2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
47 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
51 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
53 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
54 /* supports RAID CNG */
55 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
56 /* supports expanded stripe sizes of 256K, 512K and 1MB */
57 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
59 /* The OROM Support RST Caching of Volumes */
60 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
61 /* The OROM supports creating disks greater than 2TB */
62 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
63 /* The OROM supports Bad Block Management */
64 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
66 /* THe OROM Supports NVM Caching of Volumes */
67 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
68 /* The OROM supports creating volumes greater than 2TB */
69 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
70 /* originally for PMP, now it's wasted b/c. Never use this bit! */
71 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
72 /* Verify MPB contents against checksum after reading MPB */
73 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
75 /* Define all supported attributes that have to be accepted by mdadm
77 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
79 MPB_ATTRIB_2TB_DISK | \
84 MPB_ATTRIB_EXP_STRIPE_SIZE | \
87 /* Define attributes that are unused but not harmful */
88 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
90 #define MPB_SECTOR_CNT 2210
91 #define IMSM_RESERVED_SECTORS 8192
92 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2048
93 #define SECT_PER_MB_SHIFT 11
94 #define MAX_SECTOR_SIZE 4096
95 #define MULTIPLE_PPL_AREA_SIZE_IMSM (1024 * 1024) /* Size of the whole
99 /* Disk configuration info. */
100 #define IMSM_MAX_DEVICES 255
102 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
103 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
104 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
105 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
106 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
107 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
108 #define JOURNAL_DISK __cpu_to_le32(0x2000000) /* Device marked as Journaling Drive */
109 __u32 status
; /* 0xF0 - 0xF3 */
110 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
111 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
112 #define IMSM_DISK_FILLERS 3
113 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
116 /* map selector for map managment
122 /* RAID map configuration infos. */
124 __u32 pba_of_lba0_lo
; /* start address of partition */
125 __u32 blocks_per_member_lo
;/* blocks per member */
126 __u32 num_data_stripes_lo
; /* number of data stripes */
127 __u16 blocks_per_strip
;
128 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
129 #define IMSM_T_STATE_NORMAL 0
130 #define IMSM_T_STATE_UNINITIALIZED 1
131 #define IMSM_T_STATE_DEGRADED 2
132 #define IMSM_T_STATE_FAILED 3
134 #define IMSM_T_RAID0 0
135 #define IMSM_T_RAID1 1
136 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
137 __u8 num_members
; /* number of member disks */
138 __u8 num_domains
; /* number of parity domains */
139 __u8 failed_disk_num
; /* valid only when state is degraded */
141 __u32 pba_of_lba0_hi
;
142 __u32 blocks_per_member_hi
;
143 __u32 num_data_stripes_hi
;
144 __u32 filler
[4]; /* expansion area */
145 #define IMSM_ORD_REBUILD (1 << 24)
146 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
147 * top byte contains some flags
149 } __attribute__ ((packed
));
152 __u32 curr_migr_unit
;
153 __u32 checkpoint_id
; /* id to access curr_migr_unit */
154 __u8 migr_state
; /* Normal or Migrating */
156 #define MIGR_REBUILD 1
157 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
158 #define MIGR_GEN_MIGR 3
159 #define MIGR_STATE_CHANGE 4
160 #define MIGR_REPAIR 5
161 __u8 migr_type
; /* Initializing, Rebuilding, ... */
162 #define RAIDVOL_CLEAN 0
163 #define RAIDVOL_DIRTY 1
164 #define RAIDVOL_DSRECORD_VALID 2
166 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
167 __u16 verify_errors
; /* number of mismatches */
168 __u16 bad_blocks
; /* number of bad blocks during verify */
170 struct imsm_map map
[1];
171 /* here comes another one if migr_state */
172 } __attribute__ ((packed
));
175 __u8 volume
[MAX_RAID_SERIAL_LEN
];
178 #define DEV_BOOTABLE __cpu_to_le32(0x01)
179 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
180 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
181 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
182 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
183 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
184 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
185 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
186 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
187 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
188 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
189 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
190 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
191 __u32 status
; /* Persistent RaidDev status */
192 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
196 __u8 cng_master_disk
;
200 __u16 my_vol_raid_dev_num
; /* Used in Unique volume Id for this RaidDev */
206 /* Unique Volume Id of the NvCache Volume associated with this volume */
207 __u32 nvc_vol_orig_family_num
;
208 __u16 nvc_vol_raid_dev_num
;
211 #define RWH_DISTRIBUTED 1
212 #define RWH_JOURNALING_DRIVE 2
213 #define RWH_MULTIPLE_DISTRIBUTED 3
214 #define RWH_MULTIPLE_PPLS_JOURNALING_DRIVE 4
215 #define RWH_MULTIPLE_OFF 5
216 __u8 rwh_policy
; /* Raid Write Hole Policy */
217 __u8 jd_serial
[MAX_RAID_SERIAL_LEN
]; /* Journal Drive serial number */
220 #define IMSM_DEV_FILLERS 3
221 __u32 filler
[IMSM_DEV_FILLERS
];
223 } __attribute__ ((packed
));
226 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
227 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
228 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
229 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
230 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
231 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
232 __u32 attributes
; /* 0x34 - 0x37 */
233 __u8 num_disks
; /* 0x38 Number of configured disks */
234 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
235 __u8 error_log_pos
; /* 0x3A */
236 __u8 fill
[1]; /* 0x3B */
237 __u32 cache_size
; /* 0x3c - 0x40 in mb */
238 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
239 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
240 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
241 __u16 num_raid_devs_created
; /* 0x4C - 0x4D Used for generating unique
242 * volume IDs for raid_dev created in this array
245 __u16 filler1
; /* 0x4E - 0x4F */
246 #define IMSM_FILLERS 34
247 __u32 filler
[IMSM_FILLERS
]; /* 0x50 - 0xD7 RAID_MPB_FILLERS */
248 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
249 /* here comes imsm_dev[num_raid_devs] */
250 /* here comes BBM logs */
251 } __attribute__ ((packed
));
253 #define BBM_LOG_MAX_ENTRIES 254
254 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
255 #define BBM_LOG_SIGNATURE 0xabadb10c
257 struct bbm_log_block_addr
{
260 } __attribute__ ((__packed__
));
262 struct bbm_log_entry
{
263 __u8 marked_count
; /* Number of blocks marked - 1 */
264 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
265 struct bbm_log_block_addr defective_block_start
;
266 } __attribute__ ((__packed__
));
269 __u32 signature
; /* 0xABADB10C */
271 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
272 } __attribute__ ((__packed__
));
274 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
276 #define BLOCKS_PER_KB (1024/512)
278 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
280 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
282 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
283 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
284 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
287 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
288 * be recovered using srcMap */
289 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
290 * already been migrated and must
291 * be recovered from checkpoint area */
293 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of single PPL, without the header */
296 __u32 rec_status
; /* Status used to determine how to restart
297 * migration in case it aborts
299 __u32 curr_migr_unit
; /* 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 unsigned long long imsm_dev_size(struct imsm_dev
*dev
)
1208 return join_u32(dev
->size_low
, dev
->size_high
);
1211 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1213 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1216 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1218 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1221 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1223 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1226 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1228 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1231 static void set_imsm_dev_size(struct imsm_dev
*dev
, unsigned long long n
)
1233 split_ull(n
, &dev
->size_low
, &dev
->size_high
);
1236 static unsigned long long per_dev_array_size(struct imsm_map
*map
)
1238 unsigned long long array_size
= 0;
1243 array_size
= num_data_stripes(map
) * map
->blocks_per_strip
;
1244 if (get_imsm_raid_level(map
) == 1 || get_imsm_raid_level(map
) == 10)
1250 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
1252 /* find a list of used extents on the given physical device */
1253 struct extent
*rv
, *e
;
1255 int memberships
= count_memberships(dl
, super
);
1258 /* trim the reserved area for spares, so they can join any array
1259 * regardless of whether the OROM has assigned sectors from the
1260 * IMSM_RESERVED_SECTORS region
1262 if (dl
->index
== -1)
1263 reservation
= imsm_min_reserved_sectors(super
);
1265 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1267 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
1270 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1271 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1272 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1274 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1275 e
->start
= pba_of_lba0(map
);
1276 e
->size
= per_dev_array_size(map
);
1280 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1282 /* determine the start of the metadata
1283 * when no raid devices are defined use the default
1284 * ...otherwise allow the metadata to truncate the value
1285 * as is the case with older versions of imsm
1288 struct extent
*last
= &rv
[memberships
- 1];
1289 unsigned long long remainder
;
1291 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1292 /* round down to 1k block to satisfy precision of the kernel
1296 /* make sure remainder is still sane */
1297 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1298 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1299 if (reservation
> remainder
)
1300 reservation
= remainder
;
1302 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1307 /* try to determine how much space is reserved for metadata from
1308 * the last get_extents() entry, otherwise fallback to the
1311 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1317 /* for spares just return a minimal reservation which will grow
1318 * once the spare is picked up by an array
1320 if (dl
->index
== -1)
1321 return MPB_SECTOR_CNT
;
1323 e
= get_extents(super
, dl
);
1325 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1327 /* scroll to last entry */
1328 for (i
= 0; e
[i
].size
; i
++)
1331 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1338 static int is_spare(struct imsm_disk
*disk
)
1340 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1343 static int is_configured(struct imsm_disk
*disk
)
1345 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1348 static int is_failed(struct imsm_disk
*disk
)
1350 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1353 static int is_journal(struct imsm_disk
*disk
)
1355 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1358 /* round array size down to closest MB and ensure it splits evenly
1361 static unsigned long long round_size_to_mb(unsigned long long size
, unsigned int
1365 size
= (size
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
1371 static int able_to_resync(int raid_level
, int missing_disks
)
1373 int max_missing_disks
= 0;
1375 switch (raid_level
) {
1377 max_missing_disks
= 1;
1380 max_missing_disks
= 0;
1382 return missing_disks
<= max_missing_disks
;
1385 /* try to determine how much space is reserved for metadata from
1386 * the last get_extents() entry on the smallest active disk,
1387 * otherwise fallback to the default
1389 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1393 unsigned long long min_active
;
1395 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1396 struct dl
*dl
, *dl_min
= NULL
;
1402 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1405 unsigned long long blocks
= total_blocks(&dl
->disk
);
1406 if (blocks
< min_active
|| min_active
== 0) {
1408 min_active
= blocks
;
1414 /* find last lba used by subarrays on the smallest active disk */
1415 e
= get_extents(super
, dl_min
);
1418 for (i
= 0; e
[i
].size
; i
++)
1421 remainder
= min_active
- e
[i
].start
;
1424 /* to give priority to recovery we should not require full
1425 IMSM_RESERVED_SECTORS from the spare */
1426 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1428 /* if real reservation is smaller use that value */
1429 return (remainder
< rv
) ? remainder
: rv
;
1433 * Return minimum size of a spare and sector size
1434 * that can be used in this array
1436 int get_spare_criteria_imsm(struct supertype
*st
, struct spare_criteria
*c
)
1438 struct intel_super
*super
= st
->sb
;
1442 unsigned long long size
= 0;
1449 /* find first active disk in array */
1451 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1455 /* find last lba used by subarrays */
1456 e
= get_extents(super
, dl
);
1459 for (i
= 0; e
[i
].size
; i
++)
1462 size
= e
[i
-1].start
+ e
[i
-1].size
;
1465 /* add the amount of space needed for metadata */
1466 size
+= imsm_min_reserved_sectors(super
);
1468 c
->min_size
= size
* 512;
1469 c
->sector_size
= super
->sector_size
;
1474 static int is_gen_migration(struct imsm_dev
*dev
);
1476 #define IMSM_4K_DIV 8
1478 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1479 struct imsm_dev
*dev
);
1481 static void print_imsm_dev(struct intel_super
*super
,
1482 struct imsm_dev
*dev
,
1488 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1489 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1493 printf("[%.16s]:\n", dev
->volume
);
1494 printf(" UUID : %s\n", uuid
);
1495 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1497 printf(" <-- %d", get_imsm_raid_level(map2
));
1499 printf(" Members : %d", map
->num_members
);
1501 printf(" <-- %d", map2
->num_members
);
1503 printf(" Slots : [");
1504 for (i
= 0; i
< map
->num_members
; i
++) {
1505 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1506 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1511 for (i
= 0; i
< map2
->num_members
; i
++) {
1512 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1513 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1518 printf(" Failed disk : ");
1519 if (map
->failed_disk_num
== 0xff)
1522 printf("%i", map
->failed_disk_num
);
1524 slot
= get_imsm_disk_slot(map
, disk_idx
);
1526 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1527 printf(" This Slot : %d%s\n", slot
,
1528 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1530 printf(" This Slot : ?\n");
1531 printf(" Sector Size : %u\n", super
->sector_size
);
1532 sz
= imsm_dev_size(dev
);
1533 printf(" Array Size : %llu%s\n",
1534 (unsigned long long)sz
* 512 / super
->sector_size
,
1535 human_size(sz
* 512));
1536 sz
= blocks_per_member(map
);
1537 printf(" Per Dev Size : %llu%s\n",
1538 (unsigned long long)sz
* 512 / super
->sector_size
,
1539 human_size(sz
* 512));
1540 printf(" Sector Offset : %llu\n",
1542 printf(" Num Stripes : %llu\n",
1543 num_data_stripes(map
));
1544 printf(" Chunk Size : %u KiB",
1545 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1547 printf(" <-- %u KiB",
1548 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1550 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1551 printf(" Migrate State : ");
1552 if (dev
->vol
.migr_state
) {
1553 if (migr_type(dev
) == MIGR_INIT
)
1554 printf("initialize\n");
1555 else if (migr_type(dev
) == MIGR_REBUILD
)
1556 printf("rebuild\n");
1557 else if (migr_type(dev
) == MIGR_VERIFY
)
1559 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1560 printf("general migration\n");
1561 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1562 printf("state change\n");
1563 else if (migr_type(dev
) == MIGR_REPAIR
)
1566 printf("<unknown:%d>\n", migr_type(dev
));
1569 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1570 if (dev
->vol
.migr_state
) {
1571 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1573 printf(" <-- %s", map_state_str
[map
->map_state
]);
1574 printf("\n Checkpoint : %u ",
1575 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1576 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1579 printf("(%llu)", (unsigned long long)
1580 blocks_per_migr_unit(super
, dev
));
1583 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1585 printf(" RWH Policy : ");
1586 if (dev
->rwh_policy
== RWH_OFF
|| dev
->rwh_policy
== RWH_MULTIPLE_OFF
)
1588 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1589 printf("PPL distributed\n");
1590 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1591 printf("PPL journaling drive\n");
1592 else if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
1593 printf("Multiple distributed PPLs\n");
1594 else if (dev
->rwh_policy
== RWH_MULTIPLE_PPLS_JOURNALING_DRIVE
)
1595 printf("Multiple PPLs on journaling drive\n");
1597 printf("<unknown:%d>\n", dev
->rwh_policy
);
1600 static void print_imsm_disk(struct imsm_disk
*disk
,
1603 unsigned int sector_size
) {
1604 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1607 if (index
< -1 || !disk
)
1611 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1613 printf(" Disk%02d Serial : %s\n", index
, str
);
1615 printf(" Disk Serial : %s\n", str
);
1616 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1617 is_configured(disk
) ? " active" : "",
1618 is_failed(disk
) ? " failed" : "",
1619 is_journal(disk
) ? " journal" : "");
1620 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1621 sz
= total_blocks(disk
) - reserved
;
1622 printf(" Usable Size : %llu%s\n",
1623 (unsigned long long)sz
* 512 / sector_size
,
1624 human_size(sz
* 512));
1627 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1629 struct migr_record
*migr_rec
= super
->migr_rec
;
1631 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1632 migr_rec
->ckpt_area_pba
/= IMSM_4K_DIV
;
1633 migr_rec
->dest_1st_member_lba
/= IMSM_4K_DIV
;
1634 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1635 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1636 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1637 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1640 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1642 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1645 void convert_to_4k(struct intel_super
*super
)
1647 struct imsm_super
*mpb
= super
->anchor
;
1648 struct imsm_disk
*disk
;
1650 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1652 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1653 disk
= __get_imsm_disk(mpb
, i
);
1655 convert_to_4k_imsm_disk(disk
);
1657 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1658 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1659 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1661 set_imsm_dev_size(dev
, imsm_dev_size(dev
)/IMSM_4K_DIV
);
1662 dev
->vol
.curr_migr_unit
/= IMSM_4K_DIV
;
1665 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1666 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1667 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1669 if (dev
->vol
.migr_state
) {
1671 map
= get_imsm_map(dev
, MAP_1
);
1672 set_blocks_per_member(map
,
1673 blocks_per_member(map
)/IMSM_4K_DIV
);
1674 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1675 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1679 struct bbm_log
*log
= (void *)mpb
+
1680 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1683 for (i
= 0; i
< log
->entry_count
; i
++) {
1684 struct bbm_log_entry
*entry
=
1685 &log
->marked_block_entries
[i
];
1687 __u8 count
= entry
->marked_count
+ 1;
1688 unsigned long long sector
=
1689 __le48_to_cpu(&entry
->defective_block_start
);
1691 entry
->defective_block_start
=
1692 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1693 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1697 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1700 void examine_migr_rec_imsm(struct intel_super
*super
)
1702 struct migr_record
*migr_rec
= super
->migr_rec
;
1703 struct imsm_super
*mpb
= super
->anchor
;
1706 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1707 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1708 struct imsm_map
*map
;
1711 if (is_gen_migration(dev
) == 0)
1714 printf("\nMigration Record Information:");
1716 /* first map under migration */
1717 map
= get_imsm_map(dev
, MAP_0
);
1719 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1720 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1721 printf(" Empty\n ");
1722 printf("Examine one of first two disks in array\n");
1725 printf("\n Status : ");
1726 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1729 printf("Contains Data\n");
1730 printf(" Current Unit : %u\n",
1731 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1732 printf(" Family : %u\n",
1733 __le32_to_cpu(migr_rec
->family_num
));
1734 printf(" Ascending : %u\n",
1735 __le32_to_cpu(migr_rec
->ascending_migr
));
1736 printf(" Blocks Per Unit : %u\n",
1737 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1738 printf(" Dest. Depth Per Unit : %u\n",
1739 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1740 printf(" Checkpoint Area pba : %u\n",
1741 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1742 printf(" First member lba : %u\n",
1743 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1744 printf(" Total Number of Units : %u\n",
1745 __le32_to_cpu(migr_rec
->num_migr_units
));
1746 printf(" Size of volume : %u\n",
1747 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1748 printf(" Expansion space for LBA64 : %u\n",
1749 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1750 printf(" Record was read from : %u\n",
1751 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1757 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1759 struct migr_record
*migr_rec
= super
->migr_rec
;
1761 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1762 migr_rec
->ckpt_area_pba
*= IMSM_4K_DIV
;
1763 migr_rec
->dest_1st_member_lba
*= IMSM_4K_DIV
;
1764 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1765 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1766 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1767 &migr_rec
->post_migr_vol_cap
,
1768 &migr_rec
->post_migr_vol_cap_hi
);
1771 void convert_from_4k(struct intel_super
*super
)
1773 struct imsm_super
*mpb
= super
->anchor
;
1774 struct imsm_disk
*disk
;
1776 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1778 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1779 disk
= __get_imsm_disk(mpb
, i
);
1781 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1784 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1785 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1786 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1788 set_imsm_dev_size(dev
, imsm_dev_size(dev
)*IMSM_4K_DIV
);
1789 dev
->vol
.curr_migr_unit
*= IMSM_4K_DIV
;
1792 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1793 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1794 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1796 if (dev
->vol
.migr_state
) {
1798 map
= get_imsm_map(dev
, MAP_1
);
1799 set_blocks_per_member(map
,
1800 blocks_per_member(map
)*IMSM_4K_DIV
);
1801 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1802 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1806 struct bbm_log
*log
= (void *)mpb
+
1807 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1810 for (i
= 0; i
< log
->entry_count
; i
++) {
1811 struct bbm_log_entry
*entry
=
1812 &log
->marked_block_entries
[i
];
1814 __u8 count
= entry
->marked_count
+ 1;
1815 unsigned long long sector
=
1816 __le48_to_cpu(&entry
->defective_block_start
);
1818 entry
->defective_block_start
=
1819 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1820 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1824 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1827 /*******************************************************************************
1828 * function: imsm_check_attributes
1829 * Description: Function checks if features represented by attributes flags
1830 * are supported by mdadm.
1832 * attributes - Attributes read from metadata
1834 * 0 - passed attributes contains unsupported features flags
1835 * 1 - all features are supported
1836 ******************************************************************************/
1837 static int imsm_check_attributes(__u32 attributes
)
1840 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1842 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1844 not_supported
&= attributes
;
1845 if (not_supported
) {
1846 pr_err("(IMSM): Unsupported attributes : %x\n",
1847 (unsigned)__le32_to_cpu(not_supported
));
1848 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1849 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1850 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1852 if (not_supported
& MPB_ATTRIB_2TB
) {
1853 dprintf("\t\tMPB_ATTRIB_2TB\n");
1854 not_supported
^= MPB_ATTRIB_2TB
;
1856 if (not_supported
& MPB_ATTRIB_RAID0
) {
1857 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1858 not_supported
^= MPB_ATTRIB_RAID0
;
1860 if (not_supported
& MPB_ATTRIB_RAID1
) {
1861 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1862 not_supported
^= MPB_ATTRIB_RAID1
;
1864 if (not_supported
& MPB_ATTRIB_RAID10
) {
1865 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1866 not_supported
^= MPB_ATTRIB_RAID10
;
1868 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1869 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1870 not_supported
^= MPB_ATTRIB_RAID1E
;
1872 if (not_supported
& MPB_ATTRIB_RAID5
) {
1873 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1874 not_supported
^= MPB_ATTRIB_RAID5
;
1876 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1877 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1878 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1880 if (not_supported
& MPB_ATTRIB_BBM
) {
1881 dprintf("\t\tMPB_ATTRIB_BBM\n");
1882 not_supported
^= MPB_ATTRIB_BBM
;
1884 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1885 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1886 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1888 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1889 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1890 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1892 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1893 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1894 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1896 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1897 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1898 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1900 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1901 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1902 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1906 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1914 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1916 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1918 struct intel_super
*super
= st
->sb
;
1919 struct imsm_super
*mpb
= super
->anchor
;
1920 char str
[MAX_SIGNATURE_LENGTH
];
1925 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1928 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
1929 str
[MPB_SIG_LEN
-1] = '\0';
1930 printf(" Magic : %s\n", str
);
1931 printf(" Version : %s\n", get_imsm_version(mpb
));
1932 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1933 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1934 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1935 printf(" Attributes : ");
1936 if (imsm_check_attributes(mpb
->attributes
))
1937 printf("All supported\n");
1939 printf("not supported\n");
1940 getinfo_super_imsm(st
, &info
, NULL
);
1941 fname_from_uuid(st
, &info
, nbuf
, ':');
1942 printf(" UUID : %s\n", nbuf
+ 5);
1943 sum
= __le32_to_cpu(mpb
->check_sum
);
1944 printf(" Checksum : %08x %s\n", sum
,
1945 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1946 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
1947 printf(" Disks : %d\n", mpb
->num_disks
);
1948 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1949 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
1950 super
->disks
->index
, reserved
, super
->sector_size
);
1951 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
1952 struct bbm_log
*log
= super
->bbm_log
;
1955 printf("Bad Block Management Log:\n");
1956 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1957 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1958 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1960 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1962 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1964 super
->current_vol
= i
;
1965 getinfo_super_imsm(st
, &info
, NULL
);
1966 fname_from_uuid(st
, &info
, nbuf
, ':');
1967 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1969 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1970 if (i
== super
->disks
->index
)
1972 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
1973 super
->sector_size
);
1976 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1977 if (dl
->index
== -1)
1978 print_imsm_disk(&dl
->disk
, -1, reserved
,
1979 super
->sector_size
);
1981 examine_migr_rec_imsm(super
);
1984 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1986 /* We just write a generic IMSM ARRAY entry */
1989 struct intel_super
*super
= st
->sb
;
1991 if (!super
->anchor
->num_raid_devs
) {
1992 printf("ARRAY metadata=imsm\n");
1996 getinfo_super_imsm(st
, &info
, NULL
);
1997 fname_from_uuid(st
, &info
, nbuf
, ':');
1998 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
2001 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
2003 /* We just write a generic IMSM ARRAY entry */
2007 struct intel_super
*super
= st
->sb
;
2010 if (!super
->anchor
->num_raid_devs
)
2013 getinfo_super_imsm(st
, &info
, NULL
);
2014 fname_from_uuid(st
, &info
, nbuf
, ':');
2015 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2016 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2018 super
->current_vol
= i
;
2019 getinfo_super_imsm(st
, &info
, NULL
);
2020 fname_from_uuid(st
, &info
, nbuf1
, ':');
2021 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
2022 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
2026 static void export_examine_super_imsm(struct supertype
*st
)
2028 struct intel_super
*super
= st
->sb
;
2029 struct imsm_super
*mpb
= super
->anchor
;
2033 getinfo_super_imsm(st
, &info
, NULL
);
2034 fname_from_uuid(st
, &info
, nbuf
, ':');
2035 printf("MD_METADATA=imsm\n");
2036 printf("MD_LEVEL=container\n");
2037 printf("MD_UUID=%s\n", nbuf
+5);
2038 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
2041 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
2043 /* The second last sector of the device contains
2044 * the "struct imsm_super" metadata.
