361 * otherwise fallback to the default
1363 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1367 unsigned long long min_active
;
1369 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1370 struct dl
*dl
, *dl_min
= NULL
;
1376 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1379 unsigned long long blocks
= total_blocks(&dl
->disk
);
1380 if (blocks
< min_active
|| min_active
== 0) {
1382 min_active
= blocks
;
1388 /* find last lba used by subarrays on the smallest active disk */
1389 e
= get_extents(super
, dl_min
);
1392 for (i
= 0; e
[i
].size
; i
++)
1395 remainder
= min_active
- e
[i
].start
;
1398 /* to give priority to recovery we should not require full
1399 IMSM_RESERVED_SECTORS from the spare */
1400 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1402 /* if real reservation is smaller use that value */
1403 return (remainder
< rv
) ? remainder
: rv
;
1407 * Return minimum size of a spare and sector size
1408 * that can be used in this array
1410 int get_spare_criteria_imsm(struct supertype
*st
, struct spare_criteria
*c
)
1412 struct intel_super
*super
= st
->sb
;
1416 unsigned long long size
= 0;
1423 /* find first active disk in array */
1425 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1429 /* find last lba used by subarrays */
1430 e
= get_extents(super
, dl
);
1433 for (i
= 0; e
[i
].size
; i
++)
1436 size
= e
[i
-1].start
+ e
[i
-1].size
;
1439 /* add the amount of space needed for metadata */
1440 size
+= imsm_min_reserved_sectors(super
);
1442 c
->min_size
= size
* 512;
1443 c
->sector_size
= super
->sector_size
;
1448 static int is_gen_migration(struct imsm_dev
*dev
);
1450 #define IMSM_4K_DIV 8
1452 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1453 struct imsm_dev
*dev
);
1455 static void print_imsm_dev(struct intel_super
*super
,
1456 struct imsm_dev
*dev
,
1462 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1463 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1467 printf("[%.16s]:\n", dev
->volume
);
1468 printf(" UUID : %s\n", uuid
);
1469 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1471 printf(" <-- %d", get_imsm_raid_level(map2
));
1473 printf(" Members : %d", map
->num_members
);
1475 printf(" <-- %d", map2
->num_members
);
1477 printf(" Slots : [");
1478 for (i
= 0; i
< map
->num_members
; i
++) {
1479 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1480 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1485 for (i
= 0; i
< map2
->num_members
; i
++) {
1486 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1487 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1492 printf(" Failed disk : ");
1493 if (map
->failed_disk_num
== 0xff)
1496 printf("%i", map
->failed_disk_num
);
1498 slot
= get_imsm_disk_slot(map
, disk_idx
);
1500 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1501 printf(" This Slot : %d%s\n", slot
,
1502 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1504 printf(" This Slot : ?\n");
1505 printf(" Sector Size : %u\n", super
->sector_size
);
1506 sz
= __le32_to_cpu(dev
->size_high
);
1508 sz
+= __le32_to_cpu(dev
->size_low
);
1509 printf(" Array Size : %llu%s\n",
1510 (unsigned long long)sz
* 512 / super
->sector_size
,
1511 human_size(sz
* 512));
1512 sz
= blocks_per_member(map
);
1513 printf(" Per Dev Size : %llu%s\n",
1514 (unsigned long long)sz
* 512 / super
->sector_size
,
1515 human_size(sz
* 512));
1516 printf(" Sector Offset : %llu\n",
1518 printf(" Num Stripes : %llu\n",
1519 num_data_stripes(map
));
1520 printf(" Chunk Size : %u KiB",
1521 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1523 printf(" <-- %u KiB",
1524 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1526 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1527 printf(" Migrate State : ");
1528 if (dev
->vol
.migr_state
) {
1529 if (migr_type(dev
) == MIGR_INIT
)
1530 printf("initialize\n");
1531 else if (migr_type(dev
) == MIGR_REBUILD
)
1532 printf("rebuild\n");
1533 else if (migr_type(dev
) == MIGR_VERIFY
)
1535 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1536 printf("general migration\n");
1537 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1538 printf("state change\n");
1539 else if (migr_type(dev
) == MIGR_REPAIR
)
1542 printf("<unknown:%d>\n", migr_type(dev
));
1545 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1546 if (dev
->vol
.migr_state
) {
1547 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1549 printf(" <-- %s", map_state_str
[map
->map_state
]);
1550 printf("\n Checkpoint : %u ",
1551 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1552 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1555 printf("(%llu)", (unsigned long long)
1556 blocks_per_migr_unit(super
, dev
));
1559 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1561 printf(" RWH Policy : ");
1562 if (dev
->rwh_policy
== RWH_OFF
|| dev
->rwh_policy
== RWH_MULTIPLE_OFF
)
1564 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1565 printf("PPL distributed\n");
1566 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1567 printf("PPL journaling drive\n");
1568 else if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
1569 printf("Multiple distributed PPLs\n");
1570 else if (dev
->rwh_policy
== RWH_MULTIPLE_PPLS_JOURNALING_DRIVE
)
1571 printf("Multiple PPLs on journaling drive\n");
1573 printf("<unknown:%d>\n", dev
->rwh_policy
);
1576 static void print_imsm_disk(struct imsm_disk
*disk
,
1579 unsigned int sector_size
) {
1580 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1583 if (index
< -1 || !disk
)
1587 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1589 printf(" Disk%02d Serial : %s\n", index
, str
);
1591 printf(" Disk Serial : %s\n", str
);
1592 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1593 is_configured(disk
) ? " active" : "",
1594 is_failed(disk
) ? " failed" : "",
1595 is_journal(disk
) ? " journal" : "");
1596 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1597 sz
= total_blocks(disk
) - reserved
;
1598 printf(" Usable Size : %llu%s\n",
1599 (unsigned long long)sz
* 512 / sector_size
,
1600 human_size(sz
* 512));
1603 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1605 struct migr_record
*migr_rec
= super
->migr_rec
;
1607 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1608 migr_rec
->ckpt_area_pba
/= IMSM_4K_DIV
;
1609 migr_rec
->dest_1st_member_lba
/= IMSM_4K_DIV
;
1610 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1611 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1612 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1613 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1616 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1618 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1621 void convert_to_4k(struct intel_super
*super
)
1623 struct imsm_super
*mpb
= super
->anchor
;
1624 struct imsm_disk
*disk
;
1626 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1628 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1629 disk
= __get_imsm_disk(mpb
, i
);
1631 convert_to_4k_imsm_disk(disk
);
1633 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1634 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1635 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1637 split_ull((join_u32(dev
->size_low
, dev
->size_high
)/IMSM_4K_DIV
),
1638 &dev
->size_low
, &dev
->size_high
);
1639 dev
->vol
.curr_migr_unit
/= IMSM_4K_DIV
;
1642 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1643 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1644 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1646 if (dev
->vol
.migr_state
) {
1648 map
= get_imsm_map(dev
, MAP_1
);
1649 set_blocks_per_member(map
,
1650 blocks_per_member(map
)/IMSM_4K_DIV
);
1651 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1652 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1656 struct bbm_log
*log
= (void *)mpb
+
1657 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1660 for (i
= 0; i
< log
->entry_count
; i
++) {
1661 struct bbm_log_entry
*entry
=
1662 &log
->marked_block_entries
[i
];
1664 __u8 count
= entry
->marked_count
+ 1;
1665 unsigned long long sector
=
1666 __le48_to_cpu(&entry
->defective_block_start
);
1668 entry
->defective_block_start
=
1669 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1670 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1674 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1677 void examine_migr_rec_imsm(struct intel_super
*super
)
1679 struct migr_record
*migr_rec
= super
->migr_rec
;
1680 struct imsm_super
*mpb
= super
->anchor
;
1683 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1684 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1685 struct imsm_map
*map
;
1688 if (is_gen_migration(dev
) == 0)
1691 printf("\nMigration Record Information:");
1693 /* first map under migration */
1694 map
= get_imsm_map(dev
, MAP_0
);
1696 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1697 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1698 printf(" Empty\n ");
1699 printf("Examine one of first two disks in array\n");
1702 printf("\n Status : ");
1703 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1706 printf("Contains Data\n");
1707 printf(" Current Unit : %u\n",
1708 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1709 printf(" Family : %u\n",
1710 __le32_to_cpu(migr_rec
->family_num
));
1711 printf(" Ascending : %u\n",
1712 __le32_to_cpu(migr_rec
->ascending_migr
));
1713 printf(" Blocks Per Unit : %u\n",
1714 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1715 printf(" Dest. Depth Per Unit : %u\n",
1716 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1717 printf(" Checkpoint Area pba : %u\n",
1718 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1719 printf(" First member lba : %u\n",
1720 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1721 printf(" Total Number of Units : %u\n",
1722 __le32_to_cpu(migr_rec
->num_migr_units
));
1723 printf(" Size of volume : %u\n",
1724 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1725 printf(" Expansion space for LBA64 : %u\n",
1726 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1727 printf(" Record was read from : %u\n",
1728 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1734 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1736 struct migr_record
*migr_rec
= super
->migr_rec
;
1738 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1739 migr_rec
->ckpt_area_pba
*= IMSM_4K_DIV
;
1740 migr_rec
->dest_1st_member_lba
*= IMSM_4K_DIV
;
1741 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1742 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1743 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1744 &migr_rec
->post_migr_vol_cap
,
1745 &migr_rec
->post_migr_vol_cap_hi
);
1748 void convert_from_4k(struct intel_super
*super
)
1750 struct imsm_super
*mpb
= super
->anchor
;
1751 struct imsm_disk
*disk
;
1753 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1755 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1756 disk
= __get_imsm_disk(mpb
, i
);
1758 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1761 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1762 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1763 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1765 split_ull((join_u32(dev
->size_low
, dev
->size_high
)*IMSM_4K_DIV
),
1766 &dev
->size_low
, &dev
->size_high
);
1767 dev
->vol
.curr_migr_unit
*= IMSM_4K_DIV
;
1770 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1771 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1772 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1774 if (dev
->vol
.migr_state
) {
1776 map
= get_imsm_map(dev
, MAP_1
);
1777 set_blocks_per_member(map
,
1778 blocks_per_member(map
)*IMSM_4K_DIV
);
1779 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1780 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1784 struct bbm_log
*log
= (void *)mpb
+
1785 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1788 for (i
= 0; i
< log
->entry_count
; i
++) {
1789 struct bbm_log_entry
*entry
=
1790 &log
->marked_block_entries
[i
];
1792 __u8 count
= entry
->marked_count
+ 1;
1793 unsigned long long sector
=
1794 __le48_to_cpu(&entry
->defective_block_start
);
1796 entry
->defective_block_start
=
1797 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1798 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1802 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1805 /*******************************************************************************
1806 * function: imsm_check_attributes
1807 * Description: Function checks if features represented by attributes flags
1808 * are supported by mdadm.
1810 * attributes - Attributes read from metadata
1812 * 0 - passed attributes contains unsupported features flags
1813 * 1 - all features are supported
1814 ******************************************************************************/
1815 static int imsm_check_attributes(__u32 attributes
)
1818 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1820 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1822 not_supported
&= attributes
;
1823 if (not_supported
) {
1824 pr_err("(IMSM): Unsupported attributes : %x\n",
1825 (unsigned)__le32_to_cpu(not_supported
));
1826 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1827 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1828 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1830 if (not_supported
& MPB_ATTRIB_2TB
) {
1831 dprintf("\t\tMPB_ATTRIB_2TB\n");
1832 not_supported
^= MPB_ATTRIB_2TB
;
1834 if (not_supported
& MPB_ATTRIB_RAID0
) {
1835 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1836 not_supported
^= MPB_ATTRIB_RAID0
;
1838 if (not_supported
& MPB_ATTRIB_RAID1
) {
1839 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1840 not_supported
^= MPB_ATTRIB_RAID1
;
1842 if (not_supported
& MPB_ATTRIB_RAID10
) {
1843 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1844 not_supported
^= MPB_ATTRIB_RAID10
;
1846 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1847 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1848 not_supported
^= MPB_ATTRIB_RAID1E
;
1850 if (not_supported
& MPB_ATTRIB_RAID5
) {
1851 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1852 not_supported
^= MPB_ATTRIB_RAID5
;
1854 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1855 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1856 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1858 if (not_supported
& MPB_ATTRIB_BBM
) {
1859 dprintf("\t\tMPB_ATTRIB_BBM\n");
1860 not_supported
^= MPB_ATTRIB_BBM
;
1862 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1863 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1864 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1866 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1867 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1868 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1870 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1871 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1872 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1874 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1875 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1876 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1878 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1879 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1880 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1884 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1892 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1894 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1896 struct intel_super
*super
= st
->sb
;
1897 struct imsm_super
*mpb
= super
->anchor
;
1898 char str
[MAX_SIGNATURE_LENGTH
];
1903 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1906 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
1907 str
[MPB_SIG_LEN
-1] = '\0';
1908 printf(" Magic : %s\n", str
);
1909 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1910 printf(" Version : %s\n", get_imsm_version(mpb
));
1911 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1912 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1913 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1914 printf(" Attributes : ");
1915 if (imsm_check_attributes(mpb
->attributes
))
1916 printf("All supported\n");
1918 printf("not supported\n");
1919 getinfo_super_imsm(st
, &info
, NULL
);
1920 fname_from_uuid(st
, &info
, nbuf
, ':');
1921 printf(" UUID : %s\n", nbuf
+ 5);
1922 sum
= __le32_to_cpu(mpb
->check_sum
);
1923 printf(" Checksum : %08x %s\n", sum
,
1924 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1925 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
1926 printf(" Disks : %d\n", mpb
->num_disks
);
1927 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1928 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
1929 super
->disks
->index
, reserved
, super
->sector_size
);
1930 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
1931 struct bbm_log
*log
= super
->bbm_log
;
1934 printf("Bad Block Management Log:\n");
1935 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1936 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1937 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1939 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1941 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1943 super
->current_vol
= i
;
1944 getinfo_super_imsm(st
, &info
, NULL
);
1945 fname_from_uuid(st
, &info
, nbuf
, ':');
1946 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1948 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1949 if (i
== super
->disks
->index
)
1951 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
1952 super
->sector_size
);
1955 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1956 if (dl
->index
== -1)
1957 print_imsm_disk(&dl
->disk
, -1, reserved
,
1958 super
->sector_size
);
1960 examine_migr_rec_imsm(super
);
1963 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1965 /* We just write a generic IMSM ARRAY entry */
1968 struct intel_super
*super
= st
->sb
;
1970 if (!super
->anchor
->num_raid_devs
) {
1971 printf("ARRAY metadata=imsm\n");
1975 getinfo_super_imsm(st
, &info
, NULL
);
1976 fname_from_uuid(st
, &info
, nbuf
, ':');
1977 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1980 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1982 /* We just write a generic IMSM ARRAY entry */
1986 struct intel_super
*super
= st
->sb
;
1989 if (!super
->anchor
->num_raid_devs
)
1992 getinfo_super_imsm(st
, &info
, NULL
);
1993 fname_from_uuid(st
, &info
, nbuf
, ':');
1994 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1995 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1997 super
->current_vol
= i
;
1998 getinfo_super_imsm(st
, &info
, NULL
);
1999 fname_from_uuid(st
, &info
, nbuf1
, ':');
2000 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
2001 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
2005 static void export_examine_super_imsm(struct supertype
*st
)
2007 struct intel_super
*super
= st
->sb
;
2008 struct imsm_super
*mpb
= super
->anchor
;
2012 getinfo_super_imsm(st
, &info
, NULL
);
2013 fname_from_uuid(st
, &info
, nbuf
, ':');
2014 printf("MD_METADATA=imsm\n");
2015 printf("MD_LEVEL=container\n");
2016 printf("MD_UUID=%s\n", nbuf
+5);
2017 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
2020 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
2022 /* The second last sector of the device contains
2023 * the "struct imsm_super" metadata.
2024 * This contains mpb_size which is the size in bytes of the
2025 * extended metadata. This is located immediately before
2027 * We want to read all that, plus the last sector which
2028 * may contain a migration record, and write it all
2032 unsigned long long dsize
, offset
;
2034 struct imsm_super
*sb
;
2035 struct intel_super
*super
= st
->sb
;
2036 unsigned int sector_size
= super
->sector_size
;
2037 unsigned int written
= 0;
2039 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
) != 0)
2042 if (!get_dev_size(from
, NULL
, &dsize
))
2045 if (lseek64(from
, dsize
-(2*sector_size
), 0) < 0)
2047 if ((unsigned int)read(from
, buf
, sector_size
) != sector_size
)
2050 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
2053 sectors
= mpb_sectors(sb
, sector_size
) + 2;
2054 offset
= dsize
- sectors
* sector_size
;
2055 if (lseek64(from
, offset
, 0) < 0 ||
2056 lseek64(to
, offset
, 0) < 0)
2058 while (written
< sectors
* sector_size
) {
2059 int n
= sectors
*sector_size
- written
;
2062 if (read(from
, buf
, n
) != n
)
2064 if (write(to
, buf
, n
) != n
)
2075 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
2080 getinfo_super_imsm(st
, &info
, NULL
);
2081 fname_from_uuid(st
, &info
, nbuf
, ':');
2082 printf("\n UUID : %s\n", nbuf
+ 5);
2085 static void brief_detail_super_imsm(struct supertype
*st
)
2089 getinfo_super_imsm(st
, &info
, NULL
);
2090 fname_from_uuid(st
, &info
, nbuf
, ':');
2091 printf(" UUID=%s", nbuf
+ 5);
2094 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
2095 static void fd2devname(int fd
, char *name
);
2097 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2099 /* dump an unsorted list of devices attached to AHCI Intel storage
2100 * controller, as well as non-connected ports
2102 int hba_len
= strlen(hba_path
) + 1;
2107 unsigned long port_mask
= (1 << port_count
) - 1;
2109 if (port_count
> (int)sizeof(port_mask
) * 8) {
2111 pr_err("port_count %d out of range\n", port_count
);
2115 /* scroll through /sys/dev/block looking for devices attached to
2118 dir
= opendir("/sys/dev/block");
2122 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2133 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2135 path
= devt_to_devpath(makedev(major
, minor
));
2138 if (!path_attached_to_hba(path
, hba_path
)) {
2144 /* retrieve the scsi device type */
2145 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2147 pr_err("failed to allocate 'device'\n");
2151 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2152 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2154 pr_err("failed to read device type for %s\n",
2160 type
= strtoul(buf
, NULL
, 10);
2162 /* if it's not a disk print the vendor and model */
2163 if (!(type
== 0 || type
== 7 || type
== 14)) {
2166 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2167 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2168 strncpy(vendor
, buf
, sizeof(vendor
));
2169 vendor
[sizeof(vendor
) - 1] = '\0';
2170 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2171 while (isspace(*c
) || *c
== '\0')
2175 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2176 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2177 strncpy(model
, buf
, sizeof(model
));
2178 model
[sizeof(model
) - 1] = '\0';
2179 c
= (char *) &model
[sizeof(model
) - 1];
2180 while (isspace(*c
) || *c
== '\0')
2184 if (vendor
[0] && model
[0])
2185 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2187 switch (type
) { /* numbers from hald/linux/device.c */
2188 case 1: sprintf(buf
, "tape"); break;
2189 case 2: sprintf(buf
, "printer"); break;
2190 case 3: sprintf(buf
, "processor"); break;
2192 case 5: sprintf(buf
, "cdrom"); break;
2193 case 6: sprintf(buf
, "scanner"); break;
2194 case 8: sprintf(buf
, "media_changer"); break;
2195 case 9: sprintf(buf
, "comm"); break;
2196 case 12: sprintf(buf
, "raid"); break;
2197 default: sprintf(buf
, "unknown");
2203 /* chop device path to 'host%d' and calculate the port number */
2204 c
= strchr(&path
[hba_len
], '/');
2207 pr_err("%s - invalid path name\n", path
+ hba_len
);
2212 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2213 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2217 *c
= '/'; /* repair the full string */
2218 pr_err("failed to determine port number for %s\n",
2225 /* mark this port as used */
2226 port_mask
&= ~(1 << port
);
2228 /* print out the device information */
2230 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2234 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2236 printf(" Port%d : - disk info unavailable -\n", port
);
2238 fd2devname(fd
, buf
);
2239 printf(" Port%d : %s", port
, buf
);
2240 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
2241 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
2256 for (i
= 0; i
< port_count
; i
++)
2257 if (port_mask
& (1 << i
))
2258 printf(" Port%d : - no device attached -\n", i
);
2264 static int print_vmd_attached_devs(struct sys_dev
*hba
)
2272 if (hba
->type
!= SYS_DEV_VMD
)
2275 /* scroll through /sys/dev/block looking for devices attached to
2278 dir
= opendir("/sys/bus/pci/drivers/nvme");
2282 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2285 /* is 'ent' a device? check that the 'subsystem' link exists and
2286 * that its target matches 'bus'
2288 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
2290 n
= readlink(path
, link
, sizeof(link
));
2291 if (n
< 0 || n
>= (int)sizeof(link
))
2294 c
= strrchr(link
, '/');
2297 if (strncmp("pci", c
+1, strlen("pci")) != 0)
2300 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
2302 rp
= realpath(path
, NULL
);
2306 if (path_attached_to_hba(rp
, hba
->path
)) {
2307 printf(" NVMe under VMD : %s\n", rp
);
2316 static void print_found_intel_controllers(struct sys_dev
*elem
)
2318 for (; elem
; elem
= elem
->next
) {
2319 pr_err("found Intel(R) ");
2320 if (elem
->type
== SYS_DEV_SATA
)
2321 fprintf(stderr
, "SATA ");
2322 else if (elem
->type
== SYS_DEV_SAS
)
2323 fprintf(stderr
, "SAS ");
2324 else if (elem
->type
== SYS_DEV_NVME
)
2325 fprintf(stderr
, "NVMe ");
2327 if (elem
->type
== SYS_DEV_VMD
)
2328 fprintf(stderr
, "VMD domain");
2330 fprintf(stderr
, "RAID controller");
2333 fprintf(stderr
, " at %s", elem
->pci_id
);
2334 fprintf(stderr
, ".\n");
2339 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2346 if ((dir
= opendir(hba_path
)) == NULL
)
2349 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2352 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2353 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2355 if (*port_count
== 0)
2357 else if (host
< host_base
)
2360 if (host
+ 1 > *port_count
+ host_base
)
2361 *port_count
= host
+ 1 - host_base
;
2367 static void print_imsm_capability(const struct imsm_orom
*orom
)
2369 printf(" Platform : Intel(R) ");
2370 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2371 printf("Matrix Storage Manager\n");
2372 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2373 printf("Virtual RAID on CPU\n");
2375 printf("Rapid Storage Technology%s\n",
2376 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2377 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2378 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2379 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2380 printf(" RAID Levels :%s%s%s%s%s\n",
2381 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2382 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2383 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2384 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2385 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2386 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2387 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2388 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2389 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2390 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2391 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2392 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2393 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2394 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2395 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2396 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2397 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2398 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2399 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2400 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2401 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2402 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2403 printf(" 2TB volumes :%s supported\n",
2404 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2405 printf(" 2TB disks :%s supported\n",
2406 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2407 printf(" Max Disks : %d\n", orom
->tds
);
2408 printf(" Max Volumes : %d per array, %d per %s\n",
2409 orom
->vpa
, orom
->vphba
,
2410 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2414 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2416 printf("MD_FIRMWARE_TYPE=imsm\n");
2417 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2418 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2419 orom
->hotfix_ver
, orom
->build
);
2420 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2421 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2422 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2423 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2424 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2425 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2426 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2427 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2428 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2429 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2430 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2431 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2432 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2433 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2434 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2435 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2436 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2437 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2438 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2439 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2440 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2441 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2442 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2443 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2444 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2445 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2446 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2447 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2450 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2452 /* There are two components to imsm platform support, the ahci SATA
2453 * controller and the option-rom. To find the SATA controller we
2454 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2455 * controller with the Intel vendor id is present. This approach
2456 * allows mdadm to leverage the kernel's ahci detection logic, with the
2457 * caveat that if ahci.ko is not loaded mdadm will not be able to
2458 * detect platform raid capabilities. The option-rom resides in a
2459 * platform "Adapter ROM". We scan for its signature to retrieve the
2460 * platform capabilities. If raid support is disabled in the BIOS the
2461 * option-rom capability structure will not be available.
2463 struct sys_dev
*list
, *hba
;
2468 if (enumerate_only
) {
2469 if (check_env("IMSM_NO_PLATFORM"))
2471 list
= find_intel_devices();
2474 for (hba
= list
; hba
; hba
= hba
->next
) {
2475 if (find_imsm_capability(hba
)) {
2485 list
= find_intel_devices();
2488 pr_err("no active Intel(R) RAID controller found.\n");
2490 } else if (verbose
> 0)
2491 print_found_intel_controllers(list
);
2493 for (hba
= list
; hba
; hba
= hba
->next
) {
2494 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2496 if (!find_imsm_capability(hba
)) {
2498 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2499 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2500 get_sys_dev_type(hba
->type
));
2506 if (controller_path
&& result
== 1) {
2507 pr_err("no active Intel(R) RAID controller found under %s\n",
2512 const struct orom_entry
*entry
;
2514 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2515 if (entry
->type
== SYS_DEV_VMD
) {
2516 print_imsm_capability(&entry
->orom
);
2517 printf(" 3rd party NVMe :%s supported\n",
2518 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2519 for (hba
= list
; hba
; hba
= hba
->next
) {
2520 if (hba
->type
== SYS_DEV_VMD
) {
2522 printf(" I/O Controller : %s (%s)\n",
2523 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2524 if (print_vmd_attached_devs(hba
)) {
2526 pr_err("failed to get devices attached to VMD domain.\n");
2535 print_imsm_capability(&entry
->orom
);
2536 if (entry
->type
== SYS_DEV_NVME
) {
2537 for (hba
= list
; hba
; hba
= hba
->next
) {
2538 if (hba
->type
== SYS_DEV_NVME
)
2539 printf(" NVMe Device : %s\n", hba
->path
);
2545 struct devid_list
*devid
;
2546 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2547 hba
= device_by_id(devid
->devid
);
2551 printf(" I/O Controller : %s (%s)\n",
2552 hba
->path
, get_sys_dev_type(hba
->type
));
2553 if (hba
->type
== SYS_DEV_SATA
) {
2554 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2555 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2557 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2568 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2570 struct sys_dev
*list
, *hba
;
2573 list
= find_intel_devices();
2576 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2581 for (hba
= list
; hba
; hba
= hba
->next
) {
2582 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2584 if (!find_imsm_capability(hba
) && verbose
> 0) {
2586 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2587 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2593 const struct orom_entry
*entry
;
2595 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2596 if (entry
->type
== SYS_DEV_VMD
) {
2597 for (hba
= list
; hba
; hba
= hba
->next
)
2598 print_imsm_capability_export(&entry
->orom
);
2601 print_imsm_capability_export(&entry
->orom
);
2607 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2609 /* the imsm metadata format does not specify any host
2610 * identification information. We return -1 since we can never
2611 * confirm nor deny whether a given array is "meant" for this
2612 * host. We rely on compare_super and the 'family_num' fields to
2613 * exclude member disks that do not belong, and we rely on
2614 * mdadm.conf to specify the arrays that should be assembled.
2615 * Auto-assembly may still pick up "foreign" arrays.
2621 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2623 /* The uuid returned here is used for:
2624 * uuid to put into bitmap file (Create, Grow)
2625 * uuid for backup header when saving critical section (Grow)
2626 * comparing uuids when re-adding a device into an array
2627 * In these cases the uuid required is that of the data-array,
2628 * not the device-set.
2629 * uuid to recognise same set when adding a missing device back
2630 * to an array. This is a uuid for the device-set.
2632 * For each of these we can make do with a truncated
2633 * or hashed uuid rather than the original, as long as
2635 * In each case the uuid required is that of the data-array,
2636 * not the device-set.
2638 /* imsm does not track uuid's so we synthesis one using sha1 on
2639 * - The signature (Which is constant for all imsm array, but no matter)
2640 * - the orig_family_num of the container
2641 * - the index number of the volume
2642 * - the 'serial' number of the volume.
2643 * Hopefully these are all constant.