2045 * This contains mpb_size which is the size in bytes of the
2046 * extended metadata. This is located immediately before
2048 * We want to read all that, plus the last sector which
2049 * may contain a migration record, and write it all
2053 unsigned long long dsize
, offset
;
2055 struct imsm_super
*sb
;
2056 struct intel_super
*super
= st
->sb
;
2057 unsigned int sector_size
= super
->sector_size
;
2058 unsigned int written
= 0;
2060 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
) != 0)
2063 if (!get_dev_size(from
, NULL
, &dsize
))
2066 if (lseek64(from
, dsize
-(2*sector_size
), 0) < 0)
2068 if ((unsigned int)read(from
, buf
, sector_size
) != sector_size
)
2071 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
2074 sectors
= mpb_sectors(sb
, sector_size
) + 2;
2075 offset
= dsize
- sectors
* sector_size
;
2076 if (lseek64(from
, offset
, 0) < 0 ||
2077 lseek64(to
, offset
, 0) < 0)
2079 while (written
< sectors
* sector_size
) {
2080 int n
= sectors
*sector_size
- written
;
2083 if (read(from
, buf
, n
) != n
)
2085 if (write(to
, buf
, n
) != n
)
2096 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
2101 getinfo_super_imsm(st
, &info
, NULL
);
2102 fname_from_uuid(st
, &info
, nbuf
, ':');
2103 printf("\n UUID : %s\n", nbuf
+ 5);
2106 static void brief_detail_super_imsm(struct supertype
*st
)
2110 getinfo_super_imsm(st
, &info
, NULL
);
2111 fname_from_uuid(st
, &info
, nbuf
, ':');
2112 printf(" UUID=%s", nbuf
+ 5);
2115 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
2116 static void fd2devname(int fd
, char *name
);
2118 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2120 /* dump an unsorted list of devices attached to AHCI Intel storage
2121 * controller, as well as non-connected ports
2123 int hba_len
= strlen(hba_path
) + 1;
2128 unsigned long port_mask
= (1 << port_count
) - 1;
2130 if (port_count
> (int)sizeof(port_mask
) * 8) {
2132 pr_err("port_count %d out of range\n", port_count
);
2136 /* scroll through /sys/dev/block looking for devices attached to
2139 dir
= opendir("/sys/dev/block");
2143 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2154 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2156 path
= devt_to_devpath(makedev(major
, minor
));
2159 if (!path_attached_to_hba(path
, hba_path
)) {
2165 /* retrieve the scsi device type */
2166 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2168 pr_err("failed to allocate 'device'\n");
2172 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2173 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2175 pr_err("failed to read device type for %s\n",
2181 type
= strtoul(buf
, NULL
, 10);
2183 /* if it's not a disk print the vendor and model */
2184 if (!(type
== 0 || type
== 7 || type
== 14)) {
2187 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2188 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2189 strncpy(vendor
, buf
, sizeof(vendor
));
2190 vendor
[sizeof(vendor
) - 1] = '\0';
2191 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2192 while (isspace(*c
) || *c
== '\0')
2196 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2197 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2198 strncpy(model
, buf
, sizeof(model
));
2199 model
[sizeof(model
) - 1] = '\0';
2200 c
= (char *) &model
[sizeof(model
) - 1];
2201 while (isspace(*c
) || *c
== '\0')
2205 if (vendor
[0] && model
[0])
2206 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2208 switch (type
) { /* numbers from hald/linux/device.c */
2209 case 1: sprintf(buf
, "tape"); break;
2210 case 2: sprintf(buf
, "printer"); break;
2211 case 3: sprintf(buf
, "processor"); break;
2213 case 5: sprintf(buf
, "cdrom"); break;
2214 case 6: sprintf(buf
, "scanner"); break;
2215 case 8: sprintf(buf
, "media_changer"); break;
2216 case 9: sprintf(buf
, "comm"); break;
2217 case 12: sprintf(buf
, "raid"); break;
2218 default: sprintf(buf
, "unknown");
2224 /* chop device path to 'host%d' and calculate the port number */
2225 c
= strchr(&path
[hba_len
], '/');
2228 pr_err("%s - invalid path name\n", path
+ hba_len
);
2233 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2234 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2238 *c
= '/'; /* repair the full string */
2239 pr_err("failed to determine port number for %s\n",
2246 /* mark this port as used */
2247 port_mask
&= ~(1 << port
);
2249 /* print out the device information */
2251 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2255 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2257 printf(" Port%d : - disk info unavailable -\n", port
);
2259 fd2devname(fd
, buf
);
2260 printf(" Port%d : %s", port
, buf
);
2261 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
2262 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
2277 for (i
= 0; i
< port_count
; i
++)
2278 if (port_mask
& (1 << i
))
2279 printf(" Port%d : - no device attached -\n", i
);
2285 static int print_vmd_attached_devs(struct sys_dev
*hba
)
2293 if (hba
->type
!= SYS_DEV_VMD
)
2296 /* scroll through /sys/dev/block looking for devices attached to
2299 dir
= opendir("/sys/bus/pci/drivers/nvme");
2303 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2306 /* is 'ent' a device? check that the 'subsystem' link exists and
2307 * that its target matches 'bus'
2309 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
2311 n
= readlink(path
, link
, sizeof(link
));
2312 if (n
< 0 || n
>= (int)sizeof(link
))
2315 c
= strrchr(link
, '/');
2318 if (strncmp("pci", c
+1, strlen("pci")) != 0)
2321 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
2323 rp
= realpath(path
, NULL
);
2327 if (path_attached_to_hba(rp
, hba
->path
)) {
2328 printf(" NVMe under VMD : %s\n", rp
);
2337 static void print_found_intel_controllers(struct sys_dev
*elem
)
2339 for (; elem
; elem
= elem
->next
) {
2340 pr_err("found Intel(R) ");
2341 if (elem
->type
== SYS_DEV_SATA
)
2342 fprintf(stderr
, "SATA ");
2343 else if (elem
->type
== SYS_DEV_SAS
)
2344 fprintf(stderr
, "SAS ");
2345 else if (elem
->type
== SYS_DEV_NVME
)
2346 fprintf(stderr
, "NVMe ");
2348 if (elem
->type
== SYS_DEV_VMD
)
2349 fprintf(stderr
, "VMD domain");
2351 fprintf(stderr
, "RAID controller");
2354 fprintf(stderr
, " at %s", elem
->pci_id
);
2355 fprintf(stderr
, ".\n");
2360 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2367 if ((dir
= opendir(hba_path
)) == NULL
)
2370 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2373 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2374 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2376 if (*port_count
== 0)
2378 else if (host
< host_base
)
2381 if (host
+ 1 > *port_count
+ host_base
)
2382 *port_count
= host
+ 1 - host_base
;
2388 static void print_imsm_capability(const struct imsm_orom
*orom
)
2390 printf(" Platform : Intel(R) ");
2391 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2392 printf("Matrix Storage Manager\n");
2393 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2394 printf("Virtual RAID on CPU\n");
2396 printf("Rapid Storage Technology%s\n",
2397 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2398 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2399 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2400 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2401 printf(" RAID Levels :%s%s%s%s%s\n",
2402 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2403 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2404 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2405 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2406 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2407 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2408 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2409 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2410 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2411 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2412 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2413 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2414 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2415 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2416 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2417 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2418 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2419 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2420 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2421 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2422 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2423 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2424 printf(" 2TB volumes :%s supported\n",
2425 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2426 printf(" 2TB disks :%s supported\n",
2427 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2428 printf(" Max Disks : %d\n", orom
->tds
);
2429 printf(" Max Volumes : %d per array, %d per %s\n",
2430 orom
->vpa
, orom
->vphba
,
2431 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2435 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2437 printf("MD_FIRMWARE_TYPE=imsm\n");
2438 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2439 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2440 orom
->hotfix_ver
, orom
->build
);
2441 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2442 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2443 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2444 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2445 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2446 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2447 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2448 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2449 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2450 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2451 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2452 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2453 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2454 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2455 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2456 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2457 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2458 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2459 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2460 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2461 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2462 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2463 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2464 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2465 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2466 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2467 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2468 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2471 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2473 /* There are two components to imsm platform support, the ahci SATA
2474 * controller and the option-rom. To find the SATA controller we
2475 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2476 * controller with the Intel vendor id is present. This approach
2477 * allows mdadm to leverage the kernel's ahci detection logic, with the
2478 * caveat that if ahci.ko is not loaded mdadm will not be able to
2479 * detect platform raid capabilities. The option-rom resides in a
2480 * platform "Adapter ROM". We scan for its signature to retrieve the
2481 * platform capabilities. If raid support is disabled in the BIOS the
2482 * option-rom capability structure will not be available.
2484 struct sys_dev
*list
, *hba
;
2489 if (enumerate_only
) {
2490 if (check_env("IMSM_NO_PLATFORM"))
2492 list
= find_intel_devices();
2495 for (hba
= list
; hba
; hba
= hba
->next
) {
2496 if (find_imsm_capability(hba
)) {
2506 list
= find_intel_devices();
2509 pr_err("no active Intel(R) RAID controller found.\n");
2511 } else if (verbose
> 0)
2512 print_found_intel_controllers(list
);
2514 for (hba
= list
; hba
; hba
= hba
->next
) {
2515 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2517 if (!find_imsm_capability(hba
)) {
2519 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2520 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2521 get_sys_dev_type(hba
->type
));
2527 if (controller_path
&& result
== 1) {
2528 pr_err("no active Intel(R) RAID controller found under %s\n",
2533 const struct orom_entry
*entry
;
2535 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2536 if (entry
->type
== SYS_DEV_VMD
) {
2537 print_imsm_capability(&entry
->orom
);
2538 printf(" 3rd party NVMe :%s supported\n",
2539 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2540 for (hba
= list
; hba
; hba
= hba
->next
) {
2541 if (hba
->type
== SYS_DEV_VMD
) {
2543 printf(" I/O Controller : %s (%s)\n",
2544 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2545 if (print_vmd_attached_devs(hba
)) {
2547 pr_err("failed to get devices attached to VMD domain.\n");
2556 print_imsm_capability(&entry
->orom
);
2557 if (entry
->type
== SYS_DEV_NVME
) {
2558 for (hba
= list
; hba
; hba
= hba
->next
) {
2559 if (hba
->type
== SYS_DEV_NVME
)
2560 printf(" NVMe Device : %s\n", hba
->path
);
2566 struct devid_list
*devid
;
2567 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2568 hba
= device_by_id(devid
->devid
);
2572 printf(" I/O Controller : %s (%s)\n",
2573 hba
->path
, get_sys_dev_type(hba
->type
));
2574 if (hba
->type
== SYS_DEV_SATA
) {
2575 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2576 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2578 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2589 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2591 struct sys_dev
*list
, *hba
;
2594 list
= find_intel_devices();
2597 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2602 for (hba
= list
; hba
; hba
= hba
->next
) {
2603 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2605 if (!find_imsm_capability(hba
) && verbose
> 0) {
2607 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2608 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2614 const struct orom_entry
*entry
;
2616 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2617 if (entry
->type
== SYS_DEV_VMD
) {
2618 for (hba
= list
; hba
; hba
= hba
->next
)
2619 print_imsm_capability_export(&entry
->orom
);
2622 print_imsm_capability_export(&entry
->orom
);
2628 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2630 /* the imsm metadata format does not specify any host
2631 * identification information. We return -1 since we can never
2632 * confirm nor deny whether a given array is "meant" for this
2633 * host. We rely on compare_super and the 'family_num' fields to
2634 * exclude member disks that do not belong, and we rely on
2635 * mdadm.conf to specify the arrays that should be assembled.
2636 * Auto-assembly may still pick up "foreign" arrays.
2642 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2644 /* The uuid returned here is used for:
2645 * uuid to put into bitmap file (Create, Grow)
2646 * uuid for backup header when saving critical section (Grow)
2647 * comparing uuids when re-adding a device into an array
2648 * In these cases the uuid required is that of the data-array,
2649 * not the device-set.
2650 * uuid to recognise same set when adding a missing device back
2651 * to an array. This is a uuid for the device-set.
2653 * For each of these we can make do with a truncated
2654 * or hashed uuid rather than the original, as long as
2656 * In each case the uuid required is that of the data-array,
2657 * not the device-set.
2659 /* imsm does not track uuid's so we synthesis one using sha1 on
2660 * - The signature (Which is constant for all imsm array, but no matter)
2661 * - the orig_family_num of the container
2662 * - the index number of the volume
2663 * - the 'serial' number of the volume.
2664 * Hopefully these are all constant.
2666 struct intel_super
*super
= st
->sb
;
2669 struct sha1_ctx ctx
;
2670 struct imsm_dev
*dev
= NULL
;
2673 /* some mdadm versions failed to set ->orig_family_num, in which
2674 * case fall back to ->family_num. orig_family_num will be
2675 * fixed up with the first metadata update.
2677 family_num
= super
->anchor
->orig_family_num
;
2678 if (family_num
== 0)
2679 family_num
= super
->anchor
->family_num
;
2680 sha1_init_ctx(&ctx
);
2681 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2682 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2683 if (super
->current_vol
>= 0)
2684 dev
= get_imsm_dev(super
, super
->current_vol
);
2686 __u32 vol
= super
->current_vol
;
2687 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2688 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2690 sha1_finish_ctx(&ctx
, buf
);
2691 memcpy(uuid
, buf
, 4*4);
2696 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2698 __u8
*v
= get_imsm_version(mpb
);
2699 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2700 char major
[] = { 0, 0, 0 };
2701 char minor
[] = { 0 ,0, 0 };
2702 char patch
[] = { 0, 0, 0 };
2703 char *ver_parse
[] = { major
, minor
, patch
};
2707 while (*v
!= '\0' && v
< end
) {
2708 if (*v
!= '.' && j
< 2)
2709 ver_parse
[i
][j
++] = *v
;
2717 *m
= strtol(minor
, NULL
, 0);
2718 *p
= strtol(patch
, NULL
, 0);
2722 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2724 /* migr_strip_size when repairing or initializing parity */
2725 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2726 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2728 switch (get_imsm_raid_level(map
)) {
2733 return 128*1024 >> 9;
2737 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2739 /* migr_strip_size when rebuilding a degraded disk, no idea why
2740 * this is different than migr_strip_size_resync(), but it's good
2743 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2744 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2746 switch (get_imsm_raid_level(map
)) {
2749 if (map
->num_members
% map
->num_domains
== 0)
2750 return 128*1024 >> 9;
2754 return max((__u32
) 64*1024 >> 9, chunk
);
2756 return 128*1024 >> 9;
2760 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2762 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2763 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2764 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2765 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2767 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2770 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2772 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2773 int level
= get_imsm_raid_level(lo
);
2775 if (level
== 1 || level
== 10) {
2776 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2778 return hi
->num_domains
;
2780 return num_stripes_per_unit_resync(dev
);
2783 static __u8
imsm_num_data_members(struct imsm_map
*map
)
2785 /* named 'imsm_' because raid0, raid1 and raid10
2786 * counter-intuitively have the same number of data disks
2788 switch (get_imsm_raid_level(map
)) {
2790 return map
->num_members
;
2794 return map
->num_members
/2;
2796 return map
->num_members
- 1;
2798 dprintf("unsupported raid level\n");
2803 static unsigned long long calc_component_size(struct imsm_map
*map
,
2804 struct imsm_dev
*dev
)
2806 unsigned long long component_size
;
2807 unsigned long long dev_size
= imsm_dev_size(dev
);
2808 unsigned long long calc_dev_size
= 0;
2809 unsigned int member_disks
= imsm_num_data_members(map
);
2811 if (member_disks
== 0)
2814 component_size
= per_dev_array_size(map
);
2815 calc_dev_size
= component_size
* member_disks
;
2817 /* Component size is rounded to 1MB so difference between size from
2818 * metadata and size calculated from num_data_stripes equals up to
2819 * 2048 blocks per each device. If the difference is higher it means
2820 * that array size was expanded and num_data_stripes was not updated.
2822 if ((unsigned int)abs(calc_dev_size
- dev_size
) >
2823 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
2824 component_size
= dev_size
/ member_disks
;
2825 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
2826 component_size
/ map
->blocks_per_strip
,
2827 num_data_stripes(map
));
2830 return component_size
;
2833 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2835 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2836 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2838 switch(get_imsm_raid_level(map
)) {
2841 return chunk
* map
->num_domains
;
2843 return chunk
* map
->num_members
;
2849 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2851 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2852 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2853 __u32 strip
= block
/ chunk
;
2855 switch (get_imsm_raid_level(map
)) {
2858 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2859 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2861 return vol_stripe
* chunk
+ block
% chunk
;
2863 __u32 stripe
= strip
/ (map
->num_members
- 1);
2865 return stripe
* chunk
+ block
% chunk
;
2872 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2873 struct imsm_dev
*dev
)
2875 /* calculate the conversion factor between per member 'blocks'
2876 * (md/{resync,rebuild}_start) and imsm migration units, return
2877 * 0 for the 'not migrating' and 'unsupported migration' cases
2879 if (!dev
->vol
.migr_state
)
2882 switch (migr_type(dev
)) {
2883 case MIGR_GEN_MIGR
: {
2884 struct migr_record
*migr_rec
= super
->migr_rec
;
2885 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2890 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2891 __u32 stripes_per_unit
;
2892 __u32 blocks_per_unit
;
2901 /* yes, this is really the translation of migr_units to
2902 * per-member blocks in the 'resync' case
2904 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2905 migr_chunk
= migr_strip_blocks_resync(dev
);
2906 disks
= imsm_num_data_members(map
);
2907 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2908 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2909 segment
= blocks_per_unit
/ stripe
;
2910 block_rel
= blocks_per_unit
- segment
* stripe
;
2911 parity_depth
= parity_segment_depth(dev
);
2912 block_map
= map_migr_block(dev
, block_rel
);
2913 return block_map
+ parity_depth
* segment
;
2915 case MIGR_REBUILD
: {
2916 __u32 stripes_per_unit
;
2919 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2920 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2921 return migr_chunk
* stripes_per_unit
;
2923 case MIGR_STATE_CHANGE
:
2929 static int imsm_level_to_layout(int level
)
2937 return ALGORITHM_LEFT_ASYMMETRIC
;
2944 /*******************************************************************************
2945 * Function: read_imsm_migr_rec
2946 * Description: Function reads imsm migration record from last sector of disk
2948 * fd : disk descriptor
2949 * super : metadata info
2953 ******************************************************************************/
2954 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2957 unsigned int sector_size
= super
->sector_size
;
2958 unsigned long long dsize
;
2960 get_dev_size(fd
, NULL
, &dsize
);
2961 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
2963 pr_err("Cannot seek to anchor block: %s\n",
2967 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
2968 MIGR_REC_BUF_SECTORS
*sector_size
) !=
2969 MIGR_REC_BUF_SECTORS
*sector_size
) {
2970 pr_err("Cannot read migr record block: %s\n",
2975 if (sector_size
== 4096)
2976 convert_from_4k_imsm_migr_rec(super
);
2982 static struct imsm_dev
*imsm_get_device_during_migration(
2983 struct intel_super
*super
)
2986 struct intel_dev
*dv
;
2988 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2989 if (is_gen_migration(dv
->dev
))
2995 /*******************************************************************************
2996 * Function: load_imsm_migr_rec
2997 * Description: Function reads imsm migration record (it is stored at the last
3000 * super : imsm internal array info
3001 * info : general array info
3005 * -2 : no migration in progress
3006 ******************************************************************************/
3007 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
3014 struct imsm_dev
*dev
;
3015 struct imsm_map
*map
;
3018 /* find map under migration */
3019 dev
= imsm_get_device_during_migration(super
);
3020 /* nothing to load,no migration in progress?
3026 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
3027 /* read only from one of the first two slots */
3028 if ((sd
->disk
.raid_disk
< 0) ||
3029 (sd
->disk
.raid_disk
> 1))
3032 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3033 fd
= dev_open(nm
, O_RDONLY
);
3039 map
= get_imsm_map(dev
, MAP_0
);
3040 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3041 /* skip spare and failed disks
3045 /* read only from one of the first two slots */
3047 slot
= get_imsm_disk_slot(map
, dl
->index
);
3048 if (map
== NULL
|| slot
> 1 || slot
< 0)
3050 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3051 fd
= dev_open(nm
, O_RDONLY
);
3058 retval
= read_imsm_migr_rec(fd
, super
);
3066 /*******************************************************************************
3067 * function: imsm_create_metadata_checkpoint_update
3068 * Description: It creates update for checkpoint change.
3070 * super : imsm internal array info
3071 * u : pointer to prepared update
3074 * If length is equal to 0, input pointer u contains no update
3075 ******************************************************************************/
3076 static int imsm_create_metadata_checkpoint_update(
3077 struct intel_super
*super
,
3078 struct imsm_update_general_migration_checkpoint
**u
)
3081 int update_memory_size
= 0;
3083 dprintf("(enter)\n");
3089 /* size of all update data without anchor */
3090 update_memory_size
=
3091 sizeof(struct imsm_update_general_migration_checkpoint
);
3093 *u
= xcalloc(1, update_memory_size
);
3095 dprintf("error: cannot get memory\n");
3098 (*u
)->type
= update_general_migration_checkpoint
;
3099 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
3100 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3102 return update_memory_size
;
3105 static void imsm_update_metadata_locally(struct supertype
*st
,
3106 void *buf
, int len
);
3108 /*******************************************************************************
3109 * Function: write_imsm_migr_rec
3110 * Description: Function writes imsm migration record
3111 * (at the last sector of disk)
3113 * super : imsm internal array info
3117 ******************************************************************************/
3118 static int write_imsm_migr_rec(struct supertype
*st
)
3120 struct intel_super
*super
= st
->sb
;
3121 unsigned int sector_size
= super
->sector_size
;
3122 unsigned long long dsize
;
3128 struct imsm_update_general_migration_checkpoint
*u
;
3129 struct imsm_dev
*dev
;
3130 struct imsm_map
*map
;
3132 /* find map under migration */
3133 dev
= imsm_get_device_during_migration(super
);
3134 /* if no migration, write buffer anyway to clear migr_record
3135 * on disk based on first available device
3138 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3139 super
->current_vol
);
3141 map
= get_imsm_map(dev
, MAP_0
);
3143 if (sector_size
== 4096)
3144 convert_to_4k_imsm_migr_rec(super
);
3145 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3148 /* skip failed and spare devices */
3151 /* write to 2 first slots only */
3153 slot
= get_imsm_disk_slot(map
, sd
->index
);
3154 if (map
== NULL
|| slot
> 1 || slot
< 0)
3157 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3158 fd
= dev_open(nm
, O_RDWR
);
3161 get_dev_size(fd
, NULL
, &dsize
);
3162 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3164 pr_err("Cannot seek to anchor block: %s\n",
3168 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3169 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3170 MIGR_REC_BUF_SECTORS
*sector_size
) {
3171 pr_err("Cannot write migr record block: %s\n",
3178 if (sector_size
== 4096)
3179 convert_from_4k_imsm_migr_rec(super
);
3180 /* update checkpoint information in metadata */
3181 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3183 dprintf("imsm: Cannot prepare update\n");
3186 /* update metadata locally */
3187 imsm_update_metadata_locally(st
, u
, len
);
3188 /* and possibly remotely */
3189 if (st
->update_tail
) {
3190 append_metadata_update(st
, u
, len
);
3191 /* during reshape we do all work inside metadata handler
3192 * manage_reshape(), so metadata update has to be triggered
3195 flush_metadata_updates(st
);
3196 st
->update_tail
= &st
->updates
;
3207 /* spare/missing disks activations are not allowe when
3208 * array/container performs reshape operation, because
3209 * all arrays in container works on the same disks set
3211 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3214 struct intel_dev
*i_dev
;
3215 struct imsm_dev
*dev
;
3217 /* check whole container
3219 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3221 if (is_gen_migration(dev
)) {
3222 /* No repair during any migration in container
3230 static unsigned long long imsm_component_size_alignment_check(int level
,
3232 unsigned int sector_size
,
3233 unsigned long long component_size
)
3235 unsigned int component_size_alignment
;
3237 /* check component size alignment
3239 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3241 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3242 level
, chunk_size
, component_size
,
3243 component_size_alignment
);
3245 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3246 dprintf("imsm: reported component size aligned from %llu ",
3248 component_size
-= component_size_alignment
;
3249 dprintf_cont("to %llu (%i).\n",
3250 component_size
, component_size_alignment
);
3253 return component_size
;
3256 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3258 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3259 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3261 return pba_of_lba0(map
) +
3262 (num_data_stripes(map
) * map
->blocks_per_strip
);
3265 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3267 struct intel_super
*super
= st
->sb
;
3268 struct migr_record
*migr_rec
= super
->migr_rec
;
3269 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3270 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3271 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3272 struct imsm_map
*map_to_analyse
= map
;
3274 int map_disks
= info
->array
.raid_disks
;
3276 memset(info
, 0, sizeof(*info
));
3278 map_to_analyse
= prev_map
;
3280 dl
= super
->current_disk
;
3282 info
->container_member
= super
->current_vol
;
3283 info
->array
.raid_disks
= map
->num_members
;
3284 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3285 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3286 info
->array
.md_minor
= -1;
3287 info
->array
.ctime
= 0;
3288 info
->array
.utime
= 0;
3289 info
->array
.chunk_size
=
3290 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3291 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3292 info
->custom_array_size
= imsm_dev_size(dev
);
3293 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3295 if (is_gen_migration(dev
)) {
3296 info
->reshape_active
= 1;
3297 info
->new_level
= get_imsm_raid_level(map
);
3298 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3299 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3300 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3301 if (info
->delta_disks
) {
3302 /* this needs to be applied to every array
3305 info
->reshape_active
= CONTAINER_RESHAPE
;
3307 /* We shape information that we give to md might have to be
3308 * modify to cope with md's requirement for reshaping arrays.
3309 * For example, when reshaping a RAID0, md requires it to be
3310 * presented as a degraded RAID4.
3311 * Also if a RAID0 is migrating to a RAID5 we need to specify
3312 * the array as already being RAID5, but the 'before' layout
3313 * is a RAID4-like layout.
3315 switch (info
->array
.level
) {
3317 switch(info
->new_level
) {
3319 /* conversion is happening as RAID4 */
3320 info
->array
.level
= 4;
3321 info
->array
.raid_disks
+= 1;
3324 /* conversion is happening as RAID5 */
3325 info
->array
.level
= 5;
3326 info
->array
.layout
= ALGORITHM_PARITY_N
;
3327 info
->delta_disks
-= 1;
3330 /* FIXME error message */
3331 info
->array
.level
= UnSet
;
3337 info
->new_level
= UnSet
;
3338 info
->new_layout
= UnSet
;
3339 info
->new_chunk
= info
->array
.chunk_size
;
3340 info
->delta_disks
= 0;
3344 info
->disk
.major
= dl
->major
;
3345 info
->disk
.minor
= dl
->minor
;
3346 info
->disk
.number
= dl
->index
;
3347 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3351 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3352 info
->component_size
= calc_component_size(map
, dev
);
3353 info
->component_size
= imsm_component_size_alignment_check(
3355 info
->array
.chunk_size
,
3357 info
->component_size
);
3358 info
->bb
.supported
= 1;
3360 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3361 info
->recovery_start
= MaxSector
;
3363 if (info
->array
.level
== 5 &&
3364 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3365 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3366 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3367 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3368 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3369 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3371 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3373 } else if (info
->array
.level
<= 0) {
3374 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3376 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3379 info
->reshape_progress
= 0;
3380 info
->resync_start
= MaxSector
;
3381 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3382 !(info
->array
.state
& 1)) &&
3383 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3384 info
->resync_start
= 0;
3386 if (dev
->vol
.migr_state
) {
3387 switch (migr_type(dev
)) {
3390 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3392 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3394 info
->resync_start
= blocks_per_unit
* units
;
3397 case MIGR_GEN_MIGR
: {
3398 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3400 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
3401 unsigned long long array_blocks
;
3404 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3406 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
3407 (super
->migr_rec
->rec_status
==
3408 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3411 info
->reshape_progress
= blocks_per_unit
* units
;
3413 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3414 (unsigned long long)units
,
3415 (unsigned long long)blocks_per_unit
,
3416 info
->reshape_progress
);
3418 used_disks
= imsm_num_data_members(prev_map
);
3419 if (used_disks
> 0) {
3420 array_blocks
= per_dev_array_size(map
) *
3422 info
->custom_array_size
=
3423 round_size_to_mb(array_blocks
,
3429 /* we could emulate the checkpointing of
3430 * 'sync_action=check' migrations, but for now
3431 * we just immediately complete them
3434 /* this is handled by container_content_imsm() */
3435 case MIGR_STATE_CHANGE
:
3436 /* FIXME handle other migrations */
3438 /* we are not dirty, so... */
3439 info
->resync_start
= MaxSector
;
3443 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3444 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3446 info
->array
.major_version
= -1;
3447 info
->array
.minor_version
= -2;
3448 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3449 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3450 uuid_from_super_imsm(st
, info
->uuid
);
3454 for (i
=0; i
<map_disks
; i
++) {
3456 if (i
< info
->array
.raid_disks
) {
3457 struct imsm_disk
*dsk
;
3458 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3459 dsk
= get_imsm_disk(super
, j
);
3460 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3467 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3468 int failed
, int look_in_map
);
3470 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3473 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3475 if (is_gen_migration(dev
)) {
3478 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3480 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3481 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3482 if (map2
->map_state
!= map_state
) {
3483 map2
->map_state
= map_state
;
3484 super
->updates_pending
++;
3489 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3493 for (d
= super
->missing
; d
; d
= d
->next
)
3494 if (d
->index
== index
)
3499 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3501 struct intel_super
*super
= st
->sb
;
3502 struct imsm_disk
*disk
;
3503 int map_disks
= info
->array
.raid_disks
;
3504 int max_enough
= -1;
3506 struct imsm_super
*mpb
;
3508 if (super
->current_vol
>= 0) {
3509 getinfo_super_imsm_volume(st
, info
, map
);
3512 memset(info
, 0, sizeof(*info
));
3514 /* Set raid_disks to zero so that Assemble will always pull in valid
3517 info
->array
.raid_disks
= 0;
3518 info
->array
.level
= LEVEL_CONTAINER
;
3519 info
->array
.layout
= 0;
3520 info
->array
.md_minor
= -1;
3521 info
->array
.ctime
= 0; /* N/A for imsm */
3522 info
->array
.utime
= 0;
3523 info
->array
.chunk_size
= 0;
3525 info
->disk
.major
= 0;
3526 info
->disk
.minor
= 0;
3527 info
->disk
.raid_disk
= -1;
3528 info
->reshape_active
= 0;
3529 info
->array
.major_version
= -1;
3530 info
->array
.minor_version
= -2;
3531 strcpy(info
->text_version
, "imsm");
3532 info
->safe_mode_delay
= 0;
3533 info
->disk
.number
= -1;
3534 info
->disk
.state
= 0;
3536 info
->recovery_start
= MaxSector
;
3537 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3538 info
->bb
.supported
= 1;
3540 /* do we have the all the insync disks that we expect? */
3541 mpb
= super
->anchor
;
3542 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3544 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3545 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3546 int failed
, enough
, j
, missing
= 0;
3547 struct imsm_map
*map
;
3550 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3551 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3552 map
= get_imsm_map(dev
, MAP_0
);
3554 /* any newly missing disks?
3555 * (catches single-degraded vs double-degraded)
3557 for (j
= 0; j
< map
->num_members
; j
++) {
3558 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3559 __u32 idx
= ord_to_idx(ord
);
3561 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3562 info
->disk
.raid_disk
= j
;
3564 if (!(ord
& IMSM_ORD_REBUILD
) &&
3565 get_imsm_missing(super
, idx
)) {
3571 if (state
== IMSM_T_STATE_FAILED
)
3573 else if (state
== IMSM_T_STATE_DEGRADED
&&
3574 (state
!= map
->map_state
|| missing
))
3576 else /* we're normal, or already degraded */
3578 if (is_gen_migration(dev
) && missing
) {
3579 /* during general migration we need all disks
3580 * that process is running on.
3581 * No new missing disk is allowed.
3585 /* no more checks necessary
3589 /* in the missing/failed disk case check to see
3590 * if at least one array is runnable
3592 max_enough
= max(max_enough
, enough
);
3594 dprintf("enough: %d\n", max_enough
);
3595 info
->container_enough
= max_enough
;
3598 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3600 disk
= &super
->disks
->disk
;
3601 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3602 info
->component_size
= reserved
;
3603 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3604 /* we don't change info->disk.raid_disk here because
3605 * this state will be finalized in mdmon after we have
3606 * found the 'most fresh' version of the metadata
3608 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3609 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3610 0 : (1 << MD_DISK_SYNC
);
3613 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3614 * ->compare_super may have updated the 'num_raid_devs' field for spares
3616 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3617 uuid_from_super_imsm(st
, info
->uuid
);
3619 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3621 /* I don't know how to compute 'map' on imsm, so use safe default */
3624 for (i
= 0; i
< map_disks
; i
++)
3630 /* allocates memory and fills disk in mdinfo structure
3631 * for each disk in array */
3632 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3634 struct mdinfo
*mddev
;
3635 struct intel_super
*super
= st
->sb
;
3636 struct imsm_disk
*disk
;
3639 if (!super
|| !super
->disks
)
3642 mddev
= xcalloc(1, sizeof(*mddev
));
3646 tmp
= xcalloc(1, sizeof(*tmp
));
3648 tmp
->next
= mddev
->devs
;
3650 tmp
->disk
.number
= count
++;
3651 tmp
->disk
.major
= dl
->major
;
3652 tmp
->disk
.minor
= dl
->minor
;
3653 tmp
->disk
.state
= is_configured(disk
) ?
3654 (1 << MD_DISK_ACTIVE
) : 0;
3655 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3656 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3657 tmp
->disk
.raid_disk
= -1;
3663 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3664 char *update
, char *devname
, int verbose
,
3665 int uuid_set
, char *homehost
)
3667 /* For 'assemble' and 'force' we need to return non-zero if any
3668 * change was made. For others, the return value is ignored.
3669 * Update options are:
3670 * force-one : This device looks a bit old but needs to be included,
3671 * update age info appropriately.
3672 * assemble: clear any 'faulty' flag to allow this device to
3674 * force-array: Array is degraded but being forced, mark it clean
3675 * if that will be needed to assemble it.
3677 * newdev: not used ????
3678 * grow: Array has gained a new device - this is currently for
3680 * resync: mark as dirty so a resync will happen.
3681 * name: update the name - preserving the homehost
3682 * uuid: Change the uuid of the array to match watch is given
3684 * Following are not relevant for this imsm:
3685 * sparc2.2 : update from old dodgey metadata
3686 * super-minor: change the preferred_minor number
3687 * summaries: update redundant counters.
3688 * homehost: update the recorded homehost
3689 * _reshape_progress: record new reshape_progress position.
3692 struct intel_super
*super
= st
->sb
;
3693 struct imsm_super
*mpb
;
3695 /* we can only update container info */
3696 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3699 mpb
= super
->anchor
;
3701 if (strcmp(update
, "uuid") == 0) {
3702 /* We take this to mean that the family_num should be updated.
3703 * However that is much smaller than the uuid so we cannot really
3704 * allow an explicit uuid to be given. And it is hard to reliably
3706 * So if !uuid_set we know the current uuid is random and just used
3707 * the first 'int' and copy it to the other 3 positions.
3708 * Otherwise we require the 4 'int's to be the same as would be the
3709 * case if we are using a random uuid. So an explicit uuid will be
3710 * accepted as long as all for ints are the same... which shouldn't hurt
3713 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3716 if (info
->uuid
[0] != info
->uuid
[1] ||
3717 info
->uuid
[1] != info
->uuid
[2] ||
3718 info
->uuid
[2] != info
->uuid
[3])
3724 mpb
->orig_family_num
= info
->uuid
[0];
3725 } else if (strcmp(update
, "assemble") == 0)
3730 /* successful update? recompute checksum */
3732 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3737 static size_t disks_to_mpb_size(int disks
)
3741 size
= sizeof(struct imsm_super
);
3742 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3743 size
+= 2 * sizeof(struct imsm_dev
);
3744 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3745 size
+= (4 - 2) * sizeof(struct imsm_map
);
3746 /* 4 possible disk_ord_tbl's */
3747 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3748 /* maximum bbm log */
3749 size
+= sizeof(struct bbm_log
);
3754 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3755 unsigned long long data_offset
)
3757 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3760 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3763 static void free_devlist(struct intel_super
*super
)
3765 struct intel_dev
*dv
;
3767 while (super
->devlist
) {
3768 dv
= super
->devlist
->next
;
3769 free(super
->devlist
->dev
);
3770 free(super
->devlist
);
3771 super
->devlist
= dv
;
3775 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3777 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3780 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3784 * 0 same, or first was empty, and second was copied
3785 * 1 second had wrong number
3787 * 3 wrong other info
3789 struct intel_super
*first
= st
->sb
;
3790 struct intel_super
*sec
= tst
->sb
;
3797 /* in platform dependent environment test if the disks
3798 * use the same Intel hba
3799 * If not on Intel hba at all, allow anything.
3801 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3802 if (first
->hba
->type
!= sec
->hba
->type
) {
3804 "HBAs of devices do not match %s != %s\n",
3805 get_sys_dev_type(first
->hba
->type
),
3806 get_sys_dev_type(sec
->hba
->type
));
3809 if (first
->orom
!= sec
->orom
) {
3811 "HBAs of devices do not match %s != %s\n",
3812 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3817 /* if an anchor does not have num_raid_devs set then it is a free
3820 if (first
->anchor
->num_raid_devs
> 0 &&
3821 sec
->anchor
->num_raid_devs
> 0) {
3822 /* Determine if these disks might ever have been
3823 * related. Further disambiguation can only take place
3824 * in load_super_imsm_all
3826 __u32 first_family
= first
->anchor
->orig_family_num
;
3827 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3829 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3830 MAX_SIGNATURE_LENGTH
) != 0)
3833 if (first_family
== 0)
3834 first_family
= first
->anchor
->family_num
;
3835 if (sec_family
== 0)
3836 sec_family
= sec
->anchor
->family_num
;
3838 if (first_family
!= sec_family
)
3843 /* if 'first' is a spare promote it to a populated mpb with sec's
3846 if (first
->anchor
->num_raid_devs
== 0 &&
3847 sec
->anchor
->num_raid_devs
> 0) {
3849 struct intel_dev
*dv
;
3850 struct imsm_dev
*dev
;
3852 /* we need to copy raid device info from sec if an allocation
3853 * fails here we don't associate the spare
3855 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3856 dv
= xmalloc(sizeof(*dv
));
3857 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3860 dv
->next
= first
->devlist
;
3861 first
->devlist
= dv
;
3863 if (i
< sec
->anchor
->num_raid_devs
) {
3864 /* allocation failure */
3865 free_devlist(first
);
3866 pr_err("imsm: failed to associate spare\n");
3869 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3870 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3871 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3872 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3873 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3874 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3880 static void fd2devname(int fd
, char *name
)
3884 char dname
[PATH_MAX
];
3889 if (fstat(fd
, &st
) != 0)
3891 sprintf(path
, "/sys/dev/block/%d:%d",
3892 major(st
.st_rdev
), minor(st
.st_rdev
));
3894 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3899 nm
= strrchr(dname
, '/');
3902 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3906 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3909 char *name
= fd2kname(fd
);
3914 if (strncmp(name
, "nvme", 4) != 0)
3917 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3919 return load_sys(path
, buf
, buf_len
);
3922 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3924 static int imsm_read_serial(int fd
, char *devname
,
3925 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3934 memset(buf
, 0, sizeof(buf
));
3936 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3939 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3941 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3942 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3943 fd2devname(fd
, (char *) serial
);
3949 pr_err("Failed to retrieve serial for %s\n",
3954 /* trim all whitespace and non-printable characters and convert
3957 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
3960 /* ':' is reserved for use in placeholder serial
3961 * numbers for missing disks
3972 /* truncate leading characters */
3973 if (len
> MAX_RAID_SERIAL_LEN
) {
3974 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3975 len
= MAX_RAID_SERIAL_LEN
;
3978 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3979 memcpy(serial
, dest
, len
);
3984 static int serialcmp(__u8
*s1
, __u8
*s2
)
3986 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3989 static void serialcpy(__u8
*dest
, __u8
*src
)
3991 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3994 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3998 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3999 if (serialcmp(dl
->serial
, serial
) == 0)
4005 static struct imsm_disk
*
4006 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4010 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4011 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4013 if (serialcmp(disk
->serial
, serial
) == 0) {
4024 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4026 struct imsm_disk
*disk
;
4031 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4033 rv
= imsm_read_serial(fd
, devname
, serial
);
4038 dl
= xcalloc(1, sizeof(*dl
));
4041 dl
->major
= major(stb
.st_rdev
);
4042 dl
->minor
= minor(stb
.st_rdev
);
4043 dl
->next
= super
->disks
;
4044 dl
->fd
= keep_fd
? fd
: -1;
4045 assert(super
->disks
== NULL
);
4047 serialcpy(dl
->serial
, serial
);
4050 fd2devname(fd
, name
);
4052 dl
->devname
= xstrdup(devname
);
4054 dl
->devname
= xstrdup(name
);
4056 /* look up this disk's index in the current anchor */
4057 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4060 /* only set index on disks that are a member of a
4061 * populated contianer, i.e. one with raid_devs
4063 if (is_failed(&dl
->disk
))
4065 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4072 /* When migrating map0 contains the 'destination' state while map1
4073 * contains the current state. When not migrating map0 contains the
4074 * current state. This routine assumes that map[0].map_state is set to
4075 * the current array state before being called.
4077 * Migration is indicated by one of the following states
4078 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4079 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4080 * map1state=unitialized)
4081 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4083 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4084 * map1state=degraded)
4085 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4088 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4089 __u8 to_state
, int migr_type
)
4091 struct imsm_map
*dest
;
4092 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4094 dev
->vol
.migr_state
= 1;
4095 set_migr_type(dev
, migr_type
);
4096 dev
->vol
.curr_migr_unit
= 0;
4097 dest
= get_imsm_map(dev
, MAP_1
);
4099 /* duplicate and then set the target end state in map[0] */
4100 memcpy(dest
, src
, sizeof_imsm_map(src
));
4101 if (migr_type
== MIGR_GEN_MIGR
) {
4105 for (i
= 0; i
< src
->num_members
; i
++) {
4106 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4107 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4111 if (migr_type
== MIGR_GEN_MIGR
)
4112 /* Clear migration record */
4113 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4115 src
->map_state
= to_state
;
4118 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4121 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4122 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4126 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4127 * completed in the last migration.