2645 struct intel_super
*super
= st
->sb
;
2648 struct sha1_ctx ctx
;
2649 struct imsm_dev
*dev
= NULL
;
2652 /* some mdadm versions failed to set ->orig_family_num, in which
2653 * case fall back to ->family_num. orig_family_num will be
2654 * fixed up with the first metadata update.
2656 family_num
= super
->anchor
->orig_family_num
;
2657 if (family_num
== 0)
2658 family_num
= super
->anchor
->family_num
;
2659 sha1_init_ctx(&ctx
);
2660 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2661 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2662 if (super
->current_vol
>= 0)
2663 dev
= get_imsm_dev(super
, super
->current_vol
);
2665 __u32 vol
= super
->current_vol
;
2666 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2667 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2669 sha1_finish_ctx(&ctx
, buf
);
2670 memcpy(uuid
, buf
, 4*4);
2675 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2677 __u8
*v
= get_imsm_version(mpb
);
2678 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2679 char major
[] = { 0, 0, 0 };
2680 char minor
[] = { 0 ,0, 0 };
2681 char patch
[] = { 0, 0, 0 };
2682 char *ver_parse
[] = { major
, minor
, patch
};
2686 while (*v
!= '\0' && v
< end
) {
2687 if (*v
!= '.' && j
< 2)
2688 ver_parse
[i
][j
++] = *v
;
2696 *m
= strtol(minor
, NULL
, 0);
2697 *p
= strtol(patch
, NULL
, 0);
2701 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2703 /* migr_strip_size when repairing or initializing parity */
2704 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2705 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2707 switch (get_imsm_raid_level(map
)) {
2712 return 128*1024 >> 9;
2716 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2718 /* migr_strip_size when rebuilding a degraded disk, no idea why
2719 * this is different than migr_strip_size_resync(), but it's good
2722 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2723 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2725 switch (get_imsm_raid_level(map
)) {
2728 if (map
->num_members
% map
->num_domains
== 0)
2729 return 128*1024 >> 9;
2733 return max((__u32
) 64*1024 >> 9, chunk
);
2735 return 128*1024 >> 9;
2739 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2741 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2742 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2743 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2744 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2746 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2749 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2751 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2752 int level
= get_imsm_raid_level(lo
);
2754 if (level
== 1 || level
== 10) {
2755 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2757 return hi
->num_domains
;
2759 return num_stripes_per_unit_resync(dev
);
2762 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2764 /* named 'imsm_' because raid0, raid1 and raid10
2765 * counter-intuitively have the same number of data disks
2767 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2769 switch (get_imsm_raid_level(map
)) {
2771 return map
->num_members
;
2775 return map
->num_members
/2;
2777 return map
->num_members
- 1;
2779 dprintf("unsupported raid level\n");
2784 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2786 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2787 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2789 switch(get_imsm_raid_level(map
)) {
2792 return chunk
* map
->num_domains
;
2794 return chunk
* map
->num_members
;
2800 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2802 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2803 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2804 __u32 strip
= block
/ chunk
;
2806 switch (get_imsm_raid_level(map
)) {
2809 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2810 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2812 return vol_stripe
* chunk
+ block
% chunk
;
2814 __u32 stripe
= strip
/ (map
->num_members
- 1);
2816 return stripe
* chunk
+ block
% chunk
;
2823 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2824 struct imsm_dev
*dev
)
2826 /* calculate the conversion factor between per member 'blocks'
2827 * (md/{resync,rebuild}_start) and imsm migration units, return
2828 * 0 for the 'not migrating' and 'unsupported migration' cases
2830 if (!dev
->vol
.migr_state
)
2833 switch (migr_type(dev
)) {
2834 case MIGR_GEN_MIGR
: {
2835 struct migr_record
*migr_rec
= super
->migr_rec
;
2836 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2841 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2842 __u32 stripes_per_unit
;
2843 __u32 blocks_per_unit
;
2852 /* yes, this is really the translation of migr_units to
2853 * per-member blocks in the 'resync' case
2855 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2856 migr_chunk
= migr_strip_blocks_resync(dev
);
2857 disks
= imsm_num_data_members(dev
, MAP_0
);
2858 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2859 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2860 segment
= blocks_per_unit
/ stripe
;
2861 block_rel
= blocks_per_unit
- segment
* stripe
;
2862 parity_depth
= parity_segment_depth(dev
);
2863 block_map
= map_migr_block(dev
, block_rel
);
2864 return block_map
+ parity_depth
* segment
;
2866 case MIGR_REBUILD
: {
2867 __u32 stripes_per_unit
;
2870 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2871 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2872 return migr_chunk
* stripes_per_unit
;
2874 case MIGR_STATE_CHANGE
:
2880 static int imsm_level_to_layout(int level
)
2888 return ALGORITHM_LEFT_ASYMMETRIC
;
2895 /*******************************************************************************
2896 * Function: read_imsm_migr_rec
2897 * Description: Function reads imsm migration record from last sector of disk
2899 * fd : disk descriptor
2900 * super : metadata info
2904 ******************************************************************************/
2905 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2908 unsigned int sector_size
= super
->sector_size
;
2909 unsigned long long dsize
;
2911 get_dev_size(fd
, NULL
, &dsize
);
2912 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
2914 pr_err("Cannot seek to anchor block: %s\n",
2918 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
2919 MIGR_REC_BUF_SECTORS
*sector_size
) !=
2920 MIGR_REC_BUF_SECTORS
*sector_size
) {
2921 pr_err("Cannot read migr record block: %s\n",
2926 if (sector_size
== 4096)
2927 convert_from_4k_imsm_migr_rec(super
);
2933 static struct imsm_dev
*imsm_get_device_during_migration(
2934 struct intel_super
*super
)
2937 struct intel_dev
*dv
;
2939 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2940 if (is_gen_migration(dv
->dev
))
2946 /*******************************************************************************
2947 * Function: load_imsm_migr_rec
2948 * Description: Function reads imsm migration record (it is stored at the last
2951 * super : imsm internal array info
2952 * info : general array info
2956 * -2 : no migration in progress
2957 ******************************************************************************/
2958 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2965 struct imsm_dev
*dev
;
2966 struct imsm_map
*map
;
2969 /* find map under migration */
2970 dev
= imsm_get_device_during_migration(super
);
2971 /* nothing to load,no migration in progress?
2977 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2978 /* read only from one of the first two slots */
2979 if ((sd
->disk
.raid_disk
< 0) ||
2980 (sd
->disk
.raid_disk
> 1))
2983 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2984 fd
= dev_open(nm
, O_RDONLY
);
2990 map
= get_imsm_map(dev
, MAP_0
);
2991 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2992 /* skip spare and failed disks
2996 /* read only from one of the first two slots */
2998 slot
= get_imsm_disk_slot(map
, dl
->index
);
2999 if (map
== NULL
|| slot
> 1 || slot
< 0)
3001 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3002 fd
= dev_open(nm
, O_RDONLY
);
3009 retval
= read_imsm_migr_rec(fd
, super
);
3017 /*******************************************************************************
3018 * function: imsm_create_metadata_checkpoint_update
3019 * Description: It creates update for checkpoint change.
3021 * super : imsm internal array info
3022 * u : pointer to prepared update
3025 * If length is equal to 0, input pointer u contains no update
3026 ******************************************************************************/
3027 static int imsm_create_metadata_checkpoint_update(
3028 struct intel_super
*super
,
3029 struct imsm_update_general_migration_checkpoint
**u
)
3032 int update_memory_size
= 0;
3034 dprintf("(enter)\n");
3040 /* size of all update data without anchor */
3041 update_memory_size
=
3042 sizeof(struct imsm_update_general_migration_checkpoint
);
3044 *u
= xcalloc(1, update_memory_size
);
3046 dprintf("error: cannot get memory\n");
3049 (*u
)->type
= update_general_migration_checkpoint
;
3050 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
3051 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3053 return update_memory_size
;
3056 static void imsm_update_metadata_locally(struct supertype
*st
,
3057 void *buf
, int len
);
3059 /*******************************************************************************
3060 * Function: write_imsm_migr_rec
3061 * Description: Function writes imsm migration record
3062 * (at the last sector of disk)
3064 * super : imsm internal array info
3068 ******************************************************************************/
3069 static int write_imsm_migr_rec(struct supertype
*st
)
3071 struct intel_super
*super
= st
->sb
;
3072 unsigned int sector_size
= super
->sector_size
;
3073 unsigned long long dsize
;
3079 struct imsm_update_general_migration_checkpoint
*u
;
3080 struct imsm_dev
*dev
;
3081 struct imsm_map
*map
;
3083 /* find map under migration */
3084 dev
= imsm_get_device_during_migration(super
);
3085 /* if no migration, write buffer anyway to clear migr_record
3086 * on disk based on first available device
3089 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3090 super
->current_vol
);
3092 map
= get_imsm_map(dev
, MAP_0
);
3094 if (sector_size
== 4096)
3095 convert_to_4k_imsm_migr_rec(super
);
3096 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3099 /* skip failed and spare devices */
3102 /* write to 2 first slots only */
3104 slot
= get_imsm_disk_slot(map
, sd
->index
);
3105 if (map
== NULL
|| slot
> 1 || slot
< 0)
3108 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3109 fd
= dev_open(nm
, O_RDWR
);
3112 get_dev_size(fd
, NULL
, &dsize
);
3113 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3115 pr_err("Cannot seek to anchor block: %s\n",
3119 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3120 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3121 MIGR_REC_BUF_SECTORS
*sector_size
) {
3122 pr_err("Cannot write migr record block: %s\n",
3129 if (sector_size
== 4096)
3130 convert_from_4k_imsm_migr_rec(super
);
3131 /* update checkpoint information in metadata */
3132 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3134 dprintf("imsm: Cannot prepare update\n");
3137 /* update metadata locally */
3138 imsm_update_metadata_locally(st
, u
, len
);
3139 /* and possibly remotely */
3140 if (st
->update_tail
) {
3141 append_metadata_update(st
, u
, len
);
3142 /* during reshape we do all work inside metadata handler
3143 * manage_reshape(), so metadata update has to be triggered
3146 flush_metadata_updates(st
);
3147 st
->update_tail
= &st
->updates
;
3158 /* spare/missing disks activations are not allowe when
3159 * array/container performs reshape operation, because
3160 * all arrays in container works on the same disks set
3162 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3165 struct intel_dev
*i_dev
;
3166 struct imsm_dev
*dev
;
3168 /* check whole container
3170 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3172 if (is_gen_migration(dev
)) {
3173 /* No repair during any migration in container
3181 static unsigned long long imsm_component_size_aligment_check(int level
,
3183 unsigned int sector_size
,
3184 unsigned long long component_size
)
3186 unsigned int component_size_alligment
;
3188 /* check component size aligment
3190 component_size_alligment
= component_size
% (chunk_size
/sector_size
);
3192 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
3193 level
, chunk_size
, component_size
,
3194 component_size_alligment
);
3196 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
3197 dprintf("imsm: reported component size alligned from %llu ",
3199 component_size
-= component_size_alligment
;
3200 dprintf_cont("to %llu (%i).\n",
3201 component_size
, component_size_alligment
);
3204 return component_size
;
3207 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3209 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3210 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3212 return pba_of_lba0(map
) +
3213 (num_data_stripes(map
) * map
->blocks_per_strip
);
3216 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3218 struct intel_super
*super
= st
->sb
;
3219 struct migr_record
*migr_rec
= super
->migr_rec
;
3220 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3221 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3222 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3223 struct imsm_map
*map_to_analyse
= map
;
3225 int map_disks
= info
->array
.raid_disks
;
3227 memset(info
, 0, sizeof(*info
));
3229 map_to_analyse
= prev_map
;
3231 dl
= super
->current_disk
;
3233 info
->container_member
= super
->current_vol
;
3234 info
->array
.raid_disks
= map
->num_members
;
3235 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3236 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3237 info
->array
.md_minor
= -1;
3238 info
->array
.ctime
= 0;
3239 info
->array
.utime
= 0;
3240 info
->array
.chunk_size
=
3241 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3242 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3243 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
3244 info
->custom_array_size
<<= 32;
3245 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
3246 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3248 if (is_gen_migration(dev
)) {
3249 info
->reshape_active
= 1;
3250 info
->new_level
= get_imsm_raid_level(map
);
3251 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3252 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3253 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3254 if (info
->delta_disks
) {
3255 /* this needs to be applied to every array
3258 info
->reshape_active
= CONTAINER_RESHAPE
;
3260 /* We shape information that we give to md might have to be
3261 * modify to cope with md's requirement for reshaping arrays.
3262 * For example, when reshaping a RAID0, md requires it to be
3263 * presented as a degraded RAID4.
3264 * Also if a RAID0 is migrating to a RAID5 we need to specify
3265 * the array as already being RAID5, but the 'before' layout
3266 * is a RAID4-like layout.
3268 switch (info
->array
.level
) {
3270 switch(info
->new_level
) {
3272 /* conversion is happening as RAID4 */
3273 info
->array
.level
= 4;
3274 info
->array
.raid_disks
+= 1;
3277 /* conversion is happening as RAID5 */
3278 info
->array
.level
= 5;
3279 info
->array
.layout
= ALGORITHM_PARITY_N
;
3280 info
->delta_disks
-= 1;
3283 /* FIXME error message */
3284 info
->array
.level
= UnSet
;
3290 info
->new_level
= UnSet
;
3291 info
->new_layout
= UnSet
;
3292 info
->new_chunk
= info
->array
.chunk_size
;
3293 info
->delta_disks
= 0;
3297 info
->disk
.major
= dl
->major
;
3298 info
->disk
.minor
= dl
->minor
;
3299 info
->disk
.number
= dl
->index
;
3300 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3304 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3306 if (info
->array
.level
== 5) {
3307 info
->component_size
= num_data_stripes(map_to_analyse
) *
3308 map_to_analyse
->blocks_per_strip
;
3310 info
->component_size
= blocks_per_member(map_to_analyse
);
3313 info
->component_size
= imsm_component_size_aligment_check(
3315 info
->array
.chunk_size
,
3317 info
->component_size
);
3318 info
->bb
.supported
= 1;
3320 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3321 info
->recovery_start
= MaxSector
;
3323 if (info
->array
.level
== 5 &&
3324 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3325 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3326 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3327 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3328 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3329 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3331 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3333 } else if (info
->array
.level
<= 0) {
3334 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3336 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3339 info
->reshape_progress
= 0;
3340 info
->resync_start
= MaxSector
;
3341 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3342 !(info
->array
.state
& 1)) &&
3343 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3344 info
->resync_start
= 0;
3346 if (dev
->vol
.migr_state
) {
3347 switch (migr_type(dev
)) {
3350 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3352 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3354 info
->resync_start
= blocks_per_unit
* units
;
3357 case MIGR_GEN_MIGR
: {
3358 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3360 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
3361 unsigned long long array_blocks
;
3364 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3366 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
3367 (super
->migr_rec
->rec_status
==
3368 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3371 info
->reshape_progress
= blocks_per_unit
* units
;
3373 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3374 (unsigned long long)units
,
3375 (unsigned long long)blocks_per_unit
,
3376 info
->reshape_progress
);
3378 used_disks
= imsm_num_data_members(dev
, MAP_1
);
3379 if (used_disks
> 0) {
3380 array_blocks
= blocks_per_member(map
) *
3382 info
->custom_array_size
=
3383 round_size_to_mb(array_blocks
,
3389 /* we could emulate the checkpointing of
3390 * 'sync_action=check' migrations, but for now
3391 * we just immediately complete them
3394 /* this is handled by container_content_imsm() */
3395 case MIGR_STATE_CHANGE
:
3396 /* FIXME handle other migrations */
3398 /* we are not dirty, so... */
3399 info
->resync_start
= MaxSector
;
3403 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3404 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3406 info
->array
.major_version
= -1;
3407 info
->array
.minor_version
= -2;
3408 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3409 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3410 uuid_from_super_imsm(st
, info
->uuid
);
3414 for (i
=0; i
<map_disks
; i
++) {
3416 if (i
< info
->array
.raid_disks
) {
3417 struct imsm_disk
*dsk
;
3418 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3419 dsk
= get_imsm_disk(super
, j
);
3420 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3427 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3428 int failed
, int look_in_map
);
3430 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3433 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3435 if (is_gen_migration(dev
)) {
3438 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3440 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3441 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3442 if (map2
->map_state
!= map_state
) {
3443 map2
->map_state
= map_state
;
3444 super
->updates_pending
++;
3449 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3453 for (d
= super
->missing
; d
; d
= d
->next
)
3454 if (d
->index
== index
)
3459 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3461 struct intel_super
*super
= st
->sb
;
3462 struct imsm_disk
*disk
;
3463 int map_disks
= info
->array
.raid_disks
;
3464 int max_enough
= -1;
3466 struct imsm_super
*mpb
;
3468 if (super
->current_vol
>= 0) {
3469 getinfo_super_imsm_volume(st
, info
, map
);
3472 memset(info
, 0, sizeof(*info
));
3474 /* Set raid_disks to zero so that Assemble will always pull in valid
3477 info
->array
.raid_disks
= 0;
3478 info
->array
.level
= LEVEL_CONTAINER
;
3479 info
->array
.layout
= 0;
3480 info
->array
.md_minor
= -1;
3481 info
->array
.ctime
= 0; /* N/A for imsm */
3482 info
->array
.utime
= 0;
3483 info
->array
.chunk_size
= 0;
3485 info
->disk
.major
= 0;
3486 info
->disk
.minor
= 0;
3487 info
->disk
.raid_disk
= -1;
3488 info
->reshape_active
= 0;
3489 info
->array
.major_version
= -1;
3490 info
->array
.minor_version
= -2;
3491 strcpy(info
->text_version
, "imsm");
3492 info
->safe_mode_delay
= 0;
3493 info
->disk
.number
= -1;
3494 info
->disk
.state
= 0;
3496 info
->recovery_start
= MaxSector
;
3497 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3498 info
->bb
.supported
= 1;
3500 /* do we have the all the insync disks that we expect? */
3501 mpb
= super
->anchor
;
3502 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3504 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3505 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3506 int failed
, enough
, j
, missing
= 0;
3507 struct imsm_map
*map
;
3510 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3511 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3512 map
= get_imsm_map(dev
, MAP_0
);
3514 /* any newly missing disks?
3515 * (catches single-degraded vs double-degraded)
3517 for (j
= 0; j
< map
->num_members
; j
++) {
3518 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3519 __u32 idx
= ord_to_idx(ord
);
3521 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3522 info
->disk
.raid_disk
= j
;
3524 if (!(ord
& IMSM_ORD_REBUILD
) &&
3525 get_imsm_missing(super
, idx
)) {
3531 if (state
== IMSM_T_STATE_FAILED
)
3533 else if (state
== IMSM_T_STATE_DEGRADED
&&
3534 (state
!= map
->map_state
|| missing
))
3536 else /* we're normal, or already degraded */
3538 if (is_gen_migration(dev
) && missing
) {
3539 /* during general migration we need all disks
3540 * that process is running on.
3541 * No new missing disk is allowed.
3545 /* no more checks necessary
3549 /* in the missing/failed disk case check to see
3550 * if at least one array is runnable
3552 max_enough
= max(max_enough
, enough
);
3554 dprintf("enough: %d\n", max_enough
);
3555 info
->container_enough
= max_enough
;
3558 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3560 disk
= &super
->disks
->disk
;
3561 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3562 info
->component_size
= reserved
;
3563 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3564 /* we don't change info->disk.raid_disk here because
3565 * this state will be finalized in mdmon after we have
3566 * found the 'most fresh' version of the metadata
3568 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3569 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3570 0 : (1 << MD_DISK_SYNC
);
3573 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3574 * ->compare_super may have updated the 'num_raid_devs' field for spares
3576 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3577 uuid_from_super_imsm(st
, info
->uuid
);
3579 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3581 /* I don't know how to compute 'map' on imsm, so use safe default */
3584 for (i
= 0; i
< map_disks
; i
++)
3590 /* allocates memory and fills disk in mdinfo structure
3591 * for each disk in array */
3592 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3594 struct mdinfo
*mddev
;
3595 struct intel_super
*super
= st
->sb
;
3596 struct imsm_disk
*disk
;
3599 if (!super
|| !super
->disks
)
3602 mddev
= xcalloc(1, sizeof(*mddev
));
3606 tmp
= xcalloc(1, sizeof(*tmp
));
3608 tmp
->next
= mddev
->devs
;
3610 tmp
->disk
.number
= count
++;
3611 tmp
->disk
.major
= dl
->major
;
3612 tmp
->disk
.minor
= dl
->minor
;
3613 tmp
->disk
.state
= is_configured(disk
) ?
3614 (1 << MD_DISK_ACTIVE
) : 0;
3615 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3616 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3617 tmp
->disk
.raid_disk
= -1;
3623 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3624 char *update
, char *devname
, int verbose
,
3625 int uuid_set
, char *homehost
)
3627 /* For 'assemble' and 'force' we need to return non-zero if any
3628 * change was made. For others, the return value is ignored.
3629 * Update options are:
3630 * force-one : This device looks a bit old but needs to be included,
3631 * update age info appropriately.
3632 * assemble: clear any 'faulty' flag to allow this device to
3634 * force-array: Array is degraded but being forced, mark it clean
3635 * if that will be needed to assemble it.
3637 * newdev: not used ????
3638 * grow: Array has gained a new device - this is currently for
3640 * resync: mark as dirty so a resync will happen.
3641 * name: update the name - preserving the homehost
3642 * uuid: Change the uuid of the array to match watch is given
3644 * Following are not relevant for this imsm:
3645 * sparc2.2 : update from old dodgey metadata
3646 * super-minor: change the preferred_minor number
3647 * summaries: update redundant counters.
3648 * homehost: update the recorded homehost
3649 * _reshape_progress: record new reshape_progress position.
3652 struct intel_super
*super
= st
->sb
;
3653 struct imsm_super
*mpb
;
3655 /* we can only update container info */
3656 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3659 mpb
= super
->anchor
;
3661 if (strcmp(update
, "uuid") == 0) {
3662 /* We take this to mean that the family_num should be updated.
3663 * However that is much smaller than the uuid so we cannot really
3664 * allow an explicit uuid to be given. And it is hard to reliably
3666 * So if !uuid_set we know the current uuid is random and just used
3667 * the first 'int' and copy it to the other 3 positions.
3668 * Otherwise we require the 4 'int's to be the same as would be the
3669 * case if we are using a random uuid. So an explicit uuid will be
3670 * accepted as long as all for ints are the same... which shouldn't hurt
3673 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3676 if (info
->uuid
[0] != info
->uuid
[1] ||
3677 info
->uuid
[1] != info
->uuid
[2] ||
3678 info
->uuid
[2] != info
->uuid
[3])
3684 mpb
->orig_family_num
= info
->uuid
[0];
3685 } else if (strcmp(update
, "assemble") == 0)
3690 /* successful update? recompute checksum */
3692 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3697 static size_t disks_to_mpb_size(int disks
)
3701 size
= sizeof(struct imsm_super
);
3702 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3703 size
+= 2 * sizeof(struct imsm_dev
);
3704 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3705 size
+= (4 - 2) * sizeof(struct imsm_map
);
3706 /* 4 possible disk_ord_tbl's */
3707 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3708 /* maximum bbm log */
3709 size
+= sizeof(struct bbm_log
);
3714 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3715 unsigned long long data_offset
)
3717 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3720 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3723 static void free_devlist(struct intel_super
*super
)
3725 struct intel_dev
*dv
;
3727 while (super
->devlist
) {
3728 dv
= super
->devlist
->next
;
3729 free(super
->devlist
->dev
);
3730 free(super
->devlist
);
3731 super
->devlist
= dv
;
3735 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3737 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3740 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3744 * 0 same, or first was empty, and second was copied
3745 * 1 second had wrong number
3747 * 3 wrong other info
3749 struct intel_super
*first
= st
->sb
;
3750 struct intel_super
*sec
= tst
->sb
;
3757 /* in platform dependent environment test if the disks
3758 * use the same Intel hba
3759 * If not on Intel hba at all, allow anything.
3761 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3762 if (first
->hba
->type
!= sec
->hba
->type
) {
3764 "HBAs of devices do not match %s != %s\n",
3765 get_sys_dev_type(first
->hba
->type
),
3766 get_sys_dev_type(sec
->hba
->type
));
3769 if (first
->orom
!= sec
->orom
) {
3771 "HBAs of devices do not match %s != %s\n",
3772 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3777 /* if an anchor does not have num_raid_devs set then it is a free
3780 if (first
->anchor
->num_raid_devs
> 0 &&
3781 sec
->anchor
->num_raid_devs
> 0) {
3782 /* Determine if these disks might ever have been
3783 * related. Further disambiguation can only take place
3784 * in load_super_imsm_all
3786 __u32 first_family
= first
->anchor
->orig_family_num
;
3787 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3789 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3790 MAX_SIGNATURE_LENGTH
) != 0)
3793 if (first_family
== 0)
3794 first_family
= first
->anchor
->family_num
;
3795 if (sec_family
== 0)
3796 sec_family
= sec
->anchor
->family_num
;
3798 if (first_family
!= sec_family
)
3803 /* if 'first' is a spare promote it to a populated mpb with sec's
3806 if (first
->anchor
->num_raid_devs
== 0 &&
3807 sec
->anchor
->num_raid_devs
> 0) {
3809 struct intel_dev
*dv
;
3810 struct imsm_dev
*dev
;
3812 /* we need to copy raid device info from sec if an allocation
3813 * fails here we don't associate the spare
3815 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3816 dv
= xmalloc(sizeof(*dv
));
3817 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3820 dv
->next
= first
->devlist
;
3821 first
->devlist
= dv
;
3823 if (i
< sec
->anchor
->num_raid_devs
) {
3824 /* allocation failure */
3825 free_devlist(first
);
3826 pr_err("imsm: failed to associate spare\n");
3829 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3830 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3831 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3832 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3833 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3834 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3840 static void fd2devname(int fd
, char *name
)
3844 char dname
[PATH_MAX
];
3849 if (fstat(fd
, &st
) != 0)
3851 sprintf(path
, "/sys/dev/block/%d:%d",
3852 major(st
.st_rdev
), minor(st
.st_rdev
));
3854 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3859 nm
= strrchr(dname
, '/');
3862 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3866 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3869 char *name
= fd2kname(fd
);
3874 if (strncmp(name
, "nvme", 4) != 0)
3877 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3879 return load_sys(path
, buf
, buf_len
);
3882 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3884 static int imsm_read_serial(int fd
, char *devname
,
3885 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3894 memset(buf
, 0, sizeof(buf
));
3896 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3899 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3901 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3902 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3903 fd2devname(fd
, (char *) serial
);
3909 pr_err("Failed to retrieve serial for %s\n",
3914 /* trim all whitespace and non-printable characters and convert
3917 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
3920 /* ':' is reserved for use in placeholder serial
3921 * numbers for missing disks
3932 /* truncate leading characters */
3933 if (len
> MAX_RAID_SERIAL_LEN
) {
3934 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3935 len
= MAX_RAID_SERIAL_LEN
;
3938 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3939 memcpy(serial
, dest
, len
);
3944 static int serialcmp(__u8
*s1
, __u8
*s2
)
3946 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3949 static void serialcpy(__u8
*dest
, __u8
*src
)
3951 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3954 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3958 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3959 if (serialcmp(dl
->serial
, serial
) == 0)
3965 static struct imsm_disk
*
3966 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3970 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3971 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3973 if (serialcmp(disk
->serial
, serial
) == 0) {
3984 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3986 struct imsm_disk
*disk
;
3991 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3993 rv
= imsm_read_serial(fd
, devname
, serial
);
3998 dl
= xcalloc(1, sizeof(*dl
));
4001 dl
->major
= major(stb
.st_rdev
);
4002 dl
->minor
= minor(stb
.st_rdev
);
4003 dl
->next
= super
->disks
;
4004 dl
->fd
= keep_fd
? fd
: -1;
4005 assert(super
->disks
== NULL
);
4007 serialcpy(dl
->serial
, serial
);
4010 fd2devname(fd
, name
);
4012 dl
->devname
= xstrdup(devname
);
4014 dl
->devname
= xstrdup(name
);
4016 /* look up this disk's index in the current anchor */
4017 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4020 /* only set index on disks that are a member of a
4021 * populated contianer, i.e. one with raid_devs
4023 if (is_failed(&dl
->disk
))
4025 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4032 /* When migrating map0 contains the 'destination' state while map1
4033 * contains the current state. When not migrating map0 contains the
4034 * current state. This routine assumes that map[0].map_state is set to
4035 * the current array state before being called.