4129 * FIXME add support for raid-level-migration
4131 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4132 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4133 /* when final map state is other than expected
4134 * merge maps (not for migration)
4138 for (i
= 0; i
< prev
->num_members
; i
++)
4139 for (j
= 0; j
< map
->num_members
; j
++)
4140 /* during online capacity expansion
4141 * disks position can be changed
4142 * if takeover is used
4144 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4145 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4146 map
->disk_ord_tbl
[j
] |=
4147 prev
->disk_ord_tbl
[i
];
4150 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4151 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4154 dev
->vol
.migr_state
= 0;
4155 set_migr_type(dev
, 0);
4156 dev
->vol
.curr_migr_unit
= 0;
4157 map
->map_state
= map_state
;
4160 static int parse_raid_devices(struct intel_super
*super
)
4163 struct imsm_dev
*dev_new
;
4164 size_t len
, len_migr
;
4166 size_t space_needed
= 0;
4167 struct imsm_super
*mpb
= super
->anchor
;
4169 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4170 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4171 struct intel_dev
*dv
;
4173 len
= sizeof_imsm_dev(dev_iter
, 0);
4174 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4176 space_needed
+= len_migr
- len
;
4178 dv
= xmalloc(sizeof(*dv
));
4179 if (max_len
< len_migr
)
4181 if (max_len
> len_migr
)
4182 space_needed
+= max_len
- len_migr
;
4183 dev_new
= xmalloc(max_len
);
4184 imsm_copy_dev(dev_new
, dev_iter
);
4187 dv
->next
= super
->devlist
;
4188 super
->devlist
= dv
;
4191 /* ensure that super->buf is large enough when all raid devices
4194 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4197 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4198 super
->sector_size
);
4199 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4202 memcpy(buf
, super
->buf
, super
->len
);
4203 memset(buf
+ super
->len
, 0, len
- super
->len
);
4209 super
->extra_space
+= space_needed
;
4214 /*******************************************************************************
4215 * Function: check_mpb_migr_compatibility
4216 * Description: Function checks for unsupported migration features:
4217 * - migration optimization area (pba_of_lba0)
4218 * - descending reshape (ascending_migr)
4220 * super : imsm metadata information
4222 * 0 : migration is compatible
4223 * -1 : migration is not compatible
4224 ******************************************************************************/
4225 int check_mpb_migr_compatibility(struct intel_super
*super
)
4227 struct imsm_map
*map0
, *map1
;
4228 struct migr_record
*migr_rec
= super
->migr_rec
;
4231 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4232 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4235 dev_iter
->vol
.migr_state
== 1 &&
4236 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4237 /* This device is migrating */
4238 map0
= get_imsm_map(dev_iter
, MAP_0
);
4239 map1
= get_imsm_map(dev_iter
, MAP_1
);
4240 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4241 /* migration optimization area was used */
4243 if (migr_rec
->ascending_migr
== 0 &&
4244 migr_rec
->dest_depth_per_unit
> 0)
4245 /* descending reshape not supported yet */
4252 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4254 /* load_imsm_mpb - read matrix metadata
4255 * allocates super->mpb to be freed by free_imsm
4257 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4259 unsigned long long dsize
;
4260 unsigned long long sectors
;
4261 unsigned int sector_size
= super
->sector_size
;
4263 struct imsm_super
*anchor
;
4266 get_dev_size(fd
, NULL
, &dsize
);
4267 if (dsize
< 2*sector_size
) {
4269 pr_err("%s: device to small for imsm\n",
4274 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4276 pr_err("Cannot seek to anchor block on %s: %s\n",
4277 devname
, strerror(errno
));
4281 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4283 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4286 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4288 pr_err("Cannot read anchor block on %s: %s\n",
4289 devname
, strerror(errno
));
4294 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4296 pr_err("no IMSM anchor on %s\n", devname
);
4301 __free_imsm(super
, 0);
4302 /* reload capability and hba */
4304 /* capability and hba must be updated with new super allocation */
4305 find_intel_hba_capability(fd
, super
, devname
);
4306 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4307 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4309 pr_err("unable to allocate %zu byte mpb buffer\n",
4314 memcpy(super
->buf
, anchor
, sector_size
);
4316 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4319 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4320 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4321 pr_err("could not allocate migr_rec buffer\n");
4325 super
->clean_migration_record_by_mdmon
= 0;
4328 check_sum
= __gen_imsm_checksum(super
->anchor
);
4329 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4331 pr_err("IMSM checksum %x != %x on %s\n",
4333 __le32_to_cpu(super
->anchor
->check_sum
),
4341 /* read the extended mpb */
4342 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4344 pr_err("Cannot seek to extended mpb on %s: %s\n",
4345 devname
, strerror(errno
));
4349 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4350 super
->len
- sector_size
) != super
->len
- sector_size
) {
4352 pr_err("Cannot read extended mpb on %s: %s\n",
4353 devname
, strerror(errno
));
4357 check_sum
= __gen_imsm_checksum(super
->anchor
);
4358 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4360 pr_err("IMSM checksum %x != %x on %s\n",
4361 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4369 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4371 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4372 static void clear_hi(struct intel_super
*super
)
4374 struct imsm_super
*mpb
= super
->anchor
;
4376 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4378 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4379 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4380 disk
->total_blocks_hi
= 0;
4382 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4383 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4386 for (n
= 0; n
< 2; ++n
) {
4387 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4390 map
->pba_of_lba0_hi
= 0;
4391 map
->blocks_per_member_hi
= 0;
4392 map
->num_data_stripes_hi
= 0;
4398 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4402 err
= load_imsm_mpb(fd
, super
, devname
);
4405 if (super
->sector_size
== 4096)
4406 convert_from_4k(super
);
4407 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4410 err
= parse_raid_devices(super
);
4413 err
= load_bbm_log(super
);
4418 static void __free_imsm_disk(struct dl
*d
)
4430 static void free_imsm_disks(struct intel_super
*super
)
4434 while (super
->disks
) {
4436 super
->disks
= d
->next
;
4437 __free_imsm_disk(d
);
4439 while (super
->disk_mgmt_list
) {
4440 d
= super
->disk_mgmt_list
;
4441 super
->disk_mgmt_list
= d
->next
;
4442 __free_imsm_disk(d
);
4444 while (super
->missing
) {
4446 super
->missing
= d
->next
;
4447 __free_imsm_disk(d
);
4452 /* free all the pieces hanging off of a super pointer */
4453 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4455 struct intel_hba
*elem
, *next
;
4461 /* unlink capability description */
4463 if (super
->migr_rec_buf
) {
4464 free(super
->migr_rec_buf
);
4465 super
->migr_rec_buf
= NULL
;
4468 free_imsm_disks(super
);
4469 free_devlist(super
);
4473 free((void *)elem
->path
);
4479 free(super
->bbm_log
);
4483 static void free_imsm(struct intel_super
*super
)
4485 __free_imsm(super
, 1);
4486 free(super
->bb
.entries
);
4490 static void free_super_imsm(struct supertype
*st
)
4492 struct intel_super
*super
= st
->sb
;
4501 static struct intel_super
*alloc_super(void)
4503 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4505 super
->current_vol
= -1;
4506 super
->create_offset
= ~((unsigned long long) 0);
4508 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4509 sizeof(struct md_bb_entry
));
4510 if (!super
->bb
.entries
) {
4519 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4521 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4523 struct sys_dev
*hba_name
;
4526 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4531 hba_name
= find_disk_attached_hba(fd
, NULL
);
4534 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4538 rv
= attach_hba_to_super(super
, hba_name
);
4541 struct intel_hba
*hba
= super
->hba
;
4543 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4544 " but the container is assigned to Intel(R) %s %s (",
4546 get_sys_dev_type(hba_name
->type
),
4547 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4548 hba_name
->pci_id
? : "Err!",
4549 get_sys_dev_type(super
->hba
->type
),
4550 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4553 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4555 fprintf(stderr
, ", ");
4558 fprintf(stderr
, ").\n"
4559 " Mixing devices attached to different controllers is not allowed.\n");
4563 super
->orom
= find_imsm_capability(hba_name
);
4570 /* find_missing - helper routine for load_super_imsm_all that identifies
4571 * disks that have disappeared from the system. This routine relies on
4572 * the mpb being uptodate, which it is at load time.
4574 static int find_missing(struct intel_super
*super
)
4577 struct imsm_super
*mpb
= super
->anchor
;
4579 struct imsm_disk
*disk
;
4581 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4582 disk
= __get_imsm_disk(mpb
, i
);
4583 dl
= serial_to_dl(disk
->serial
, super
);
4587 dl
= xmalloc(sizeof(*dl
));
4591 dl
->devname
= xstrdup("missing");
4593 serialcpy(dl
->serial
, disk
->serial
);
4596 dl
->next
= super
->missing
;
4597 super
->missing
= dl
;
4603 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4605 struct intel_disk
*idisk
= disk_list
;
4608 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4610 idisk
= idisk
->next
;
4616 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4617 struct intel_super
*super
,
4618 struct intel_disk
**disk_list
)
4620 struct imsm_disk
*d
= &super
->disks
->disk
;
4621 struct imsm_super
*mpb
= super
->anchor
;
4624 for (i
= 0; i
< tbl_size
; i
++) {
4625 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4626 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4628 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4629 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4630 dprintf("mpb from %d:%d matches %d:%d\n",
4631 super
->disks
->major
,
4632 super
->disks
->minor
,
4633 table
[i
]->disks
->major
,
4634 table
[i
]->disks
->minor
);
4638 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4639 is_configured(d
) == is_configured(tbl_d
)) &&
4640 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4641 /* current version of the mpb is a
4642 * better candidate than the one in
4643 * super_table, but copy over "cross
4644 * generational" status
4646 struct intel_disk
*idisk
;
4648 dprintf("mpb from %d:%d replaces %d:%d\n",
4649 super
->disks
->major
,
4650 super
->disks
->minor
,
4651 table
[i
]->disks
->major
,
4652 table
[i
]->disks
->minor
);
4654 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4655 if (idisk
&& is_failed(&idisk
->disk
))
4656 tbl_d
->status
|= FAILED_DISK
;
4659 struct intel_disk
*idisk
;
4660 struct imsm_disk
*disk
;
4662 /* tbl_mpb is more up to date, but copy
4663 * over cross generational status before
4666 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4667 if (disk
&& is_failed(disk
))
4668 d
->status
|= FAILED_DISK
;
4670 idisk
= disk_list_get(d
->serial
, *disk_list
);
4673 if (disk
&& is_configured(disk
))
4674 idisk
->disk
.status
|= CONFIGURED_DISK
;
4677 dprintf("mpb from %d:%d prefer %d:%d\n",
4678 super
->disks
->major
,
4679 super
->disks
->minor
,
4680 table
[i
]->disks
->major
,
4681 table
[i
]->disks
->minor
);
4689 table
[tbl_size
++] = super
;
4693 /* update/extend the merged list of imsm_disk records */
4694 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4695 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4696 struct intel_disk
*idisk
;
4698 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4700 idisk
->disk
.status
|= disk
->status
;
4701 if (is_configured(&idisk
->disk
) ||
4702 is_failed(&idisk
->disk
))
4703 idisk
->disk
.status
&= ~(SPARE_DISK
);
4705 idisk
= xcalloc(1, sizeof(*idisk
));
4706 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4707 idisk
->disk
= *disk
;
4708 idisk
->next
= *disk_list
;
4712 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4719 static struct intel_super
*
4720 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4723 struct imsm_super
*mpb
= super
->anchor
;
4727 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4728 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4729 struct intel_disk
*idisk
;
4731 idisk
= disk_list_get(disk
->serial
, disk_list
);
4733 if (idisk
->owner
== owner
||
4734 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4737 dprintf("'%.16s' owner %d != %d\n",
4738 disk
->serial
, idisk
->owner
,
4741 dprintf("unknown disk %x [%d]: %.16s\n",
4742 __le32_to_cpu(mpb
->family_num
), i
,
4748 if (ok_count
== mpb
->num_disks
)
4753 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4755 struct intel_super
*s
;
4757 for (s
= super_list
; s
; s
= s
->next
) {
4758 if (family_num
!= s
->anchor
->family_num
)
4760 pr_err("Conflict, offlining family %#x on '%s'\n",
4761 __le32_to_cpu(family_num
), s
->disks
->devname
);
4765 static struct intel_super
*
4766 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4768 struct intel_super
*super_table
[len
];
4769 struct intel_disk
*disk_list
= NULL
;
4770 struct intel_super
*champion
, *spare
;
4771 struct intel_super
*s
, **del
;
4776 memset(super_table
, 0, sizeof(super_table
));
4777 for (s
= *super_list
; s
; s
= s
->next
)
4778 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4780 for (i
= 0; i
< tbl_size
; i
++) {
4781 struct imsm_disk
*d
;
4782 struct intel_disk
*idisk
;
4783 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4786 d
= &s
->disks
->disk
;
4788 /* 'd' must appear in merged disk list for its
4789 * configuration to be valid
4791 idisk
= disk_list_get(d
->serial
, disk_list
);
4792 if (idisk
&& idisk
->owner
== i
)
4793 s
= validate_members(s
, disk_list
, i
);
4798 dprintf("marking family: %#x from %d:%d offline\n",
4800 super_table
[i
]->disks
->major
,
4801 super_table
[i
]->disks
->minor
);
4805 /* This is where the mdadm implementation differs from the Windows
4806 * driver which has no strict concept of a container. We can only
4807 * assemble one family from a container, so when returning a prodigal
4808 * array member to this system the code will not be able to disambiguate
4809 * the container contents that should be assembled ("foreign" versus
4810 * "local"). It requires user intervention to set the orig_family_num
4811 * to a new value to establish a new container. The Windows driver in
4812 * this situation fixes up the volume name in place and manages the
4813 * foreign array as an independent entity.
4818 for (i
= 0; i
< tbl_size
; i
++) {
4819 struct intel_super
*tbl_ent
= super_table
[i
];
4825 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4830 if (s
&& !is_spare
) {
4831 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4833 } else if (!s
&& !is_spare
)
4846 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4847 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4849 /* collect all dl's onto 'champion', and update them to
4850 * champion's version of the status
4852 for (s
= *super_list
; s
; s
= s
->next
) {
4853 struct imsm_super
*mpb
= champion
->anchor
;
4854 struct dl
*dl
= s
->disks
;
4859 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4861 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4862 struct imsm_disk
*disk
;
4864 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4867 /* only set index on disks that are a member of
4868 * a populated contianer, i.e. one with
4871 if (is_failed(&dl
->disk
))
4873 else if (is_spare(&dl
->disk
))
4879 if (i
>= mpb
->num_disks
) {
4880 struct intel_disk
*idisk
;
4882 idisk
= disk_list_get(dl
->serial
, disk_list
);
4883 if (idisk
&& is_spare(&idisk
->disk
) &&
4884 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4892 dl
->next
= champion
->disks
;
4893 champion
->disks
= dl
;
4897 /* delete 'champion' from super_list */
4898 for (del
= super_list
; *del
; ) {
4899 if (*del
== champion
) {
4900 *del
= (*del
)->next
;
4903 del
= &(*del
)->next
;
4905 champion
->next
= NULL
;
4909 struct intel_disk
*idisk
= disk_list
;
4911 disk_list
= disk_list
->next
;
4919 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4920 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4921 int major
, int minor
, int keep_fd
);
4923 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4924 int *max
, int keep_fd
);
4926 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4927 char *devname
, struct md_list
*devlist
,
4930 struct intel_super
*super_list
= NULL
;
4931 struct intel_super
*super
= NULL
;
4936 /* 'fd' is an opened container */
4937 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4939 /* get super block from devlist devices */
4940 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4943 /* all mpbs enter, maybe one leaves */
4944 super
= imsm_thunderdome(&super_list
, i
);
4950 if (find_missing(super
) != 0) {
4956 /* load migration record */
4957 err
= load_imsm_migr_rec(super
, NULL
);
4959 /* migration is in progress,
4960 * but migr_rec cannot be loaded,
4966 /* Check migration compatibility */
4967 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
4968 pr_err("Unsupported migration detected");
4970 fprintf(stderr
, " on %s\n", devname
);
4972 fprintf(stderr
, " (IMSM).\n");
4981 while (super_list
) {
4982 struct intel_super
*s
= super_list
;
4984 super_list
= super_list
->next
;
4993 strcpy(st
->container_devnm
, fd2devnm(fd
));
4995 st
->container_devnm
[0] = 0;
4996 if (err
== 0 && st
->ss
== NULL
) {
4997 st
->ss
= &super_imsm
;
4998 st
->minor_version
= 0;
4999 st
->max_devs
= IMSM_MAX_DEVICES
;
5005 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5006 int *max
, int keep_fd
)
5008 struct md_list
*tmpdev
;
5012 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5013 if (tmpdev
->used
!= 1)
5015 if (tmpdev
->container
== 1) {
5017 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5019 pr_err("cannot open device %s: %s\n",
5020 tmpdev
->devname
, strerror(errno
));
5024 err
= get_sra_super_block(fd
, super_list
,
5025 tmpdev
->devname
, &lmax
,
5034 int major
= major(tmpdev
->st_rdev
);
5035 int minor
= minor(tmpdev
->st_rdev
);
5036 err
= get_super_block(super_list
,
5053 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5054 int major
, int minor
, int keep_fd
)
5056 struct intel_super
*s
;
5068 sprintf(nm
, "%d:%d", major
, minor
);
5069 dfd
= dev_open(nm
, O_RDWR
);
5075 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5076 find_intel_hba_capability(dfd
, s
, devname
);
5077 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5079 /* retry the load if we might have raced against mdmon */
5080 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5081 for (retry
= 0; retry
< 3; retry
++) {
5083 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5089 s
->next
= *super_list
;
5097 if (dfd
>= 0 && !keep_fd
)
5104 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5111 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5115 if (sra
->array
.major_version
!= -1 ||
5116 sra
->array
.minor_version
!= -2 ||
5117 strcmp(sra
->text_version
, "imsm") != 0) {
5122 devnm
= fd2devnm(fd
);
5123 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5124 if (get_super_block(super_list
, devnm
, devname
,
5125 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5136 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5138 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5141 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5143 struct intel_super
*super
;
5147 if (test_partition(fd
))
5148 /* IMSM not allowed on partitions */
5151 free_super_imsm(st
);
5153 super
= alloc_super();
5154 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5157 /* Load hba and capabilities if they exist.
5158 * But do not preclude loading metadata in case capabilities or hba are
5159 * non-compliant and ignore_hw_compat is set.
5161 rv
= find_intel_hba_capability(fd
, super
, devname
);
5162 /* no orom/efi or non-intel hba of the disk */
5163 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5165 pr_err("No OROM/EFI properties for %s\n", devname
);
5169 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5171 /* retry the load if we might have raced against mdmon */
5173 struct mdstat_ent
*mdstat
= NULL
;
5174 char *name
= fd2kname(fd
);
5177 mdstat
= mdstat_by_component(name
);
5179 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5180 for (retry
= 0; retry
< 3; retry
++) {
5182 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5188 free_mdstat(mdstat
);
5193 pr_err("Failed to load all information sections on %s\n", devname
);
5199 if (st
->ss
== NULL
) {
5200 st
->ss
= &super_imsm
;
5201 st
->minor_version
= 0;
5202 st
->max_devs
= IMSM_MAX_DEVICES
;
5205 /* load migration record */
5206 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5207 /* Check for unsupported migration features */
5208 if (check_mpb_migr_compatibility(super
) != 0) {
5209 pr_err("Unsupported migration detected");
5211 fprintf(stderr
, " on %s\n", devname
);
5213 fprintf(stderr
, " (IMSM).\n");
5221 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5223 if (info
->level
== 1)
5225 return info
->chunk_size
>> 9;
5228 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5229 unsigned long long size
)
5231 if (info
->level
== 1)
5234 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5237 static void imsm_update_version_info(struct intel_super
*super
)
5239 /* update the version and attributes */
5240 struct imsm_super
*mpb
= super
->anchor
;
5242 struct imsm_dev
*dev
;
5243 struct imsm_map
*map
;
5246 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5247 dev
= get_imsm_dev(super
, i
);
5248 map
= get_imsm_map(dev
, MAP_0
);
5249 if (__le32_to_cpu(dev
->size_high
) > 0)
5250 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5252 /* FIXME detect when an array spans a port multiplier */
5254 mpb
->attributes
|= MPB_ATTRIB_PM
;
5257 if (mpb
->num_raid_devs
> 1 ||
5258 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5259 version
= MPB_VERSION_ATTRIBS
;
5260 switch (get_imsm_raid_level(map
)) {
5261 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5262 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5263 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5264 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5267 if (map
->num_members
>= 5)
5268 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5269 else if (dev
->status
== DEV_CLONE_N_GO
)
5270 version
= MPB_VERSION_CNG
;
5271 else if (get_imsm_raid_level(map
) == 5)
5272 version
= MPB_VERSION_RAID5
;
5273 else if (map
->num_members
>= 3)
5274 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5275 else if (get_imsm_raid_level(map
) == 1)
5276 version
= MPB_VERSION_RAID1
;
5278 version
= MPB_VERSION_RAID0
;
5280 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5284 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5286 struct imsm_super
*mpb
= super
->anchor
;
5287 char *reason
= NULL
;
5290 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
5291 reason
= "must be 16 characters or less";
5293 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5294 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5296 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5297 reason
= "already exists";
5302 if (reason
&& !quiet
)
5303 pr_err("imsm volume name %s\n", reason
);
5308 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5309 struct shape
*s
, char *name
,
5310 char *homehost
, int *uuid
,
5311 long long data_offset
)
5313 /* We are creating a volume inside a pre-existing container.
5314 * so st->sb is already set.
5316 struct intel_super
*super
= st
->sb
;
5317 unsigned int sector_size
= super
->sector_size
;
5318 struct imsm_super
*mpb
= super
->anchor
;
5319 struct intel_dev
*dv
;
5320 struct imsm_dev
*dev
;
5321 struct imsm_vol
*vol
;
5322 struct imsm_map
*map
;
5323 int idx
= mpb
->num_raid_devs
;
5326 unsigned long long array_blocks
;
5327 size_t size_old
, size_new
;
5328 unsigned long long num_data_stripes
;
5329 unsigned int data_disks
;
5330 unsigned long long size_per_member
;
5332 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5333 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5337 /* ensure the mpb is large enough for the new data */
5338 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5339 size_new
= disks_to_mpb_size(info
->nr_disks
);
5340 if (size_new
> size_old
) {
5342 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5344 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5345 pr_err("could not allocate new mpb\n");
5348 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5349 MIGR_REC_BUF_SECTORS
*
5350 MAX_SECTOR_SIZE
) != 0) {
5351 pr_err("could not allocate migr_rec buffer\n");
5357 memcpy(mpb_new
, mpb
, size_old
);
5360 super
->anchor
= mpb_new
;
5361 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5362 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5363 super
->len
= size_round
;
5365 super
->current_vol
= idx
;
5367 /* handle 'failed_disks' by either:
5368 * a) create dummy disk entries in the table if this the first
5369 * volume in the array. We add them here as this is the only
5370 * opportunity to add them. add_to_super_imsm_volume()
5371 * handles the non-failed disks and continues incrementing
5373 * b) validate that 'failed_disks' matches the current number
5374 * of missing disks if the container is populated
5376 if (super
->current_vol
== 0) {
5378 for (i
= 0; i
< info
->failed_disks
; i
++) {
5379 struct imsm_disk
*disk
;
5382 disk
= __get_imsm_disk(mpb
, i
);
5383 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5384 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5385 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5386 "missing:%d", (__u8
)i
);
5388 find_missing(super
);
5393 for (d
= super
->missing
; d
; d
= d
->next
)
5395 if (info
->failed_disks
> missing
) {
5396 pr_err("unable to add 'missing' disk to container\n");
5401 if (!check_name(super
, name
, 0))
5403 dv
= xmalloc(sizeof(*dv
));
5404 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5406 * Explicitly allow truncating to not confuse gcc's
5407 * -Werror=stringop-truncation
5409 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5410 memcpy(dev
->volume
, name
, namelen
);
5411 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5412 info
->layout
, info
->chunk_size
,
5413 s
->size
* BLOCKS_PER_KB
);
5414 data_disks
= get_data_disks(info
->level
, info
->layout
,
5416 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5417 size_per_member
= array_blocks
/ data_disks
;
5419 set_imsm_dev_size(dev
, array_blocks
);
5420 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5422 vol
->migr_state
= 0;
5423 set_migr_type(dev
, MIGR_INIT
);
5424 vol
->dirty
= !info
->state
;
5425 vol
->curr_migr_unit
= 0;
5426 map
= get_imsm_map(dev
, MAP_0
);
5427 set_pba_of_lba0(map
, super
->create_offset
);
5428 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5429 map
->failed_disk_num
= ~0;
5430 if (info
->level
> 0)
5431 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5432 : IMSM_T_STATE_UNINITIALIZED
);
5434 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5435 IMSM_T_STATE_NORMAL
;
5438 if (info
->level
== 1 && info
->raid_disks
> 2) {
5441 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5445 map
->raid_level
= info
->level
;
5446 if (info
->level
== 10) {
5447 map
->raid_level
= 1;
5448 map
->num_domains
= info
->raid_disks
/ 2;
5449 } else if (info
->level
== 1)
5450 map
->num_domains
= info
->raid_disks
;
5452 map
->num_domains
= 1;
5454 /* info->size is only int so use the 'size' parameter instead */
5455 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5456 num_data_stripes
/= map
->num_domains
;
5457 set_num_data_stripes(map
, num_data_stripes
);
5459 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5460 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5464 map
->num_members
= info
->raid_disks
;
5465 for (i
= 0; i
< map
->num_members
; i
++) {
5466 /* initialized in add_to_super */
5467 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5469 mpb
->num_raid_devs
++;
5470 mpb
->num_raid_devs_created
++;
5471 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5473 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5474 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5475 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5476 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5480 pr_err("imsm does not support consistency policy %s\n",
5481 map_num(consistency_policies
, s
->consistency_policy
));
5486 dv
->index
= super
->current_vol
;
5487 dv
->next
= super
->devlist
;
5488 super
->devlist
= dv
;
5490 imsm_update_version_info(super
);
5495 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5496 struct shape
*s
, char *name
,
5497 char *homehost
, int *uuid
,
5498 unsigned long long data_offset
)
5500 /* This is primarily called by Create when creating a new array.
5501 * We will then get add_to_super called for each component, and then
5502 * write_init_super called to write it out to each device.
5503 * For IMSM, Create can create on fresh devices or on a pre-existing
5505 * To create on a pre-existing array a different method will be called.
5506 * This one is just for fresh drives.