4037 * Migration is indicated by one of the following states
4038 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4039 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4040 * map1state=unitialized)
4041 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4043 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4044 * map1state=degraded)
4045 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4048 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4049 __u8 to_state
, int migr_type
)
4051 struct imsm_map
*dest
;
4052 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4054 dev
->vol
.migr_state
= 1;
4055 set_migr_type(dev
, migr_type
);
4056 dev
->vol
.curr_migr_unit
= 0;
4057 dest
= get_imsm_map(dev
, MAP_1
);
4059 /* duplicate and then set the target end state in map[0] */
4060 memcpy(dest
, src
, sizeof_imsm_map(src
));
4061 if (migr_type
== MIGR_GEN_MIGR
) {
4065 for (i
= 0; i
< src
->num_members
; i
++) {
4066 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4067 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4071 if (migr_type
== MIGR_GEN_MIGR
)
4072 /* Clear migration record */
4073 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4075 src
->map_state
= to_state
;
4078 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4081 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4082 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4086 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4087 * completed in the last migration.
4089 * FIXME add support for raid-level-migration
4091 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4092 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4093 /* when final map state is other than expected
4094 * merge maps (not for migration)
4098 for (i
= 0; i
< prev
->num_members
; i
++)
4099 for (j
= 0; j
< map
->num_members
; j
++)
4100 /* during online capacity expansion
4101 * disks position can be changed
4102 * if takeover is used
4104 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4105 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4106 map
->disk_ord_tbl
[j
] |=
4107 prev
->disk_ord_tbl
[i
];
4110 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4111 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4114 dev
->vol
.migr_state
= 0;
4115 set_migr_type(dev
, 0);
4116 dev
->vol
.curr_migr_unit
= 0;
4117 map
->map_state
= map_state
;
4120 static int parse_raid_devices(struct intel_super
*super
)
4123 struct imsm_dev
*dev_new
;
4124 size_t len
, len_migr
;
4126 size_t space_needed
= 0;
4127 struct imsm_super
*mpb
= super
->anchor
;
4129 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4130 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4131 struct intel_dev
*dv
;
4133 len
= sizeof_imsm_dev(dev_iter
, 0);
4134 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4136 space_needed
+= len_migr
- len
;
4138 dv
= xmalloc(sizeof(*dv
));
4139 if (max_len
< len_migr
)
4141 if (max_len
> len_migr
)
4142 space_needed
+= max_len
- len_migr
;
4143 dev_new
= xmalloc(max_len
);
4144 imsm_copy_dev(dev_new
, dev_iter
);
4147 dv
->next
= super
->devlist
;
4148 super
->devlist
= dv
;
4151 /* ensure that super->buf is large enough when all raid devices
4154 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4157 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4158 super
->sector_size
);
4159 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4162 memcpy(buf
, super
->buf
, super
->len
);
4163 memset(buf
+ super
->len
, 0, len
- super
->len
);
4169 super
->extra_space
+= space_needed
;
4174 /*******************************************************************************
4175 * Function: check_mpb_migr_compatibility
4176 * Description: Function checks for unsupported migration features:
4177 * - migration optimization area (pba_of_lba0)
4178 * - descending reshape (ascending_migr)
4180 * super : imsm metadata information
4182 * 0 : migration is compatible
4183 * -1 : migration is not compatible
4184 ******************************************************************************/
4185 int check_mpb_migr_compatibility(struct intel_super
*super
)
4187 struct imsm_map
*map0
, *map1
;
4188 struct migr_record
*migr_rec
= super
->migr_rec
;
4191 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4192 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4195 dev_iter
->vol
.migr_state
== 1 &&
4196 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4197 /* This device is migrating */
4198 map0
= get_imsm_map(dev_iter
, MAP_0
);
4199 map1
= get_imsm_map(dev_iter
, MAP_1
);
4200 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4201 /* migration optimization area was used */
4203 if (migr_rec
->ascending_migr
== 0 &&
4204 migr_rec
->dest_depth_per_unit
> 0)
4205 /* descending reshape not supported yet */
4212 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4214 /* load_imsm_mpb - read matrix metadata
4215 * allocates super->mpb to be freed by free_imsm
4217 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4219 unsigned long long dsize
;
4220 unsigned long long sectors
;
4221 unsigned int sector_size
= super
->sector_size
;
4223 struct imsm_super
*anchor
;
4226 get_dev_size(fd
, NULL
, &dsize
);
4227 if (dsize
< 2*sector_size
) {
4229 pr_err("%s: device to small for imsm\n",
4234 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4236 pr_err("Cannot seek to anchor block on %s: %s\n",
4237 devname
, strerror(errno
));
4241 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4243 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4246 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4248 pr_err("Cannot read anchor block on %s: %s\n",
4249 devname
, strerror(errno
));
4254 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4256 pr_err("no IMSM anchor on %s\n", devname
);
4261 __free_imsm(super
, 0);
4262 /* reload capability and hba */
4264 /* capability and hba must be updated with new super allocation */
4265 find_intel_hba_capability(fd
, super
, devname
);
4266 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4267 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4269 pr_err("unable to allocate %zu byte mpb buffer\n",
4274 memcpy(super
->buf
, anchor
, sector_size
);
4276 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4279 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4280 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4281 pr_err("could not allocate migr_rec buffer\n");
4285 super
->clean_migration_record_by_mdmon
= 0;
4288 check_sum
= __gen_imsm_checksum(super
->anchor
);
4289 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4291 pr_err("IMSM checksum %x != %x on %s\n",
4293 __le32_to_cpu(super
->anchor
->check_sum
),
4301 /* read the extended mpb */
4302 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4304 pr_err("Cannot seek to extended mpb on %s: %s\n",
4305 devname
, strerror(errno
));
4309 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4310 super
->len
- sector_size
) != super
->len
- sector_size
) {
4312 pr_err("Cannot read extended mpb on %s: %s\n",
4313 devname
, strerror(errno
));
4317 check_sum
= __gen_imsm_checksum(super
->anchor
);
4318 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4320 pr_err("IMSM checksum %x != %x on %s\n",
4321 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4329 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4331 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4332 static void clear_hi(struct intel_super
*super
)
4334 struct imsm_super
*mpb
= super
->anchor
;
4336 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4338 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4339 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4340 disk
->total_blocks_hi
= 0;
4342 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4343 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4346 for (n
= 0; n
< 2; ++n
) {
4347 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4350 map
->pba_of_lba0_hi
= 0;
4351 map
->blocks_per_member_hi
= 0;
4352 map
->num_data_stripes_hi
= 0;
4358 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4362 err
= load_imsm_mpb(fd
, super
, devname
);
4365 if (super
->sector_size
== 4096)
4366 convert_from_4k(super
);
4367 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4370 err
= parse_raid_devices(super
);
4373 err
= load_bbm_log(super
);
4378 static void __free_imsm_disk(struct dl
*d
)
4390 static void free_imsm_disks(struct intel_super
*super
)
4394 while (super
->disks
) {
4396 super
->disks
= d
->next
;
4397 __free_imsm_disk(d
);
4399 while (super
->disk_mgmt_list
) {
4400 d
= super
->disk_mgmt_list
;
4401 super
->disk_mgmt_list
= d
->next
;
4402 __free_imsm_disk(d
);
4404 while (super
->missing
) {
4406 super
->missing
= d
->next
;
4407 __free_imsm_disk(d
);
4412 /* free all the pieces hanging off of a super pointer */
4413 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4415 struct intel_hba
*elem
, *next
;
4421 /* unlink capability description */
4423 if (super
->migr_rec_buf
) {
4424 free(super
->migr_rec_buf
);
4425 super
->migr_rec_buf
= NULL
;
4428 free_imsm_disks(super
);
4429 free_devlist(super
);
4433 free((void *)elem
->path
);
4439 free(super
->bbm_log
);
4443 static void free_imsm(struct intel_super
*super
)
4445 __free_imsm(super
, 1);
4446 free(super
->bb
.entries
);
4450 static void free_super_imsm(struct supertype
*st
)
4452 struct intel_super
*super
= st
->sb
;
4461 static struct intel_super
*alloc_super(void)
4463 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4465 super
->current_vol
= -1;
4466 super
->create_offset
= ~((unsigned long long) 0);
4468 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4469 sizeof(struct md_bb_entry
));
4470 if (!super
->bb
.entries
) {
4479 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4481 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4483 struct sys_dev
*hba_name
;
4486 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4491 hba_name
= find_disk_attached_hba(fd
, NULL
);
4494 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4498 rv
= attach_hba_to_super(super
, hba_name
);
4501 struct intel_hba
*hba
= super
->hba
;
4503 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4504 " but the container is assigned to Intel(R) %s %s (",
4506 get_sys_dev_type(hba_name
->type
),
4507 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4508 hba_name
->pci_id
? : "Err!",
4509 get_sys_dev_type(super
->hba
->type
),
4510 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4513 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4515 fprintf(stderr
, ", ");
4518 fprintf(stderr
, ").\n"
4519 " Mixing devices attached to different controllers is not allowed.\n");
4523 super
->orom
= find_imsm_capability(hba_name
);
4530 /* find_missing - helper routine for load_super_imsm_all that identifies
4531 * disks that have disappeared from the system. This routine relies on
4532 * the mpb being uptodate, which it is at load time.
4534 static int find_missing(struct intel_super
*super
)
4537 struct imsm_super
*mpb
= super
->anchor
;
4539 struct imsm_disk
*disk
;
4541 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4542 disk
= __get_imsm_disk(mpb
, i
);
4543 dl
= serial_to_dl(disk
->serial
, super
);
4547 dl
= xmalloc(sizeof(*dl
));
4551 dl
->devname
= xstrdup("missing");
4553 serialcpy(dl
->serial
, disk
->serial
);
4556 dl
->next
= super
->missing
;
4557 super
->missing
= dl
;
4563 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4565 struct intel_disk
*idisk
= disk_list
;
4568 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4570 idisk
= idisk
->next
;
4576 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4577 struct intel_super
*super
,
4578 struct intel_disk
**disk_list
)
4580 struct imsm_disk
*d
= &super
->disks
->disk
;
4581 struct imsm_super
*mpb
= super
->anchor
;
4584 for (i
= 0; i
< tbl_size
; i
++) {
4585 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4586 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4588 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4589 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4590 dprintf("mpb from %d:%d matches %d:%d\n",
4591 super
->disks
->major
,
4592 super
->disks
->minor
,
4593 table
[i
]->disks
->major
,
4594 table
[i
]->disks
->minor
);
4598 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4599 is_configured(d
) == is_configured(tbl_d
)) &&
4600 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4601 /* current version of the mpb is a
4602 * better candidate than the one in
4603 * super_table, but copy over "cross
4604 * generational" status
4606 struct intel_disk
*idisk
;
4608 dprintf("mpb from %d:%d replaces %d:%d\n",
4609 super
->disks
->major
,
4610 super
->disks
->minor
,
4611 table
[i
]->disks
->major
,
4612 table
[i
]->disks
->minor
);
4614 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4615 if (idisk
&& is_failed(&idisk
->disk
))
4616 tbl_d
->status
|= FAILED_DISK
;
4619 struct intel_disk
*idisk
;
4620 struct imsm_disk
*disk
;
4622 /* tbl_mpb is more up to date, but copy
4623 * over cross generational status before
4626 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4627 if (disk
&& is_failed(disk
))
4628 d
->status
|= FAILED_DISK
;
4630 idisk
= disk_list_get(d
->serial
, *disk_list
);
4633 if (disk
&& is_configured(disk
))
4634 idisk
->disk
.status
|= CONFIGURED_DISK
;
4637 dprintf("mpb from %d:%d prefer %d:%d\n",
4638 super
->disks
->major
,
4639 super
->disks
->minor
,
4640 table
[i
]->disks
->major
,
4641 table
[i
]->disks
->minor
);
4649 table
[tbl_size
++] = super
;
4653 /* update/extend the merged list of imsm_disk records */
4654 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4655 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4656 struct intel_disk
*idisk
;
4658 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4660 idisk
->disk
.status
|= disk
->status
;
4661 if (is_configured(&idisk
->disk
) ||
4662 is_failed(&idisk
->disk
))
4663 idisk
->disk
.status
&= ~(SPARE_DISK
);
4665 idisk
= xcalloc(1, sizeof(*idisk
));
4666 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4667 idisk
->disk
= *disk
;
4668 idisk
->next
= *disk_list
;
4672 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4679 static struct intel_super
*
4680 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4683 struct imsm_super
*mpb
= super
->anchor
;
4687 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4688 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4689 struct intel_disk
*idisk
;
4691 idisk
= disk_list_get(disk
->serial
, disk_list
);
4693 if (idisk
->owner
== owner
||
4694 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4697 dprintf("'%.16s' owner %d != %d\n",
4698 disk
->serial
, idisk
->owner
,
4701 dprintf("unknown disk %x [%d]: %.16s\n",
4702 __le32_to_cpu(mpb
->family_num
), i
,
4708 if (ok_count
== mpb
->num_disks
)
4713 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4715 struct intel_super
*s
;
4717 for (s
= super_list
; s
; s
= s
->next
) {
4718 if (family_num
!= s
->anchor
->family_num
)
4720 pr_err("Conflict, offlining family %#x on '%s'\n",
4721 __le32_to_cpu(family_num
), s
->disks
->devname
);
4725 static struct intel_super
*
4726 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4728 struct intel_super
*super_table
[len
];
4729 struct intel_disk
*disk_list
= NULL
;
4730 struct intel_super
*champion
, *spare
;
4731 struct intel_super
*s
, **del
;
4736 memset(super_table
, 0, sizeof(super_table
));
4737 for (s
= *super_list
; s
; s
= s
->next
)
4738 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4740 for (i
= 0; i
< tbl_size
; i
++) {
4741 struct imsm_disk
*d
;
4742 struct intel_disk
*idisk
;
4743 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4746 d
= &s
->disks
->disk
;
4748 /* 'd' must appear in merged disk list for its
4749 * configuration to be valid
4751 idisk
= disk_list_get(d
->serial
, disk_list
);
4752 if (idisk
&& idisk
->owner
== i
)
4753 s
= validate_members(s
, disk_list
, i
);
4758 dprintf("marking family: %#x from %d:%d offline\n",
4760 super_table
[i
]->disks
->major
,
4761 super_table
[i
]->disks
->minor
);
4765 /* This is where the mdadm implementation differs from the Windows
4766 * driver which has no strict concept of a container. We can only
4767 * assemble one family from a container, so when returning a prodigal
4768 * array member to this system the code will not be able to disambiguate
4769 * the container contents that should be assembled ("foreign" versus
4770 * "local"). It requires user intervention to set the orig_family_num
4771 * to a new value to establish a new container. The Windows driver in
4772 * this situation fixes up the volume name in place and manages the
4773 * foreign array as an independent entity.
4778 for (i
= 0; i
< tbl_size
; i
++) {
4779 struct intel_super
*tbl_ent
= super_table
[i
];
4785 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4790 if (s
&& !is_spare
) {
4791 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4793 } else if (!s
&& !is_spare
)
4806 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4807 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4809 /* collect all dl's onto 'champion', and update them to
4810 * champion's version of the status
4812 for (s
= *super_list
; s
; s
= s
->next
) {
4813 struct imsm_super
*mpb
= champion
->anchor
;
4814 struct dl
*dl
= s
->disks
;
4819 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4821 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4822 struct imsm_disk
*disk
;
4824 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4827 /* only set index on disks that are a member of
4828 * a populated contianer, i.e. one with
4831 if (is_failed(&dl
->disk
))
4833 else if (is_spare(&dl
->disk
))
4839 if (i
>= mpb
->num_disks
) {
4840 struct intel_disk
*idisk
;
4842 idisk
= disk_list_get(dl
->serial
, disk_list
);
4843 if (idisk
&& is_spare(&idisk
->disk
) &&
4844 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4852 dl
->next
= champion
->disks
;
4853 champion
->disks
= dl
;
4857 /* delete 'champion' from super_list */
4858 for (del
= super_list
; *del
; ) {
4859 if (*del
== champion
) {
4860 *del
= (*del
)->next
;
4863 del
= &(*del
)->next
;
4865 champion
->next
= NULL
;
4869 struct intel_disk
*idisk
= disk_list
;
4871 disk_list
= disk_list
->next
;
4879 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4880 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4881 int major
, int minor
, int keep_fd
);
4883 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4884 int *max
, int keep_fd
);
4886 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4887 char *devname
, struct md_list
*devlist
,
4890 struct intel_super
*super_list
= NULL
;
4891 struct intel_super
*super
= NULL
;
4896 /* 'fd' is an opened container */
4897 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4899 /* get super block from devlist devices */
4900 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4903 /* all mpbs enter, maybe one leaves */
4904 super
= imsm_thunderdome(&super_list
, i
);
4910 if (find_missing(super
) != 0) {
4916 /* load migration record */
4917 err
= load_imsm_migr_rec(super
, NULL
);
4919 /* migration is in progress,
4920 * but migr_rec cannot be loaded,
4926 /* Check migration compatibility */
4927 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
4928 pr_err("Unsupported migration detected");
4930 fprintf(stderr
, " on %s\n", devname
);
4932 fprintf(stderr
, " (IMSM).\n");
4941 while (super_list
) {
4942 struct intel_super
*s
= super_list
;
4944 super_list
= super_list
->next
;
4953 strcpy(st
->container_devnm
, fd2devnm(fd
));
4955 st
->container_devnm
[0] = 0;
4956 if (err
== 0 && st
->ss
== NULL
) {
4957 st
->ss
= &super_imsm
;
4958 st
->minor_version
= 0;
4959 st
->max_devs
= IMSM_MAX_DEVICES
;
4965 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4966 int *max
, int keep_fd
)
4968 struct md_list
*tmpdev
;
4972 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4973 if (tmpdev
->used
!= 1)
4975 if (tmpdev
->container
== 1) {
4977 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4979 pr_err("cannot open device %s: %s\n",
4980 tmpdev
->devname
, strerror(errno
));
4984 err
= get_sra_super_block(fd
, super_list
,
4985 tmpdev
->devname
, &lmax
,
4994 int major
= major(tmpdev
->st_rdev
);
4995 int minor
= minor(tmpdev
->st_rdev
);
4996 err
= get_super_block(super_list
,
5013 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5014 int major
, int minor
, int keep_fd
)
5016 struct intel_super
*s
;
5028 sprintf(nm
, "%d:%d", major
, minor
);
5029 dfd
= dev_open(nm
, O_RDWR
);
5035 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5036 find_intel_hba_capability(dfd
, s
, devname
);
5037 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5039 /* retry the load if we might have raced against mdmon */
5040 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5041 for (retry
= 0; retry
< 3; retry
++) {
5043 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5049 s
->next
= *super_list
;
5057 if (dfd
>= 0 && !keep_fd
)
5064 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5071 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5075 if (sra
->array
.major_version
!= -1 ||
5076 sra
->array
.minor_version
!= -2 ||
5077 strcmp(sra
->text_version
, "imsm") != 0) {
5082 devnm
= fd2devnm(fd
);
5083 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5084 if (get_super_block(super_list
, devnm
, devname
,
5085 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5096 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5098 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5101 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5103 struct intel_super
*super
;
5107 if (test_partition(fd
))
5108 /* IMSM not allowed on partitions */
5111 free_super_imsm(st
);
5113 super
= alloc_super();
5114 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5117 /* Load hba and capabilities if they exist.
5118 * But do not preclude loading metadata in case capabilities or hba are
5119 * non-compliant and ignore_hw_compat is set.
5121 rv
= find_intel_hba_capability(fd
, super
, devname
);
5122 /* no orom/efi or non-intel hba of the disk */
5123 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5125 pr_err("No OROM/EFI properties for %s\n", devname
);
5129 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5131 /* retry the load if we might have raced against mdmon */
5133 struct mdstat_ent
*mdstat
= NULL
;
5134 char *name
= fd2kname(fd
);
5137 mdstat
= mdstat_by_component(name
);
5139 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5140 for (retry
= 0; retry
< 3; retry
++) {
5142 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5148 free_mdstat(mdstat
);
5153 pr_err("Failed to load all information sections on %s\n", devname
);
5159 if (st
->ss
== NULL
) {
5160 st
->ss
= &super_imsm
;
5161 st
->minor_version
= 0;
5162 st
->max_devs
= IMSM_MAX_DEVICES
;
5165 /* load migration record */
5166 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5167 /* Check for unsupported migration features */
5168 if (check_mpb_migr_compatibility(super
) != 0) {
5169 pr_err("Unsupported migration detected");
5171 fprintf(stderr
, " on %s\n", devname
);
5173 fprintf(stderr
, " (IMSM).\n");
5181 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5183 if (info
->level
== 1)
5185 return info
->chunk_size
>> 9;
5188 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5189 unsigned long long size
)
5191 if (info
->level
== 1)
5194 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5197 static void imsm_update_version_info(struct intel_super
*super
)
5199 /* update the version and attributes */
5200 struct imsm_super
*mpb
= super
->anchor
;
5202 struct imsm_dev
*dev
;
5203 struct imsm_map
*map
;
5206 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5207 dev
= get_imsm_dev(super
, i
);
5208 map
= get_imsm_map(dev
, MAP_0
);
5209 if (__le32_to_cpu(dev
->size_high
) > 0)
5210 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5212 /* FIXME detect when an array spans a port multiplier */
5214 mpb
->attributes
|= MPB_ATTRIB_PM
;
5217 if (mpb
->num_raid_devs
> 1 ||
5218 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5219 version
= MPB_VERSION_ATTRIBS
;
5220 switch (get_imsm_raid_level(map
)) {
5221 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5222 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5223 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5224 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5227 if (map
->num_members
>= 5)
5228 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5229 else if (dev
->status
== DEV_CLONE_N_GO
)
5230 version
= MPB_VERSION_CNG
;
5231 else if (get_imsm_raid_level(map
) == 5)
5232 version
= MPB_VERSION_RAID5
;
5233 else if (map
->num_members
>= 3)
5234 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5235 else if (get_imsm_raid_level(map
) == 1)
5236 version
= MPB_VERSION_RAID1
;
5238 version
= MPB_VERSION_RAID0
;
5240 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5244 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5246 struct imsm_super
*mpb
= super
->anchor
;
5247 char *reason
= NULL
;
5250 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
5251 reason
= "must be 16 characters or less";
5253 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5254 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5256 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5257 reason
= "already exists";
5262 if (reason
&& !quiet
)
5263 pr_err("imsm volume name %s\n", reason
);
5268 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5269 struct shape
*s
, char *name
,
5270 char *homehost
, int *uuid
,
5271 long long data_offset
)
5273 /* We are creating a volume inside a pre-existing container.
5274 * so st->sb is already set.
5276 struct intel_super
*super
= st
->sb
;
5277 unsigned int sector_size
= super
->sector_size
;
5278 struct imsm_super
*mpb
= super
->anchor
;
5279 struct intel_dev
*dv
;
5280 struct imsm_dev
*dev
;
5281 struct imsm_vol
*vol
;
5282 struct imsm_map
*map
;
5283 int idx
= mpb
->num_raid_devs
;
5285 unsigned long long array_blocks
;
5286 size_t size_old
, size_new
;
5287 unsigned long long num_data_stripes
;
5288 unsigned int data_disks
;
5289 unsigned long long size_per_member
;
5291 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5292 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5296 /* ensure the mpb is large enough for the new data */
5297 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5298 size_new
= disks_to_mpb_size(info
->nr_disks
);
5299 if (size_new
> size_old
) {
5301 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5303 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5304 pr_err("could not allocate new mpb\n");
5307 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5308 MIGR_REC_BUF_SECTORS
*
5309 MAX_SECTOR_SIZE
) != 0) {
5310 pr_err("could not allocate migr_rec buffer\n");
5316 memcpy(mpb_new
, mpb
, size_old
);
5319 super
->anchor
= mpb_new
;
5320 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5321 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5322 super
->len
= size_round
;
5324 super
->current_vol
= idx
;
5326 /* handle 'failed_disks' by either:
5327 * a) create dummy disk entries in the table if this the first
5328 * volume in the array. We add them here as this is the only
5329 * opportunity to add them. add_to_super_imsm_volume()
5330 * handles the non-failed disks and continues incrementing
5332 * b) validate that 'failed_disks' matches the current number
5333 * of missing disks if the container is populated
5335 if (super
->current_vol
== 0) {
5337 for (i
= 0; i
< info
->failed_disks
; i
++) {
5338 struct imsm_disk
*disk
;
5341 disk
= __get_imsm_disk(mpb
, i
);
5342 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5343 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5344 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5345 "missing:%d", (__u8
)i
);
5347 find_missing(super
);
5352 for (d
= super
->missing
; d
; d
= d
->next
)
5354 if (info
->failed_disks
> missing
) {
5355 pr_err("unable to add 'missing' disk to container\n");
5360 if (!check_name(super
, name
, 0))
5362 dv
= xmalloc(sizeof(*dv
));
5363 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5364 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
5365 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5366 info
->layout
, info
->chunk_size
,
5367 s
->size
* BLOCKS_PER_KB
);
5368 data_disks
= get_data_disks(info
->level
, info
->layout
,
5370 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5371 size_per_member
= array_blocks
/ data_disks
;
5373 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
5374 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
5375 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5377 vol
->migr_state
= 0;
5378 set_migr_type(dev
, MIGR_INIT
);
5379 vol
->dirty
= !info
->state
;
5380 vol
->curr_migr_unit
= 0;
5381 map
= get_imsm_map(dev
, MAP_0
);
5382 set_pba_of_lba0(map
, super
->create_offset
);
5383 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5386 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5387 map
->failed_disk_num
= ~0;
5388 if (info
->level
> 0)
5389 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5390 : IMSM_T_STATE_UNINITIALIZED
);
5392 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5393 IMSM_T_STATE_NORMAL
;
5396 if (info
->level
== 1 && info
->raid_disks
> 2) {
5399 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5403 map
->raid_level
= info
->level
;
5404 if (info
->level
== 10) {
5405 map
->raid_level
= 1;
5406 map
->num_domains
= info
->raid_disks
/ 2;
5407 } else if (info
->level
== 1)
5408 map
->num_domains
= info
->raid_disks
;
5410 map
->num_domains
= 1;
5412 /* info->size is only int so use the 'size' parameter instead */
5413 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5414 num_data_stripes
/= map
->num_domains
;
5415 set_num_data_stripes(map
, num_data_stripes
);
5417 map
->num_members
= info
->raid_disks
;
5418 for (i
= 0; i
< map
->num_members
; i
++) {
5419 /* initialized in add_to_super */
5420 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5422 mpb
->num_raid_devs
++;
5423 mpb
->num_raid_devs_created
++;
5424 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5426 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5427 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5428 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5429 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5433 pr_err("imsm does not support consistency policy %s\n",
5434 map_num(consistency_policies
, s
->consistency_policy
));
5439 dv
->index
= super
->current_vol
;
5440 dv
->next
= super
->devlist
;
5441 super
->devlist
= dv
;
5443 imsm_update_version_info(super
);
5448 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5449 struct shape
*s
, char *name
,
5450 char *homehost
, int *uuid
,
5451 unsigned long long data_offset
)
5453 /* This is primarily called by Create when creating a new array.
5454 * We will then get add_to_super called for each component, and then
5455 * write_init_super called to write it out to each device.
5456 * For IMSM, Create can create on fresh devices or on a pre-existing
5458 * To create on a pre-existing array a different method will be called.
5459 * This one is just for fresh drives.
5461 struct intel_super
*super
;
5462 struct imsm_super
*mpb
;
5466 if (data_offset
!= INVALID_SECTORS
) {
5467 pr_err("data-offset not supported by imsm\n");
5472 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5476 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5478 mpb_size
= MAX_SECTOR_SIZE
;
5480 super
= alloc_super();
5482 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5487 pr_err("could not allocate superblock\n");
5490 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5491 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5492 pr_err("could not allocate migr_rec buffer\n");
5497 memset(super
->buf
, 0, mpb_size
);
5499 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5503 /* zeroing superblock */
5507 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5509 version
= (char *) mpb
->sig
;
5510 strcpy(version
, MPB_SIGNATURE
);
5511 version
+= strlen(MPB_SIGNATURE
);
5512 strcpy(version
, MPB_VERSION_RAID0
);
5517 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5519 unsigned int member_sector_size
;
5522 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5526 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5528 if (member_sector_size
!= super
->sector_size
)
5533 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5534 int fd
, char *devname
)
5536 struct intel_super
*super
= st
->sb
;
5537 struct imsm_super
*mpb
= super
->anchor
;
5538 struct imsm_disk
*_disk
;
5539 struct imsm_dev
*dev
;
5540 struct imsm_map
*map
;
5544 dev
= get_imsm_dev(super
, super
->current_vol
);
5545 map
= get_imsm_map(dev
, MAP_0
);
5547 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5548 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5554 /* we're doing autolayout so grab the pre-marked (in
5555 * validate_geometry) raid_disk
5557 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5558 if (dl
->raiddisk
== dk
->raid_disk
)
5561 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5562 if (dl
->major
== dk
->major
&&
5563 dl
->minor
== dk
->minor
)
5568 pr_err("%s is not a member of the same container\n", devname
);
5572 if (!drive_validate_sector_size(super
, dl
)) {
5573 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5577 /* add a pristine spare to the metadata */
5578 if (dl
->index
< 0) {
5579 dl
->index
= super
->anchor
->num_disks
;
5580 super
->anchor
->num_disks
++;
5582 /* Check the device has not already been added */
5583 slot
= get_imsm_disk_slot(map
, dl
->index
);
5585 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5586 pr_err("%s has been included in this array twice\n",
5590 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5591 dl
->disk
.status
= CONFIGURED_DISK
;
5593 /* update size of 'missing' disks to be at least as large as the
5594 * largest acitve member (we only have dummy missing disks when
5595 * creating the first volume)
5597 if (super
->current_vol
== 0) {
5598 for (df
= super
->missing
; df
; df
= df
->next
) {
5599 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5600 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5601 _disk
= __get_imsm_disk(mpb
, df
->index
);
5606 /* refresh unset/failed slots to point to valid 'missing' entries */
5607 for (df
= super
->missing
; df
; df
= df
->next
)
5608 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5609 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5611 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5613 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5614 if (is_gen_migration(dev
)) {
5615 struct imsm_map
*map2
= get_imsm_map(dev
,
5617 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5618 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5619 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5622 if ((unsigned)df
->index
==
5624 set_imsm_ord_tbl_ent(map2
,
5630 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5634 /* if we are creating the first raid device update the family number */
5635 if (super
->current_vol
== 0) {
5637 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5639 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5640 if (!_dev
|| !_disk
) {
5641 pr_err("BUG mpb setup error\n");
5647 sum
+= __gen_imsm_checksum(mpb
);
5648 mpb
->family_num
= __cpu_to_le32(sum
);
5649 mpb
->orig_family_num
= mpb
->family_num
;
5651 super
->current_disk
= dl
;
5656 * Function marks disk as spare and restores disk serial
5657 * in case it was previously marked as failed by takeover operation
5659 * -1 : critical error
5660 * 0 : disk is marked as spare but serial is not set
5663 int mark_spare(struct dl
*disk
)
5665 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5672 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5673 /* Restore disk serial number, because takeover marks disk
5674 * as failed and adds to serial ':0' before it becomes
5677 serialcpy(disk
->serial
, serial
);
5678 serialcpy(disk
->disk
.serial
, serial
);
5681 disk
->disk
.status
= SPARE_DISK
;
5687 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5688 int fd
, char *devname
,
5689 unsigned long long data_offset
)
5691 struct intel_super
*super
= st
->sb
;
5693 unsigned long long size
;
5694 unsigned int member_sector_size
;
5699 /* If we are on an RAID enabled platform check that the disk is
5700 * attached to the raid controller.
5701 * We do not need to test disks attachment for container based additions,
5702 * they shall be already tested when container was created/assembled.
5704 rv
= find_intel_hba_capability(fd
, super
, devname
);
5705 /* no orom/efi or non-intel hba of the disk */
5707 dprintf("capability: %p fd: %d ret: %d\n",
5708 super
->orom
, fd
, rv
);
5712 if (super
->current_vol
>= 0)
5713 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5716 dd
= xcalloc(sizeof(*dd
), 1);
5717 dd
->major
= major(stb
.st_rdev
);
5718 dd
->minor
= minor(stb
.st_rdev
);
5719 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5722 dd
->action
= DISK_ADD
;
5723 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5725 pr_err("failed to retrieve scsi serial, aborting\n");
5731 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5732 (super
->hba
->type
== SYS_DEV_VMD
))) {
5734 char *devpath
= diskfd_to_devpath(fd
);
5735 char controller_path
[PATH_MAX
];
5738 pr_err("failed to get devpath, aborting\n");
5745 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5748 if (devpath_to_vendor(controller_path
) == 0x8086) {
5750 * If Intel's NVMe drive has serial ended with
5751 * "-A","-B","-1" or "-2" it means that this is "x8"
5752 * device (double drive on single PCIe card).
5753 * User should be warned about potential data loss.
5755 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5756 /* Skip empty character at the end */
5757 if (dd
->serial
[i
] == 0)
5760 if (((dd
->serial
[i
] == 'A') ||
5761 (dd
->serial
[i
] == 'B') ||
5762 (dd
->serial
[i
] == '1') ||
5763 (dd
->serial
[i
] == '2')) &&
5764 (dd
->serial
[i
-1] == '-'))
5765 pr_err("\tThe action you are about to take may put your data at risk.\n"
5766 "\tPlease note that x8 devices may consist of two separate x4 devices "
5767 "located on a single PCIe port.\n"
5768 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5771 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5772 !imsm_orom_has_tpv_support(super
->orom
)) {
5773 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5774 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
5781 get_dev_size(fd
, NULL
, &size
);
5782 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5784 if (super
->sector_size
== 0) {
5785 /* this a first device, so sector_size is not set yet */
5786 super
->sector_size
= member_sector_size
;
5789 /* clear migr_rec when adding disk to container */
5790 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5791 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5793 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5794 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5795 MIGR_REC_BUF_SECTORS
*member_sector_size
)
5796 perror("Write migr_rec failed");
5800 serialcpy(dd
->disk
.serial
, dd
->serial
);
5801 set_total_blocks(&dd
->disk
, size
);
5802 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5803 struct imsm_super
*mpb
= super
->anchor
;
5804 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5807 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5808 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5810 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5812 if (st
->update_tail
) {
5813 dd
->next
= super
->disk_mgmt_list
;
5814 super
->disk_mgmt_list
= dd
;
5816 dd
->next
= super
->disks
;
5818 super
->updates_pending
++;
5824 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5826 struct intel_super
*super
= st
->sb
;
5829 /* remove from super works only in mdmon - for communication
5830 * manager - monitor. Check if communication memory buffer
5833 if (!st
->update_tail
) {
5834 pr_err("shall be used in mdmon context only\n");
5837 dd
= xcalloc(1, sizeof(*dd
));
5838 dd
->major
= dk
->major
;
5839 dd
->minor
= dk
->minor
;
5842 dd
->action
= DISK_REMOVE
;
5844 dd
->next
= super
->disk_mgmt_list
;
5845 super
->disk_mgmt_list
= dd
;
5850 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5853 char buf
[MAX_SECTOR_SIZE
];
5854 struct imsm_super anchor
;
5855 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5857 /* spare records have their own family number and do not have any defined raid
5860 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5862 struct imsm_super
*mpb
= super
->anchor
;
5863 struct imsm_super
*spare
= &spare_record
.anchor
;
5867 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5868 spare
->generation_num
= __cpu_to_le32(1UL);
5869 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5870 spare
->num_disks
= 1;
5871 spare
->num_raid_devs
= 0;
5872 spare
->cache_size
= mpb
->cache_size
;
5873 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5875 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5876 MPB_SIGNATURE MPB_VERSION_RAID0
);
5878 for (d
= super
->disks
; d
; d
= d
->next
) {
5882 spare
->disk
[0] = d
->disk
;
5883 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5884 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5886 if (super
->sector_size
== 4096)
5887 convert_to_4k_imsm_disk(&spare
->disk
[0]);
5889 sum
= __gen_imsm_checksum(spare
);
5890 spare
->family_num
= __cpu_to_le32(sum
);
5891 spare
->orig_family_num
= 0;
5892 sum
= __gen_imsm_checksum(spare
);
5893 spare
->check_sum
= __cpu_to_le32(sum
);
5895 if (store_imsm_mpb(d
->fd
, spare
)) {
5896 pr_err("failed for device %d:%d %s\n",
5897 d
->major
, d
->minor
, strerror(errno
));
5909 static int write_super_imsm(struct supertype
*st
, int doclose
)
5911 struct intel_super
*super
= st
->sb
;
5912 unsigned int sector_size
= super
->sector_size
;
5913 struct imsm_super
*mpb
= super
->anchor
;
5919 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5921 int clear_migration_record
= 1;
5924 /* 'generation' is incremented everytime the metadata is written */
5925 generation
= __le32_to_cpu(mpb
->generation_num
);
5927 mpb
->generation_num
= __cpu_to_le32(generation
);
5929 /* fix up cases where previous mdadm releases failed to set
5932 if (mpb
->orig_family_num
== 0)
5933 mpb
->orig_family_num
= mpb
->family_num
;
5935 for (d
= super
->disks
; d
; d
= d
->next
) {
5939 mpb
->disk
[d
->index
] = d
->disk
;
5943 for (d
= super
->missing
; d
; d
= d
->next
) {
5944 mpb
->disk
[d
->index
] = d
->disk
;
5947 mpb
->num_disks
= num_disks
;
5948 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5950 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5951 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5952 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5954 imsm_copy_dev(dev
, dev2
);
5955 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5957 if (is_gen_migration(dev2
))
5958 clear_migration_record
= 0;
5961 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
5964 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
5965 mpb
->attributes
|= MPB_ATTRIB_BBM
;
5967 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
5969 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
5970 mpb_size
+= bbm_log_size
;
5971 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5974 assert(super
->len
== 0 || mpb_size
<= super
->len
);
5977 /* recalculate checksum */
5978 sum
= __gen_imsm_checksum(mpb
);
5979 mpb
->check_sum
= __cpu_to_le32(sum
);
5981 if (super
->clean_migration_record_by_mdmon
) {
5982 clear_migration_record
= 1;
5983 super
->clean_migration_record_by_mdmon
= 0;
5985 if (clear_migration_record
)
5986 memset(super
->migr_rec_buf
, 0,
5987 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5989 if (sector_size
== 4096)
5990 convert_to_4k(super
);
5992 /* write the mpb for disks that compose raid devices */
5993 for (d
= super
->disks
; d
; d
= d
->next
) {
5994 if (d
->index
< 0 || is_failed(&d
->disk
))
5997 if (clear_migration_record
) {
5998 unsigned long long dsize
;
6000 get_dev_size(d
->fd
, NULL
, &dsize
);
6001 if (lseek64(d
->fd
, dsize
- sector_size
,
6003 if ((unsigned int)write(d
->fd
,
6004 super
->migr_rec_buf
,
6005 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6006 MIGR_REC_BUF_SECTORS
*sector_size
)
6007 perror("Write migr_rec failed");
6011 if (store_imsm_mpb(d
->fd
, mpb
))
6013 "failed for device %d:%d (fd: %d)%s\n",
6015 d
->fd
, strerror(errno
));
6024 return write_super_imsm_spares(super
, doclose
);
6029 static int create_array(struct supertype
*st
, int dev_idx
)
6032 struct imsm_update_create_array
*u
;
6033 struct intel_super
*super
= st
->sb
;
6034 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6035 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6036 struct disk_info
*inf
;
6037 struct imsm_disk
*disk
;
6040 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6041 sizeof(*inf
) * map
->num_members
;
6043 u
->type
= update_create_array
;
6044 u
->dev_idx
= dev_idx
;
6045 imsm_copy_dev(&u
->dev
, dev
);
6046 inf
= get_disk_info(u
);
6047 for (i
= 0; i
< map
->num_members
; i
++) {
6048 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6050 disk
= get_imsm_disk(super
, idx
);
6052 disk
= get_imsm_missing(super
, idx
);
6053 serialcpy(inf
[i
].serial
, disk
->serial
);
6055 append_metadata_update(st
, u
, len
);
6060 static int mgmt_disk(struct supertype
*st
)
6062 struct intel_super
*super
= st
->sb
;
6064 struct imsm_update_add_remove_disk
*u
;
6066 if (!super
->disk_mgmt_list
)
6071 u
->type
= update_add_remove_disk
;
6072 append_metadata_update(st
, u
, len
);
6077 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6079 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6081 struct ppl_header
*ppl_hdr
= buf
;
6084 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6086 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6088 perror("Failed to seek to PPL header location");
6092 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6094 perror("Write PPL header failed");
6103 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6105 struct intel_super
*super
= st
->sb
;
6107 struct ppl_header
*ppl_hdr
;
6110 /* first clear entire ppl space */
6111 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6115 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6117 pr_err("Failed to allocate PPL header buffer\n");
6121 memset(buf
, 0, PPL_HEADER_SIZE
);
6123 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6124 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6126 if (info
->mismatch_cnt
) {
6128 * We are overwriting an invalid ppl. Make one entry with wrong
6129 * checksum to prevent the kernel from skipping resync.
6131 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6132 ppl_hdr
->entries
[0].checksum
= ~0;
6135 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6141 static int is_rebuilding(struct imsm_dev
*dev
);
6143 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6144 struct mdinfo
*disk
)
6146 struct intel_super
*super
= st
->sb
;
6148 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6150 struct ppl_header
*ppl_hdr
= NULL
;
6152 struct imsm_dev
*dev
;
6155 unsigned long long ppl_offset
= 0;
6156 unsigned long long prev_gen_num
= 0;
6158 if (disk
->disk
.raid_disk
< 0)
6161 dev
= get_imsm_dev(super
, info
->container_member
);
6162 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6163 d
= get_imsm_dl_disk(super
, idx
);
6165 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6168 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6169 pr_err("Failed to allocate PPL header buffer\n");
6175 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6178 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6180 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6182 perror("Failed to seek to PPL header location");
6187 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6188 perror("Read PPL header failed");
6195 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6196 ppl_hdr
->checksum
= 0;
6198 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6199 dprintf("Wrong PPL header checksum on %s\n",
6204 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6205 /* previous was newest, it was already checked */
6209 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6210 super
->anchor
->orig_family_num
)) {
6211 dprintf("Wrong PPL header signature on %s\n",
6218 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6220 ppl_offset
+= PPL_HEADER_SIZE
;
6221 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6223 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6226 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6238 * Update metadata to use mutliple PPLs area (1MB).
6239 * This is done once for all RAID members
6241 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6242 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6244 struct mdinfo
*member_dev
;
6246 sprintf(subarray
, "%d", info
->container_member
);
6248 if (mdmon_running(st
->container_devnm
))
6249 st
->update_tail
= &st
->updates
;
6251 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6252 pr_err("Failed to update subarray %s\n",
6255 if (st
->update_tail
)
6256 flush_metadata_updates(st
);
6258 st
->ss
->sync_metadata(st
);
6259 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6260 for (member_dev
= info
->devs
; member_dev
;
6261 member_dev
= member_dev
->next
)
6262 member_dev
->ppl_size
=
6263 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6268 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6270 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6271 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6272 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6273 (is_rebuilding(dev
) &&
6274 dev
->vol
.curr_migr_unit
== 0 &&
6275 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6276 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6278 info
->mismatch_cnt
++;
6279 } else if (ret
== 0 &&
6280 ppl_hdr
->entries_count
== 0 &&
6281 is_rebuilding(dev
) &&
6282 info
->resync_start
== 0) {
6284 * The header has no entries - add a single empty entry and
6285 * rewrite the header to prevent the kernel from going into
6286 * resync after an interrupted rebuild.
6288 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6289 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6297 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6299 struct intel_super
*super
= st
->sb
;
6303 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6304 info
->array
.level
!= 5)
6307 for (d
= super
->disks
; d
; d
= d
->next
) {
6308 if (d
->index
< 0 || is_failed(&d
->disk
))
6311 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6319 static int write_init_super_imsm(struct supertype
*st
)
6321 struct intel_super
*super
= st
->sb
;
6322 int current_vol
= super
->current_vol
;
6326 getinfo_super_imsm(st
, &info
, NULL
);
6328 /* we are done with current_vol reset it to point st at the container */
6329 super
->current_vol
= -1;
6331 if (st
->update_tail
) {
6332 /* queue the recently created array / added disk
6333 * as a metadata update */
6335 /* determine if we are creating a volume or adding a disk */
6336 if (current_vol
< 0) {
6337 /* in the mgmt (add/remove) disk case we are running
6338 * in mdmon context, so don't close fd's
6342 rv
= write_init_ppl_imsm_all(st
, &info
);
6344 rv
= create_array(st
, current_vol
);
6348 for (d
= super
->disks
; d
; d
= d
->next
)
6349 Kill(d
->devname
, NULL
, 0, -1, 1);
6350 if (current_vol
>= 0)
6351 rv
= write_init_ppl_imsm_all(st
, &info
);
6353 rv
= write_super_imsm(st
, 1);
6359 static int store_super_imsm(struct supertype
*st
, int fd
)
6361 struct intel_super
*super
= st
->sb
;
6362 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6367 if (super
->sector_size
== 4096)
6368 convert_to_4k(super
);
6369 return store_imsm_mpb(fd
, mpb
);
6372 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6373 int layout
, int raiddisks
, int chunk
,
6374 unsigned long long size
,
6375 unsigned long long data_offset
,
6377 unsigned long long *freesize
,
6381 unsigned long long ldsize
;
6382 struct intel_super
*super
;
6385 if (level
!= LEVEL_CONTAINER
)
6390 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6393 pr_err("imsm: Cannot open %s: %s\n",
6394 dev
, strerror(errno
));
6397 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6402 /* capabilities retrieve could be possible
6403 * note that there is no fd for the disks in array.
6405 super
= alloc_super();
6410 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6416 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6420 fd2devname(fd
, str
);
6421 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6422 fd
, str
, super
->orom
, rv
, raiddisks
);
6424 /* no orom/efi or non-intel hba of the disk */
6431 if (raiddisks
> super
->orom
->tds
) {
6433 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6434 raiddisks
, super
->orom
->tds
);
6438 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6439 (ldsize
>> 9) >> 32 > 0) {
6441 pr_err("%s exceeds maximum platform supported size\n", dev
);
6447 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6453 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6455 const unsigned long long base_start
= e
[*idx
].start
;
6456 unsigned long long end
= base_start
+ e
[*idx
].size
;
6459 if (base_start
== end
)
6463 for (i
= *idx
; i
< num_extents
; i
++) {
6464 /* extend overlapping extents */
6465 if (e
[i
].start
>= base_start
&&
6466 e
[i
].start
<= end
) {
6469 if (e
[i
].start
+ e
[i
].size
> end
)
6470 end
= e
[i
].start
+ e
[i
].size
;
6471 } else if (e
[i
].start
> end
) {
6477 return end
- base_start
;
6480 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6482 /* build a composite disk with all known extents and generate a new
6483 * 'maxsize' given the "all disks in an array must share a common start
6484 * offset" constraint
6486 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6490 unsigned long long pos
;
6491 unsigned long long start
= 0;
6492 unsigned long long maxsize
;
6493 unsigned long reserve
;
6495 /* coalesce and sort all extents. also, check to see if we need to
6496 * reserve space between member arrays
6499 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6502 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6505 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6510 while (i
< sum_extents
) {
6511 e
[j
].start
= e
[i
].start
;
6512 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6514 if (e
[j
-1].size
== 0)
6523 unsigned long long esize
;
6525 esize
= e
[i
].start
- pos
;
6526 if (esize
>= maxsize
) {
6531 pos
= e
[i
].start
+ e
[i
].size
;
6533 } while (e
[i
-1].size
);
6539 /* FIXME assumes volume at offset 0 is the first volume in a
6542 if (start_extent
> 0)
6543 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6547 if (maxsize
< reserve
)
6550 super
->create_offset
= ~((unsigned long long) 0);
6551 if (start
+ reserve
> super
->create_offset
)
6552 return 0; /* start overflows create_offset */
6553 super
->create_offset
= start
+ reserve
;
6555 return maxsize
- reserve
;
6558 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6560 if (level
< 0 || level
== 6 || level
== 4)
6563 /* if we have an orom prevent invalid raid levels */
6566 case 0: return imsm_orom_has_raid0(orom
);
6569 return imsm_orom_has_raid1e(orom
);
6570 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6571 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6572 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6575 return 1; /* not on an Intel RAID platform so anything goes */
6581 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6582 int dpa
, int verbose
)
6584 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6585 struct mdstat_ent
*memb
;
6591 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6592 if (memb
->metadata_version
&&
6593 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6594 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6595 !is_subarray(memb
->metadata_version
+9) &&
6597 struct dev_member
*dev
= memb
->members
;
6599 while(dev
&& (fd
< 0)) {
6600 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6601 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6603 fd
= open(path
, O_RDONLY
, 0);
6604 if (num
<= 0 || fd
< 0) {
6605 pr_vrb("Cannot open %s: %s\n",
6606 dev
->name
, strerror(errno
));
6612 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6613 struct mdstat_ent
*vol
;
6614 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6615 if (vol
->active
> 0 &&
6616 vol
->metadata_version
&&
6617 is_container_member(vol
, memb
->devnm
)) {
6622 if (*devlist
&& (found
< dpa
)) {
6623 dv
= xcalloc(1, sizeof(*dv
));
6624 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6625 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6628 dv
->next
= *devlist
;
6636 free_mdstat(mdstat
);
6641 static struct md_list
*
6642 get_loop_devices(void)
6645 struct md_list
*devlist
= NULL
;
6648 for(i
= 0; i
< 12; i
++) {
6649 dv
= xcalloc(1, sizeof(*dv
));
6650 dv
->devname
= xmalloc(40);
6651 sprintf(dv
->devname
, "/dev/loop%d", i
);
6659 static struct md_list
*
6660 get_devices(const char *hba_path
)
6662 struct md_list
*devlist
= NULL
;
6669 devlist
= get_loop_devices();
6672 /* scroll through /sys/dev/block looking for devices attached to
6675 dir
= opendir("/sys/dev/block");
6676 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6681 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6683 path
= devt_to_devpath(makedev(major
, minor
));
6686 if (!path_attached_to_hba(path
, hba_path
)) {
6693 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6695 fd2devname(fd
, buf
);
6698 pr_err("cannot open device: %s\n",
6703 dv
= xcalloc(1, sizeof(*dv
));
6704 dv
->devname
= xstrdup(buf
);
6711 devlist
= devlist
->next
;
6721 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6722 int verbose
, int *found
)
6724 struct md_list
*tmpdev
;
6726 struct supertype
*st
;
6728 /* first walk the list of devices to find a consistent set
6729 * that match the criterea, if that is possible.
6730 * We flag the ones we like with 'used'.
6733 st
= match_metadata_desc_imsm("imsm");
6735 pr_vrb("cannot allocate memory for imsm supertype\n");
6739 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6740 char *devname
= tmpdev
->devname
;
6742 struct supertype
*tst
;
6744 if (tmpdev
->used
> 1)
6746 tst
= dup_super(st
);
6748 pr_vrb("cannot allocate memory for imsm supertype\n");
6751 tmpdev
->container
= 0;
6752 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6754 dprintf("cannot open device %s: %s\n",
6755 devname
, strerror(errno
));
6757 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
6759 } else if (must_be_container(dfd
)) {
6760 struct supertype
*cst
;
6761 cst
= super_by_fd(dfd
, NULL
);
6763 dprintf("cannot recognize container type %s\n",
6766 } else if (tst
->ss
!= st
->ss
) {
6767 dprintf("non-imsm container - ignore it: %s\n",
6770 } else if (!tst
->ss
->load_container
||
6771 tst
->ss
->load_container(tst
, dfd
, NULL
))
6774 tmpdev
->container
= 1;
6777 cst
->ss
->free_super(cst
);
6779 tmpdev
->st_rdev
= rdev
;
6780 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6781 dprintf("no RAID superblock on %s\n",
6784 } else if (tst
->ss
->compare_super
== NULL
) {
6785 dprintf("Cannot assemble %s metadata on %s\n",
6786 tst
->ss
->name
, devname
);
6792 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6793 /* Ignore unrecognised devices during auto-assembly */
6798 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6800 if (st
->minor_version
== -1)
6801 st
->minor_version
= tst
->minor_version
;
6803 if (memcmp(info
.uuid
, uuid_zero
,
6804 sizeof(int[4])) == 0) {
6805 /* this is a floating spare. It cannot define
6806 * an array unless there are no more arrays of
6807 * this type to be found. It can be included
6808 * in an array of this type though.
6814 if (st
->ss
!= tst
->ss
||
6815 st
->minor_version
!= tst
->minor_version
||
6816 st
->ss
->compare_super(st
, tst
) != 0) {
6817 /* Some mismatch. If exactly one array matches this host,
6818 * we can resolve on that one.
6819 * Or, if we are auto assembling, we just ignore the second
6822 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6828 dprintf("found: devname: %s\n", devname
);
6832 tst
->ss
->free_super(tst
);
6836 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6837 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6838 for (iter
= head
; iter
; iter
= iter
->next
) {
6839 dprintf("content->text_version: %s vol\n",
6840 iter
->text_version
);
6841 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6842 /* do not assemble arrays with unsupported
6844 dprintf("Cannot activate member %s.\n",
6845 iter
->text_version
);
6852 dprintf("No valid super block on device list: err: %d %p\n",
6856 dprintf("no more devices to examine\n");
6859 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6860 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6862 if (count
< tmpdev
->found
)
6865 count
-= tmpdev
->found
;
6868 if (tmpdev
->used
== 1)
6873 st
->ss
->free_super(st
);
6877 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
6880 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6882 const struct orom_entry
*entry
;
6883 struct devid_list
*dv
, *devid_list
;
6888 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
6889 if (strstr(idev
->path
, hba_path
))
6893 if (!idev
|| !idev
->dev_id
)
6896 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
6898 if (!entry
|| !entry
->devid_list
)
6901 devid_list
= entry
->devid_list
;
6902 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
6903 struct md_list
*devlist
;
6904 struct sys_dev
*device
= NULL
;
6909 device
= device_by_id_and_path(dv
->devid
, hba_path
);
6911 device
= device_by_id(dv
->devid
);
6914 hpath
= device
->path
;
6918 devlist
= get_devices(hpath
);
6919 /* if no intel devices return zero volumes */
6920 if (devlist
== NULL
)
6923 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
6925 dprintf("path: %s active arrays: %d\n", hpath
, count
);
6926 if (devlist
== NULL
)
6930 count
+= count_volumes_list(devlist
,
6934 dprintf("found %d count: %d\n", found
, count
);
6937 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
6940 struct md_list
*dv
= devlist
;
6941 devlist
= devlist
->next
;
6949 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
6953 if (hba
->type
== SYS_DEV_VMD
) {
6954 struct sys_dev
*dev
;
6957 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
6958 if (dev
->type
== SYS_DEV_VMD
)
6959 count
+= __count_volumes(dev
->path
, dpa
,
6964 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
6967 static int imsm_default_chunk(const struct imsm_orom
*orom
)
6969 /* up to 512 if the plaform supports it, otherwise the platform max.