5508 struct intel_super
*super
;
5509 struct imsm_super
*mpb
;
5513 if (data_offset
!= INVALID_SECTORS
) {
5514 pr_err("data-offset not supported by imsm\n");
5519 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5523 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5525 mpb_size
= MAX_SECTOR_SIZE
;
5527 super
= alloc_super();
5529 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5534 pr_err("could not allocate superblock\n");
5537 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5538 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5539 pr_err("could not allocate migr_rec buffer\n");
5544 memset(super
->buf
, 0, mpb_size
);
5546 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5550 /* zeroing superblock */
5554 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5556 version
= (char *) mpb
->sig
;
5557 strcpy(version
, MPB_SIGNATURE
);
5558 version
+= strlen(MPB_SIGNATURE
);
5559 strcpy(version
, MPB_VERSION_RAID0
);
5564 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5566 unsigned int member_sector_size
;
5569 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5573 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5575 if (member_sector_size
!= super
->sector_size
)
5580 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5581 int fd
, char *devname
)
5583 struct intel_super
*super
= st
->sb
;
5584 struct imsm_super
*mpb
= super
->anchor
;
5585 struct imsm_disk
*_disk
;
5586 struct imsm_dev
*dev
;
5587 struct imsm_map
*map
;
5591 dev
= get_imsm_dev(super
, super
->current_vol
);
5592 map
= get_imsm_map(dev
, MAP_0
);
5594 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5595 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5601 /* we're doing autolayout so grab the pre-marked (in
5602 * validate_geometry) raid_disk
5604 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5605 if (dl
->raiddisk
== dk
->raid_disk
)
5608 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5609 if (dl
->major
== dk
->major
&&
5610 dl
->minor
== dk
->minor
)
5615 pr_err("%s is not a member of the same container\n", devname
);
5619 if (!drive_validate_sector_size(super
, dl
)) {
5620 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5624 /* add a pristine spare to the metadata */
5625 if (dl
->index
< 0) {
5626 dl
->index
= super
->anchor
->num_disks
;
5627 super
->anchor
->num_disks
++;
5629 /* Check the device has not already been added */
5630 slot
= get_imsm_disk_slot(map
, dl
->index
);
5632 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5633 pr_err("%s has been included in this array twice\n",
5637 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5638 dl
->disk
.status
= CONFIGURED_DISK
;
5640 /* update size of 'missing' disks to be at least as large as the
5641 * largest acitve member (we only have dummy missing disks when
5642 * creating the first volume)
5644 if (super
->current_vol
== 0) {
5645 for (df
= super
->missing
; df
; df
= df
->next
) {
5646 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5647 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5648 _disk
= __get_imsm_disk(mpb
, df
->index
);
5653 /* refresh unset/failed slots to point to valid 'missing' entries */
5654 for (df
= super
->missing
; df
; df
= df
->next
)
5655 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5656 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5658 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5660 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5661 if (is_gen_migration(dev
)) {
5662 struct imsm_map
*map2
= get_imsm_map(dev
,
5664 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5665 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5666 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5669 if ((unsigned)df
->index
==
5671 set_imsm_ord_tbl_ent(map2
,
5677 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5681 /* if we are creating the first raid device update the family number */
5682 if (super
->current_vol
== 0) {
5684 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5686 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5687 if (!_dev
|| !_disk
) {
5688 pr_err("BUG mpb setup error\n");
5694 sum
+= __gen_imsm_checksum(mpb
);
5695 mpb
->family_num
= __cpu_to_le32(sum
);
5696 mpb
->orig_family_num
= mpb
->family_num
;
5698 super
->current_disk
= dl
;
5703 * Function marks disk as spare and restores disk serial
5704 * in case it was previously marked as failed by takeover operation
5706 * -1 : critical error
5707 * 0 : disk is marked as spare but serial is not set
5710 int mark_spare(struct dl
*disk
)
5712 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5719 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5720 /* Restore disk serial number, because takeover marks disk
5721 * as failed and adds to serial ':0' before it becomes
5724 serialcpy(disk
->serial
, serial
);
5725 serialcpy(disk
->disk
.serial
, serial
);
5728 disk
->disk
.status
= SPARE_DISK
;
5734 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5735 int fd
, char *devname
,
5736 unsigned long long data_offset
)
5738 struct intel_super
*super
= st
->sb
;
5740 unsigned long long size
;
5741 unsigned int member_sector_size
;
5746 /* If we are on an RAID enabled platform check that the disk is
5747 * attached to the raid controller.
5748 * We do not need to test disks attachment for container based additions,
5749 * they shall be already tested when container was created/assembled.
5751 rv
= find_intel_hba_capability(fd
, super
, devname
);
5752 /* no orom/efi or non-intel hba of the disk */
5754 dprintf("capability: %p fd: %d ret: %d\n",
5755 super
->orom
, fd
, rv
);
5759 if (super
->current_vol
>= 0)
5760 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5763 dd
= xcalloc(sizeof(*dd
), 1);
5764 dd
->major
= major(stb
.st_rdev
);
5765 dd
->minor
= minor(stb
.st_rdev
);
5766 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5769 dd
->action
= DISK_ADD
;
5770 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5772 pr_err("failed to retrieve scsi serial, aborting\n");
5778 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5779 (super
->hba
->type
== SYS_DEV_VMD
))) {
5781 char *devpath
= diskfd_to_devpath(fd
);
5782 char controller_path
[PATH_MAX
];
5785 pr_err("failed to get devpath, aborting\n");
5792 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5795 if (devpath_to_vendor(controller_path
) == 0x8086) {
5797 * If Intel's NVMe drive has serial ended with
5798 * "-A","-B","-1" or "-2" it means that this is "x8"
5799 * device (double drive on single PCIe card).
5800 * User should be warned about potential data loss.
5802 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5803 /* Skip empty character at the end */
5804 if (dd
->serial
[i
] == 0)
5807 if (((dd
->serial
[i
] == 'A') ||
5808 (dd
->serial
[i
] == 'B') ||
5809 (dd
->serial
[i
] == '1') ||
5810 (dd
->serial
[i
] == '2')) &&
5811 (dd
->serial
[i
-1] == '-'))
5812 pr_err("\tThe action you are about to take may put your data at risk.\n"
5813 "\tPlease note that x8 devices may consist of two separate x4 devices "
5814 "located on a single PCIe port.\n"
5815 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5818 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5819 !imsm_orom_has_tpv_support(super
->orom
)) {
5820 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5821 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
5828 get_dev_size(fd
, NULL
, &size
);
5829 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5831 if (super
->sector_size
== 0) {
5832 /* this a first device, so sector_size is not set yet */
5833 super
->sector_size
= member_sector_size
;
5836 /* clear migr_rec when adding disk to container */
5837 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5838 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5840 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5841 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5842 MIGR_REC_BUF_SECTORS
*member_sector_size
)
5843 perror("Write migr_rec failed");
5847 serialcpy(dd
->disk
.serial
, dd
->serial
);
5848 set_total_blocks(&dd
->disk
, size
);
5849 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5850 struct imsm_super
*mpb
= super
->anchor
;
5851 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5854 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5855 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5857 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5859 if (st
->update_tail
) {
5860 dd
->next
= super
->disk_mgmt_list
;
5861 super
->disk_mgmt_list
= dd
;
5863 dd
->next
= super
->disks
;
5865 super
->updates_pending
++;
5871 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5873 struct intel_super
*super
= st
->sb
;
5876 /* remove from super works only in mdmon - for communication
5877 * manager - monitor. Check if communication memory buffer
5880 if (!st
->update_tail
) {
5881 pr_err("shall be used in mdmon context only\n");
5884 dd
= xcalloc(1, sizeof(*dd
));
5885 dd
->major
= dk
->major
;
5886 dd
->minor
= dk
->minor
;
5889 dd
->action
= DISK_REMOVE
;
5891 dd
->next
= super
->disk_mgmt_list
;
5892 super
->disk_mgmt_list
= dd
;
5897 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5900 char buf
[MAX_SECTOR_SIZE
];
5901 struct imsm_super anchor
;
5902 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5904 /* spare records have their own family number and do not have any defined raid
5907 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5909 struct imsm_super
*mpb
= super
->anchor
;
5910 struct imsm_super
*spare
= &spare_record
.anchor
;
5914 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5915 spare
->generation_num
= __cpu_to_le32(1UL);
5916 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5917 spare
->num_disks
= 1;
5918 spare
->num_raid_devs
= 0;
5919 spare
->cache_size
= mpb
->cache_size
;
5920 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5922 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5923 MPB_SIGNATURE MPB_VERSION_RAID0
);
5925 for (d
= super
->disks
; d
; d
= d
->next
) {
5929 spare
->disk
[0] = d
->disk
;
5930 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5931 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5933 if (super
->sector_size
== 4096)
5934 convert_to_4k_imsm_disk(&spare
->disk
[0]);
5936 sum
= __gen_imsm_checksum(spare
);
5937 spare
->family_num
= __cpu_to_le32(sum
);
5938 spare
->orig_family_num
= 0;
5939 sum
= __gen_imsm_checksum(spare
);
5940 spare
->check_sum
= __cpu_to_le32(sum
);
5942 if (store_imsm_mpb(d
->fd
, spare
)) {
5943 pr_err("failed for device %d:%d %s\n",
5944 d
->major
, d
->minor
, strerror(errno
));
5956 static int write_super_imsm(struct supertype
*st
, int doclose
)
5958 struct intel_super
*super
= st
->sb
;
5959 unsigned int sector_size
= super
->sector_size
;
5960 struct imsm_super
*mpb
= super
->anchor
;
5966 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5968 int clear_migration_record
= 1;
5971 /* 'generation' is incremented everytime the metadata is written */
5972 generation
= __le32_to_cpu(mpb
->generation_num
);
5974 mpb
->generation_num
= __cpu_to_le32(generation
);
5976 /* fix up cases where previous mdadm releases failed to set
5979 if (mpb
->orig_family_num
== 0)
5980 mpb
->orig_family_num
= mpb
->family_num
;
5982 for (d
= super
->disks
; d
; d
= d
->next
) {
5986 mpb
->disk
[d
->index
] = d
->disk
;
5990 for (d
= super
->missing
; d
; d
= d
->next
) {
5991 mpb
->disk
[d
->index
] = d
->disk
;
5994 mpb
->num_disks
= num_disks
;
5995 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5997 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5998 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5999 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6001 imsm_copy_dev(dev
, dev2
);
6002 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6004 if (is_gen_migration(dev2
))
6005 clear_migration_record
= 0;
6008 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6011 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6012 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6014 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6016 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6017 mpb_size
+= bbm_log_size
;
6018 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6021 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6024 /* recalculate checksum */
6025 sum
= __gen_imsm_checksum(mpb
);
6026 mpb
->check_sum
= __cpu_to_le32(sum
);
6028 if (super
->clean_migration_record_by_mdmon
) {
6029 clear_migration_record
= 1;
6030 super
->clean_migration_record_by_mdmon
= 0;
6032 if (clear_migration_record
)
6033 memset(super
->migr_rec_buf
, 0,
6034 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6036 if (sector_size
== 4096)
6037 convert_to_4k(super
);
6039 /* write the mpb for disks that compose raid devices */
6040 for (d
= super
->disks
; d
; d
= d
->next
) {
6041 if (d
->index
< 0 || is_failed(&d
->disk
))
6044 if (clear_migration_record
) {
6045 unsigned long long dsize
;
6047 get_dev_size(d
->fd
, NULL
, &dsize
);
6048 if (lseek64(d
->fd
, dsize
- sector_size
,
6050 if ((unsigned int)write(d
->fd
,
6051 super
->migr_rec_buf
,
6052 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6053 MIGR_REC_BUF_SECTORS
*sector_size
)
6054 perror("Write migr_rec failed");
6058 if (store_imsm_mpb(d
->fd
, mpb
))
6060 "failed for device %d:%d (fd: %d)%s\n",
6062 d
->fd
, strerror(errno
));
6071 return write_super_imsm_spares(super
, doclose
);
6076 static int create_array(struct supertype
*st
, int dev_idx
)
6079 struct imsm_update_create_array
*u
;
6080 struct intel_super
*super
= st
->sb
;
6081 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6082 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6083 struct disk_info
*inf
;
6084 struct imsm_disk
*disk
;
6087 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6088 sizeof(*inf
) * map
->num_members
;
6090 u
->type
= update_create_array
;
6091 u
->dev_idx
= dev_idx
;
6092 imsm_copy_dev(&u
->dev
, dev
);
6093 inf
= get_disk_info(u
);
6094 for (i
= 0; i
< map
->num_members
; i
++) {
6095 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6097 disk
= get_imsm_disk(super
, idx
);
6099 disk
= get_imsm_missing(super
, idx
);
6100 serialcpy(inf
[i
].serial
, disk
->serial
);
6102 append_metadata_update(st
, u
, len
);
6107 static int mgmt_disk(struct supertype
*st
)
6109 struct intel_super
*super
= st
->sb
;
6111 struct imsm_update_add_remove_disk
*u
;
6113 if (!super
->disk_mgmt_list
)
6118 u
->type
= update_add_remove_disk
;
6119 append_metadata_update(st
, u
, len
);
6124 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6126 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6128 struct ppl_header
*ppl_hdr
= buf
;
6131 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6133 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6135 perror("Failed to seek to PPL header location");
6139 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6141 perror("Write PPL header failed");
6150 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6152 struct intel_super
*super
= st
->sb
;
6154 struct ppl_header
*ppl_hdr
;
6157 /* first clear entire ppl space */
6158 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6162 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6164 pr_err("Failed to allocate PPL header buffer\n");
6168 memset(buf
, 0, PPL_HEADER_SIZE
);
6170 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6171 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6173 if (info
->mismatch_cnt
) {
6175 * We are overwriting an invalid ppl. Make one entry with wrong
6176 * checksum to prevent the kernel from skipping resync.
6178 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6179 ppl_hdr
->entries
[0].checksum
= ~0;
6182 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6188 static int is_rebuilding(struct imsm_dev
*dev
);
6190 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6191 struct mdinfo
*disk
)
6193 struct intel_super
*super
= st
->sb
;
6195 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6197 struct ppl_header
*ppl_hdr
= NULL
;
6199 struct imsm_dev
*dev
;
6202 unsigned long long ppl_offset
= 0;
6203 unsigned long long prev_gen_num
= 0;
6205 if (disk
->disk
.raid_disk
< 0)
6208 dev
= get_imsm_dev(super
, info
->container_member
);
6209 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6210 d
= get_imsm_dl_disk(super
, idx
);
6212 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6215 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6216 pr_err("Failed to allocate PPL header buffer\n");
6222 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6225 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6227 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6229 perror("Failed to seek to PPL header location");
6234 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6235 perror("Read PPL header failed");
6242 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6243 ppl_hdr
->checksum
= 0;
6245 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6246 dprintf("Wrong PPL header checksum on %s\n",
6251 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6252 /* previous was newest, it was already checked */
6256 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6257 super
->anchor
->orig_family_num
)) {
6258 dprintf("Wrong PPL header signature on %s\n",
6265 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6267 ppl_offset
+= PPL_HEADER_SIZE
;
6268 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6270 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6273 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6285 * Update metadata to use mutliple PPLs area (1MB).
6286 * This is done once for all RAID members
6288 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6289 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6291 struct mdinfo
*member_dev
;
6293 sprintf(subarray
, "%d", info
->container_member
);
6295 if (mdmon_running(st
->container_devnm
))
6296 st
->update_tail
= &st
->updates
;
6298 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6299 pr_err("Failed to update subarray %s\n",
6302 if (st
->update_tail
)
6303 flush_metadata_updates(st
);
6305 st
->ss
->sync_metadata(st
);
6306 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6307 for (member_dev
= info
->devs
; member_dev
;
6308 member_dev
= member_dev
->next
)
6309 member_dev
->ppl_size
=
6310 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6315 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6317 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6318 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6319 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6320 (is_rebuilding(dev
) &&
6321 dev
->vol
.curr_migr_unit
== 0 &&
6322 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6323 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6325 info
->mismatch_cnt
++;
6326 } else if (ret
== 0 &&
6327 ppl_hdr
->entries_count
== 0 &&
6328 is_rebuilding(dev
) &&
6329 info
->resync_start
== 0) {
6331 * The header has no entries - add a single empty entry and
6332 * rewrite the header to prevent the kernel from going into
6333 * resync after an interrupted rebuild.
6335 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6336 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6344 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6346 struct intel_super
*super
= st
->sb
;
6350 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6351 info
->array
.level
!= 5)
6354 for (d
= super
->disks
; d
; d
= d
->next
) {
6355 if (d
->index
< 0 || is_failed(&d
->disk
))
6358 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6366 static int write_init_super_imsm(struct supertype
*st
)
6368 struct intel_super
*super
= st
->sb
;
6369 int current_vol
= super
->current_vol
;
6373 getinfo_super_imsm(st
, &info
, NULL
);
6375 /* we are done with current_vol reset it to point st at the container */
6376 super
->current_vol
= -1;
6378 if (st
->update_tail
) {
6379 /* queue the recently created array / added disk
6380 * as a metadata update */
6382 /* determine if we are creating a volume or adding a disk */
6383 if (current_vol
< 0) {
6384 /* in the mgmt (add/remove) disk case we are running
6385 * in mdmon context, so don't close fd's
6389 rv
= write_init_ppl_imsm_all(st
, &info
);
6391 rv
= create_array(st
, current_vol
);
6395 for (d
= super
->disks
; d
; d
= d
->next
)
6396 Kill(d
->devname
, NULL
, 0, -1, 1);
6397 if (current_vol
>= 0)
6398 rv
= write_init_ppl_imsm_all(st
, &info
);
6400 rv
= write_super_imsm(st
, 1);
6406 static int store_super_imsm(struct supertype
*st
, int fd
)
6408 struct intel_super
*super
= st
->sb
;
6409 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6414 if (super
->sector_size
== 4096)
6415 convert_to_4k(super
);
6416 return store_imsm_mpb(fd
, mpb
);
6419 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6420 int layout
, int raiddisks
, int chunk
,
6421 unsigned long long size
,
6422 unsigned long long data_offset
,
6424 unsigned long long *freesize
,
6428 unsigned long long ldsize
;
6429 struct intel_super
*super
;
6432 if (level
!= LEVEL_CONTAINER
)
6437 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6440 pr_err("imsm: Cannot open %s: %s\n",
6441 dev
, strerror(errno
));
6444 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6449 /* capabilities retrieve could be possible
6450 * note that there is no fd for the disks in array.
6452 super
= alloc_super();
6457 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6463 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6467 fd2devname(fd
, str
);
6468 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6469 fd
, str
, super
->orom
, rv
, raiddisks
);
6471 /* no orom/efi or non-intel hba of the disk */
6478 if (raiddisks
> super
->orom
->tds
) {
6480 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6481 raiddisks
, super
->orom
->tds
);
6485 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6486 (ldsize
>> 9) >> 32 > 0) {
6488 pr_err("%s exceeds maximum platform supported size\n", dev
);
6494 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6500 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6502 const unsigned long long base_start
= e
[*idx
].start
;
6503 unsigned long long end
= base_start
+ e
[*idx
].size
;
6506 if (base_start
== end
)
6510 for (i
= *idx
; i
< num_extents
; i
++) {
6511 /* extend overlapping extents */
6512 if (e
[i
].start
>= base_start
&&
6513 e
[i
].start
<= end
) {
6516 if (e
[i
].start
+ e
[i
].size
> end
)
6517 end
= e
[i
].start
+ e
[i
].size
;
6518 } else if (e
[i
].start
> end
) {
6524 return end
- base_start
;
6527 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6529 /* build a composite disk with all known extents and generate a new
6530 * 'maxsize' given the "all disks in an array must share a common start
6531 * offset" constraint
6533 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6537 unsigned long long pos
;
6538 unsigned long long start
= 0;
6539 unsigned long long maxsize
;
6540 unsigned long reserve
;
6542 /* coalesce and sort all extents. also, check to see if we need to
6543 * reserve space between member arrays
6546 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6549 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6552 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6557 while (i
< sum_extents
) {
6558 e
[j
].start
= e
[i
].start
;
6559 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6561 if (e
[j
-1].size
== 0)
6570 unsigned long long esize
;
6572 esize
= e
[i
].start
- pos
;
6573 if (esize
>= maxsize
) {
6578 pos
= e
[i
].start
+ e
[i
].size
;
6580 } while (e
[i
-1].size
);
6586 /* FIXME assumes volume at offset 0 is the first volume in a
6589 if (start_extent
> 0)
6590 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6594 if (maxsize
< reserve
)
6597 super
->create_offset
= ~((unsigned long long) 0);
6598 if (start
+ reserve
> super
->create_offset
)
6599 return 0; /* start overflows create_offset */
6600 super
->create_offset
= start
+ reserve
;
6602 return maxsize
- reserve
;
6605 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6607 if (level
< 0 || level
== 6 || level
== 4)
6610 /* if we have an orom prevent invalid raid levels */
6613 case 0: return imsm_orom_has_raid0(orom
);
6616 return imsm_orom_has_raid1e(orom
);
6617 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6618 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6619 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6622 return 1; /* not on an Intel RAID platform so anything goes */
6628 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6629 int dpa
, int verbose
)
6631 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6632 struct mdstat_ent
*memb
;
6638 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6639 if (memb
->metadata_version
&&
6640 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6641 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6642 !is_subarray(memb
->metadata_version
+9) &&
6644 struct dev_member
*dev
= memb
->members
;
6646 while(dev
&& (fd
< 0)) {
6647 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6648 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6650 fd
= open(path
, O_RDONLY
, 0);
6651 if (num
<= 0 || fd
< 0) {
6652 pr_vrb("Cannot open %s: %s\n",
6653 dev
->name
, strerror(errno
));
6659 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6660 struct mdstat_ent
*vol
;
6661 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6662 if (vol
->active
> 0 &&
6663 vol
->metadata_version
&&
6664 is_container_member(vol
, memb
->devnm
)) {
6669 if (*devlist
&& (found
< dpa
)) {
6670 dv
= xcalloc(1, sizeof(*dv
));
6671 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6672 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6675 dv
->next
= *devlist
;
6683 free_mdstat(mdstat
);
6688 static struct md_list
*
6689 get_loop_devices(void)
6692 struct md_list
*devlist
= NULL
;
6695 for(i
= 0; i
< 12; i
++) {
6696 dv
= xcalloc(1, sizeof(*dv
));
6697 dv
->devname
= xmalloc(40);
6698 sprintf(dv
->devname
, "/dev/loop%d", i
);
6706 static struct md_list
*
6707 get_devices(const char *hba_path
)
6709 struct md_list
*devlist
= NULL
;
6716 devlist
= get_loop_devices();
6719 /* scroll through /sys/dev/block looking for devices attached to
6722 dir
= opendir("/sys/dev/block");
6723 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6728 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6730 path
= devt_to_devpath(makedev(major
, minor
));
6733 if (!path_attached_to_hba(path
, hba_path
)) {
6740 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6742 fd2devname(fd
, buf
);
6745 pr_err("cannot open device: %s\n",
6750 dv
= xcalloc(1, sizeof(*dv
));
6751 dv
->devname
= xstrdup(buf
);
6758 devlist
= devlist
->next
;
6768 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6769 int verbose
, int *found
)
6771 struct md_list
*tmpdev
;
6773 struct supertype
*st
;
6775 /* first walk the list of devices to find a consistent set
6776 * that match the criterea, if that is possible.
6777 * We flag the ones we like with 'used'.
6780 st
= match_metadata_desc_imsm("imsm");
6782 pr_vrb("cannot allocate memory for imsm supertype\n");
6786 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6787 char *devname
= tmpdev
->devname
;
6789 struct supertype
*tst
;
6791 if (tmpdev
->used
> 1)
6793 tst
= dup_super(st
);
6795 pr_vrb("cannot allocate memory for imsm supertype\n");
6798 tmpdev
->container
= 0;
6799 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6801 dprintf("cannot open device %s: %s\n",
6802 devname
, strerror(errno
));
6804 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
6806 } else if (must_be_container(dfd
)) {
6807 struct supertype
*cst
;
6808 cst
= super_by_fd(dfd
, NULL
);
6810 dprintf("cannot recognize container type %s\n",
6813 } else if (tst
->ss
!= st
->ss
) {
6814 dprintf("non-imsm container - ignore it: %s\n",
6817 } else if (!tst
->ss
->load_container
||
6818 tst
->ss
->load_container(tst
, dfd
, NULL
))
6821 tmpdev
->container
= 1;
6824 cst
->ss
->free_super(cst
);
6826 tmpdev
->st_rdev
= rdev
;
6827 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6828 dprintf("no RAID superblock on %s\n",
6831 } else if (tst
->ss
->compare_super
== NULL
) {
6832 dprintf("Cannot assemble %s metadata on %s\n",
6833 tst
->ss
->name
, devname
);
6839 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6840 /* Ignore unrecognised devices during auto-assembly */
6845 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6847 if (st
->minor_version
== -1)
6848 st
->minor_version
= tst
->minor_version
;
6850 if (memcmp(info
.uuid
, uuid_zero
,
6851 sizeof(int[4])) == 0) {
6852 /* this is a floating spare. It cannot define
6853 * an array unless there are no more arrays of
6854 * this type to be found. It can be included
6855 * in an array of this type though.