6970 * 128 if no platform detected
6972 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
6974 return min(512, (1 << fs
));
6978 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
6979 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
6981 /* check/set platform and metadata limits/defaults */
6982 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
6983 pr_vrb("platform supports a maximum of %d disks per array\n",
6988 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6989 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
6990 pr_vrb("platform does not support raid%d with %d disk%s\n",
6991 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
6995 if (*chunk
== 0 || *chunk
== UnSet
)
6996 *chunk
= imsm_default_chunk(super
->orom
);
6998 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
6999 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7003 if (layout
!= imsm_level_to_layout(level
)) {
7005 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7006 else if (level
== 10)
7007 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7009 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7014 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7015 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7016 pr_vrb("platform does not support a volume size over 2TB\n");
7023 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7024 * FIX ME add ahci details
7026 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7027 int layout
, int raiddisks
, int *chunk
,
7028 unsigned long long size
,
7029 unsigned long long data_offset
,
7031 unsigned long long *freesize
,
7035 struct intel_super
*super
= st
->sb
;
7036 struct imsm_super
*mpb
;
7038 unsigned long long pos
= 0;
7039 unsigned long long maxsize
;
7043 /* We must have the container info already read in. */
7047 mpb
= super
->anchor
;
7049 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7050 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
7054 /* General test: make sure there is space for
7055 * 'raiddisks' device extents of size 'size' at a given
7058 unsigned long long minsize
= size
;
7059 unsigned long long start_offset
= MaxSector
;
7062 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7063 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7068 e
= get_extents(super
, dl
);
7071 unsigned long long esize
;
7072 esize
= e
[i
].start
- pos
;
7073 if (esize
>= minsize
)
7075 if (found
&& start_offset
== MaxSector
) {
7078 } else if (found
&& pos
!= start_offset
) {
7082 pos
= e
[i
].start
+ e
[i
].size
;
7084 } while (e
[i
-1].size
);
7089 if (dcnt
< raiddisks
) {
7091 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7098 /* This device must be a member of the set */
7099 if (!stat_is_blkdev(dev
, &rdev
))
7101 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7102 if (dl
->major
== (int)major(rdev
) &&
7103 dl
->minor
== (int)minor(rdev
))
7108 pr_err("%s is not in the same imsm set\n", dev
);
7110 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7111 /* If a volume is present then the current creation attempt
7112 * cannot incorporate new spares because the orom may not
7113 * understand this configuration (all member disks must be
7114 * members of each array in the container).
7116 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7117 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7119 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7120 mpb
->num_disks
!= raiddisks
) {
7121 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7125 /* retrieve the largest free space block */
7126 e
= get_extents(super
, dl
);
7131 unsigned long long esize
;
7133 esize
= e
[i
].start
- pos
;
7134 if (esize
>= maxsize
)
7136 pos
= e
[i
].start
+ e
[i
].size
;
7138 } while (e
[i
-1].size
);
7143 pr_err("unable to determine free space for: %s\n",
7147 if (maxsize
< size
) {
7149 pr_err("%s not enough space (%llu < %llu)\n",
7150 dev
, maxsize
, size
);
7154 /* count total number of extents for merge */
7156 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7158 i
+= dl
->extent_cnt
;
7160 maxsize
= merge_extents(super
, i
);
7162 if (!check_env("IMSM_NO_PLATFORM") &&
7163 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7164 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7168 if (maxsize
< size
|| maxsize
== 0) {
7171 pr_err("no free space left on device. Aborting...\n");
7173 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7179 *freesize
= maxsize
;
7182 int count
= count_volumes(super
->hba
,
7183 super
->orom
->dpa
, verbose
);
7184 if (super
->orom
->vphba
<= count
) {
7185 pr_vrb("platform does not support more than %d raid volumes.\n",
7186 super
->orom
->vphba
);
7193 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7194 unsigned long long size
, int chunk
,
7195 unsigned long long *freesize
)
7197 struct intel_super
*super
= st
->sb
;
7198 struct imsm_super
*mpb
= super
->anchor
;
7203 unsigned long long maxsize
;
7204 unsigned long long minsize
;
7208 /* find the largest common start free region of the possible disks */
7212 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7218 /* don't activate new spares if we are orom constrained
7219 * and there is already a volume active in the container
7221 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7224 e
= get_extents(super
, dl
);
7227 for (i
= 1; e
[i
-1].size
; i
++)
7235 maxsize
= merge_extents(super
, extent_cnt
);
7239 minsize
= chunk
* 2;
7241 if (cnt
< raiddisks
||
7242 (super
->orom
&& used
&& used
!= raiddisks
) ||
7243 maxsize
< minsize
||
7245 pr_err("not enough devices with space to create array.\n");
7246 return 0; /* No enough free spaces large enough */
7257 if (!check_env("IMSM_NO_PLATFORM") &&
7258 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7259 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7263 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7265 dl
->raiddisk
= cnt
++;
7269 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7274 static int reserve_space(struct supertype
*st
, int raiddisks
,
7275 unsigned long long size
, int chunk
,
7276 unsigned long long *freesize
)
7278 struct intel_super
*super
= st
->sb
;
7283 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7286 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7288 dl
->raiddisk
= cnt
++;
7295 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7296 int raiddisks
, int *chunk
, unsigned long long size
,
7297 unsigned long long data_offset
,
7298 char *dev
, unsigned long long *freesize
,
7299 int consistency_policy
, int verbose
)
7306 * if given unused devices create a container
7307 * if given given devices in a container create a member volume
7309 if (level
== LEVEL_CONTAINER
) {
7310 /* Must be a fresh device to add to a container */
7311 return validate_geometry_imsm_container(st
, level
, layout
,
7321 struct intel_super
*super
= st
->sb
;
7322 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7323 raiddisks
, chunk
, size
,
7326 /* we are being asked to automatically layout a
7327 * new volume based on the current contents of
7328 * the container. If the the parameters can be
7329 * satisfied reserve_space will record the disks,
7330 * start offset, and size of the volume to be
7331 * created. add_to_super and getinfo_super
7332 * detect when autolayout is in progress.
7334 /* assuming that freesize is always given when array is
7336 if (super
->orom
&& freesize
) {
7338 count
= count_volumes(super
->hba
,
7339 super
->orom
->dpa
, verbose
);
7340 if (super
->orom
->vphba
<= count
) {
7341 pr_vrb("platform does not support more than %d raid volumes.\n",
7342 super
->orom
->vphba
);
7347 return reserve_space(st
, raiddisks
, size
,
7353 /* creating in a given container */
7354 return validate_geometry_imsm_volume(st
, level
, layout
,
7355 raiddisks
, chunk
, size
,
7357 dev
, freesize
, verbose
);
7360 /* This device needs to be a device in an 'imsm' container */
7361 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7364 pr_err("Cannot create this array on device %s\n",
7369 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7371 pr_err("Cannot open %s: %s\n",
7372 dev
, strerror(errno
));
7375 /* Well, it is in use by someone, maybe an 'imsm' container. */
7376 cfd
= open_container(fd
);
7380 pr_err("Cannot use %s: It is busy\n",
7384 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7385 if (sra
&& sra
->array
.major_version
== -1 &&
7386 strcmp(sra
->text_version
, "imsm") == 0)
7390 /* This is a member of a imsm container. Load the container
7391 * and try to create a volume
7393 struct intel_super
*super
;
7395 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7397 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7399 return validate_geometry_imsm_volume(st
, level
, layout
,
7401 size
, data_offset
, dev
,
7408 pr_err("failed container membership check\n");
7414 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7416 struct intel_super
*super
= st
->sb
;
7418 if (level
&& *level
== UnSet
)
7419 *level
= LEVEL_CONTAINER
;
7421 if (level
&& layout
&& *layout
== UnSet
)
7422 *layout
= imsm_level_to_layout(*level
);
7424 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7425 *chunk
= imsm_default_chunk(super
->orom
);
7428 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7430 static int kill_subarray_imsm(struct supertype
*st
)
7432 /* remove the subarray currently referenced by ->current_vol */
7434 struct intel_dev
**dp
;
7435 struct intel_super
*super
= st
->sb
;
7436 __u8 current_vol
= super
->current_vol
;
7437 struct imsm_super
*mpb
= super
->anchor
;
7439 if (super
->current_vol
< 0)
7441 super
->current_vol
= -1; /* invalidate subarray cursor */
7443 /* block deletions that would change the uuid of active subarrays
7445 * FIXME when immutable ids are available, but note that we'll
7446 * also need to fixup the invalidated/active subarray indexes in
7449 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7452 if (i
< current_vol
)
7454 sprintf(subarray
, "%u", i
);
7455 if (is_subarray_active(subarray
, st
->devnm
)) {
7456 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7463 if (st
->update_tail
) {
7464 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7466 u
->type
= update_kill_array
;
7467 u
->dev_idx
= current_vol
;
7468 append_metadata_update(st
, u
, sizeof(*u
));
7473 for (dp
= &super
->devlist
; *dp
;)
7474 if ((*dp
)->index
== current_vol
) {
7477 handle_missing(super
, (*dp
)->dev
);
7478 if ((*dp
)->index
> current_vol
)
7483 /* no more raid devices, all active components are now spares,
7484 * but of course failed are still failed
7486 if (--mpb
->num_raid_devs
== 0) {
7489 for (d
= super
->disks
; d
; d
= d
->next
)
7494 super
->updates_pending
++;
7499 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7500 char *update
, struct mddev_ident
*ident
)
7502 /* update the subarray currently referenced by ->current_vol */
7503 struct intel_super
*super
= st
->sb
;
7504 struct imsm_super
*mpb
= super
->anchor
;
7506 if (strcmp(update
, "name") == 0) {
7507 char *name
= ident
->name
;
7511 if (is_subarray_active(subarray
, st
->devnm
)) {
7512 pr_err("Unable to update name of active subarray\n");
7516 if (!check_name(super
, name
, 0))
7519 vol
= strtoul(subarray
, &ep
, 10);
7520 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7523 if (st
->update_tail
) {
7524 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7526 u
->type
= update_rename_array
;
7528 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7529 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7530 append_metadata_update(st
, u
, sizeof(*u
));
7532 struct imsm_dev
*dev
;
7535 dev
= get_imsm_dev(super
, vol
);
7536 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
7537 dev
->volume
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7538 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7539 dev
= get_imsm_dev(super
, i
);
7540 handle_missing(super
, dev
);
7542 super
->updates_pending
++;
7544 } else if (strcmp(update
, "ppl") == 0 ||
7545 strcmp(update
, "no-ppl") == 0) {
7548 int vol
= strtoul(subarray
, &ep
, 10);
7550 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7553 if (strcmp(update
, "ppl") == 0)
7554 new_policy
= RWH_MULTIPLE_DISTRIBUTED
;
7556 new_policy
= RWH_MULTIPLE_OFF
;
7558 if (st
->update_tail
) {
7559 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7561 u
->type
= update_rwh_policy
;
7563 u
->new_policy
= new_policy
;
7564 append_metadata_update(st
, u
, sizeof(*u
));
7566 struct imsm_dev
*dev
;
7568 dev
= get_imsm_dev(super
, vol
);
7569 dev
->rwh_policy
= new_policy
;
7570 super
->updates_pending
++;
7578 static int is_gen_migration(struct imsm_dev
*dev
)
7583 if (!dev
->vol
.migr_state
)
7586 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7592 static int is_rebuilding(struct imsm_dev
*dev
)
7594 struct imsm_map
*migr_map
;
7596 if (!dev
->vol
.migr_state
)
7599 if (migr_type(dev
) != MIGR_REBUILD
)
7602 migr_map
= get_imsm_map(dev
, MAP_1
);
7604 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7610 static int is_initializing(struct imsm_dev
*dev
)
7612 struct imsm_map
*migr_map
;
7614 if (!dev
->vol
.migr_state
)
7617 if (migr_type(dev
) != MIGR_INIT
)
7620 migr_map
= get_imsm_map(dev
, MAP_1
);
7622 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7628 static void update_recovery_start(struct intel_super
*super
,
7629 struct imsm_dev
*dev
,
7630 struct mdinfo
*array
)
7632 struct mdinfo
*rebuild
= NULL
;
7636 if (!is_rebuilding(dev
))
7639 /* Find the rebuild target, but punt on the dual rebuild case */
7640 for (d
= array
->devs
; d
; d
= d
->next
)
7641 if (d
->recovery_start
== 0) {
7648 /* (?) none of the disks are marked with
7649 * IMSM_ORD_REBUILD, so assume they are missing and the
7650 * disk_ord_tbl was not correctly updated
7652 dprintf("failed to locate out-of-sync disk\n");
7656 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7657 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7660 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7662 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7664 /* Given a container loaded by load_super_imsm_all,
7665 * extract information about all the arrays into
7667 * If 'subarray' is given, just extract info about that array.
7669 * For each imsm_dev create an mdinfo, fill it in,
7670 * then look for matching devices in super->disks
7671 * and create appropriate device mdinfo.
7673 struct intel_super
*super
= st
->sb
;
7674 struct imsm_super
*mpb
= super
->anchor
;
7675 struct mdinfo
*rest
= NULL
;
7679 int spare_disks
= 0;
7681 /* do not assemble arrays when not all attributes are supported */
7682 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7684 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7687 /* count spare devices, not used in maps
7689 for (d
= super
->disks
; d
; d
= d
->next
)
7693 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7694 struct imsm_dev
*dev
;
7695 struct imsm_map
*map
;
7696 struct imsm_map
*map2
;
7697 struct mdinfo
*this;
7704 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7707 dev
= get_imsm_dev(super
, i
);
7708 map
= get_imsm_map(dev
, MAP_0
);
7709 map2
= get_imsm_map(dev
, MAP_1
);
7710 level
= get_imsm_raid_level(map
);
7712 /* do not publish arrays that are in the middle of an
7713 * unsupported migration
7715 if (dev
->vol
.migr_state
&&
7716 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7717 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7721 /* do not publish arrays that are not support by controller's
7725 this = xmalloc(sizeof(*this));
7727 super
->current_vol
= i
;
7728 getinfo_super_imsm_volume(st
, this, NULL
);
7730 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7731 /* mdadm does not support all metadata features- set the bit in all arrays state */
7732 if (!validate_geometry_imsm_orom(super
,
7733 level
, /* RAID level */
7734 imsm_level_to_layout(level
),
7735 map
->num_members
, /* raid disks */
7736 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
7738 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7740 this->array
.state
|=
7741 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7742 (1<<MD_SB_BLOCK_VOLUME
);
7745 /* if array has bad blocks, set suitable bit in all arrays state */
7747 this->array
.state
|=
7748 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7749 (1<<MD_SB_BLOCK_VOLUME
);
7751 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7752 unsigned long long recovery_start
;
7753 struct mdinfo
*info_d
;
7761 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7762 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7763 for (d
= super
->disks
; d
; d
= d
->next
)
7764 if (d
->index
== idx
)
7767 recovery_start
= MaxSector
;
7770 if (d
&& is_failed(&d
->disk
))
7772 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
7776 * if we skip some disks the array will be assmebled degraded;
7777 * reset resync start to avoid a dirty-degraded
7778 * situation when performing the intial sync
7783 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
7784 if ((!able_to_resync(level
, missing
) ||
7785 recovery_start
== 0))
7786 this->resync_start
= MaxSector
;
7789 * FIXME handle dirty degraded
7796 info_d
= xcalloc(1, sizeof(*info_d
));
7797 info_d
->next
= this->devs
;
7798 this->devs
= info_d
;
7800 info_d
->disk
.number
= d
->index
;
7801 info_d
->disk
.major
= d
->major
;
7802 info_d
->disk
.minor
= d
->minor
;
7803 info_d
->disk
.raid_disk
= slot
;
7804 info_d
->recovery_start
= recovery_start
;
7806 if (slot
< map2
->num_members
)
7807 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7809 this->array
.spare_disks
++;
7811 if (slot
< map
->num_members
)
7812 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7814 this->array
.spare_disks
++;
7816 if (info_d
->recovery_start
== MaxSector
)
7817 this->array
.working_disks
++;
7819 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7820 info_d
->data_offset
= pba_of_lba0(map
);
7822 if (map
->raid_level
== 5) {
7823 info_d
->component_size
=
7824 num_data_stripes(map
) *
7825 map
->blocks_per_strip
;
7826 info_d
->ppl_sector
= this->ppl_sector
;
7827 info_d
->ppl_size
= this->ppl_size
;
7828 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
7829 recovery_start
== 0)
7830 this->resync_start
= 0;
7832 info_d
->component_size
= blocks_per_member(map
);
7835 info_d
->bb
.supported
= 1;
7836 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7837 info_d
->data_offset
,
7838 info_d
->component_size
,
7841 /* now that the disk list is up-to-date fixup recovery_start */
7842 update_recovery_start(super
, dev
, this);
7843 this->array
.spare_disks
+= spare_disks
;
7845 /* check for reshape */
7846 if (this->reshape_active
== 1)
7847 recover_backup_imsm(st
, this);
7854 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7855 int failed
, int look_in_map
)
7857 struct imsm_map
*map
;
7859 map
= get_imsm_map(dev
, look_in_map
);
7862 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
7863 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
7865 switch (get_imsm_raid_level(map
)) {
7867 return IMSM_T_STATE_FAILED
;
7870 if (failed
< map
->num_members
)
7871 return IMSM_T_STATE_DEGRADED
;
7873 return IMSM_T_STATE_FAILED
;
7878 * check to see if any mirrors have failed, otherwise we
7879 * are degraded. Even numbered slots are mirrored on
7883 /* gcc -Os complains that this is unused */
7884 int insync
= insync
;
7886 for (i
= 0; i
< map
->num_members
; i
++) {
7887 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
7888 int idx
= ord_to_idx(ord
);
7889 struct imsm_disk
*disk
;
7891 /* reset the potential in-sync count on even-numbered
7892 * slots. num_copies is always 2 for imsm raid10
7897 disk
= get_imsm_disk(super
, idx
);
7898 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7901 /* no in-sync disks left in this mirror the
7905 return IMSM_T_STATE_FAILED
;
7908 return IMSM_T_STATE_DEGRADED
;
7912 return IMSM_T_STATE_DEGRADED
;
7914 return IMSM_T_STATE_FAILED
;
7920 return map
->map_state
;
7923 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
7928 struct imsm_disk
*disk
;
7929 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7930 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
7931 struct imsm_map
*map_for_loop
;
7936 /* at the beginning of migration we set IMSM_ORD_REBUILD on
7937 * disks that are being rebuilt. New failures are recorded to
7938 * map[0]. So we look through all the disks we started with and
7939 * see if any failures are still present, or if any new ones
7943 if (prev
&& (map
->num_members
< prev
->num_members
))
7944 map_for_loop
= prev
;
7946 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
7948 /* when MAP_X is passed both maps failures are counted
7951 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
7952 i
< prev
->num_members
) {
7953 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
7954 idx_1
= ord_to_idx(ord
);
7956 disk
= get_imsm_disk(super
, idx_1
);
7957 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7960 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
7961 i
< map
->num_members
) {
7962 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
7963 idx
= ord_to_idx(ord
);
7966 disk
= get_imsm_disk(super
, idx
);
7967 if (!disk
|| is_failed(disk
) ||
7968 ord
& IMSM_ORD_REBUILD
)
7977 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
7980 struct intel_super
*super
= c
->sb
;
7981 struct imsm_super
*mpb
= super
->anchor
;
7982 struct imsm_update_prealloc_bb_mem u
;
7984 if (atoi(inst
) >= mpb
->num_raid_devs
) {
7985 pr_err("subarry index %d, out of range\n", atoi(inst
));
7989 dprintf("imsm: open_new %s\n", inst
);
7990 a
->info
.container_member
= atoi(inst
);
7992 u
.type
= update_prealloc_badblocks_mem
;
7993 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
7998 static int is_resyncing(struct imsm_dev
*dev
)
8000 struct imsm_map
*migr_map
;
8002 if (!dev
->vol
.migr_state
)
8005 if (migr_type(dev
) == MIGR_INIT
||
8006 migr_type(dev
) == MIGR_REPAIR
)
8009 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8012 migr_map
= get_imsm_map(dev
, MAP_1
);
8014 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8015 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8021 /* return true if we recorded new information */
8022 static int mark_failure(struct intel_super
*super
,
8023 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8027 struct imsm_map
*map
;
8028 char buf
[MAX_RAID_SERIAL_LEN
+3];
8029 unsigned int len
, shift
= 0;
8031 /* new failures are always set in map[0] */
8032 map
= get_imsm_map(dev
, MAP_0
);
8034 slot
= get_imsm_disk_slot(map
, idx
);
8038 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8039 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8042 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8043 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8045 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8046 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8047 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
8049 disk
->status
|= FAILED_DISK
;
8050 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8051 /* mark failures in second map if second map exists and this disk
8053 * This is valid for migration, initialization and rebuild
8055 if (dev
->vol
.migr_state
) {
8056 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8057 int slot2
= get_imsm_disk_slot(map2
, idx
);
8059 if (slot2
< map2
->num_members
&& slot2
>= 0)
8060 set_imsm_ord_tbl_ent(map2
, slot2
,
8061 idx
| IMSM_ORD_REBUILD
);
8063 if (map
->failed_disk_num
== 0xff)
8064 map
->failed_disk_num
= slot
;
8066 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8071 static void mark_missing(struct intel_super
*super
,
8072 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8074 mark_failure(super
, dev
, disk
, idx
);
8076 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8079 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8080 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8083 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8087 if (!super
->missing
)
8090 /* When orom adds replacement for missing disk it does
8091 * not remove entry of missing disk, but just updates map with
8092 * new added disk. So it is not enough just to test if there is
8093 * any missing disk, we have to look if there are any failed disks
8094 * in map to stop migration */
8096 dprintf("imsm: mark missing\n");
8097 /* end process for initialization and rebuild only
8099 if (is_gen_migration(dev
) == 0) {
8100 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8104 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8105 struct imsm_map
*map1
;
8106 int i
, ord
, ord_map1
;
8109 for (i
= 0; i
< map
->num_members
; i
++) {
8110 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8111 if (!(ord
& IMSM_ORD_REBUILD
))
8114 map1
= get_imsm_map(dev
, MAP_1
);
8118 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8119 if (ord_map1
& IMSM_ORD_REBUILD
)
8124 map_state
= imsm_check_degraded(super
, dev
,
8126 end_migration(dev
, super
, map_state
);
8130 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8131 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8132 super
->updates_pending
++;
8135 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8138 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8139 unsigned long long array_blocks
;
8140 struct imsm_map
*map
;
8142 if (used_disks
== 0) {
8143 /* when problems occures
8144 * return current array_blocks value
8146 array_blocks
= __le32_to_cpu(dev
->size_high
);
8147 array_blocks
= array_blocks
<< 32;
8148 array_blocks
+= __le32_to_cpu(dev
->size_low
);
8150 return array_blocks
;
8153 /* set array size in metadata
8155 if (new_size
<= 0) {
8156 /* OLCE size change is caused by added disks
8158 map
= get_imsm_map(dev
, MAP_0
);
8159 array_blocks
= blocks_per_member(map
) * used_disks
;
8161 /* Online Volume Size Change
8162 * Using available free space
8164 array_blocks
= new_size
;
8167 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8168 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
8169 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
8171 return array_blocks
;
8174 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8176 static void imsm_progress_container_reshape(struct intel_super
*super
)
8178 /* if no device has a migr_state, but some device has a
8179 * different number of members than the previous device, start
8180 * changing the number of devices in this device to match
8183 struct imsm_super
*mpb
= super
->anchor
;
8184 int prev_disks
= -1;
8188 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8189 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8190 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8191 struct imsm_map
*map2
;
8192 int prev_num_members
;
8194 if (dev
->vol
.migr_state
)
8197 if (prev_disks
== -1)
8198 prev_disks
= map
->num_members
;
8199 if (prev_disks
== map
->num_members
)
8202 /* OK, this array needs to enter reshape mode.
8203 * i.e it needs a migr_state
8206 copy_map_size
= sizeof_imsm_map(map
);
8207 prev_num_members
= map
->num_members
;
8208 map
->num_members
= prev_disks
;
8209 dev
->vol
.migr_state
= 1;
8210 dev
->vol
.curr_migr_unit
= 0;
8211 set_migr_type(dev
, MIGR_GEN_MIGR
);
8212 for (i
= prev_num_members
;
8213 i
< map
->num_members
; i
++)
8214 set_imsm_ord_tbl_ent(map
, i
, i
);
8215 map2
= get_imsm_map(dev
, MAP_1
);
8216 /* Copy the current map */
8217 memcpy(map2
, map
, copy_map_size
);
8218 map2
->num_members
= prev_num_members
;
8220 imsm_set_array_size(dev
, -1);
8221 super
->clean_migration_record_by_mdmon
= 1;
8222 super
->updates_pending
++;
8226 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8227 * states are handled in imsm_set_disk() with one exception, when a
8228 * resync is stopped due to a new failure this routine will set the
8229 * 'degraded' state for the array.
8231 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8233 int inst
= a
->info
.container_member
;
8234 struct intel_super
*super
= a
->container
->sb
;
8235 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8236 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8237 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8238 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8239 __u32 blocks_per_unit
;
8241 if (dev
->vol
.migr_state
&&
8242 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8243 /* array state change is blocked due to reshape action
8245 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8246 * - finish the reshape (if last_checkpoint is big and action != reshape)
8247 * - update curr_migr_unit
8249 if (a
->curr_action
== reshape
) {
8250 /* still reshaping, maybe update curr_migr_unit */
8251 goto mark_checkpoint
;
8253 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8254 /* for some reason we aborted the reshape.