6861 if (st
->ss
!= tst
->ss
||
6862 st
->minor_version
!= tst
->minor_version
||
6863 st
->ss
->compare_super(st
, tst
) != 0) {
6864 /* Some mismatch. If exactly one array matches this host,
6865 * we can resolve on that one.
6866 * Or, if we are auto assembling, we just ignore the second
6869 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6875 dprintf("found: devname: %s\n", devname
);
6879 tst
->ss
->free_super(tst
);
6883 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6884 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6885 for (iter
= head
; iter
; iter
= iter
->next
) {
6886 dprintf("content->text_version: %s vol\n",
6887 iter
->text_version
);
6888 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6889 /* do not assemble arrays with unsupported
6891 dprintf("Cannot activate member %s.\n",
6892 iter
->text_version
);
6899 dprintf("No valid super block on device list: err: %d %p\n",
6903 dprintf("no more devices to examine\n");
6906 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6907 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6909 if (count
< tmpdev
->found
)
6912 count
-= tmpdev
->found
;
6915 if (tmpdev
->used
== 1)
6920 st
->ss
->free_super(st
);
6924 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
6927 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6929 const struct orom_entry
*entry
;
6930 struct devid_list
*dv
, *devid_list
;
6935 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
6936 if (strstr(idev
->path
, hba_path
))
6940 if (!idev
|| !idev
->dev_id
)
6943 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
6945 if (!entry
|| !entry
->devid_list
)
6948 devid_list
= entry
->devid_list
;
6949 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
6950 struct md_list
*devlist
;
6951 struct sys_dev
*device
= NULL
;
6956 device
= device_by_id_and_path(dv
->devid
, hba_path
);
6958 device
= device_by_id(dv
->devid
);
6961 hpath
= device
->path
;
6965 devlist
= get_devices(hpath
);
6966 /* if no intel devices return zero volumes */
6967 if (devlist
== NULL
)
6970 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
6972 dprintf("path: %s active arrays: %d\n", hpath
, count
);
6973 if (devlist
== NULL
)
6977 count
+= count_volumes_list(devlist
,
6981 dprintf("found %d count: %d\n", found
, count
);
6984 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
6987 struct md_list
*dv
= devlist
;
6988 devlist
= devlist
->next
;
6996 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7000 if (hba
->type
== SYS_DEV_VMD
) {
7001 struct sys_dev
*dev
;
7004 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7005 if (dev
->type
== SYS_DEV_VMD
)
7006 count
+= __count_volumes(dev
->path
, dpa
,
7011 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7014 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7016 /* up to 512 if the plaform supports it, otherwise the platform max.
7017 * 128 if no platform detected
7019 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7021 return min(512, (1 << fs
));
7025 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7026 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7028 /* check/set platform and metadata limits/defaults */
7029 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7030 pr_vrb("platform supports a maximum of %d disks per array\n",
7035 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7036 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7037 pr_vrb("platform does not support raid%d with %d disk%s\n",
7038 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7042 if (*chunk
== 0 || *chunk
== UnSet
)
7043 *chunk
= imsm_default_chunk(super
->orom
);
7045 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7046 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7050 if (layout
!= imsm_level_to_layout(level
)) {
7052 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7053 else if (level
== 10)
7054 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7056 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7061 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7062 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7063 pr_vrb("platform does not support a volume size over 2TB\n");
7070 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7071 * FIX ME add ahci details
7073 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7074 int layout
, int raiddisks
, int *chunk
,
7075 unsigned long long size
,
7076 unsigned long long data_offset
,
7078 unsigned long long *freesize
,
7082 struct intel_super
*super
= st
->sb
;
7083 struct imsm_super
*mpb
;
7085 unsigned long long pos
= 0;
7086 unsigned long long maxsize
;
7090 /* We must have the container info already read in. */
7094 mpb
= super
->anchor
;
7096 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7097 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7101 /* General test: make sure there is space for
7102 * 'raiddisks' device extents of size 'size' at a given
7105 unsigned long long minsize
= size
;
7106 unsigned long long start_offset
= MaxSector
;
7109 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7110 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7115 e
= get_extents(super
, dl
);
7118 unsigned long long esize
;
7119 esize
= e
[i
].start
- pos
;
7120 if (esize
>= minsize
)
7122 if (found
&& start_offset
== MaxSector
) {
7125 } else if (found
&& pos
!= start_offset
) {
7129 pos
= e
[i
].start
+ e
[i
].size
;
7131 } while (e
[i
-1].size
);
7136 if (dcnt
< raiddisks
) {
7138 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7145 /* This device must be a member of the set */
7146 if (!stat_is_blkdev(dev
, &rdev
))
7148 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7149 if (dl
->major
== (int)major(rdev
) &&
7150 dl
->minor
== (int)minor(rdev
))
7155 pr_err("%s is not in the same imsm set\n", dev
);
7157 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7158 /* If a volume is present then the current creation attempt
7159 * cannot incorporate new spares because the orom may not
7160 * understand this configuration (all member disks must be
7161 * members of each array in the container).
7163 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7164 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7166 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7167 mpb
->num_disks
!= raiddisks
) {
7168 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7172 /* retrieve the largest free space block */
7173 e
= get_extents(super
, dl
);
7178 unsigned long long esize
;
7180 esize
= e
[i
].start
- pos
;
7181 if (esize
>= maxsize
)
7183 pos
= e
[i
].start
+ e
[i
].size
;
7185 } while (e
[i
-1].size
);
7190 pr_err("unable to determine free space for: %s\n",
7194 if (maxsize
< size
) {
7196 pr_err("%s not enough space (%llu < %llu)\n",
7197 dev
, maxsize
, size
);
7201 /* count total number of extents for merge */
7203 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7205 i
+= dl
->extent_cnt
;
7207 maxsize
= merge_extents(super
, i
);
7209 if (!check_env("IMSM_NO_PLATFORM") &&
7210 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7211 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7215 if (maxsize
< size
|| maxsize
== 0) {
7218 pr_err("no free space left on device. Aborting...\n");
7220 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7226 *freesize
= maxsize
;
7229 int count
= count_volumes(super
->hba
,
7230 super
->orom
->dpa
, verbose
);
7231 if (super
->orom
->vphba
<= count
) {
7232 pr_vrb("platform does not support more than %d raid volumes.\n",
7233 super
->orom
->vphba
);
7240 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7241 unsigned long long size
, int chunk
,
7242 unsigned long long *freesize
)
7244 struct intel_super
*super
= st
->sb
;
7245 struct imsm_super
*mpb
= super
->anchor
;
7250 unsigned long long maxsize
;
7251 unsigned long long minsize
;
7255 /* find the largest common start free region of the possible disks */
7259 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7265 /* don't activate new spares if we are orom constrained
7266 * and there is already a volume active in the container
7268 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7271 e
= get_extents(super
, dl
);
7274 for (i
= 1; e
[i
-1].size
; i
++)
7282 maxsize
= merge_extents(super
, extent_cnt
);
7286 minsize
= chunk
* 2;
7288 if (cnt
< raiddisks
||
7289 (super
->orom
&& used
&& used
!= raiddisks
) ||
7290 maxsize
< minsize
||
7292 pr_err("not enough devices with space to create array.\n");
7293 return 0; /* No enough free spaces large enough */
7304 if (!check_env("IMSM_NO_PLATFORM") &&
7305 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7306 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7310 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7312 dl
->raiddisk
= cnt
++;
7316 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7321 static int reserve_space(struct supertype
*st
, int raiddisks
,
7322 unsigned long long size
, int chunk
,
7323 unsigned long long *freesize
)
7325 struct intel_super
*super
= st
->sb
;
7330 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7333 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7335 dl
->raiddisk
= cnt
++;
7342 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7343 int raiddisks
, int *chunk
, unsigned long long size
,
7344 unsigned long long data_offset
,
7345 char *dev
, unsigned long long *freesize
,
7346 int consistency_policy
, int verbose
)
7353 * if given unused devices create a container
7354 * if given given devices in a container create a member volume
7356 if (level
== LEVEL_CONTAINER
) {
7357 /* Must be a fresh device to add to a container */
7358 return validate_geometry_imsm_container(st
, level
, layout
,
7366 if (size
&& ((size
< 1024) || (*chunk
!= UnSet
&&
7367 size
< (unsigned long long) *chunk
))) {
7368 pr_err("Given size must be greater than 1M and chunk size.\n");
7369 /* Depends on algorithm in Create.c :
7370 * if container was given (dev == NULL) return -1,
7371 * if block device was given ( dev != NULL) return 0.
7373 return dev
? -1 : 0;
7378 struct intel_super
*super
= st
->sb
;
7379 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7380 raiddisks
, chunk
, size
,
7383 /* we are being asked to automatically layout a
7384 * new volume based on the current contents of
7385 * the container. If the the parameters can be
7386 * satisfied reserve_space will record the disks,
7387 * start offset, and size of the volume to be
7388 * created. add_to_super and getinfo_super
7389 * detect when autolayout is in progress.
7391 /* assuming that freesize is always given when array is
7393 if (super
->orom
&& freesize
) {
7395 count
= count_volumes(super
->hba
,
7396 super
->orom
->dpa
, verbose
);
7397 if (super
->orom
->vphba
<= count
) {
7398 pr_vrb("platform does not support more than %d raid volumes.\n",
7399 super
->orom
->vphba
);
7404 return reserve_space(st
, raiddisks
, size
,
7410 /* creating in a given container */
7411 return validate_geometry_imsm_volume(st
, level
, layout
,
7412 raiddisks
, chunk
, size
,
7414 dev
, freesize
, verbose
);
7417 /* This device needs to be a device in an 'imsm' container */
7418 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7421 pr_err("Cannot create this array on device %s\n",
7426 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7428 pr_err("Cannot open %s: %s\n",
7429 dev
, strerror(errno
));
7432 /* Well, it is in use by someone, maybe an 'imsm' container. */
7433 cfd
= open_container(fd
);
7437 pr_err("Cannot use %s: It is busy\n",
7441 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7442 if (sra
&& sra
->array
.major_version
== -1 &&
7443 strcmp(sra
->text_version
, "imsm") == 0)
7447 /* This is a member of a imsm container. Load the container
7448 * and try to create a volume
7450 struct intel_super
*super
;
7452 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7454 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7456 return validate_geometry_imsm_volume(st
, level
, layout
,
7458 size
, data_offset
, dev
,
7465 pr_err("failed container membership check\n");
7471 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7473 struct intel_super
*super
= st
->sb
;
7475 if (level
&& *level
== UnSet
)
7476 *level
= LEVEL_CONTAINER
;
7478 if (level
&& layout
&& *layout
== UnSet
)
7479 *layout
= imsm_level_to_layout(*level
);
7481 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7482 *chunk
= imsm_default_chunk(super
->orom
);
7485 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7487 static int kill_subarray_imsm(struct supertype
*st
)
7489 /* remove the subarray currently referenced by ->current_vol */
7491 struct intel_dev
**dp
;
7492 struct intel_super
*super
= st
->sb
;
7493 __u8 current_vol
= super
->current_vol
;
7494 struct imsm_super
*mpb
= super
->anchor
;
7496 if (super
->current_vol
< 0)
7498 super
->current_vol
= -1; /* invalidate subarray cursor */
7500 /* block deletions that would change the uuid of active subarrays
7502 * FIXME when immutable ids are available, but note that we'll
7503 * also need to fixup the invalidated/active subarray indexes in
7506 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7509 if (i
< current_vol
)
7511 sprintf(subarray
, "%u", i
);
7512 if (is_subarray_active(subarray
, st
->devnm
)) {
7513 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7520 if (st
->update_tail
) {
7521 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7523 u
->type
= update_kill_array
;
7524 u
->dev_idx
= current_vol
;
7525 append_metadata_update(st
, u
, sizeof(*u
));
7530 for (dp
= &super
->devlist
; *dp
;)
7531 if ((*dp
)->index
== current_vol
) {
7534 handle_missing(super
, (*dp
)->dev
);
7535 if ((*dp
)->index
> current_vol
)
7540 /* no more raid devices, all active components are now spares,
7541 * but of course failed are still failed
7543 if (--mpb
->num_raid_devs
== 0) {
7546 for (d
= super
->disks
; d
; d
= d
->next
)
7551 super
->updates_pending
++;
7556 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7557 char *update
, struct mddev_ident
*ident
)
7559 /* update the subarray currently referenced by ->current_vol */
7560 struct intel_super
*super
= st
->sb
;
7561 struct imsm_super
*mpb
= super
->anchor
;
7563 if (strcmp(update
, "name") == 0) {
7564 char *name
= ident
->name
;
7568 if (is_subarray_active(subarray
, st
->devnm
)) {
7569 pr_err("Unable to update name of active subarray\n");
7573 if (!check_name(super
, name
, 0))
7576 vol
= strtoul(subarray
, &ep
, 10);
7577 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7580 if (st
->update_tail
) {
7581 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7583 u
->type
= update_rename_array
;
7585 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7586 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7587 append_metadata_update(st
, u
, sizeof(*u
));
7589 struct imsm_dev
*dev
;
7592 dev
= get_imsm_dev(super
, vol
);
7593 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
7594 dev
->volume
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7595 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7596 dev
= get_imsm_dev(super
, i
);
7597 handle_missing(super
, dev
);
7599 super
->updates_pending
++;
7601 } else if (strcmp(update
, "ppl") == 0 ||
7602 strcmp(update
, "no-ppl") == 0) {
7605 int vol
= strtoul(subarray
, &ep
, 10);
7607 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7610 if (strcmp(update
, "ppl") == 0)
7611 new_policy
= RWH_MULTIPLE_DISTRIBUTED
;
7613 new_policy
= RWH_MULTIPLE_OFF
;
7615 if (st
->update_tail
) {
7616 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7618 u
->type
= update_rwh_policy
;
7620 u
->new_policy
= new_policy
;
7621 append_metadata_update(st
, u
, sizeof(*u
));
7623 struct imsm_dev
*dev
;
7625 dev
= get_imsm_dev(super
, vol
);
7626 dev
->rwh_policy
= new_policy
;
7627 super
->updates_pending
++;
7635 static int is_gen_migration(struct imsm_dev
*dev
)
7640 if (!dev
->vol
.migr_state
)
7643 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7649 static int is_rebuilding(struct imsm_dev
*dev
)
7651 struct imsm_map
*migr_map
;
7653 if (!dev
->vol
.migr_state
)
7656 if (migr_type(dev
) != MIGR_REBUILD
)
7659 migr_map
= get_imsm_map(dev
, MAP_1
);
7661 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7667 static int is_initializing(struct imsm_dev
*dev
)
7669 struct imsm_map
*migr_map
;
7671 if (!dev
->vol
.migr_state
)
7674 if (migr_type(dev
) != MIGR_INIT
)
7677 migr_map
= get_imsm_map(dev
, MAP_1
);
7679 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7685 static void update_recovery_start(struct intel_super
*super
,
7686 struct imsm_dev
*dev
,
7687 struct mdinfo
*array
)
7689 struct mdinfo
*rebuild
= NULL
;
7693 if (!is_rebuilding(dev
))
7696 /* Find the rebuild target, but punt on the dual rebuild case */
7697 for (d
= array
->devs
; d
; d
= d
->next
)
7698 if (d
->recovery_start
== 0) {
7705 /* (?) none of the disks are marked with
7706 * IMSM_ORD_REBUILD, so assume they are missing and the
7707 * disk_ord_tbl was not correctly updated
7709 dprintf("failed to locate out-of-sync disk\n");
7713 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7714 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7717 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7719 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7721 /* Given a container loaded by load_super_imsm_all,
7722 * extract information about all the arrays into
7724 * If 'subarray' is given, just extract info about that array.
7726 * For each imsm_dev create an mdinfo, fill it in,
7727 * then look for matching devices in super->disks
7728 * and create appropriate device mdinfo.
7730 struct intel_super
*super
= st
->sb
;
7731 struct imsm_super
*mpb
= super
->anchor
;
7732 struct mdinfo
*rest
= NULL
;
7736 int spare_disks
= 0;
7738 /* do not assemble arrays when not all attributes are supported */
7739 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7741 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7744 /* count spare devices, not used in maps
7746 for (d
= super
->disks
; d
; d
= d
->next
)
7750 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7751 struct imsm_dev
*dev
;
7752 struct imsm_map
*map
;
7753 struct imsm_map
*map2
;
7754 struct mdinfo
*this;
7761 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7764 dev
= get_imsm_dev(super
, i
);
7765 map
= get_imsm_map(dev
, MAP_0
);
7766 map2
= get_imsm_map(dev
, MAP_1
);
7767 level
= get_imsm_raid_level(map
);
7769 /* do not publish arrays that are in the middle of an
7770 * unsupported migration
7772 if (dev
->vol
.migr_state
&&
7773 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7774 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7778 /* do not publish arrays that are not support by controller's
7782 this = xmalloc(sizeof(*this));
7784 super
->current_vol
= i
;
7785 getinfo_super_imsm_volume(st
, this, NULL
);
7787 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7788 /* mdadm does not support all metadata features- set the bit in all arrays state */
7789 if (!validate_geometry_imsm_orom(super
,
7790 level
, /* RAID level */
7791 imsm_level_to_layout(level
),
7792 map
->num_members
, /* raid disks */
7793 &chunk
, imsm_dev_size(dev
),
7795 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7797 this->array
.state
|=
7798 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7799 (1<<MD_SB_BLOCK_VOLUME
);
7802 /* if array has bad blocks, set suitable bit in all arrays state */
7804 this->array
.state
|=
7805 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7806 (1<<MD_SB_BLOCK_VOLUME
);
7808 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7809 unsigned long long recovery_start
;
7810 struct mdinfo
*info_d
;
7818 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7819 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7820 for (d
= super
->disks
; d
; d
= d
->next
)
7821 if (d
->index
== idx
)
7824 recovery_start
= MaxSector
;
7827 if (d
&& is_failed(&d
->disk
))
7829 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
7833 * if we skip some disks the array will be assmebled degraded;
7834 * reset resync start to avoid a dirty-degraded
7835 * situation when performing the intial sync
7840 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
7841 if ((!able_to_resync(level
, missing
) ||
7842 recovery_start
== 0))
7843 this->resync_start
= MaxSector
;
7846 * FIXME handle dirty degraded
7853 info_d
= xcalloc(1, sizeof(*info_d
));
7854 info_d
->next
= this->devs
;
7855 this->devs
= info_d
;
7857 info_d
->disk
.number
= d
->index
;
7858 info_d
->disk
.major
= d
->major
;
7859 info_d
->disk
.minor
= d
->minor
;
7860 info_d
->disk
.raid_disk
= slot
;
7861 info_d
->recovery_start
= recovery_start
;
7863 if (slot
< map2
->num_members
)
7864 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7866 this->array
.spare_disks
++;
7868 if (slot
< map
->num_members
)
7869 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7871 this->array
.spare_disks
++;
7873 if (info_d
->recovery_start
== MaxSector
)
7874 this->array
.working_disks
++;
7876 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7877 info_d
->data_offset
= pba_of_lba0(map
);
7878 info_d
->component_size
= calc_component_size(map
, dev
);
7880 if (map
->raid_level
== 5) {
7881 info_d
->ppl_sector
= this->ppl_sector
;
7882 info_d
->ppl_size
= this->ppl_size
;
7883 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
7884 recovery_start
== 0)
7885 this->resync_start
= 0;
7888 info_d
->bb
.supported
= 1;
7889 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7890 info_d
->data_offset
,
7891 info_d
->component_size
,
7894 /* now that the disk list is up-to-date fixup recovery_start */
7895 update_recovery_start(super
, dev
, this);
7896 this->array
.spare_disks
+= spare_disks
;
7898 /* check for reshape */
7899 if (this->reshape_active
== 1)
7900 recover_backup_imsm(st
, this);
7907 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7908 int failed
, int look_in_map
)
7910 struct imsm_map
*map
;
7912 map
= get_imsm_map(dev
, look_in_map
);
7915 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
7916 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
7918 switch (get_imsm_raid_level(map
)) {
7920 return IMSM_T_STATE_FAILED
;
7923 if (failed
< map
->num_members
)
7924 return IMSM_T_STATE_DEGRADED
;
7926 return IMSM_T_STATE_FAILED
;
7931 * check to see if any mirrors have failed, otherwise we
7932 * are degraded. Even numbered slots are mirrored on
7936 /* gcc -Os complains that this is unused */
7937 int insync
= insync
;
7939 for (i
= 0; i
< map
->num_members
; i
++) {
7940 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
7941 int idx
= ord_to_idx(ord
);
7942 struct imsm_disk
*disk
;
7944 /* reset the potential in-sync count on even-numbered
7945 * slots. num_copies is always 2 for imsm raid10
7950 disk
= get_imsm_disk(super
, idx
);
7951 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7954 /* no in-sync disks left in this mirror the
7958 return IMSM_T_STATE_FAILED
;
7961 return IMSM_T_STATE_DEGRADED
;
7965 return IMSM_T_STATE_DEGRADED
;
7967 return IMSM_T_STATE_FAILED
;
7973 return map
->map_state
;
7976 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
7981 struct imsm_disk
*disk
;
7982 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7983 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
7984 struct imsm_map
*map_for_loop
;
7989 /* at the beginning of migration we set IMSM_ORD_REBUILD on
7990 * disks that are being rebuilt. New failures are recorded to
7991 * map[0]. So we look through all the disks we started with and
7992 * see if any failures are still present, or if any new ones
7996 if (prev
&& (map
->num_members
< prev
->num_members
))
7997 map_for_loop
= prev
;
7999 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8001 /* when MAP_X is passed both maps failures are counted
8004 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8005 i
< prev
->num_members
) {
8006 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8007 idx_1
= ord_to_idx(ord
);
8009 disk
= get_imsm_disk(super
, idx_1
);
8010 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8013 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8014 i
< map
->num_members
) {
8015 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8016 idx
= ord_to_idx(ord
);
8019 disk
= get_imsm_disk(super
, idx
);
8020 if (!disk
|| is_failed(disk
) ||
8021 ord
& IMSM_ORD_REBUILD
)
8030 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8033 struct intel_super
*super
= c
->sb
;
8034 struct imsm_super
*mpb
= super
->anchor
;
8035 struct imsm_update_prealloc_bb_mem u
;
8037 if (atoi(inst
) >= mpb
->num_raid_devs
) {
8038 pr_err("subarry index %d, out of range\n", atoi(inst
));
8042 dprintf("imsm: open_new %s\n", inst
);
8043 a
->info
.container_member
= atoi(inst
);
8045 u
.type
= update_prealloc_badblocks_mem
;
8046 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8051 static int is_resyncing(struct imsm_dev
*dev
)
8053 struct imsm_map
*migr_map
;
8055 if (!dev
->vol
.migr_state
)
8058 if (migr_type(dev
) == MIGR_INIT
||
8059 migr_type(dev
) == MIGR_REPAIR
)
8062 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8065 migr_map
= get_imsm_map(dev
, MAP_1
);
8067 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8068 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8074 /* return true if we recorded new information */
8075 static int mark_failure(struct intel_super
*super
,
8076 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8080 struct imsm_map
*map
;
8081 char buf
[MAX_RAID_SERIAL_LEN
+3];
8082 unsigned int len
, shift
= 0;
8084 /* new failures are always set in map[0] */
8085 map
= get_imsm_map(dev
, MAP_0
);
8087 slot
= get_imsm_disk_slot(map
, idx
);
8091 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8092 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8095 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8096 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8098 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8099 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8100 memcpy(disk
->serial
, &buf
[shift
], len
+ 1 - shift
);
8102 disk
->status
|= FAILED_DISK
;
8103 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8104 /* mark failures in second map if second map exists and this disk
8106 * This is valid for migration, initialization and rebuild
8108 if (dev
->vol
.migr_state
) {
8109 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8110 int slot2
= get_imsm_disk_slot(map2
, idx
);
8112 if (slot2
< map2
->num_members
&& slot2
>= 0)
8113 set_imsm_ord_tbl_ent(map2
, slot2
,
8114 idx
| IMSM_ORD_REBUILD
);
8116 if (map
->failed_disk_num
== 0xff)
8117 map
->failed_disk_num
= slot
;
8119 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8124 static void mark_missing(struct intel_super
*super
,
8125 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8127 mark_failure(super
, dev
, disk
, idx
);
8129 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8132 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8133 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8136 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8140 if (!super
->missing
)
8143 /* When orom adds replacement for missing disk it does
8144 * not remove entry of missing disk, but just updates map with
8145 * new added disk. So it is not enough just to test if there is
8146 * any missing disk, we have to look if there are any failed disks
8147 * in map to stop migration */
8149 dprintf("imsm: mark missing\n");
8150 /* end process for initialization and rebuild only
8152 if (is_gen_migration(dev
) == 0) {
8153 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8157 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8158 struct imsm_map
*map1
;
8159 int i
, ord
, ord_map1
;
8162 for (i
= 0; i
< map
->num_members
; i
++) {
8163 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8164 if (!(ord
& IMSM_ORD_REBUILD
))
8167 map1
= get_imsm_map(dev
, MAP_1
);
8171 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8172 if (ord_map1
& IMSM_ORD_REBUILD
)
8177 map_state
= imsm_check_degraded(super
, dev
,
8179 end_migration(dev
, super
, map_state
);
8183 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8184 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8185 super
->updates_pending
++;
8188 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8191 unsigned long long array_blocks
;
8192 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8193 int used_disks
= imsm_num_data_members(map
);
8195 if (used_disks
== 0) {
8196 /* when problems occures
8197 * return current array_blocks value
8199 array_blocks
= imsm_dev_size(dev
);
8201 return array_blocks
;
8204 /* set array size in metadata
8207 /* OLCE size change is caused by added disks
8209 array_blocks
= per_dev_array_size(map
) * used_disks
;
8211 /* Online Volume Size Change
8212 * Using available free space
8214 array_blocks
= new_size
;
8216 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8217 set_imsm_dev_size(dev
, array_blocks
);
8219 return array_blocks
;
8222 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8224 static void imsm_progress_container_reshape(struct intel_super
*super
)
8226 /* if no device has a migr_state, but some device has a
8227 * different number of members than the previous device, start
8228 * changing the number of devices in this device to match
8231 struct imsm_super
*mpb
= super
->anchor
;
8232 int prev_disks
= -1;
8236 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8237 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8238 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8239 struct imsm_map
*map2
;
8240 int prev_num_members
;
8242 if (dev
->vol
.migr_state
)
8245 if (prev_disks
== -1)
8246 prev_disks
= map
->num_members
;
8247 if (prev_disks
== map
->num_members
)
8250 /* OK, this array needs to enter reshape mode.
8251 * i.e it needs a migr_state
8254 copy_map_size
= sizeof_imsm_map(map
);
8255 prev_num_members
= map
->num_members
;
8256 map
->num_members
= prev_disks
;
8257 dev
->vol
.migr_state
= 1;
8258 dev
->vol
.curr_migr_unit
= 0;
8259 set_migr_type(dev
, MIGR_GEN_MIGR
);
8260 for (i
= prev_num_members
;
8261 i
< map
->num_members
; i
++)
8262 set_imsm_ord_tbl_ent(map
, i
, i
);
8263 map2
= get_imsm_map(dev
, MAP_1
);
8264 /* Copy the current map */
8265 memcpy(map2
, map
, copy_map_size
);
8266 map2
->num_members
= prev_num_members
;
8268 imsm_set_array_size(dev
, -1);
8269 super
->clean_migration_record_by_mdmon
= 1;
8270 super
->updates_pending
++;
8274 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8275 * states are handled in imsm_set_disk() with one exception, when a
8276 * resync is stopped due to a new failure this routine will set the
8277 * 'degraded' state for the array.
8279 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8281 int inst
= a
->info
.container_member
;
8282 struct intel_super
*super
= a
->container
->sb
;
8283 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8284 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8285 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8286 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8287 __u32 blocks_per_unit
;
8289 if (dev
->vol
.migr_state
&&
8290 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8291 /* array state change is blocked due to reshape action
8293 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8294 * - finish the reshape (if last_checkpoint is big and action != reshape)
8295 * - update curr_migr_unit
8297 if (a
->curr_action
== reshape
) {
8298 /* still reshaping, maybe update curr_migr_unit */
8299 goto mark_checkpoint
;
8301 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8302 /* for some reason we aborted the reshape.
8304 * disable automatic metadata rollback
8305 * user action is required to recover process
8308 struct imsm_map
*map2
=
8309 get_imsm_map(dev
, MAP_1
);
8310 dev
->vol
.migr_state
= 0;
8311 set_migr_type(dev
, 0);
8312 dev
->vol
.curr_migr_unit
= 0;
8314 sizeof_imsm_map(map2
));
8315 super
->updates_pending
++;
8318 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8319 unsigned long long array_blocks
;
8323 used_disks
= imsm_num_data_members(map
);
8324 if (used_disks
> 0) {
8326 per_dev_array_size(map
) *
8329 round_size_to_mb(array_blocks
,
8331 a
->info
.custom_array_size
= array_blocks
;
8332 /* encourage manager to update array
8336 a
->check_reshape
= 1;
8338 /* finalize online capacity expansion/reshape */
8339 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8341 mdi
->disk
.raid_disk
,
8344 imsm_progress_container_reshape(super
);
8349 /* before we activate this array handle any missing disks */
8350 if (consistent
== 2)
8351 handle_missing(super
, dev
);
8353 if (consistent
== 2 &&
8354 (!is_resync_complete(&a
->info
) ||
8355 map_state
!= IMSM_T_STATE_NORMAL
||
8356 dev
->vol
.migr_state
))
8359 if (is_resync_complete(&a
->info
)) {
8360 /* complete intialization / resync,
8361 * recovery and interrupted recovery is completed in
8364 if (is_resyncing(dev
)) {
8365 dprintf("imsm: mark resync done\n");
8366 end_migration(dev
, super
, map_state
);
8367 super
->updates_pending
++;
8368 a
->last_checkpoint
= 0;
8370 } else if ((!is_resyncing(dev
) && !failed
) &&
8371 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8372 /* mark the start of the init process if nothing is failed */
8373 dprintf("imsm: mark resync start\n");
8374 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8375 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8377 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8378 super
->updates_pending
++;
8382 /* skip checkpointing for general migration,
8383 * it is controlled in mdadm
8385 if (is_gen_migration(dev
))
8386 goto skip_mark_checkpoint
;
8388 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8389 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8390 if (blocks_per_unit
) {
8394 units
= a
->last_checkpoint
/ blocks_per_unit
;
8397 /* check that we did not overflow 32-bits, and that
8398 * curr_migr_unit needs updating
8400 if (units32
== units
&&
8402 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8403 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8404 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8405 super
->updates_pending
++;
8409 skip_mark_checkpoint
:
8410 /* mark dirty / clean */
8411 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8412 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8413 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8415 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8417 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8418 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8419 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8420 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8422 super
->updates_pending
++;
8428 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8430 int inst
= a
->info
.container_member
;
8431 struct intel_super
*super
= a
->container
->sb
;
8432 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8433 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8435 if (slot
> map
->num_members
) {
8436 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8437 slot
, map
->num_members
- 1);
8444 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8447 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8449 int inst
= a
->info
.container_member
;
8450 struct intel_super
*super
= a
->container
->sb
;
8451 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8452 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8453 struct imsm_disk
*disk
;
8455 int recovery_not_finished
= 0;
8459 int rebuild_done
= 0;
8462 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8466 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8467 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8469 /* check for new failures */
8470 if (state
& DS_FAULTY
) {
8471 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8472 super
->updates_pending
++;
8475 /* check if in_sync */
8476 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8477 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8479 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8481 super
->updates_pending
++;
8484 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8485 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8487 /* check if recovery complete, newly degraded, or failed */
8488 dprintf("imsm: Detected transition to state ");
8489 switch (map_state
) {
8490 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8491 dprintf("normal: ");
8492 if (is_rebuilding(dev
)) {
8493 dprintf_cont("while rebuilding");
8494 /* check if recovery is really finished */
8495 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8496 if (mdi
->recovery_start
!= MaxSector
) {
8497 recovery_not_finished
= 1;
8500 if (recovery_not_finished
) {
8502 dprintf("Rebuild has not finished yet, state not changed");
8503 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8504 a
->last_checkpoint
= mdi
->recovery_start
;
8505 super
->updates_pending
++;
8509 end_migration(dev
, super
, map_state
);
8510 map
= get_imsm_map(dev
, MAP_0
);
8511 map
->failed_disk_num
= ~0;
8512 super
->updates_pending
++;
8513 a
->last_checkpoint
= 0;
8516 if (is_gen_migration(dev
)) {
8517 dprintf_cont("while general migration");
8518 if (a
->last_checkpoint
>= a
->info
.component_size
)
8519 end_migration(dev
, super
, map_state
);
8521 map
->map_state
= map_state
;
8522 map
= get_imsm_map(dev
, MAP_0
);
8523 map
->failed_disk_num
= ~0;
8524 super
->updates_pending
++;
8528 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8529 dprintf_cont("degraded: ");
8530 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8531 dprintf_cont("mark degraded");
8532 map
->map_state
= map_state
;
8533 super
->updates_pending
++;
8534 a
->last_checkpoint
= 0;
8537 if (is_rebuilding(dev
)) {
8538 dprintf_cont("while rebuilding.");
8539 if (map
->map_state
!= map_state
) {
8540 dprintf_cont(" Map state change");
8541 end_migration(dev
, super
, map_state
);
8542 super
->updates_pending
++;
8543 } else if (!rebuild_done
) {
8547 /* check if recovery is really finished */
8548 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8549 if (mdi
->recovery_start
!= MaxSector
) {
8550 recovery_not_finished
= 1;
8553 if (recovery_not_finished
) {
8555 dprintf("Rebuild has not finished yet, state not changed");
8556 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8557 a
->last_checkpoint
=
8558 mdi
->recovery_start
;
8559 super
->updates_pending
++;
8564 dprintf_cont(" Rebuild done, still degraded");
8565 dev
->vol
.migr_state
= 0;
8566 set_migr_type(dev
, 0);
8567 dev
->vol
.curr_migr_unit
= 0;
8569 for (i
= 0; i
< map
->num_members
; i
++) {
8570 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8572 if (idx
& IMSM_ORD_REBUILD
)
8573 map
->failed_disk_num
= i
;
8575 super
->updates_pending
++;
8578 if (is_gen_migration(dev
)) {
8579 dprintf_cont("while general migration");
8580 if (a
->last_checkpoint
>= a
->info
.component_size
)
8581 end_migration(dev
, super
, map_state
);
8583 map
->map_state
= map_state
;
8584 manage_second_map(super
, dev
);
8586 super
->updates_pending
++;
8589 if (is_initializing(dev
)) {
8590 dprintf_cont("while initialization.");
8591 map
->map_state
= map_state
;
8592 super
->updates_pending
++;
8596 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8597 dprintf_cont("failed: ");
8598 if (is_gen_migration(dev
)) {
8599 dprintf_cont("while general migration");
8600 map
->map_state
= map_state
;
8601 super
->updates_pending
++;
8604 if (map
->map_state
!= map_state
) {
8605 dprintf_cont("mark failed");
8606 end_migration(dev
, super
, map_state
);
8607 super
->updates_pending
++;
8608 a
->last_checkpoint
= 0;
8613 dprintf_cont("state %i\n", map_state
);
8618 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8621 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8622 unsigned long long dsize
;
8623 unsigned long long sectors
;
8624 unsigned int sector_size
;
8626 get_dev_sector_size(fd
, NULL
, §or_size
);
8627 get_dev_size(fd
, NULL
, &dsize
);
8629 if (mpb_size
> sector_size
) {
8630 /* -1 to account for anchor */
8631 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8633 /* write the extended mpb to the sectors preceeding the anchor */
8634 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8638 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8639 sector_size
* sectors
) != sector_size
* sectors
)
8643 /* first block is stored on second to last sector of the disk */
8644 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8647 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8653 static void imsm_sync_metadata(struct supertype
*container
)
8655 struct intel_super
*super
= container
->sb
;
8657 dprintf("sync metadata: %d\n", super
->updates_pending
);
8658 if (!super
->updates_pending
)
8661 write_super_imsm(container
, 0);
8663 super
->updates_pending
= 0;
8666 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8668 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8669 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8672 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8676 if (dl
&& is_failed(&dl
->disk
))
8680 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8685 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8686 struct active_array
*a
, int activate_new
,
8687 struct mdinfo
*additional_test_list
)
8689 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8690 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8691 struct imsm_super
*mpb
= super
->anchor
;
8692 struct imsm_map
*map
;
8693 unsigned long long pos
;
8698 __u32 array_start
= 0;
8699 __u32 array_end
= 0;
8701 struct mdinfo
*test_list
;
8703 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8704 /* If in this array, skip */
8705 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8706 if (d
->state_fd
>= 0 &&
8707 d
->disk
.major
== dl
->major
&&
8708 d
->disk
.minor
== dl
->minor
) {
8709 dprintf("%x:%x already in array\n",
8710 dl
->major
, dl
->minor
);
8715 test_list
= additional_test_list
;
8717 if (test_list
->disk
.major
== dl
->major
&&
8718 test_list
->disk
.minor
== dl
->minor
) {
8719 dprintf("%x:%x already in additional test list\n",
8720 dl
->major
, dl
->minor
);
8723 test_list
= test_list
->next
;
8728 /* skip in use or failed drives */
8729 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8731 dprintf("%x:%x status (failed: %d index: %d)\n",
8732 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8736 /* skip pure spares when we are looking for partially
8737 * assimilated drives
8739 if (dl
->index
== -1 && !activate_new
)
8742 if (!drive_validate_sector_size(super
, dl
))
8745 /* Does this unused device have the requisite free space?
8746 * It needs to be able to cover all member volumes
8748 ex
= get_extents(super
, dl
);
8750 dprintf("cannot get extents\n");
8753 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8754 dev
= get_imsm_dev(super
, i
);
8755 map
= get_imsm_map(dev
, MAP_0
);
8757 /* check if this disk is already a member of
8760 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8766 array_start
= pba_of_lba0(map
);
8767 array_end
= array_start
+
8768 per_dev_array_size(map
) - 1;
8771 /* check that we can start at pba_of_lba0 with
8772 * num_data_stripes*blocks_per_stripe of space
8774 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8778 pos
= ex
[j
].start
+ ex
[j
].size
;
8780 } while (ex
[j
-1].size
);
8787 if (i
< mpb
->num_raid_devs
) {
8788 dprintf("%x:%x does not have %u to %u available\n",
8789 dl
->major
, dl
->minor
, array_start
, array_end
);
8799 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8801 struct imsm_dev
*dev2
;
8802 struct imsm_map
*map
;
8808 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8810 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8811 if (state
== IMSM_T_STATE_FAILED
) {
8812 map
= get_imsm_map(dev2
, MAP_0
);
8815 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8817 * Check if failed disks are deleted from intel
8818 * disk list or are marked to be deleted
8820 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8821 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8823 * Do not rebuild the array if failed disks
8824 * from failed sub-array are not removed from
8828 is_failed(&idisk
->disk
) &&
8829 (idisk
->action
!= DISK_REMOVE
))
8837 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8838 struct metadata_update
**updates
)
8841 * Find a device with unused free space and use it to replace a
8842 * failed/vacant region in an array. We replace failed regions one a
8843 * array at a time. The result is that a new spare disk will be added
8844 * to the first failed array and after the monitor has finished
8845 * propagating failures the remainder will be consumed.
8847 * FIXME add a capability for mdmon to request spares from another
8851 struct intel_super
*super
= a
->container
->sb
;
8852 int inst
= a
->info
.container_member
;
8853 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8854 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8855 int failed
= a
->info
.array
.raid_disks
;
8856 struct mdinfo
*rv
= NULL
;
8859 struct metadata_update
*mu
;
8861 struct imsm_update_activate_spare
*u
;
8866 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8867 if ((d
->curr_state
& DS_FAULTY
) &&
8869 /* wait for Removal to happen */
8871 if (d
->state_fd
>= 0)
8875 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8876 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8878 if (imsm_reshape_blocks_arrays_changes(super
))
8881 /* Cannot activate another spare if rebuild is in progress already
8883 if (is_rebuilding(dev
)) {
8884 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8888 if (a
->info
.array
.level
== 4)
8889 /* No repair for takeovered array
8890 * imsm doesn't support raid4
8894 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8895 IMSM_T_STATE_DEGRADED
)
8898 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8899 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8904 * If there are any failed disks check state of the other volume.
8905 * Block rebuild if the another one is failed until failed disks
8906 * are removed from container.
8909 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8910 MAX_RAID_SERIAL_LEN
, dev
->volume
);
8911 /* check if states of the other volumes allow for rebuild */
8912 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
8914 allowed
= imsm_rebuild_allowed(a
->container
,
8922 /* For each slot, if it is not working, find a spare */
8923 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
8924 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8925 if (d
->disk
.raid_disk
== i
)
8927 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
8928 if (d
&& (d
->state_fd
>= 0))
8932 * OK, this device needs recovery. Try to re-add the
8933 * previous occupant of this slot, if this fails see if
8934 * we can continue the assimilation of a spare that was
8935 * partially assimilated, finally try to activate a new
8938 dl
= imsm_readd(super
, i
, a
);
8940 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
8942 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
8946 /* found a usable disk with enough space */
8947 di
= xcalloc(1, sizeof(*di
));
8949 /* dl->index will be -1 in the case we are activating a
8950 * pristine spare. imsm_process_update() will create a
8951 * new index in this case. Once a disk is found to be
8952 * failed in all member arrays it is kicked from the
8955 di
->disk
.number
= dl
->index
;
8957 /* (ab)use di->devs to store a pointer to the device
8960 di
->devs
= (struct mdinfo
*) dl
;
8962 di
->disk
.raid_disk
= i
;
8963 di
->disk
.major
= dl
->major
;
8964 di
->disk
.minor
= dl
->minor
;
8966 di
->recovery_start
= 0;
8967 di
->data_offset
= pba_of_lba0(map
);
8968 di
->component_size
= a
->info
.component_size
;
8969 di
->container_member
= inst
;
8970 di
->bb
.supported
= 1;
8971 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
8972 di
->ppl_sector
= get_ppl_sector(super
, inst
);
8973 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
8975 super
->random
= random32();
8979 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
8980 i
, di
->data_offset
);
8984 /* No spares found */
8986 /* Now 'rv' has a list of devices to return.
8987 * Create a metadata_update record to update the
8988 * disk_ord_tbl for the array
8990 mu
= xmalloc(sizeof(*mu
));
8991 mu
->buf
= xcalloc(num_spares
,
8992 sizeof(struct imsm_update_activate_spare
));
8994 mu
->space_list
= NULL
;
8995 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
8996 mu
->next
= *updates
;
8997 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
8999 for (di
= rv
; di
; di
= di
->next
) {
9000 u
->type
= update_activate_spare
;
9001 u
->dl
= (struct dl
*) di
->devs
;
9003 u
->slot
= di
->disk
.raid_disk
;
9014 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9016 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9017 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9018 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9019 struct disk_info
*inf
= get_disk_info(u
);
9020 struct imsm_disk
*disk
;
9024 for (i
= 0; i
< map
->num_members
; i
++) {
9025 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9026 for (j
= 0; j
< new_map
->num_members
; j
++)
9027 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9034 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9038 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9039 if (dl
->major
== major
&& dl
->minor
== minor
)
9044 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9050 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9051 if (dl
->major
== major
&& dl
->minor
== minor
) {
9054 prev
->next
= dl
->next
;
9056 super
->disks
= dl
->next
;
9058 __free_imsm_disk(dl
);
9059 dprintf("removed %x:%x\n", major
, minor
);
9067 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9069 static int add_remove_disk_update(struct intel_super
*super
)
9071 int check_degraded
= 0;
9074 /* add/remove some spares to/from the metadata/contrainer */
9075 while (super
->disk_mgmt_list
) {
9076 struct dl
*disk_cfg
;
9078 disk_cfg
= super
->disk_mgmt_list
;
9079 super
->disk_mgmt_list
= disk_cfg
->next
;
9080 disk_cfg
->next
= NULL
;
9082 if (disk_cfg
->action
== DISK_ADD
) {
9083 disk_cfg
->next
= super
->disks
;
9084 super
->disks
= disk_cfg
;
9086 dprintf("added %x:%x\n",
9087 disk_cfg
->major
, disk_cfg
->minor
);
9088 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9089 dprintf("Disk remove action processed: %x.%x\n",
9090 disk_cfg
->major
, disk_cfg
->minor
);
9091 disk
= get_disk_super(super
,
9095 /* store action status */
9096 disk
->action
= DISK_REMOVE
;
9097 /* remove spare disks only */
9098 if (disk
->index
== -1) {
9099 remove_disk_super(super
,
9104 /* release allocate disk structure */
9105 __free_imsm_disk(disk_cfg
);
9108 return check_degraded
;
9111 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9112 struct intel_super
*super
,
9115 struct intel_dev
*id
;
9116 void **tofree
= NULL
;
9119 dprintf("(enter)\n");
9120 if (u
->subdev
< 0 || u
->subdev
> 1) {
9121 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9124 if (space_list
== NULL
|| *space_list
== NULL
) {
9125 dprintf("imsm: Error: Memory is not allocated\n");
9129 for (id
= super
->devlist
; id
; id
= id
->next
) {
9130 if (id
->index
== (unsigned)u
->subdev
) {
9131 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9132 struct imsm_map
*map
;
9133 struct imsm_dev
*new_dev
=
9134 (struct imsm_dev
*)*space_list
;
9135 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9137 struct dl
*new_disk
;
9139 if (new_dev
== NULL
)
9141 *space_list
= **space_list
;
9142 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9143 map
= get_imsm_map(new_dev
, MAP_0
);
9145 dprintf("imsm: Error: migration in progress");
9149 to_state
= map
->map_state
;
9150 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9152 /* this should not happen */
9153 if (u
->new_disks
[0] < 0) {
9154 map
->failed_disk_num
=
9155 map
->num_members
- 1;
9156 to_state
= IMSM_T_STATE_DEGRADED
;
9158 to_state
= IMSM_T_STATE_NORMAL
;
9160 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9161 if (u
->new_level
> -1)
9162 map
->raid_level
= u
->new_level
;
9163 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9164 if ((u
->new_level
== 5) &&
9165 (migr_map
->raid_level
== 0)) {
9166 int ord
= map
->num_members
- 1;
9167 migr_map
->num_members
--;
9168 if (u
->new_disks
[0] < 0)
9169 ord
|= IMSM_ORD_REBUILD
;
9170 set_imsm_ord_tbl_ent(map
,
9171 map
->num_members
- 1,
9175 tofree
= (void **)dev
;
9177 /* update chunk size
9179 if (u
->new_chunksize
> 0) {
9180 unsigned long long num_data_stripes
;
9181 struct imsm_map
*dest_map
=
9182 get_imsm_map(dev
, MAP_0
);
9184 imsm_num_data_members(dest_map
);
9186 if (used_disks
== 0)
9189 map
->blocks_per_strip
=
9190 __cpu_to_le16(u
->new_chunksize
* 2);
9192 imsm_dev_size(dev
) / used_disks
;
9193 num_data_stripes
/= map
->blocks_per_strip
;
9194 num_data_stripes
/= map
->num_domains
;
9195 set_num_data_stripes(map
, num_data_stripes
);
9198 /* ensure blocks_per_member has valid value
9200 set_blocks_per_member(map
,
9201 per_dev_array_size(map
) +
9202 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9206 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9207 migr_map
->raid_level
== map
->raid_level
)
9210 if (u
->new_disks
[0] >= 0) {
9213 new_disk
= get_disk_super(super
,
9214 major(u
->new_disks
[0]),
9215 minor(u
->new_disks
[0]));
9216 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9217 major(u
->new_disks
[0]),
9218 minor(u
->new_disks
[0]),
9219 new_disk
, new_disk
->index
);
9220 if (new_disk
== NULL
)
9221 goto error_disk_add
;
9223 new_disk
->index
= map
->num_members
- 1;
9224 /* slot to fill in autolayout
9226 new_disk
->raiddisk
= new_disk
->index
;
9227 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9228 new_disk
->disk
.status
&= ~SPARE_DISK
;
9230 goto error_disk_add
;
9233 *tofree
= *space_list
;
9234 /* calculate new size
9236 imsm_set_array_size(new_dev
, -1);
9243 *space_list
= tofree
;
9247 dprintf("Error: imsm: Cannot find disk.\n");
9251 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9252 struct intel_super
*super
)
9254 struct intel_dev
*id
;
9257 dprintf("(enter)\n");
9258 if (u
->subdev
< 0 || u
->subdev
> 1) {
9259 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9263 for (id
= super
->devlist
; id
; id
= id
->next
) {
9264 if (id
->index
== (unsigned)u
->subdev
) {
9265 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9266 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9267 int used_disks
= imsm_num_data_members(map
);
9268 unsigned long long blocks_per_member
;
9269 unsigned long long num_data_stripes
;
9270 unsigned long long new_size_per_disk
;
9272 if (used_disks
== 0)
9275 /* calculate new size
9277 new_size_per_disk
= u
->new_size
/ used_disks
;
9278 blocks_per_member
= new_size_per_disk
+
9279 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9280 num_data_stripes
= new_size_per_disk
/
9281 map
->blocks_per_strip
;
9282 num_data_stripes
/= map
->num_domains
;
9283 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9284 u
->new_size
, new_size_per_disk
,
9286 set_blocks_per_member(map
, blocks_per_member
);
9287 set_num_data_stripes(map
, num_data_stripes
);
9288 imsm_set_array_size(dev
, u
->new_size
);
9298 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9299 struct intel_super
*super
,
9300 struct active_array
*active_array
)
9302 struct imsm_super
*mpb
= super
->anchor
;
9303 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9304 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9305 struct imsm_map
*migr_map
;
9306 struct active_array
*a
;
9307 struct imsm_disk
*disk
;
9314 int second_map_created
= 0;
9316 for (; u
; u
= u
->next
) {
9317 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9322 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9327 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9332 /* count failures (excluding rebuilds and the victim)
9333 * to determine map[0] state
9336 for (i
= 0; i
< map
->num_members
; i
++) {
9339 disk
= get_imsm_disk(super
,
9340 get_imsm_disk_idx(dev
, i
, MAP_X
));
9341 if (!disk
|| is_failed(disk
))
9345 /* adding a pristine spare, assign a new index */
9346 if (dl
->index
< 0) {
9347 dl
->index
= super
->anchor
->num_disks
;
9348 super
->anchor
->num_disks
++;
9351 disk
->status
|= CONFIGURED_DISK
;
9352 disk
->status
&= ~SPARE_DISK
;
9355 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9356 if (!second_map_created
) {
9357 second_map_created
= 1;
9358 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9359 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9361 map
->map_state
= to_state
;
9362 migr_map
= get_imsm_map(dev
, MAP_1
);
9363 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9364 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9365 dl
->index
| IMSM_ORD_REBUILD
);
9367 /* update the family_num to mark a new container
9368 * generation, being careful to record the existing
9369 * family_num in orig_family_num to clean up after
9370 * earlier mdadm versions that neglected to set it.