8256 * disable automatic metadata rollback
8257 * user action is required to recover process
8260 struct imsm_map
*map2
=
8261 get_imsm_map(dev
, MAP_1
);
8262 dev
->vol
.migr_state
= 0;
8263 set_migr_type(dev
, 0);
8264 dev
->vol
.curr_migr_unit
= 0;
8266 sizeof_imsm_map(map2
));
8267 super
->updates_pending
++;
8270 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8271 unsigned long long array_blocks
;
8275 used_disks
= imsm_num_data_members(dev
, MAP_0
);
8276 if (used_disks
> 0) {
8278 blocks_per_member(map
) *
8281 round_size_to_mb(array_blocks
,
8283 a
->info
.custom_array_size
= array_blocks
;
8284 /* encourage manager to update array
8288 a
->check_reshape
= 1;
8290 /* finalize online capacity expansion/reshape */
8291 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8293 mdi
->disk
.raid_disk
,
8296 imsm_progress_container_reshape(super
);
8301 /* before we activate this array handle any missing disks */
8302 if (consistent
== 2)
8303 handle_missing(super
, dev
);
8305 if (consistent
== 2 &&
8306 (!is_resync_complete(&a
->info
) ||
8307 map_state
!= IMSM_T_STATE_NORMAL
||
8308 dev
->vol
.migr_state
))
8311 if (is_resync_complete(&a
->info
)) {
8312 /* complete intialization / resync,
8313 * recovery and interrupted recovery is completed in
8316 if (is_resyncing(dev
)) {
8317 dprintf("imsm: mark resync done\n");
8318 end_migration(dev
, super
, map_state
);
8319 super
->updates_pending
++;
8320 a
->last_checkpoint
= 0;
8322 } else if ((!is_resyncing(dev
) && !failed
) &&
8323 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8324 /* mark the start of the init process if nothing is failed */
8325 dprintf("imsm: mark resync start\n");
8326 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8327 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8329 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8330 super
->updates_pending
++;
8334 /* skip checkpointing for general migration,
8335 * it is controlled in mdadm
8337 if (is_gen_migration(dev
))
8338 goto skip_mark_checkpoint
;
8340 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8341 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8342 if (blocks_per_unit
) {
8346 units
= a
->last_checkpoint
/ blocks_per_unit
;
8349 /* check that we did not overflow 32-bits, and that
8350 * curr_migr_unit needs updating
8352 if (units32
== units
&&
8354 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8355 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8356 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8357 super
->updates_pending
++;
8361 skip_mark_checkpoint
:
8362 /* mark dirty / clean */
8363 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8364 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8365 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8367 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8369 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8370 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8371 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8372 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8374 super
->updates_pending
++;
8380 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8382 int inst
= a
->info
.container_member
;
8383 struct intel_super
*super
= a
->container
->sb
;
8384 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8385 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8387 if (slot
> map
->num_members
) {
8388 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8389 slot
, map
->num_members
- 1);
8396 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8399 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8401 int inst
= a
->info
.container_member
;
8402 struct intel_super
*super
= a
->container
->sb
;
8403 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8404 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8405 struct imsm_disk
*disk
;
8407 int recovery_not_finished
= 0;
8411 int rebuild_done
= 0;
8414 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8418 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8419 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8421 /* check for new failures */
8422 if (state
& DS_FAULTY
) {
8423 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8424 super
->updates_pending
++;
8427 /* check if in_sync */
8428 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8429 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8431 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8433 super
->updates_pending
++;
8436 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8437 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8439 /* check if recovery complete, newly degraded, or failed */
8440 dprintf("imsm: Detected transition to state ");
8441 switch (map_state
) {
8442 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8443 dprintf("normal: ");
8444 if (is_rebuilding(dev
)) {
8445 dprintf_cont("while rebuilding");
8446 /* check if recovery is really finished */
8447 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8448 if (mdi
->recovery_start
!= MaxSector
) {
8449 recovery_not_finished
= 1;
8452 if (recovery_not_finished
) {
8454 dprintf("Rebuild has not finished yet, state not changed");
8455 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8456 a
->last_checkpoint
= mdi
->recovery_start
;
8457 super
->updates_pending
++;
8461 end_migration(dev
, super
, map_state
);
8462 map
= get_imsm_map(dev
, MAP_0
);
8463 map
->failed_disk_num
= ~0;
8464 super
->updates_pending
++;
8465 a
->last_checkpoint
= 0;
8468 if (is_gen_migration(dev
)) {
8469 dprintf_cont("while general migration");
8470 if (a
->last_checkpoint
>= a
->info
.component_size
)
8471 end_migration(dev
, super
, map_state
);
8473 map
->map_state
= map_state
;
8474 map
= get_imsm_map(dev
, MAP_0
);
8475 map
->failed_disk_num
= ~0;
8476 super
->updates_pending
++;
8480 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8481 dprintf_cont("degraded: ");
8482 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8483 dprintf_cont("mark degraded");
8484 map
->map_state
= map_state
;
8485 super
->updates_pending
++;
8486 a
->last_checkpoint
= 0;
8489 if (is_rebuilding(dev
)) {
8490 dprintf_cont("while rebuilding.");
8491 if (map
->map_state
!= map_state
) {
8492 dprintf_cont(" Map state change");
8493 end_migration(dev
, super
, map_state
);
8494 super
->updates_pending
++;
8495 } else if (!rebuild_done
) {
8499 /* check if recovery is really finished */
8500 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8501 if (mdi
->recovery_start
!= MaxSector
) {
8502 recovery_not_finished
= 1;
8505 if (recovery_not_finished
) {
8507 dprintf("Rebuild has not finished yet, state not changed");
8508 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8509 a
->last_checkpoint
=
8510 mdi
->recovery_start
;
8511 super
->updates_pending
++;
8516 dprintf_cont(" Rebuild done, still degraded");
8517 dev
->vol
.migr_state
= 0;
8518 set_migr_type(dev
, 0);
8519 dev
->vol
.curr_migr_unit
= 0;
8521 for (i
= 0; i
< map
->num_members
; i
++) {
8522 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8524 if (idx
& IMSM_ORD_REBUILD
)
8525 map
->failed_disk_num
= i
;
8527 super
->updates_pending
++;
8530 if (is_gen_migration(dev
)) {
8531 dprintf_cont("while general migration");
8532 if (a
->last_checkpoint
>= a
->info
.component_size
)
8533 end_migration(dev
, super
, map_state
);
8535 map
->map_state
= map_state
;
8536 manage_second_map(super
, dev
);
8538 super
->updates_pending
++;
8541 if (is_initializing(dev
)) {
8542 dprintf_cont("while initialization.");
8543 map
->map_state
= map_state
;
8544 super
->updates_pending
++;
8548 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8549 dprintf_cont("failed: ");
8550 if (is_gen_migration(dev
)) {
8551 dprintf_cont("while general migration");
8552 map
->map_state
= map_state
;
8553 super
->updates_pending
++;
8556 if (map
->map_state
!= map_state
) {
8557 dprintf_cont("mark failed");
8558 end_migration(dev
, super
, map_state
);
8559 super
->updates_pending
++;
8560 a
->last_checkpoint
= 0;
8565 dprintf_cont("state %i\n", map_state
);
8570 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8573 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8574 unsigned long long dsize
;
8575 unsigned long long sectors
;
8576 unsigned int sector_size
;
8578 get_dev_sector_size(fd
, NULL
, §or_size
);
8579 get_dev_size(fd
, NULL
, &dsize
);
8581 if (mpb_size
> sector_size
) {
8582 /* -1 to account for anchor */
8583 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8585 /* write the extended mpb to the sectors preceeding the anchor */
8586 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8590 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8591 sector_size
* sectors
) != sector_size
* sectors
)
8595 /* first block is stored on second to last sector of the disk */
8596 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8599 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8605 static void imsm_sync_metadata(struct supertype
*container
)
8607 struct intel_super
*super
= container
->sb
;
8609 dprintf("sync metadata: %d\n", super
->updates_pending
);
8610 if (!super
->updates_pending
)
8613 write_super_imsm(container
, 0);
8615 super
->updates_pending
= 0;
8618 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8620 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8621 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8624 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8628 if (dl
&& is_failed(&dl
->disk
))
8632 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8637 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8638 struct active_array
*a
, int activate_new
,
8639 struct mdinfo
*additional_test_list
)
8641 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8642 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8643 struct imsm_super
*mpb
= super
->anchor
;
8644 struct imsm_map
*map
;
8645 unsigned long long pos
;
8650 __u32 array_start
= 0;
8651 __u32 array_end
= 0;
8653 struct mdinfo
*test_list
;
8655 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8656 /* If in this array, skip */
8657 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8658 if (d
->state_fd
>= 0 &&
8659 d
->disk
.major
== dl
->major
&&
8660 d
->disk
.minor
== dl
->minor
) {
8661 dprintf("%x:%x already in array\n",
8662 dl
->major
, dl
->minor
);
8667 test_list
= additional_test_list
;
8669 if (test_list
->disk
.major
== dl
->major
&&
8670 test_list
->disk
.minor
== dl
->minor
) {
8671 dprintf("%x:%x already in additional test list\n",
8672 dl
->major
, dl
->minor
);
8675 test_list
= test_list
->next
;
8680 /* skip in use or failed drives */
8681 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8683 dprintf("%x:%x status (failed: %d index: %d)\n",
8684 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8688 /* skip pure spares when we are looking for partially
8689 * assimilated drives
8691 if (dl
->index
== -1 && !activate_new
)
8694 if (!drive_validate_sector_size(super
, dl
))
8697 /* Does this unused device have the requisite free space?
8698 * It needs to be able to cover all member volumes
8700 ex
= get_extents(super
, dl
);
8702 dprintf("cannot get extents\n");
8705 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8706 dev
= get_imsm_dev(super
, i
);
8707 map
= get_imsm_map(dev
, MAP_0
);
8709 /* check if this disk is already a member of
8712 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8718 array_start
= pba_of_lba0(map
);
8719 array_end
= array_start
+
8720 blocks_per_member(map
) - 1;
8723 /* check that we can start at pba_of_lba0 with
8724 * blocks_per_member of space
8726 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8730 pos
= ex
[j
].start
+ ex
[j
].size
;
8732 } while (ex
[j
-1].size
);
8739 if (i
< mpb
->num_raid_devs
) {
8740 dprintf("%x:%x does not have %u to %u available\n",
8741 dl
->major
, dl
->minor
, array_start
, array_end
);
8751 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8753 struct imsm_dev
*dev2
;
8754 struct imsm_map
*map
;
8760 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8762 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8763 if (state
== IMSM_T_STATE_FAILED
) {
8764 map
= get_imsm_map(dev2
, MAP_0
);
8767 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8769 * Check if failed disks are deleted from intel
8770 * disk list or are marked to be deleted
8772 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8773 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8775 * Do not rebuild the array if failed disks
8776 * from failed sub-array are not removed from
8780 is_failed(&idisk
->disk
) &&
8781 (idisk
->action
!= DISK_REMOVE
))
8789 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8790 struct metadata_update
**updates
)
8793 * Find a device with unused free space and use it to replace a
8794 * failed/vacant region in an array. We replace failed regions one a
8795 * array at a time. The result is that a new spare disk will be added
8796 * to the first failed array and after the monitor has finished
8797 * propagating failures the remainder will be consumed.
8799 * FIXME add a capability for mdmon to request spares from another
8803 struct intel_super
*super
= a
->container
->sb
;
8804 int inst
= a
->info
.container_member
;
8805 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8806 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8807 int failed
= a
->info
.array
.raid_disks
;
8808 struct mdinfo
*rv
= NULL
;
8811 struct metadata_update
*mu
;
8813 struct imsm_update_activate_spare
*u
;
8818 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8819 if ((d
->curr_state
& DS_FAULTY
) &&
8821 /* wait for Removal to happen */
8823 if (d
->state_fd
>= 0)
8827 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8828 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8830 if (imsm_reshape_blocks_arrays_changes(super
))
8833 /* Cannot activate another spare if rebuild is in progress already
8835 if (is_rebuilding(dev
)) {
8836 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8840 if (a
->info
.array
.level
== 4)
8841 /* No repair for takeovered array
8842 * imsm doesn't support raid4
8846 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8847 IMSM_T_STATE_DEGRADED
)
8850 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8851 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8856 * If there are any failed disks check state of the other volume.
8857 * Block rebuild if the another one is failed until failed disks
8858 * are removed from container.
8861 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8862 MAX_RAID_SERIAL_LEN
, dev
->volume
);
8863 /* check if states of the other volumes allow for rebuild */
8864 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
8866 allowed
= imsm_rebuild_allowed(a
->container
,
8874 /* For each slot, if it is not working, find a spare */
8875 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
8876 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8877 if (d
->disk
.raid_disk
== i
)
8879 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
8880 if (d
&& (d
->state_fd
>= 0))
8884 * OK, this device needs recovery. Try to re-add the
8885 * previous occupant of this slot, if this fails see if
8886 * we can continue the assimilation of a spare that was
8887 * partially assimilated, finally try to activate a new
8890 dl
= imsm_readd(super
, i
, a
);
8892 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
8894 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
8898 /* found a usable disk with enough space */
8899 di
= xcalloc(1, sizeof(*di
));
8901 /* dl->index will be -1 in the case we are activating a
8902 * pristine spare. imsm_process_update() will create a
8903 * new index in this case. Once a disk is found to be
8904 * failed in all member arrays it is kicked from the
8907 di
->disk
.number
= dl
->index
;
8909 /* (ab)use di->devs to store a pointer to the device
8912 di
->devs
= (struct mdinfo
*) dl
;
8914 di
->disk
.raid_disk
= i
;
8915 di
->disk
.major
= dl
->major
;
8916 di
->disk
.minor
= dl
->minor
;
8918 di
->recovery_start
= 0;
8919 di
->data_offset
= pba_of_lba0(map
);
8920 di
->component_size
= a
->info
.component_size
;
8921 di
->container_member
= inst
;
8922 di
->bb
.supported
= 1;
8923 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
8924 di
->ppl_sector
= get_ppl_sector(super
, inst
);
8925 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
8927 super
->random
= random32();
8931 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
8932 i
, di
->data_offset
);
8936 /* No spares found */
8938 /* Now 'rv' has a list of devices to return.
8939 * Create a metadata_update record to update the
8940 * disk_ord_tbl for the array
8942 mu
= xmalloc(sizeof(*mu
));
8943 mu
->buf
= xcalloc(num_spares
,
8944 sizeof(struct imsm_update_activate_spare
));
8946 mu
->space_list
= NULL
;
8947 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
8948 mu
->next
= *updates
;
8949 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
8951 for (di
= rv
; di
; di
= di
->next
) {
8952 u
->type
= update_activate_spare
;
8953 u
->dl
= (struct dl
*) di
->devs
;
8955 u
->slot
= di
->disk
.raid_disk
;
8966 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
8968 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
8969 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8970 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8971 struct disk_info
*inf
= get_disk_info(u
);
8972 struct imsm_disk
*disk
;
8976 for (i
= 0; i
< map
->num_members
; i
++) {
8977 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
8978 for (j
= 0; j
< new_map
->num_members
; j
++)
8979 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
8986 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
8990 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8991 if (dl
->major
== major
&& dl
->minor
== minor
)
8996 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9002 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9003 if (dl
->major
== major
&& dl
->minor
== minor
) {
9006 prev
->next
= dl
->next
;
9008 super
->disks
= dl
->next
;
9010 __free_imsm_disk(dl
);
9011 dprintf("removed %x:%x\n", major
, minor
);
9019 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9021 static int add_remove_disk_update(struct intel_super
*super
)
9023 int check_degraded
= 0;
9026 /* add/remove some spares to/from the metadata/contrainer */
9027 while (super
->disk_mgmt_list
) {
9028 struct dl
*disk_cfg
;
9030 disk_cfg
= super
->disk_mgmt_list
;
9031 super
->disk_mgmt_list
= disk_cfg
->next
;
9032 disk_cfg
->next
= NULL
;
9034 if (disk_cfg
->action
== DISK_ADD
) {
9035 disk_cfg
->next
= super
->disks
;
9036 super
->disks
= disk_cfg
;
9038 dprintf("added %x:%x\n",
9039 disk_cfg
->major
, disk_cfg
->minor
);
9040 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9041 dprintf("Disk remove action processed: %x.%x\n",
9042 disk_cfg
->major
, disk_cfg
->minor
);
9043 disk
= get_disk_super(super
,
9047 /* store action status */
9048 disk
->action
= DISK_REMOVE
;
9049 /* remove spare disks only */
9050 if (disk
->index
== -1) {
9051 remove_disk_super(super
,
9056 /* release allocate disk structure */
9057 __free_imsm_disk(disk_cfg
);
9060 return check_degraded
;
9063 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9064 struct intel_super
*super
,
9067 struct intel_dev
*id
;
9068 void **tofree
= NULL
;
9071 dprintf("(enter)\n");
9072 if (u
->subdev
< 0 || u
->subdev
> 1) {
9073 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9076 if (space_list
== NULL
|| *space_list
== NULL
) {
9077 dprintf("imsm: Error: Memory is not allocated\n");
9081 for (id
= super
->devlist
; id
; id
= id
->next
) {
9082 if (id
->index
== (unsigned)u
->subdev
) {
9083 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9084 struct imsm_map
*map
;
9085 struct imsm_dev
*new_dev
=
9086 (struct imsm_dev
*)*space_list
;
9087 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9089 struct dl
*new_disk
;
9091 if (new_dev
== NULL
)
9093 *space_list
= **space_list
;
9094 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9095 map
= get_imsm_map(new_dev
, MAP_0
);
9097 dprintf("imsm: Error: migration in progress");
9101 to_state
= map
->map_state
;
9102 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9104 /* this should not happen */
9105 if (u
->new_disks
[0] < 0) {
9106 map
->failed_disk_num
=
9107 map
->num_members
- 1;
9108 to_state
= IMSM_T_STATE_DEGRADED
;
9110 to_state
= IMSM_T_STATE_NORMAL
;
9112 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9113 if (u
->new_level
> -1)
9114 map
->raid_level
= u
->new_level
;
9115 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9116 if ((u
->new_level
== 5) &&
9117 (migr_map
->raid_level
== 0)) {
9118 int ord
= map
->num_members
- 1;
9119 migr_map
->num_members
--;
9120 if (u
->new_disks
[0] < 0)
9121 ord
|= IMSM_ORD_REBUILD
;
9122 set_imsm_ord_tbl_ent(map
,
9123 map
->num_members
- 1,
9127 tofree
= (void **)dev
;
9129 /* update chunk size
9131 if (u
->new_chunksize
> 0) {
9132 unsigned long long num_data_stripes
;
9134 imsm_num_data_members(dev
, MAP_0
);
9136 if (used_disks
== 0)
9139 map
->blocks_per_strip
=
9140 __cpu_to_le16(u
->new_chunksize
* 2);
9142 (join_u32(dev
->size_low
, dev
->size_high
)
9144 num_data_stripes
/= map
->blocks_per_strip
;
9145 num_data_stripes
/= map
->num_domains
;
9146 set_num_data_stripes(map
, num_data_stripes
);
9151 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9152 migr_map
->raid_level
== map
->raid_level
)
9155 if (u
->new_disks
[0] >= 0) {
9158 new_disk
= get_disk_super(super
,
9159 major(u
->new_disks
[0]),
9160 minor(u
->new_disks
[0]));
9161 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9162 major(u
->new_disks
[0]),
9163 minor(u
->new_disks
[0]),
9164 new_disk
, new_disk
->index
);
9165 if (new_disk
== NULL
)
9166 goto error_disk_add
;
9168 new_disk
->index
= map
->num_members
- 1;
9169 /* slot to fill in autolayout
9171 new_disk
->raiddisk
= new_disk
->index
;
9172 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9173 new_disk
->disk
.status
&= ~SPARE_DISK
;
9175 goto error_disk_add
;
9178 *tofree
= *space_list
;
9179 /* calculate new size
9181 imsm_set_array_size(new_dev
, -1);
9188 *space_list
= tofree
;
9192 dprintf("Error: imsm: Cannot find disk.\n");
9196 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9197 struct intel_super
*super
)
9199 struct intel_dev
*id
;
9202 dprintf("(enter)\n");
9203 if (u
->subdev
< 0 || u
->subdev
> 1) {
9204 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9208 for (id
= super
->devlist
; id
; id
= id
->next
) {
9209 if (id
->index
== (unsigned)u
->subdev
) {
9210 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9211 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9212 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
9213 unsigned long long blocks_per_member
;
9214 unsigned long long num_data_stripes
;
9216 /* calculate new size
9218 blocks_per_member
= u
->new_size
/ used_disks
;
9219 num_data_stripes
= blocks_per_member
/
9220 map
->blocks_per_strip
;
9221 num_data_stripes
/= map
->num_domains
;
9222 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9223 u
->new_size
, blocks_per_member
,
9225 set_blocks_per_member(map
, blocks_per_member
);
9226 set_num_data_stripes(map
, num_data_stripes
);
9227 imsm_set_array_size(dev
, u
->new_size
);
9237 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9238 struct intel_super
*super
,
9239 struct active_array
*active_array
)
9241 struct imsm_super
*mpb
= super
->anchor
;
9242 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9243 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9244 struct imsm_map
*migr_map
;
9245 struct active_array
*a
;
9246 struct imsm_disk
*disk
;
9253 int second_map_created
= 0;
9255 for (; u
; u
= u
->next
) {
9256 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9261 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9266 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9271 /* count failures (excluding rebuilds and the victim)
9272 * to determine map[0] state
9275 for (i
= 0; i
< map
->num_members
; i
++) {
9278 disk
= get_imsm_disk(super
,
9279 get_imsm_disk_idx(dev
, i
, MAP_X
));
9280 if (!disk
|| is_failed(disk
))
9284 /* adding a pristine spare, assign a new index */
9285 if (dl
->index
< 0) {
9286 dl
->index
= super
->anchor
->num_disks
;
9287 super
->anchor
->num_disks
++;
9290 disk
->status
|= CONFIGURED_DISK
;
9291 disk
->status
&= ~SPARE_DISK
;
9294 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9295 if (!second_map_created
) {
9296 second_map_created
= 1;
9297 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9298 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9300 map
->map_state
= to_state
;
9301 migr_map
= get_imsm_map(dev
, MAP_1
);
9302 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9303 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9304 dl
->index
| IMSM_ORD_REBUILD
);
9306 /* update the family_num to mark a new container
9307 * generation, being careful to record the existing
9308 * family_num in orig_family_num to clean up after
9309 * earlier mdadm versions that neglected to set it.
9311 if (mpb
->orig_family_num
== 0)
9312 mpb
->orig_family_num
= mpb
->family_num
;
9313 mpb
->family_num
+= super
->random
;
9315 /* count arrays using the victim in the metadata */
9317 for (a
= active_array
; a
; a
= a
->next
) {
9318 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9319 map
= get_imsm_map(dev
, MAP_0
);
9321 if (get_imsm_disk_slot(map
, victim
) >= 0)
9325 /* delete the victim if it is no longer being
9331 /* We know that 'manager' isn't touching anything,
9332 * so it is safe to delete
9334 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9335 if ((*dlp
)->index
== victim
)
9338 /* victim may be on the missing list */
9340 for (dlp
= &super
->missing
; *dlp
;
9341 dlp
= &(*dlp
)->next
)
9342 if ((*dlp
)->index
== victim
)
9344 imsm_delete(super
, dlp
, victim
);
9351 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9352 struct intel_super
*super
,
9355 struct dl
*new_disk
;
9356 struct intel_dev
*id
;
9358 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9359 int disk_count
= u
->old_raid_disks
;
9360 void **tofree
= NULL
;
9361 int devices_to_reshape
= 1;
9362 struct imsm_super
*mpb
= super
->anchor
;
9364 unsigned int dev_id
;
9366 dprintf("(enter)\n");
9368 /* enable spares to use in array */
9369 for (i
= 0; i
< delta_disks
; i
++) {
9370 new_disk
= get_disk_super(super
,
9371 major(u
->new_disks
[i
]),
9372 minor(u
->new_disks
[i
]));
9373 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9374 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9375 new_disk
, new_disk
->index
);
9376 if (new_disk
== NULL
||
9377 (new_disk
->index
>= 0 &&
9378 new_disk
->index
< u
->old_raid_disks
))
9379 goto update_reshape_exit
;
9380 new_disk
->index
= disk_count
++;
9381 /* slot to fill in autolayout
9383 new_disk
->raiddisk
= new_disk
->index
;
9384 new_disk
->disk
.status
|=
9386 new_disk
->disk
.status
&= ~SPARE_DISK
;
9389 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9390 mpb
->num_raid_devs
);
9391 /* manage changes in volume
9393 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9394 void **sp
= *space_list
;
9395 struct imsm_dev
*newdev
;
9396 struct imsm_map
*newmap
, *oldmap
;
9398 for (id
= super
->devlist
; id
; id
= id
->next
) {
9399 if (id
->index
== dev_id
)
9408 /* Copy the dev, but not (all of) the map */
9409 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9410 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9411 newmap
= get_imsm_map(newdev
, MAP_0
);
9412 /* Copy the current map */
9413 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9414 /* update one device only
9416 if (devices_to_reshape
) {
9417 dprintf("imsm: modifying subdev: %i\n",
9419 devices_to_reshape
--;
9420 newdev
->vol
.migr_state
= 1;
9421 newdev
->vol
.curr_migr_unit
= 0;
9422 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9423 newmap
->num_members
= u
->new_raid_disks
;
9424 for (i
= 0; i
< delta_disks
; i
++) {
9425 set_imsm_ord_tbl_ent(newmap
,
9426 u
->old_raid_disks
+ i
,
9427 u
->old_raid_disks
+ i
);
9429 /* New map is correct, now need to save old map
9431 newmap
= get_imsm_map(newdev
, MAP_1
);
9432 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9434 imsm_set_array_size(newdev
, -1);
9437 sp
= (void **)id
->dev
;
9442 /* Clear migration record */
9443 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9446 *space_list
= tofree
;
9449 update_reshape_exit
:
9454 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9455 struct intel_super
*super
,
9458 struct imsm_dev
*dev
= NULL
;
9459 struct intel_dev
*dv
;
9460 struct imsm_dev
*dev_new
;
9461 struct imsm_map
*map
;
9465 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9466 if (dv
->index
== (unsigned int)u
->subarray
) {
9474 map
= get_imsm_map(dev
, MAP_0
);
9476 if (u
->direction
== R10_TO_R0
) {
9477 unsigned long long num_data_stripes
;
9479 map
->num_domains
= 1;
9480 num_data_stripes
= blocks_per_member(map
);
9481 num_data_stripes
/= map
->blocks_per_strip
;
9482 num_data_stripes
/= map
->num_domains
;
9483 set_num_data_stripes(map
, num_data_stripes
);
9485 /* Number of failed disks must be half of initial disk number */
9486 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9487 (map
->num_members
/ 2))
9490 /* iterate through devices to mark removed disks as spare */
9491 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9492 if (dm
->disk
.status
& FAILED_DISK
) {
9493 int idx
= dm
->index
;
9494 /* update indexes on the disk list */
9495 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9496 the index values will end up being correct.... NB */
9497 for (du
= super
->disks
; du
; du
= du
->next
)
9498 if (du
->index
> idx
)
9500 /* mark as spare disk */
9505 map
->num_members
= map
->num_members
/ 2;
9506 map
->map_state
= IMSM_T_STATE_NORMAL
;
9507 map
->num_domains
= 1;
9508 map
->raid_level
= 0;
9509 map
->failed_disk_num
= -1;
9512 if (u
->direction
== R0_TO_R10
) {
9514 /* update slots in current disk list */
9515 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9519 /* create new *missing* disks */
9520 for (i
= 0; i
< map
->num_members
; i
++) {
9521 space
= *space_list
;
9524 *space_list
= *space
;
9526 memcpy(du
, super
->disks
, sizeof(*du
));
9530 du
->index
= (i
* 2) + 1;
9531 sprintf((char *)du
->disk
.serial
,
9532 " MISSING_%d", du
->index
);
9533 sprintf((char *)du
->serial
,
9534 "MISSING_%d", du
->index
);
9535 du
->next
= super
->missing
;
9536 super
->missing
= du
;
9538 /* create new dev and map */
9539 space
= *space_list
;
9542 *space_list
= *space
;
9543 dev_new
= (void *)space
;
9544 memcpy(dev_new
, dev
, sizeof(*dev
));
9545 /* update new map */
9546 map
= get_imsm_map(dev_new
, MAP_0
);
9547 map
->num_members
= map
->num_members
* 2;
9548 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9549 map
->num_domains
= 2;
9550 map
->raid_level
= 1;
9551 /* replace dev<->dev_new */
9554 /* update disk order table */
9555 for (du
= super
->disks
; du
; du
= du
->next
)
9557 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9558 for (du
= super
->missing
; du
; du
= du
->next
)
9559 if (du
->index
>= 0) {
9560 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9561 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9567 static void imsm_process_update(struct supertype
*st
,
9568 struct metadata_update
*update
)
9571 * crack open the metadata_update envelope to find the update record
9572 * update can be one of:
9573 * update_reshape_container_disks - all the arrays in the container
9574 * are being reshaped to have more devices. We need to mark
9575 * the arrays for general migration and convert selected spares
9576 * into active devices.
9577 * update_activate_spare - a spare device has replaced a failed
9578 * device in an array, update the disk_ord_tbl. If this disk is
9579 * present in all member arrays then also clear the SPARE_DISK
9581 * update_create_array
9583 * update_rename_array
9584 * update_add_remove_disk
9586 struct intel_super
*super
= st
->sb
;
9587 struct imsm_super
*mpb
;
9588 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9590 /* update requires a larger buf but the allocation failed */
9591 if (super
->next_len
&& !super
->next_buf
) {
9592 super
->next_len
= 0;
9596 if (super
->next_buf
) {
9597 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9599 super
->len
= super
->next_len
;
9600 super
->buf
= super
->next_buf
;
9602 super
->next_len
= 0;
9603 super
->next_buf
= NULL
;
9606 mpb
= super
->anchor
;
9609 case update_general_migration_checkpoint
: {
9610 struct intel_dev
*id
;
9611 struct imsm_update_general_migration_checkpoint
*u
=
9612 (void *)update
->buf
;
9614 dprintf("called for update_general_migration_checkpoint\n");
9616 /* find device under general migration */
9617 for (id
= super
->devlist
; id
; id
= id
->next
) {
9618 if (is_gen_migration(id
->dev
)) {
9619 id
->dev
->vol
.curr_migr_unit
=
9620 __cpu_to_le32(u
->curr_migr_unit
);
9621 super
->updates_pending
++;
9626 case update_takeover
: {
9627 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9628 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9629 imsm_update_version_info(super
);
9630 super
->updates_pending
++;
9635 case update_reshape_container_disks
: {
9636 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9637 if (apply_reshape_container_disks_update(
9638 u
, super
, &update
->space_list
))
9639 super
->updates_pending
++;
9642 case update_reshape_migration
: {
9643 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9644 if (apply_reshape_migration_update(
9645 u
, super
, &update
->space_list
))
9646 super
->updates_pending
++;
9649 case update_size_change
: {
9650 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9651 if (apply_size_change_update(u
, super
))
9652 super
->updates_pending
++;
9655 case update_activate_spare
: {
9656 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9657 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9658 super
->updates_pending
++;
9661 case update_create_array
: {
9662 /* someone wants to create a new array, we need to be aware of
9663 * a few races/collisions:
9664 * 1/ 'Create' called by two separate instances of mdadm
9665 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9666 * devices that have since been assimilated via
9668 * In the event this update can not be carried out mdadm will
9669 * (FIX ME) notice that its update did not take hold.
9671 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9672 struct intel_dev
*dv
;
9673 struct imsm_dev
*dev
;
9674 struct imsm_map
*map
, *new_map
;
9675 unsigned long long start
, end
;
9676 unsigned long long new_start
, new_end
;
9678 struct disk_info
*inf
;
9681 /* handle racing creates: first come first serve */
9682 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9683 dprintf("subarray %d already defined\n", u
->dev_idx
);
9687 /* check update is next in sequence */
9688 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9689 dprintf("can not create array %d expected index %d\n",
9690 u
->dev_idx
, mpb
->num_raid_devs
);
9694 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9695 new_start
= pba_of_lba0(new_map
);
9696 new_end
= new_start
+ blocks_per_member(new_map
);
9697 inf
= get_disk_info(u
);
9699 /* handle activate_spare versus create race:
9700 * check to make sure that overlapping arrays do not include
9703 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9704 dev
= get_imsm_dev(super
, i
);
9705 map
= get_imsm_map(dev
, MAP_0
);
9706 start
= pba_of_lba0(map
);
9707 end
= start
+ blocks_per_member(map
);
9708 if ((new_start
>= start
&& new_start
<= end
) ||
9709 (start
>= new_start
&& start
<= new_end
))
9714 if (disks_overlap(super
, i
, u
)) {
9715 dprintf("arrays overlap\n");
9720 /* check that prepare update was successful */
9721 if (!update
->space
) {
9722 dprintf("prepare update failed\n");
9726 /* check that all disks are still active before committing
9727 * changes. FIXME: could we instead handle this by creating a
9728 * degraded array? That's probably not what the user expects,
9729 * so better to drop this update on the floor.
9731 for (i
= 0; i
< new_map
->num_members
; i
++) {
9732 dl
= serial_to_dl(inf
[i
].serial
, super
);
9734 dprintf("disk disappeared\n");
9739 super
->updates_pending
++;
9741 /* convert spares to members and fixup ord_tbl */
9742 for (i
= 0; i
< new_map
->num_members
; i
++) {
9743 dl
= serial_to_dl(inf
[i
].serial
, super
);
9744 if (dl
->index
== -1) {
9745 dl
->index
= mpb
->num_disks
;
9747 dl
->disk
.status
|= CONFIGURED_DISK
;
9748 dl
->disk
.status
&= ~SPARE_DISK
;
9750 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9755 update
->space
= NULL
;
9756 imsm_copy_dev(dev
, &u
->dev
);
9757 dv
->index
= u
->dev_idx
;
9758 dv
->next
= super
->devlist
;
9759 super
->devlist
= dv
;
9760 mpb
->num_raid_devs
++;
9762 imsm_update_version_info(super
);
9765 /* mdmon knows how to release update->space, but not
9766 * ((struct intel_dev *) update->space)->dev
9768 if (update
->space
) {
9774 case update_kill_array
: {
9775 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9776 int victim
= u
->dev_idx
;
9777 struct active_array
*a
;
9778 struct intel_dev
**dp
;
9779 struct imsm_dev
*dev
;
9781 /* sanity check that we are not affecting the uuid of
9782 * active arrays, or deleting an active array
9784 * FIXME when immutable ids are available, but note that
9785 * we'll also need to fixup the invalidated/active
9786 * subarray indexes in mdstat
9788 for (a
= st
->arrays
; a
; a
= a
->next
)
9789 if (a
->info
.container_member
>= victim
)
9791 /* by definition if mdmon is running at least one array
9792 * is active in the container, so checking
9793 * mpb->num_raid_devs is just extra paranoia
9795 dev
= get_imsm_dev(super
, victim
);
9796 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9797 dprintf("failed to delete subarray-%d\n", victim
);
9801 for (dp
= &super
->devlist
; *dp
;)
9802 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9805 if ((*dp
)->index
> (unsigned)victim
)
9809 mpb
->num_raid_devs
--;
9810 super
->updates_pending
++;
9813 case update_rename_array
: {
9814 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9815 char name
[MAX_RAID_SERIAL_LEN
+1];
9816 int target
= u
->dev_idx
;
9817 struct active_array
*a
;
9818 struct imsm_dev
*dev
;
9820 /* sanity check that we are not affecting the uuid of
9823 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9824 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9825 for (a
= st
->arrays
; a
; a
= a
->next
)
9826 if (a
->info
.container_member
== target
)
9828 dev
= get_imsm_dev(super
, u
->dev_idx
);
9829 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9830 dprintf("failed to rename subarray-%d\n", target
);
9834 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
9835 super
->updates_pending
++;
9838 case update_add_remove_disk
: {
9839 /* we may be able to repair some arrays if disks are
9840 * being added, check the status of add_remove_disk
9841 * if discs has been added.
9843 if (add_remove_disk_update(super
)) {
9844 struct active_array
*a
;
9846 super
->updates_pending
++;
9847 for (a
= st
->arrays
; a
; a
= a
->next
)
9848 a
->check_degraded
= 1;
9852 case update_prealloc_badblocks_mem
:
9854 case update_rwh_policy
: {
9855 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
9856 int target
= u
->dev_idx
;
9857 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
9859 dprintf("could not find subarray-%d\n", target
);
9863 if (dev
->rwh_policy
!= u
->new_policy
) {
9864 dev
->rwh_policy
= u
->new_policy
;
9865 super
->updates_pending
++;
9870 pr_err("error: unsuported process update type:(type: %d)\n", type
);
9874 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
9876 static int imsm_prepare_update(struct supertype
*st
,
9877 struct metadata_update
*update
)
9880 * Allocate space to hold new disk entries, raid-device entries or a new
9881 * mpb if necessary. The manager synchronously waits for updates to
9882 * complete in the monitor, so new mpb buffers allocated here can be
9883 * integrated by the monitor thread without worrying about live pointers
9884 * in the manager thread.
9886 enum imsm_update_type type
;
9887 struct intel_super
*super
= st
->sb
;
9888 unsigned int sector_size
= super
->sector_size
;
9889 struct imsm_super
*mpb
= super
->anchor
;
9893 if (update
->len
< (int)sizeof(type
))
9896 type
= *(enum imsm_update_type
*) update
->buf
;
9899 case update_general_migration_checkpoint
:
9900 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
9902 dprintf("called for update_general_migration_checkpoint\n");
9904 case update_takeover
: {
9905 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9906 if (update
->len
< (int)sizeof(*u
))
9908 if (u
->direction
== R0_TO_R10
) {
9909 void **tail
= (void **)&update
->space_list
;
9910 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
9911 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9912 int num_members
= map
->num_members
;
9915 /* allocate memory for added disks */
9916 for (i
= 0; i
< num_members
; i
++) {
9917 size
= sizeof(struct dl
);
9918 space
= xmalloc(size
);
9923 /* allocate memory for new device */
9924 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
9925 (num_members
* sizeof(__u32
));
9926 space
= xmalloc(size
);
9930 len
= disks_to_mpb_size(num_members
* 2);
9935 case update_reshape_container_disks
: {
9936 /* Every raid device in the container is about to
9937 * gain some more devices, and we will enter a
9939 * So each 'imsm_map' will be bigger, and the imsm_vol
9940 * will now hold 2 of them.
9941 * Thus we need new 'struct imsm_dev' allocations sized
9942 * as sizeof_imsm_dev but with more devices in both maps.
9944 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9945 struct intel_dev
*dl
;
9946 void **space_tail
= (void**)&update
->space_list
;
9948 if (update
->len
< (int)sizeof(*u
))
9951 dprintf("for update_reshape\n");
9953 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
9954 int size
= sizeof_imsm_dev(dl
->dev
, 1);
9956 if (u
->new_raid_disks
> u
->old_raid_disks
)
9957 size
+= sizeof(__u32
)*2*
9958 (u
->new_raid_disks
- u
->old_raid_disks
);
9965 len
= disks_to_mpb_size(u
->new_raid_disks
);
9966 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9969 case update_reshape_migration
: {
9970 /* for migration level 0->5 we need to add disks
9971 * so the same as for container operation we will copy
9972 * device to the bigger location.
9973 * in memory prepared device and new disk area are prepared
9974 * for usage in process update
9976 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9977 struct intel_dev
*id
;
9978 void **space_tail
= (void **)&update
->space_list
;
9981 int current_level
= -1;
9983 if (update
->len
< (int)sizeof(*u
))
9986 dprintf("for update_reshape\n");
9988 /* add space for bigger array in update
9990 for (id
= super
->devlist
; id
; id
= id
->next
) {
9991 if (id
->index
== (unsigned)u
->subdev
) {
9992 size
= sizeof_imsm_dev(id
->dev
, 1);
9993 if (u
->new_raid_disks
> u
->old_raid_disks
)
9994 size
+= sizeof(__u32
)*2*
9995 (u
->new_raid_disks
- u
->old_raid_disks
);
10003 if (update
->space_list
== NULL
)
10006 /* add space for disk in update
10008 size
= sizeof(struct dl
);
10012 *space_tail
= NULL
;
10014 /* add spare device to update
10016 for (id
= super
->devlist
; id
; id
= id
->next
)
10017 if (id
->index
== (unsigned)u
->subdev
) {
10018 struct imsm_dev
*dev
;
10019 struct imsm_map
*map
;
10021 dev
= get_imsm_dev(super
, u
->subdev
);
10022 map
= get_imsm_map(dev
, MAP_0
);
10023 current_level
= map
->raid_level
;
10026 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10027 struct mdinfo
*spares
;
10029 spares
= get_spares_for_grow(st
);
10032 struct mdinfo
*dev
;
10034 dev
= spares
->devs
;
10037 makedev(dev
->disk
.major
,
10039 dl
= get_disk_super(super
,
10042 dl
->index
= u
->old_raid_disks
;
10045 sysfs_free(spares
);
10048 len
= disks_to_mpb_size(u
->new_raid_disks
);
10049 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10052 case update_size_change
: {
10053 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10057 case update_activate_spare
: {
10058 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10062 case update_create_array
: {
10063 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10064 struct intel_dev
*dv
;
10065 struct imsm_dev
*dev
= &u
->dev
;
10066 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10068 struct disk_info
*inf
;
10072 if (update
->len
< (int)sizeof(*u
))
10075 inf
= get_disk_info(u
);
10076 len
= sizeof_imsm_dev(dev
, 1);
10077 /* allocate a new super->devlist entry */
10078 dv
= xmalloc(sizeof(*dv
));
10079 dv
->dev
= xmalloc(len
);
10080 update
->space
= dv
;
10082 /* count how many spares will be converted to members */
10083 for (i
= 0; i
< map
->num_members
; i
++) {
10084 dl
= serial_to_dl(inf
[i
].serial
, super
);
10086 /* hmm maybe it failed?, nothing we can do about
10091 if (count_memberships(dl
, super
) == 0)
10094 len
+= activate
* sizeof(struct imsm_disk
);
10097 case update_kill_array
: {
10098 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10102 case update_rename_array
: {
10103 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10107 case update_add_remove_disk
:
10108 /* no update->len needed */
10110 case update_prealloc_badblocks_mem
:
10111 super
->extra_space
+= sizeof(struct bbm_log
) -
10112 get_imsm_bbm_log_size(super
->bbm_log
);
10114 case update_rwh_policy
: {
10115 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10123 /* check if we need a larger metadata buffer */
10124 if (super
->next_buf
)
10125 buf_len
= super
->next_len
;
10127 buf_len
= super
->len
;
10129 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10130 /* ok we need a larger buf than what is currently allocated
10131 * if this allocation fails process_update will notice that
10132 * ->next_len is set and ->next_buf is NULL
10134 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10135 super
->extra_space
+ len
, sector_size
);
10136 if (super
->next_buf
)
10137 free(super
->next_buf
);
10139 super
->next_len
= buf_len
;
10140 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10141 memset(super
->next_buf
, 0, buf_len
);
10143 super
->next_buf
= NULL
;
10148 /* must be called while manager is quiesced */
10149 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10151 struct imsm_super
*mpb
= super
->anchor
;
10153 struct imsm_dev
*dev
;
10154 struct imsm_map
*map
;
10155 unsigned int i
, j
, num_members
;
10156 __u32 ord
, ord_map0
;
10157 struct bbm_log
*log
= super
->bbm_log
;
10159 dprintf("deleting device[%d] from imsm_super\n", index
);
10161 /* shift all indexes down one */
10162 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10163 if (iter
->index
> (int)index
)
10165 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10166 if (iter
->index
> (int)index
)
10169 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10170 dev
= get_imsm_dev(super
, i
);
10171 map
= get_imsm_map(dev
, MAP_0
);
10172 num_members
= map
->num_members
;
10173 for (j
= 0; j
< num_members
; j
++) {
10174 /* update ord entries being careful not to propagate
10175 * ord-flags to the first map
10177 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10178 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10180 if (ord_to_idx(ord
) <= index
)
10183 map
= get_imsm_map(dev
, MAP_0
);
10184 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10185 map
= get_imsm_map(dev
, MAP_1
);
10187 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10191 for (i
= 0; i
< log
->entry_count
; i
++) {
10192 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10194 if (entry
->disk_ordinal
<= index
)
10196 entry
->disk_ordinal
--;
10200 super
->updates_pending
++;
10202 struct dl
*dl
= *dlp
;
10204 *dlp
= (*dlp
)->next
;
10205 __free_imsm_disk(dl
);
10209 static void close_targets(int *targets
, int new_disks
)
10216 for (i
= 0; i
< new_disks
; i
++) {
10217 if (targets
[i
] >= 0) {
10224 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10225 struct intel_super
*super
,
10226 struct imsm_dev
*dev
)
10232 struct imsm_map
*map
;
10235 ret_val
= raid_disks
/2;
10236 /* check map if all disks pairs not failed
10239 map
= get_imsm_map(dev
, MAP_0
);
10240 for (i
= 0; i
< ret_val
; i
++) {
10241 int degradation
= 0;
10242 if (get_imsm_disk(super
, i
) == NULL
)
10244 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10246 if (degradation
== 2)
10249 map
= get_imsm_map(dev
, MAP_1
);
10250 /* if there is no second map
10251 * result can be returned
10255 /* check degradation in second map
10257 for (i
= 0; i
< ret_val
; i
++) {
10258 int degradation
= 0;
10259 if (get_imsm_disk(super
, i
) == NULL
)
10261 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10263 if (degradation
== 2)
10277 /*******************************************************************************
10278 * Function: open_backup_targets
10279 * Description: Function opens file descriptors for all devices given in
10282 * info : general array info
10283 * raid_disks : number of disks
10284 * raid_fds : table of device's file descriptors
10285 * super : intel super for raid10 degradation check
10286 * dev : intel device for raid10 degradation check
10290 ******************************************************************************/
10291 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10292 struct intel_super
*super
, struct imsm_dev
*dev
)
10298 for (i
= 0; i
< raid_disks
; i
++)
10301 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10304 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10305 dprintf("disk is faulty!!\n");
10309 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10312 dn
= map_dev(sd
->disk
.major
,
10313 sd
->disk
.minor
, 1);
10314 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10315 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10316 pr_err("cannot open component\n");
10321 /* check if maximum array degradation level is not exceeded
10323 if ((raid_disks
- opened
) >
10324 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10326 pr_err("Not enough disks can be opened.\n");
10327 close_targets(raid_fds
, raid_disks
);
10333 /*******************************************************************************
10334 * Function: validate_container_imsm
10335 * Description: This routine validates container after assemble,
10336 * eg. if devices in container are under the same controller.
10339 * info : linked list with info about devices used in array
10343 ******************************************************************************/
10344 int validate_container_imsm(struct mdinfo
*info
)
10346 if (check_env("IMSM_NO_PLATFORM"))
10349 struct sys_dev
*idev
;
10350 struct sys_dev
*hba
= NULL
;
10351 struct sys_dev
*intel_devices
= find_intel_devices();
10352 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10353 info
->disk
.minor
));
10355 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10356 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10365 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10366 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10370 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10371 struct mdinfo
*dev
;
10373 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10374 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10376 struct sys_dev
*hba2
= NULL
;
10377 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10378 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10386 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10387 get_orom_by_device_id(hba2
->dev_id
);
10389 if (hba2
&& hba
->type
!= hba2
->type
) {
10390 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10391 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10395 if (orom
!= orom2
) {
10396 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10397 " This operation is not supported and can lead to data loss.\n");
10402 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10403 " This operation is not supported and can lead to data loss.\n");
10411 /*******************************************************************************
10412 * Function: imsm_record_badblock
10413 * Description: This routine stores new bad block record in BBM log
10416 * a : array containing a bad block
10417 * slot : disk number containing a bad block
10418 * sector : bad block sector
10419 * length : bad block sectors range
10423 ******************************************************************************/
10424 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10425 unsigned long long sector
, int length
)
10427 struct intel_super
*super
= a
->container
->sb
;
10431 ord
= imsm_disk_slot_to_ord(a
, slot
);
10435 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10438 super
->updates_pending
++;
10442 /*******************************************************************************
10443 * Function: imsm_clear_badblock
10444 * Description: This routine clears bad block record from BBM log
10447 * a : array containing a bad block
10448 * slot : disk number containing a bad block
10449 * sector : bad block sector
10450 * length : bad block sectors range
10454 ******************************************************************************/
10455 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10456 unsigned long long sector
, int length
)
10458 struct intel_super
*super
= a
->container
->sb
;
10462 ord
= imsm_disk_slot_to_ord(a
, slot
);
10466 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10468 super
->updates_pending
++;
10472 /*******************************************************************************
10473 * Function: imsm_get_badblocks
10474 * Description: This routine get list of bad blocks for an array
10478 * slot : disk number
10480 * bb : structure containing bad blocks
10482 ******************************************************************************/
10483 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10485 int inst
= a
->info
.container_member
;
10486 struct intel_super
*super
= a
->container
->sb
;
10487 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10488 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10491 ord
= imsm_disk_slot_to_ord(a
, slot
);
10495 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10496 blocks_per_member(map
), &super
->bb
);
10500 /*******************************************************************************
10501 * Function: examine_badblocks_imsm
10502 * Description: Prints list of bad blocks on a disk to the standard output
10505 * st : metadata handler
10506 * fd : open file descriptor for device
10507 * devname : device name
10511 ******************************************************************************/
10512 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10514 struct intel_super
*super
= st
->sb
;
10515 struct bbm_log
*log
= super
->bbm_log
;
10516 struct dl
*d
= NULL
;
10519 for (d
= super
->disks
; d
; d
= d
->next
) {
10520 if (strcmp(d
->devname
, devname
) == 0)
10524 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10525 pr_err("%s doesn't appear to be part of a raid array\n",
10532 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10534 for (i
= 0; i
< log
->entry_count
; i
++) {
10535 if (entry
[i
].disk_ordinal
== d
->index
) {
10536 unsigned long long sector
= __le48_to_cpu(
10537 &entry
[i
].defective_block_start
);
10538 int cnt
= entry
[i
].marked_count
+ 1;
10541 printf("Bad-blocks on %s:\n", devname
);
10545 printf("%20llu for %d sectors\n", sector
, cnt
);
10551 printf("No bad-blocks list configured on %s\n", devname
);
10555 /*******************************************************************************
10556 * Function: init_migr_record_imsm
10557 * Description: Function inits imsm migration record
10559 * super : imsm internal array info
10560 * dev : device under migration
10561 * info : general array info to find the smallest device
10564 ******************************************************************************/
10565 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10566 struct mdinfo
*info
)
10568 struct intel_super
*super
= st
->sb
;
10569 struct migr_record
*migr_rec
= super
->migr_rec
;
10570 int new_data_disks
;
10571 unsigned long long dsize
, dev_sectors
;
10572 long long unsigned min_dev_sectors
= -1LLU;
10576 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10577 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10578 unsigned long long num_migr_units
;
10579 unsigned long long array_blocks
;
10581 memset(migr_rec
, 0, sizeof(struct migr_record
));
10582 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10584 /* only ascending reshape supported now */
10585 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10587 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10588 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10589 migr_rec
->dest_depth_per_unit
*=
10590 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10591 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
10592 migr_rec
->blocks_per_unit
=
10593 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10594 migr_rec
->dest_depth_per_unit
=
10595 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10596 array_blocks
= info
->component_size
* new_data_disks
;
10598 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10600 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10602 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
10604 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10605 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10607 /* Find the smallest dev */
10608 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10609 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10610 fd
= dev_open(nm
, O_RDONLY
);
10613 get_dev_size(fd
, NULL
, &dsize
);
10614 dev_sectors
= dsize
/ 512;
10615 if (dev_sectors
< min_dev_sectors
)
10616 min_dev_sectors
= dev_sectors
;
10619 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
10620 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10622 write_imsm_migr_rec(st
);
10627 /*******************************************************************************
10628 * Function: save_backup_imsm
10629 * Description: Function saves critical data stripes to Migration Copy Area
10630 * and updates the current migration unit status.
10631 * Use restore_stripes() to form a destination stripe,
10632 * and to write it to the Copy Area.
10634 * st : supertype information
10635 * dev : imsm device that backup is saved for
10636 * info : general array info
10637 * buf : input buffer
10638 * length : length of data to backup (blocks_per_unit)
10642 ******************************************************************************/
10643 int save_backup_imsm(struct supertype
*st
,
10644 struct imsm_dev
*dev
,
10645 struct mdinfo
*info
,
10650 struct intel_super
*super
= st
->sb
;
10651 unsigned long long *target_offsets
;
10654 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10655 int new_disks
= map_dest
->num_members
;
10656 int dest_layout
= 0;
10658 unsigned long long start
;
10659 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
10661 targets
= xmalloc(new_disks
* sizeof(int));
10663 for (i
= 0; i
< new_disks
; i
++)
10666 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10668 start
= info
->reshape_progress
* 512;
10669 for (i
= 0; i
< new_disks
; i
++) {
10670 target_offsets
[i
] = (unsigned long long)
10671 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
10672 /* move back copy area adderss, it will be moved forward
10673 * in restore_stripes() using start input variable
10675 target_offsets
[i
] -= start
/data_disks
;
10678 if (open_backup_targets(info
, new_disks
, targets
,
10682 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10683 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10685 if (restore_stripes(targets
, /* list of dest devices */
10686 target_offsets
, /* migration record offsets */
10689 map_dest
->raid_level
,
10691 -1, /* source backup file descriptor */
10692 0, /* input buf offset
10693 * always 0 buf is already offseted */
10697 pr_err("Error restoring stripes\n");
10705 close_targets(targets
, new_disks
);
10708 free(target_offsets
);
10713 /*******************************************************************************
10714 * Function: save_checkpoint_imsm
10715 * Description: Function called for current unit status update
10716 * in the migration record. It writes it to disk.
10718 * super : imsm internal array info
10719 * info : general array info
10723 * 2: failure, means no valid migration record
10724 * / no general migration in progress /
10725 ******************************************************************************/
10726 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10728 struct intel_super
*super
= st
->sb
;
10729 unsigned long long blocks_per_unit
;
10730 unsigned long long curr_migr_unit
;
10732 if (load_imsm_migr_rec(super
, info
) != 0) {
10733 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10737 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10738 if (blocks_per_unit
== 0) {
10739 dprintf("imsm: no migration in progress.\n");
10742 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10743 /* check if array is alligned to copy area
10744 * if it is not alligned, add one to current migration unit value
10745 * this can happend on array reshape finish only
10747 if (info
->reshape_progress
% blocks_per_unit
)
10750 super
->migr_rec
->curr_migr_unit
=
10751 __cpu_to_le32(curr_migr_unit
);
10752 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10753 super
->migr_rec
->dest_1st_member_lba
=
10754 __cpu_to_le32(curr_migr_unit
*
10755 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
10756 if (write_imsm_migr_rec(st
) < 0) {
10757 dprintf("imsm: Cannot write migration record outside backup area\n");
10764 /*******************************************************************************
10765 * Function: recover_backup_imsm
10766 * Description: Function recovers critical data from the Migration Copy Area
10767 * while assembling an array.
10769 * super : imsm internal array info
10770 * info : general array info
10772 * 0 : success (or there is no data to recover)
10774 ******************************************************************************/
10775 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10777 struct intel_super
*super
= st
->sb
;
10778 struct migr_record
*migr_rec
= super
->migr_rec
;
10779 struct imsm_map
*map_dest
;
10780 struct intel_dev
*id
= NULL
;
10781 unsigned long long read_offset
;
10782 unsigned long long write_offset
;
10784 int *targets
= NULL
;
10785 int new_disks
, i
, err
;
10788 unsigned int sector_size
= super
->sector_size
;
10789 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
10790 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
10792 int skipped_disks
= 0;
10794 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10798 /* recover data only during assemblation */
10799 if (strncmp(buffer
, "inactive", 8) != 0)
10801 /* no data to recover */
10802 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10804 if (curr_migr_unit
>= num_migr_units
)
10807 /* find device during reshape */
10808 for (id
= super
->devlist
; id
; id
= id
->next
)
10809 if (is_gen_migration(id
->dev
))
10814 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10815 new_disks
= map_dest
->num_members
;
10817 read_offset
= (unsigned long long)
10818 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
10820 write_offset
= ((unsigned long long)
10821 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
10822 pba_of_lba0(map_dest
)) * 512;
10824 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10825 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10827 targets
= xcalloc(new_disks
, sizeof(int));
10829 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10830 pr_err("Cannot open some devices belonging to array.\n");
10834 for (i
= 0; i
< new_disks
; i
++) {
10835 if (targets
[i
] < 0) {
10839 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
10840 pr_err("Cannot seek to block: %s\n",
10845 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
10846 pr_err("Cannot read copy area block: %s\n",
10851 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
10852 pr_err("Cannot seek to block: %s\n",
10857 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
10858 pr_err("Cannot restore block: %s\n",
10865 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
10869 pr_err("Cannot restore data from backup. Too many failed disks\n");
10873 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
10874 /* ignore error == 2, this can mean end of reshape here
10876 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
10882 for (i
= 0; i
< new_disks
; i
++)
10891 static char disk_by_path
[] = "/dev/disk/by-path/";
10893 static const char *imsm_get_disk_controller_domain(const char *path
)
10895 char disk_path
[PATH_MAX
];
10899 strcpy(disk_path
, disk_by_path
);
10900 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
10901 if (stat(disk_path
, &st
) == 0) {
10902 struct sys_dev
* hba
;
10905 path
= devt_to_devpath(st
.st_rdev
);
10908 hba
= find_disk_attached_hba(-1, path
);
10909 if (hba
&& hba
->type
== SYS_DEV_SAS
)
10911 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
10913 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
10915 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
10919 dprintf("path: %s hba: %s attached: %s\n",
10920 path
, (hba
) ? hba
->path
: "NULL", drv
);
10926 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
10928 static char devnm
[32];
10929 char subdev_name
[20];
10930 struct mdstat_ent
*mdstat
;
10932 sprintf(subdev_name
, "%d", subdev
);
10933 mdstat
= mdstat_by_subdev(subdev_name
, container
);
10937 strcpy(devnm
, mdstat
->devnm
);
10938 free_mdstat(mdstat
);
10942 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
10943 struct geo_params
*geo
,
10944 int *old_raid_disks
,
10947 /* currently we only support increasing the number of devices
10948 * for a container. This increases the number of device for each
10949 * member array. They must all be RAID0 or RAID5.
10952 struct mdinfo
*info
, *member
;
10953 int devices_that_can_grow
= 0;
10955 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
10957 if (geo
->size
> 0 ||
10958 geo
->level
!= UnSet
||
10959 geo
->layout
!= UnSet
||
10960 geo
->chunksize
!= 0 ||
10961 geo
->raid_disks
== UnSet
) {
10962 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
10966 if (direction
== ROLLBACK_METADATA_CHANGES
) {
10967 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
10971 info
= container_content_imsm(st
, NULL
);
10972 for (member
= info
; member
; member
= member
->next
) {
10975 dprintf("imsm: checking device_num: %i\n",
10976 member
->container_member
);
10978 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
10979 /* we work on container for Online Capacity Expansion
10980 * only so raid_disks has to grow
10982 dprintf("imsm: for container operation raid disks increase is required\n");
10986 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
10987 /* we cannot use this container with other raid level
10989 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
10990 info
->array
.level
);
10993 /* check for platform support
10994 * for this raid level configuration
10996 struct intel_super
*super
= st
->sb
;
10997 if (!is_raid_level_supported(super
->orom
,
10998 member
->array
.level
,
10999 geo
->raid_disks
)) {
11000 dprintf("platform does not support raid%d with %d disk%s\n",
11003 geo
->raid_disks
> 1 ? "s" : "");
11006 /* check if component size is aligned to chunk size
11008 if (info
->component_size
%
11009 (info
->array
.chunk_size
/512)) {
11010 dprintf("Component size is not aligned to chunk size\n");
11015 if (*old_raid_disks
&&
11016 info
->array
.raid_disks
!= *old_raid_disks
)
11018 *old_raid_disks
= info
->array
.raid_disks
;
11020 /* All raid5 and raid0 volumes in container
11021 * have to be ready for Online Capacity Expansion
11022 * so they need to be assembled. We have already
11023 * checked that no recovery etc is happening.
11025 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11026 st
->container_devnm
);
11027 if (result
== NULL
) {
11028 dprintf("imsm: cannot find array\n");
11031 devices_that_can_grow
++;
11034 if (!member
&& devices_that_can_grow
)
11038 dprintf("Container operation allowed\n");
11040 dprintf("Error: %i\n", ret_val
);
11045 /* Function: get_spares_for_grow
11046 * Description: Allocates memory and creates list of spare devices
11047 * avaliable in container. Checks if spare drive size is acceptable.
11048 * Parameters: Pointer to the supertype structure
11049 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11052 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11054 struct spare_criteria sc
;
11056 get_spare_criteria_imsm(st
, &sc
);
11057 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11060 /******************************************************************************
11061 * function: imsm_create_metadata_update_for_reshape
11062 * Function creates update for whole IMSM container.
11064 ******************************************************************************/
11065 static int imsm_create_metadata_update_for_reshape(
11066 struct supertype
*st
,
11067 struct geo_params
*geo
,
11068 int old_raid_disks
,
11069 struct imsm_update_reshape
**updatep
)
11071 struct intel_super
*super
= st
->sb
;
11072 struct imsm_super
*mpb
= super
->anchor
;
11073 int update_memory_size
;
11074 struct imsm_update_reshape
*u
;
11075 struct mdinfo
*spares
;
11078 struct mdinfo
*dev
;
11080 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11082 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11084 /* size of all update data without anchor */
11085 update_memory_size
= sizeof(struct imsm_update_reshape
);
11087 /* now add space for spare disks that we need to add. */
11088 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11090 u
= xcalloc(1, update_memory_size
);
11091 u
->type
= update_reshape_container_disks
;
11092 u
->old_raid_disks
= old_raid_disks
;
11093 u
->new_raid_disks
= geo
->raid_disks
;
11095 /* now get spare disks list
11097 spares
= get_spares_for_grow(st
);
11099 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11100 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11105 /* we have got spares
11106 * update disk list in imsm_disk list table in anchor
11108 dprintf("imsm: %i spares are available.\n\n",
11109 spares
->array
.spare_disks
);
11111 dev
= spares
->devs
;
11112 for (i
= 0; i
< delta_disks
; i
++) {
11117 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11119 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11120 dl
->index
= mpb
->num_disks
;
11128 sysfs_free(spares
);
11130 dprintf("imsm: reshape update preparation :");
11131 if (i
== delta_disks
) {
11132 dprintf_cont(" OK\n");
11134 return update_memory_size
;
11137 dprintf_cont(" Error\n");
11142 /******************************************************************************
11143 * function: imsm_create_metadata_update_for_size_change()
11144 * Creates update for IMSM array for array size change.
11146 ******************************************************************************/
11147 static int imsm_create_metadata_update_for_size_change(
11148 struct supertype
*st
,
11149 struct geo_params
*geo
,
11150 struct imsm_update_size_change
**updatep
)
11152 struct intel_super
*super
= st
->sb
;
11153 int update_memory_size
;
11154 struct imsm_update_size_change
*u
;
11156 dprintf("(enter) New size = %llu\n", geo
->size
);
11158 /* size of all update data without anchor */
11159 update_memory_size
= sizeof(struct imsm_update_size_change
);
11161 u
= xcalloc(1, update_memory_size
);
11162 u
->type
= update_size_change
;
11163 u
->subdev
= super
->current_vol
;
11164 u
->new_size
= geo
->size
;
11166 dprintf("imsm: reshape update preparation : OK\n");
11169 return update_memory_size
;
11172 /******************************************************************************
11173 * function: imsm_create_metadata_update_for_migration()
11174 * Creates update for IMSM array.
11176 ******************************************************************************/
11177 static int imsm_create_metadata_update_for_migration(
11178 struct supertype
*st
,
11179 struct geo_params
*geo
,
11180 struct imsm_update_reshape_migration
**updatep
)
11182 struct intel_super
*super
= st
->sb
;
11183 int update_memory_size
;
11184 struct imsm_update_reshape_migration
*u
;
11185 struct imsm_dev
*dev
;
11186 int previous_level
= -1;
11188 dprintf("(enter) New Level = %i\n", geo
->level
);
11190 /* size of all update data without anchor */
11191 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11193 u
= xcalloc(1, update_memory_size
);
11194 u
->type
= update_reshape_migration
;
11195 u
->subdev
= super
->current_vol
;
11196 u
->new_level
= geo
->level
;
11197 u
->new_layout
= geo
->layout
;
11198 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11199 u
->new_disks
[0] = -1;
11200 u
->new_chunksize
= -1;
11202 dev
= get_imsm_dev(super
, u
->subdev
);
11204 struct imsm_map
*map
;
11206 map
= get_imsm_map(dev
, MAP_0
);
11208 int current_chunk_size
=
11209 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11211 if (geo
->chunksize
!= current_chunk_size
) {
11212 u
->new_chunksize
= geo
->chunksize
/ 1024;
11213 dprintf("imsm: chunk size change from %i to %i\n",
11214 current_chunk_size
, u
->new_chunksize
);
11216 previous_level
= map
->raid_level
;
11219 if (geo
->level
== 5 && previous_level
== 0) {
11220 struct mdinfo
*spares
= NULL
;
11222 u
->new_raid_disks
++;
11223 spares
= get_spares_for_grow(st
);
11224 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11226 sysfs_free(spares
);
11227 update_memory_size
= 0;
11228 pr_err("cannot get spare device for requested migration\n");
11231 sysfs_free(spares
);
11233 dprintf("imsm: reshape update preparation : OK\n");
11236 return update_memory_size
;
11239 static void imsm_update_metadata_locally(struct supertype
*st
,
11240 void *buf
, int len
)
11242 struct metadata_update mu
;
11247 mu
.space_list
= NULL
;
11249 if (imsm_prepare_update(st
, &mu
))
11250 imsm_process_update(st
, &mu
);
11252 while (mu
.space_list
) {
11253 void **space
= mu
.space_list
;
11254 mu
.space_list
= *space
;
11259 /***************************************************************************
11260 * Function: imsm_analyze_change
11261 * Description: Function analyze change for single volume
11262 * and validate if transition is supported
11263 * Parameters: Geometry parameters, supertype structure,
11264 * metadata change direction (apply/rollback)
11265 * Returns: Operation type code on success, -1 if fail
11266 ****************************************************************************/
11267 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11268 struct geo_params
*geo
,
11271 struct mdinfo info
;
11273 int check_devs
= 0;
11275 /* number of added/removed disks in operation result */
11276 int devNumChange
= 0;
11277 /* imsm compatible layout value for array geometry verification */
11278 int imsm_layout
= -1;
11280 struct imsm_dev
*dev
;
11281 struct intel_super
*super
;
11282 unsigned long long current_size
;
11283 unsigned long long free_size
;
11284 unsigned long long max_size
;
11287 getinfo_super_imsm_volume(st
, &info
, NULL
);
11288 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11289 geo
->level
!= UnSet
) {
11290 switch (info
.array
.level
) {
11292 if (geo
->level
== 5) {
11293 change
= CH_MIGRATION
;
11294 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11295 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11297 goto analyse_change_exit
;
11299 imsm_layout
= geo
->layout
;
11301 devNumChange
= 1; /* parity disk added */
11302 } else if (geo
->level
== 10) {
11303 change
= CH_TAKEOVER
;
11305 devNumChange
= 2; /* two mirrors added */
11306 imsm_layout
= 0x102; /* imsm supported layout */
11311 if (geo
->level
== 0) {
11312 change
= CH_TAKEOVER
;
11314 devNumChange
= -(geo
->raid_disks
/2);
11315 imsm_layout
= 0; /* imsm raid0 layout */
11319 if (change
== -1) {
11320 pr_err("Error. Level Migration from %d to %d not supported!\n",
11321 info
.array
.level
, geo
->level
);
11322 goto analyse_change_exit
;
11325 geo
->level
= info
.array
.level
;
11327 if (geo
->layout
!= info
.array
.layout
&&
11328 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11329 change
= CH_MIGRATION
;
11330 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11331 geo
->layout
== 5) {
11332 /* reshape 5 -> 4 */
11333 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11334 geo
->layout
== 0) {
11335 /* reshape 4 -> 5 */
11339 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11340 info
.array
.layout
, geo
->layout
);
11342 goto analyse_change_exit
;
11345 geo
->layout
= info
.array
.layout
;
11346 if (imsm_layout
== -1)
11347 imsm_layout
= info
.array
.layout
;
11350 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11351 geo
->chunksize
!= info
.array
.chunk_size
) {
11352 if (info
.array
.level
== 10) {
11353 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11355 goto analyse_change_exit
;
11356 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11357 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11358 geo
->chunksize
/1024, info
.component_size
/2);
11360 goto analyse_change_exit
;
11362 change
= CH_MIGRATION
;
11364 geo
->chunksize
= info
.array
.chunk_size
;
11367 chunk
= geo
->chunksize
/ 1024;
11370 dev
= get_imsm_dev(super
, super
->current_vol
);
11371 data_disks
= imsm_num_data_members(dev
, MAP_0
);
11372 /* compute current size per disk member
11374 current_size
= info
.custom_array_size
/ data_disks
;
11376 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11377 /* align component size
11379 geo
->size
= imsm_component_size_aligment_check(
11380 get_imsm_raid_level(dev
->vol
.map
),
11381 chunk
* 1024, super
->sector_size
,
11383 if (geo
->size
== 0) {
11384 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11386 goto analyse_change_exit
;
11390 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11391 if (change
!= -1) {
11392 pr_err("Error. Size change should be the only one at a time.\n");
11394 goto analyse_change_exit
;
11396 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11397 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11398 super
->current_vol
, st
->devnm
);
11399 goto analyse_change_exit
;
11401 /* check the maximum available size
11403 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11404 0, chunk
, &free_size
);
11406 /* Cannot find maximum available space
11410 max_size
= free_size
+ current_size
;
11411 /* align component size
11413 max_size
= imsm_component_size_aligment_check(
11414 get_imsm_raid_level(dev
->vol
.map
),
11415 chunk
* 1024, super
->sector_size
,
11418 if (geo
->size
== MAX_SIZE
) {
11419 /* requested size change to the maximum available size
11421 if (max_size
== 0) {
11422 pr_err("Error. Cannot find maximum available space.\n");
11424 goto analyse_change_exit
;
11426 geo
->size
= max_size
;
11429 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11430 /* accept size for rollback only
11433 /* round size due to metadata compatibility
11435 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11436 << SECT_PER_MB_SHIFT
;
11437 dprintf("Prepare update for size change to %llu\n",
11439 if (current_size
>= geo
->size
) {
11440 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11441 current_size
, geo
->size
);
11442 goto analyse_change_exit
;
11444 if (max_size
&& geo
->size
> max_size
) {
11445 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11446 max_size
, geo
->size
);
11447 goto analyse_change_exit
;
11450 geo
->size
*= data_disks
;
11451 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11452 change
= CH_ARRAY_SIZE
;
11454 if (!validate_geometry_imsm(st
,
11457 geo
->raid_disks
+ devNumChange
,
11459 geo
->size
, INVALID_SECTORS
,
11460 0, 0, info
.consistency_policy
, 1))
11464 struct intel_super
*super
= st
->sb
;
11465 struct imsm_super
*mpb
= super
->anchor
;
11467 if (mpb
->num_raid_devs
> 1) {
11468 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11474 analyse_change_exit
:
11475 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11476 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11477 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11483 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11485 struct intel_super
*super
= st
->sb
;
11486 struct imsm_update_takeover
*u
;
11488 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11490 u
->type
= update_takeover
;
11491 u
->subarray
= super
->current_vol
;
11493 /* 10->0 transition */
11494 if (geo
->level
== 0)
11495 u
->direction
= R10_TO_R0
;
11497 /* 0->10 transition */
11498 if (geo
->level
== 10)
11499 u
->direction
= R0_TO_R10
;
11501 /* update metadata locally */
11502 imsm_update_metadata_locally(st
, u
,
11503 sizeof(struct imsm_update_takeover
));
11504 /* and possibly remotely */
11505 if (st
->update_tail
)
11506 append_metadata_update(st
, u
,
11507 sizeof(struct imsm_update_takeover
));
11514 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11516 int layout
, int chunksize
, int raid_disks
,
11517 int delta_disks
, char *backup
, char *dev
,
11518 int direction
, int verbose
)
11521 struct geo_params geo
;
11523 dprintf("(enter)\n");
11525 memset(&geo
, 0, sizeof(struct geo_params
));
11527 geo
.dev_name
= dev
;
11528 strcpy(geo
.devnm
, st
->devnm
);
11531 geo
.layout
= layout
;
11532 geo
.chunksize
= chunksize
;
11533 geo
.raid_disks
= raid_disks
;
11534 if (delta_disks
!= UnSet
)
11535 geo
.raid_disks
+= delta_disks
;
11537 dprintf("for level : %i\n", geo
.level
);
11538 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11540 if (experimental() == 0)
11543 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11544 /* On container level we can only increase number of devices. */
11545 dprintf("imsm: info: Container operation\n");
11546 int old_raid_disks
= 0;
11548 if (imsm_reshape_is_allowed_on_container(
11549 st
, &geo
, &old_raid_disks
, direction
)) {
11550 struct imsm_update_reshape
*u
= NULL
;
11553 len
= imsm_create_metadata_update_for_reshape(
11554 st
, &geo
, old_raid_disks
, &u
);
11557 dprintf("imsm: Cannot prepare update\n");
11558 goto exit_imsm_reshape_super
;
11562 /* update metadata locally */
11563 imsm_update_metadata_locally(st
, u
, len
);
11564 /* and possibly remotely */
11565 if (st
->update_tail
)
11566 append_metadata_update(st
, u
, len
);
11571 pr_err("(imsm) Operation is not allowed on this container\n");
11574 /* On volume level we support following operations
11575 * - takeover: raid10 -> raid0; raid0 -> raid10
11576 * - chunk size migration
11577 * - migration: raid5 -> raid0; raid0 -> raid5
11579 struct intel_super
*super
= st
->sb
;
11580 struct intel_dev
*dev
= super
->devlist
;
11582 dprintf("imsm: info: Volume operation\n");
11583 /* find requested device */
11586 imsm_find_array_devnm_by_subdev(
11587 dev
->index
, st
->container_devnm
);
11588 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11593 pr_err("Cannot find %s (%s) subarray\n",
11594 geo
.dev_name
, geo
.devnm
);
11595 goto exit_imsm_reshape_super
;
11597 super
->current_vol
= dev
->index
;
11598 change
= imsm_analyze_change(st
, &geo
, direction
);
11601 ret_val
= imsm_takeover(st
, &geo
);
11603 case CH_MIGRATION
: {
11604 struct imsm_update_reshape_migration
*u
= NULL
;
11606 imsm_create_metadata_update_for_migration(
11609 dprintf("imsm: Cannot prepare update\n");
11613 /* update metadata locally */
11614 imsm_update_metadata_locally(st
, u
, len
);
11615 /* and possibly remotely */
11616 if (st
->update_tail
)
11617 append_metadata_update(st
, u
, len
);
11622 case CH_ARRAY_SIZE
: {
11623 struct imsm_update_size_change
*u
= NULL
;
11625 imsm_create_metadata_update_for_size_change(
11628 dprintf("imsm: Cannot prepare update\n");
11632 /* update metadata locally */
11633 imsm_update_metadata_locally(st
, u
, len
);
11634 /* and possibly remotely */
11635 if (st
->update_tail
)
11636 append_metadata_update(st
, u
, len
);
11646 exit_imsm_reshape_super
:
11647 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11651 #define COMPLETED_OK 0
11652 #define COMPLETED_NONE 1
11653 #define COMPLETED_DELAYED 2
11655 static int read_completed(int fd
, unsigned long long *val
)
11660 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11664 ret
= COMPLETED_OK
;
11665 if (strncmp(buf
, "none", 4) == 0) {
11666 ret
= COMPLETED_NONE
;
11667 } else if (strncmp(buf
, "delayed", 7) == 0) {
11668 ret
= COMPLETED_DELAYED
;
11671 *val
= strtoull(buf
, &ep
, 0);
11672 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11678 /*******************************************************************************
11679 * Function: wait_for_reshape_imsm
11680 * Description: Function writes new sync_max value and waits until
11681 * reshape process reach new position
11683 * sra : general array info
11684 * ndata : number of disks in new array's layout
11687 * 1 : there is no reshape in progress,
11689 ******************************************************************************/
11690 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11692 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11694 unsigned long long completed
;
11695 /* to_complete : new sync_max position */
11696 unsigned long long to_complete
= sra
->reshape_progress
;
11697 unsigned long long position_to_set
= to_complete
/ ndata
;
11700 dprintf("cannot open reshape_position\n");
11705 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11707 dprintf("cannot read reshape_position (no reshape in progres)\n");
11716 if (completed
> position_to_set
) {
11717 dprintf("wrong next position to set %llu (%llu)\n",
11718 to_complete
, position_to_set
);
11722 dprintf("Position set: %llu\n", position_to_set
);
11723 if (sysfs_set_num(sra
, NULL
, "sync_max",
11724 position_to_set
) != 0) {
11725 dprintf("cannot set reshape position to %llu\n",
11734 int timeout
= 3000;
11736 sysfs_wait(fd
, &timeout
);
11737 if (sysfs_get_str(sra
, NULL
, "sync_action",
11739 strncmp(action
, "reshape", 7) != 0) {
11740 if (strncmp(action
, "idle", 4) == 0)
11746 rc
= read_completed(fd
, &completed
);
11748 dprintf("cannot read reshape_position (in loop)\n");
11751 } else if (rc
== COMPLETED_NONE
)
11753 } while (completed
< position_to_set
);
11759 /*******************************************************************************
11760 * Function: check_degradation_change
11761 * Description: Check that array hasn't become failed.
11763 * info : for sysfs access
11764 * sources : source disks descriptors
11765 * degraded: previous degradation level
11767 * degradation level
11768 ******************************************************************************/
11769 int check_degradation_change(struct mdinfo
*info
,
11773 unsigned long long new_degraded
;
11776 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11777 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11778 /* check each device to ensure it is still working */
11781 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11782 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11784 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11787 if (sysfs_get_str(info
,
11788 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11789 strstr(sbuf
, "faulty") ||
11790 strstr(sbuf
, "in_sync") == NULL
) {
11791 /* this device is dead */
11792 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11793 if (sd
->disk
.raid_disk
>= 0 &&
11794 sources
[sd
->disk
.raid_disk
] >= 0) {
11796 sd
->disk
.raid_disk
]);
11797 sources
[sd
->disk
.raid_disk
] =
11806 return new_degraded
;
11809 /*******************************************************************************
11810 * Function: imsm_manage_reshape
11811 * Description: Function finds array under reshape and it manages reshape
11812 * process. It creates stripes backups (if required) and sets
11815 * afd : Backup handle (nattive) - not used
11816 * sra : general array info
11817 * reshape : reshape parameters - not used
11818 * st : supertype structure
11819 * blocks : size of critical section [blocks]
11820 * fds : table of source device descriptor
11821 * offsets : start of array (offest per devices)
11823 * destfd : table of destination device descriptor
11824 * destoffsets : table of destination offsets (per device)
11826 * 1 : success, reshape is done
11828 ******************************************************************************/
11829 static int imsm_manage_reshape(
11830 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
11831 struct supertype
*st
, unsigned long backup_blocks
,
11832 int *fds
, unsigned long long *offsets
,
11833 int dests
, int *destfd
, unsigned long long *destoffsets
)
11836 struct intel_super
*super
= st
->sb
;
11837 struct intel_dev
*dv
;
11838 unsigned int sector_size
= super
->sector_size
;
11839 struct imsm_dev
*dev
= NULL
;
11840 struct imsm_map
*map_src
;
11841 int migr_vol_qan
= 0;
11842 int ndata
, odata
; /* [bytes] */
11843 int chunk
; /* [bytes] */
11844 struct migr_record
*migr_rec
;
11846 unsigned int buf_size
; /* [bytes] */
11847 unsigned long long max_position
; /* array size [bytes] */
11848 unsigned long long next_step
; /* [blocks]/[bytes] */
11849 unsigned long long old_data_stripe_length
;
11850 unsigned long long start_src
; /* [bytes] */
11851 unsigned long long start
; /* [bytes] */
11852 unsigned long long start_buf_shift
; /* [bytes] */
11854 int source_layout
= 0;
11859 if (!fds
|| !offsets
)
11862 /* Find volume during the reshape */
11863 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
11864 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
11865 dv
->dev
->vol
.migr_state
== 1) {
11870 /* Only one volume can migrate at the same time */
11871 if (migr_vol_qan
!= 1) {
11872 pr_err("%s", migr_vol_qan
?
11873 "Number of migrating volumes greater than 1\n" :
11874 "There is no volume during migrationg\n");
11878 map_src
= get_imsm_map(dev
, MAP_1
);
11879 if (map_src
== NULL
)
11882 ndata
= imsm_num_data_members(dev
, MAP_0
);
11883 odata
= imsm_num_data_members(dev
, MAP_1
);
11885 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
11886 old_data_stripe_length
= odata
* chunk
;
11888 migr_rec
= super
->migr_rec
;
11890 /* initialize migration record for start condition */
11891 if (sra
->reshape_progress
== 0)
11892 init_migr_record_imsm(st
, dev
, sra
);
11894 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
11895 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
11898 /* Save checkpoint to update migration record for current
11899 * reshape position (in md). It can be farther than current
11900 * reshape position in metadata.
11902 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11903 /* ignore error == 2, this can mean end of reshape here
11905 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
11910 /* size for data */
11911 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
11912 /* extend buffer size for parity disk */
11913 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11914 /* add space for stripe aligment */
11915 buf_size
+= old_data_stripe_length
;
11916 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
11917 dprintf("imsm: Cannot allocate checkpoint buffer\n");
11921 max_position
= sra
->component_size
* ndata
;
11922 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
11924 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
11925 __le32_to_cpu(migr_rec
->num_migr_units
)) {
11926 /* current reshape position [blocks] */
11927 unsigned long long current_position
=
11928 __le32_to_cpu(migr_rec
->blocks_per_unit
)
11929 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
11930 unsigned long long border
;
11932 /* Check that array hasn't become failed.
11934 degraded
= check_degradation_change(sra
, fds
, degraded
);
11935 if (degraded
> 1) {
11936 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
11940 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
11942 if ((current_position
+ next_step
) > max_position
)
11943 next_step
= max_position
- current_position
;
11945 start
= current_position
* 512;
11947 /* align reading start to old geometry */
11948 start_buf_shift
= start
% old_data_stripe_length
;
11949 start_src
= start
- start_buf_shift
;
11951 border
= (start_src
/ odata
) - (start
/ ndata
);
11953 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
11954 /* save critical stripes to buf
11955 * start - start address of current unit
11956 * to backup [bytes]
11957 * start_src - start address of current unit
11958 * to backup alligned to source array
11961 unsigned long long next_step_filler
;
11962 unsigned long long copy_length
= next_step
* 512;
11964 /* allign copy area length to stripe in old geometry */
11965 next_step_filler
= ((copy_length
+ start_buf_shift
)
11966 % old_data_stripe_length
);
11967 if (next_step_filler
)
11968 next_step_filler
= (old_data_stripe_length
11969 - next_step_filler
);
11970 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
11971 start
, start_src
, copy_length
,
11972 start_buf_shift
, next_step_filler
);
11974 if (save_stripes(fds
, offsets
, map_src
->num_members
,
11975 chunk
, map_src
->raid_level
,
11976 source_layout
, 0, NULL
, start_src
,
11978 next_step_filler
+ start_buf_shift
,
11980 dprintf("imsm: Cannot save stripes to buffer\n");
11983 /* Convert data to destination format and store it
11984 * in backup general migration area
11986 if (save_backup_imsm(st
, dev
, sra
,
11987 buf
+ start_buf_shift
, copy_length
)) {
11988 dprintf("imsm: Cannot save stripes to target devices\n");
11991 if (save_checkpoint_imsm(st
, sra
,
11992 UNIT_SRC_IN_CP_AREA
)) {
11993 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
11997 /* set next step to use whole border area */
11998 border
/= next_step
;
12000 next_step
*= border
;
12002 /* When data backed up, checkpoint stored,
12003 * kick the kernel to reshape unit of data
12005 next_step
= next_step
+ sra
->reshape_progress
;
12006 /* limit next step to array max position */
12007 if (next_step
> max_position
)
12008 next_step
= max_position
;
12009 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12010 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12011 sra
->reshape_progress
= next_step
;
12013 /* wait until reshape finish */
12014 if (wait_for_reshape_imsm(sra
, ndata
)) {
12015 dprintf("wait_for_reshape_imsm returned error!\n");
12021 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12022 /* ignore error == 2, this can mean end of reshape here
12024 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12030 /* clear migr_rec on disks after successful migration */
12033 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12034 for (d
= super
->disks
; d
; d
= d
->next
) {
12035 if (d
->index
< 0 || is_failed(&d
->disk
))
12037 unsigned long long dsize
;
12039 get_dev_size(d
->fd
, NULL
, &dsize
);
12040 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12042 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12043 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12044 MIGR_REC_BUF_SECTORS
*sector_size
)
12045 perror("Write migr_rec failed");
12049 /* return '1' if done */
12053 /* See Grow.c: abort_reshape() for further explanation */
12054 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12055 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12056 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12061 struct superswitch super_imsm
= {
12062 .examine_super
= examine_super_imsm
,
12063 .brief_examine_super
= brief_examine_super_imsm
,
12064 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12065 .export_examine_super
= export_examine_super_imsm
,
12066 .detail_super
= detail_super_imsm
,
12067 .brief_detail_super
= brief_detail_super_imsm
,
12068 .write_init_super
= write_init_super_imsm
,
12069 .validate_geometry
= validate_geometry_imsm
,
12070 .add_to_super
= add_to_super_imsm
,
12071 .remove_from_super
= remove_from_super_imsm
,
12072 .detail_platform
= detail_platform_imsm
,
12073 .export_detail_platform
= export_detail_platform_imsm
,
12074 .kill_subarray
= kill_subarray_imsm
,
12075 .update_subarray
= update_subarray_imsm
,
12076 .load_container
= load_container_imsm
,
12077 .default_geometry
= default_geometry_imsm
,
12078 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12079 .reshape_super
= imsm_reshape_super
,
12080 .manage_reshape
= imsm_manage_reshape
,
12081 .recover_backup
= recover_backup_imsm
,
12082 .copy_metadata
= copy_metadata_imsm
,
12083 .examine_badblocks
= examine_badblocks_imsm
,
12084 .match_home
= match_home_imsm
,
12085 .uuid_from_super
= uuid_from_super_imsm
,
12086 .getinfo_super
= getinfo_super_imsm
,
12087 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12088 .update_super
= update_super_imsm
,
12090 .avail_size
= avail_size_imsm
,
12091 .get_spare_criteria
= get_spare_criteria_imsm
,
12093 .compare_super
= compare_super_imsm
,
12095 .load_super
= load_super_imsm
,
12096 .init_super
= init_super_imsm
,
12097 .store_super
= store_super_imsm
,
12098 .free_super
= free_super_imsm
,
12099 .match_metadata_desc
= match_metadata_desc_imsm
,
12100 .container_content
= container_content_imsm
,
12101 .validate_container
= validate_container_imsm
,
12103 .write_init_ppl
= write_init_ppl_imsm
,
12104 .validate_ppl
= validate_ppl_imsm
,
12110 .open_new
= imsm_open_new
,
12111 .set_array_state
= imsm_set_array_state
,
12112 .set_disk
= imsm_set_disk
,
12113 .sync_metadata
= imsm_sync_metadata
,
12114 .activate_spare
= imsm_activate_spare
,
12115 .process_update
= imsm_process_update
,
12116 .prepare_update
= imsm_prepare_update
,
12117 .record_bad_block
= imsm_record_badblock
,
12118 .clear_bad_block
= imsm_clear_badblock
,
12119 .get_bad_blocks
= imsm_get_badblocks
,