9372 if (mpb
->orig_family_num
== 0)
9373 mpb
->orig_family_num
= mpb
->family_num
;
9374 mpb
->family_num
+= super
->random
;
9376 /* count arrays using the victim in the metadata */
9378 for (a
= active_array
; a
; a
= a
->next
) {
9379 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9380 map
= get_imsm_map(dev
, MAP_0
);
9382 if (get_imsm_disk_slot(map
, victim
) >= 0)
9386 /* delete the victim if it is no longer being
9392 /* We know that 'manager' isn't touching anything,
9393 * so it is safe to delete
9395 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9396 if ((*dlp
)->index
== victim
)
9399 /* victim may be on the missing list */
9401 for (dlp
= &super
->missing
; *dlp
;
9402 dlp
= &(*dlp
)->next
)
9403 if ((*dlp
)->index
== victim
)
9405 imsm_delete(super
, dlp
, victim
);
9412 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9413 struct intel_super
*super
,
9416 struct dl
*new_disk
;
9417 struct intel_dev
*id
;
9419 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9420 int disk_count
= u
->old_raid_disks
;
9421 void **tofree
= NULL
;
9422 int devices_to_reshape
= 1;
9423 struct imsm_super
*mpb
= super
->anchor
;
9425 unsigned int dev_id
;
9427 dprintf("(enter)\n");
9429 /* enable spares to use in array */
9430 for (i
= 0; i
< delta_disks
; i
++) {
9431 new_disk
= get_disk_super(super
,
9432 major(u
->new_disks
[i
]),
9433 minor(u
->new_disks
[i
]));
9434 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9435 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9436 new_disk
, new_disk
->index
);
9437 if (new_disk
== NULL
||
9438 (new_disk
->index
>= 0 &&
9439 new_disk
->index
< u
->old_raid_disks
))
9440 goto update_reshape_exit
;
9441 new_disk
->index
= disk_count
++;
9442 /* slot to fill in autolayout
9444 new_disk
->raiddisk
= new_disk
->index
;
9445 new_disk
->disk
.status
|=
9447 new_disk
->disk
.status
&= ~SPARE_DISK
;
9450 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9451 mpb
->num_raid_devs
);
9452 /* manage changes in volume
9454 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9455 void **sp
= *space_list
;
9456 struct imsm_dev
*newdev
;
9457 struct imsm_map
*newmap
, *oldmap
;
9459 for (id
= super
->devlist
; id
; id
= id
->next
) {
9460 if (id
->index
== dev_id
)
9469 /* Copy the dev, but not (all of) the map */
9470 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9471 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9472 newmap
= get_imsm_map(newdev
, MAP_0
);
9473 /* Copy the current map */
9474 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9475 /* update one device only
9477 if (devices_to_reshape
) {
9478 dprintf("imsm: modifying subdev: %i\n",
9480 devices_to_reshape
--;
9481 newdev
->vol
.migr_state
= 1;
9482 newdev
->vol
.curr_migr_unit
= 0;
9483 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9484 newmap
->num_members
= u
->new_raid_disks
;
9485 for (i
= 0; i
< delta_disks
; i
++) {
9486 set_imsm_ord_tbl_ent(newmap
,
9487 u
->old_raid_disks
+ i
,
9488 u
->old_raid_disks
+ i
);
9490 /* New map is correct, now need to save old map
9492 newmap
= get_imsm_map(newdev
, MAP_1
);
9493 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9495 imsm_set_array_size(newdev
, -1);
9498 sp
= (void **)id
->dev
;
9503 /* Clear migration record */
9504 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9507 *space_list
= tofree
;
9510 update_reshape_exit
:
9515 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9516 struct intel_super
*super
,
9519 struct imsm_dev
*dev
= NULL
;
9520 struct intel_dev
*dv
;
9521 struct imsm_dev
*dev_new
;
9522 struct imsm_map
*map
;
9526 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9527 if (dv
->index
== (unsigned int)u
->subarray
) {
9535 map
= get_imsm_map(dev
, MAP_0
);
9537 if (u
->direction
== R10_TO_R0
) {
9538 unsigned long long num_data_stripes
;
9540 map
->num_domains
= 1;
9541 num_data_stripes
= imsm_dev_size(dev
) / 2;
9542 num_data_stripes
/= map
->blocks_per_strip
;
9543 num_data_stripes
/= map
->num_domains
;
9544 set_num_data_stripes(map
, num_data_stripes
);
9546 /* Number of failed disks must be half of initial disk number */
9547 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9548 (map
->num_members
/ 2))
9551 /* iterate through devices to mark removed disks as spare */
9552 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9553 if (dm
->disk
.status
& FAILED_DISK
) {
9554 int idx
= dm
->index
;
9555 /* update indexes on the disk list */
9556 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9557 the index values will end up being correct.... NB */
9558 for (du
= super
->disks
; du
; du
= du
->next
)
9559 if (du
->index
> idx
)
9561 /* mark as spare disk */
9566 map
->num_members
= map
->num_members
/ 2;
9567 map
->map_state
= IMSM_T_STATE_NORMAL
;
9568 map
->num_domains
= 1;
9569 map
->raid_level
= 0;
9570 map
->failed_disk_num
= -1;
9573 if (u
->direction
== R0_TO_R10
) {
9575 /* update slots in current disk list */
9576 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9580 /* create new *missing* disks */
9581 for (i
= 0; i
< map
->num_members
; i
++) {
9582 space
= *space_list
;
9585 *space_list
= *space
;
9587 memcpy(du
, super
->disks
, sizeof(*du
));
9591 du
->index
= (i
* 2) + 1;
9592 sprintf((char *)du
->disk
.serial
,
9593 " MISSING_%d", du
->index
);
9594 sprintf((char *)du
->serial
,
9595 "MISSING_%d", du
->index
);
9596 du
->next
= super
->missing
;
9597 super
->missing
= du
;
9599 /* create new dev and map */
9600 space
= *space_list
;
9603 *space_list
= *space
;
9604 dev_new
= (void *)space
;
9605 memcpy(dev_new
, dev
, sizeof(*dev
));
9606 /* update new map */
9607 map
= get_imsm_map(dev_new
, MAP_0
);
9608 map
->num_members
= map
->num_members
* 2;
9609 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9610 map
->num_domains
= 2;
9611 map
->raid_level
= 1;
9612 /* replace dev<->dev_new */
9615 /* update disk order table */
9616 for (du
= super
->disks
; du
; du
= du
->next
)
9618 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9619 for (du
= super
->missing
; du
; du
= du
->next
)
9620 if (du
->index
>= 0) {
9621 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9622 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9628 static void imsm_process_update(struct supertype
*st
,
9629 struct metadata_update
*update
)
9632 * crack open the metadata_update envelope to find the update record
9633 * update can be one of:
9634 * update_reshape_container_disks - all the arrays in the container
9635 * are being reshaped to have more devices. We need to mark
9636 * the arrays for general migration and convert selected spares
9637 * into active devices.
9638 * update_activate_spare - a spare device has replaced a failed
9639 * device in an array, update the disk_ord_tbl. If this disk is
9640 * present in all member arrays then also clear the SPARE_DISK
9642 * update_create_array
9644 * update_rename_array
9645 * update_add_remove_disk
9647 struct intel_super
*super
= st
->sb
;
9648 struct imsm_super
*mpb
;
9649 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9651 /* update requires a larger buf but the allocation failed */
9652 if (super
->next_len
&& !super
->next_buf
) {
9653 super
->next_len
= 0;
9657 if (super
->next_buf
) {
9658 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9660 super
->len
= super
->next_len
;
9661 super
->buf
= super
->next_buf
;
9663 super
->next_len
= 0;
9664 super
->next_buf
= NULL
;
9667 mpb
= super
->anchor
;
9670 case update_general_migration_checkpoint
: {
9671 struct intel_dev
*id
;
9672 struct imsm_update_general_migration_checkpoint
*u
=
9673 (void *)update
->buf
;
9675 dprintf("called for update_general_migration_checkpoint\n");
9677 /* find device under general migration */
9678 for (id
= super
->devlist
; id
; id
= id
->next
) {
9679 if (is_gen_migration(id
->dev
)) {
9680 id
->dev
->vol
.curr_migr_unit
=
9681 __cpu_to_le32(u
->curr_migr_unit
);
9682 super
->updates_pending
++;
9687 case update_takeover
: {
9688 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9689 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9690 imsm_update_version_info(super
);
9691 super
->updates_pending
++;
9696 case update_reshape_container_disks
: {
9697 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9698 if (apply_reshape_container_disks_update(
9699 u
, super
, &update
->space_list
))
9700 super
->updates_pending
++;
9703 case update_reshape_migration
: {
9704 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9705 if (apply_reshape_migration_update(
9706 u
, super
, &update
->space_list
))
9707 super
->updates_pending
++;
9710 case update_size_change
: {
9711 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9712 if (apply_size_change_update(u
, super
))
9713 super
->updates_pending
++;
9716 case update_activate_spare
: {
9717 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9718 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9719 super
->updates_pending
++;
9722 case update_create_array
: {
9723 /* someone wants to create a new array, we need to be aware of
9724 * a few races/collisions:
9725 * 1/ 'Create' called by two separate instances of mdadm
9726 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9727 * devices that have since been assimilated via
9729 * In the event this update can not be carried out mdadm will
9730 * (FIX ME) notice that its update did not take hold.
9732 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9733 struct intel_dev
*dv
;
9734 struct imsm_dev
*dev
;
9735 struct imsm_map
*map
, *new_map
;
9736 unsigned long long start
, end
;
9737 unsigned long long new_start
, new_end
;
9739 struct disk_info
*inf
;
9742 /* handle racing creates: first come first serve */
9743 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9744 dprintf("subarray %d already defined\n", u
->dev_idx
);
9748 /* check update is next in sequence */
9749 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9750 dprintf("can not create array %d expected index %d\n",
9751 u
->dev_idx
, mpb
->num_raid_devs
);
9755 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9756 new_start
= pba_of_lba0(new_map
);
9757 new_end
= new_start
+ per_dev_array_size(new_map
);
9758 inf
= get_disk_info(u
);
9760 /* handle activate_spare versus create race:
9761 * check to make sure that overlapping arrays do not include
9764 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9765 dev
= get_imsm_dev(super
, i
);
9766 map
= get_imsm_map(dev
, MAP_0
);
9767 start
= pba_of_lba0(map
);
9768 end
= start
+ per_dev_array_size(map
);
9769 if ((new_start
>= start
&& new_start
<= end
) ||
9770 (start
>= new_start
&& start
<= new_end
))
9775 if (disks_overlap(super
, i
, u
)) {
9776 dprintf("arrays overlap\n");
9781 /* check that prepare update was successful */
9782 if (!update
->space
) {
9783 dprintf("prepare update failed\n");
9787 /* check that all disks are still active before committing
9788 * changes. FIXME: could we instead handle this by creating a
9789 * degraded array? That's probably not what the user expects,
9790 * so better to drop this update on the floor.
9792 for (i
= 0; i
< new_map
->num_members
; i
++) {
9793 dl
= serial_to_dl(inf
[i
].serial
, super
);
9795 dprintf("disk disappeared\n");
9800 super
->updates_pending
++;
9802 /* convert spares to members and fixup ord_tbl */
9803 for (i
= 0; i
< new_map
->num_members
; i
++) {
9804 dl
= serial_to_dl(inf
[i
].serial
, super
);
9805 if (dl
->index
== -1) {
9806 dl
->index
= mpb
->num_disks
;
9808 dl
->disk
.status
|= CONFIGURED_DISK
;
9809 dl
->disk
.status
&= ~SPARE_DISK
;
9811 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9816 update
->space
= NULL
;
9817 imsm_copy_dev(dev
, &u
->dev
);
9818 dv
->index
= u
->dev_idx
;
9819 dv
->next
= super
->devlist
;
9820 super
->devlist
= dv
;
9821 mpb
->num_raid_devs
++;
9823 imsm_update_version_info(super
);
9826 /* mdmon knows how to release update->space, but not
9827 * ((struct intel_dev *) update->space)->dev
9829 if (update
->space
) {
9835 case update_kill_array
: {
9836 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9837 int victim
= u
->dev_idx
;
9838 struct active_array
*a
;
9839 struct intel_dev
**dp
;
9840 struct imsm_dev
*dev
;
9842 /* sanity check that we are not affecting the uuid of
9843 * active arrays, or deleting an active array
9845 * FIXME when immutable ids are available, but note that
9846 * we'll also need to fixup the invalidated/active
9847 * subarray indexes in mdstat
9849 for (a
= st
->arrays
; a
; a
= a
->next
)
9850 if (a
->info
.container_member
>= victim
)
9852 /* by definition if mdmon is running at least one array
9853 * is active in the container, so checking
9854 * mpb->num_raid_devs is just extra paranoia
9856 dev
= get_imsm_dev(super
, victim
);
9857 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9858 dprintf("failed to delete subarray-%d\n", victim
);
9862 for (dp
= &super
->devlist
; *dp
;)
9863 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9866 if ((*dp
)->index
> (unsigned)victim
)
9870 mpb
->num_raid_devs
--;
9871 super
->updates_pending
++;
9874 case update_rename_array
: {
9875 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9876 char name
[MAX_RAID_SERIAL_LEN
+1];
9877 int target
= u
->dev_idx
;
9878 struct active_array
*a
;
9879 struct imsm_dev
*dev
;
9881 /* sanity check that we are not affecting the uuid of
9884 memset(name
, 0, sizeof(name
));
9885 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9886 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9887 for (a
= st
->arrays
; a
; a
= a
->next
)
9888 if (a
->info
.container_member
== target
)
9890 dev
= get_imsm_dev(super
, u
->dev_idx
);
9891 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9892 dprintf("failed to rename subarray-%d\n", target
);
9896 memcpy(dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
9897 super
->updates_pending
++;
9900 case update_add_remove_disk
: {
9901 /* we may be able to repair some arrays if disks are
9902 * being added, check the status of add_remove_disk
9903 * if discs has been added.
9905 if (add_remove_disk_update(super
)) {
9906 struct active_array
*a
;
9908 super
->updates_pending
++;
9909 for (a
= st
->arrays
; a
; a
= a
->next
)
9910 a
->check_degraded
= 1;
9914 case update_prealloc_badblocks_mem
:
9916 case update_rwh_policy
: {
9917 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
9918 int target
= u
->dev_idx
;
9919 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
9921 dprintf("could not find subarray-%d\n", target
);
9925 if (dev
->rwh_policy
!= u
->new_policy
) {
9926 dev
->rwh_policy
= u
->new_policy
;
9927 super
->updates_pending
++;
9932 pr_err("error: unsuported process update type:(type: %d)\n", type
);
9936 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
9938 static int imsm_prepare_update(struct supertype
*st
,
9939 struct metadata_update
*update
)
9942 * Allocate space to hold new disk entries, raid-device entries or a new
9943 * mpb if necessary. The manager synchronously waits for updates to
9944 * complete in the monitor, so new mpb buffers allocated here can be
9945 * integrated by the monitor thread without worrying about live pointers
9946 * in the manager thread.
9948 enum imsm_update_type type
;
9949 struct intel_super
*super
= st
->sb
;
9950 unsigned int sector_size
= super
->sector_size
;
9951 struct imsm_super
*mpb
= super
->anchor
;
9955 if (update
->len
< (int)sizeof(type
))
9958 type
= *(enum imsm_update_type
*) update
->buf
;
9961 case update_general_migration_checkpoint
:
9962 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
9964 dprintf("called for update_general_migration_checkpoint\n");
9966 case update_takeover
: {
9967 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9968 if (update
->len
< (int)sizeof(*u
))
9970 if (u
->direction
== R0_TO_R10
) {
9971 void **tail
= (void **)&update
->space_list
;
9972 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
9973 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9974 int num_members
= map
->num_members
;
9977 /* allocate memory for added disks */
9978 for (i
= 0; i
< num_members
; i
++) {
9979 size
= sizeof(struct dl
);
9980 space
= xmalloc(size
);
9985 /* allocate memory for new device */
9986 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
9987 (num_members
* sizeof(__u32
));
9988 space
= xmalloc(size
);
9992 len
= disks_to_mpb_size(num_members
* 2);
9997 case update_reshape_container_disks
: {
9998 /* Every raid device in the container is about to
9999 * gain some more devices, and we will enter a
10001 * So each 'imsm_map' will be bigger, and the imsm_vol
10002 * will now hold 2 of them.
10003 * Thus we need new 'struct imsm_dev' allocations sized
10004 * as sizeof_imsm_dev but with more devices in both maps.
10006 struct imsm_update_reshape
*u
= (void *)update
->buf
;
10007 struct intel_dev
*dl
;
10008 void **space_tail
= (void**)&update
->space_list
;
10010 if (update
->len
< (int)sizeof(*u
))
10013 dprintf("for update_reshape\n");
10015 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
10016 int size
= sizeof_imsm_dev(dl
->dev
, 1);
10018 if (u
->new_raid_disks
> u
->old_raid_disks
)
10019 size
+= sizeof(__u32
)*2*
10020 (u
->new_raid_disks
- u
->old_raid_disks
);
10024 *space_tail
= NULL
;
10027 len
= disks_to_mpb_size(u
->new_raid_disks
);
10028 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10031 case update_reshape_migration
: {
10032 /* for migration level 0->5 we need to add disks
10033 * so the same as for container operation we will copy
10034 * device to the bigger location.
10035 * in memory prepared device and new disk area are prepared
10036 * for usage in process update
10038 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
10039 struct intel_dev
*id
;
10040 void **space_tail
= (void **)&update
->space_list
;
10043 int current_level
= -1;
10045 if (update
->len
< (int)sizeof(*u
))
10048 dprintf("for update_reshape\n");
10050 /* add space for bigger array in update
10052 for (id
= super
->devlist
; id
; id
= id
->next
) {
10053 if (id
->index
== (unsigned)u
->subdev
) {
10054 size
= sizeof_imsm_dev(id
->dev
, 1);
10055 if (u
->new_raid_disks
> u
->old_raid_disks
)
10056 size
+= sizeof(__u32
)*2*
10057 (u
->new_raid_disks
- u
->old_raid_disks
);
10061 *space_tail
= NULL
;
10065 if (update
->space_list
== NULL
)
10068 /* add space for disk in update
10070 size
= sizeof(struct dl
);
10074 *space_tail
= NULL
;
10076 /* add spare device to update
10078 for (id
= super
->devlist
; id
; id
= id
->next
)
10079 if (id
->index
== (unsigned)u
->subdev
) {
10080 struct imsm_dev
*dev
;
10081 struct imsm_map
*map
;
10083 dev
= get_imsm_dev(super
, u
->subdev
);
10084 map
= get_imsm_map(dev
, MAP_0
);
10085 current_level
= map
->raid_level
;
10088 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
10089 struct mdinfo
*spares
;
10091 spares
= get_spares_for_grow(st
);
10094 struct mdinfo
*dev
;
10096 dev
= spares
->devs
;
10099 makedev(dev
->disk
.major
,
10101 dl
= get_disk_super(super
,
10104 dl
->index
= u
->old_raid_disks
;
10107 sysfs_free(spares
);
10110 len
= disks_to_mpb_size(u
->new_raid_disks
);
10111 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
10114 case update_size_change
: {
10115 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
10119 case update_activate_spare
: {
10120 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
10124 case update_create_array
: {
10125 struct imsm_update_create_array
*u
= (void *) update
->buf
;
10126 struct intel_dev
*dv
;
10127 struct imsm_dev
*dev
= &u
->dev
;
10128 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10130 struct disk_info
*inf
;
10134 if (update
->len
< (int)sizeof(*u
))
10137 inf
= get_disk_info(u
);
10138 len
= sizeof_imsm_dev(dev
, 1);
10139 /* allocate a new super->devlist entry */
10140 dv
= xmalloc(sizeof(*dv
));
10141 dv
->dev
= xmalloc(len
);
10142 update
->space
= dv
;
10144 /* count how many spares will be converted to members */
10145 for (i
= 0; i
< map
->num_members
; i
++) {
10146 dl
= serial_to_dl(inf
[i
].serial
, super
);
10148 /* hmm maybe it failed?, nothing we can do about
10153 if (count_memberships(dl
, super
) == 0)
10156 len
+= activate
* sizeof(struct imsm_disk
);
10159 case update_kill_array
: {
10160 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
10164 case update_rename_array
: {
10165 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
10169 case update_add_remove_disk
:
10170 /* no update->len needed */
10172 case update_prealloc_badblocks_mem
:
10173 super
->extra_space
+= sizeof(struct bbm_log
) -
10174 get_imsm_bbm_log_size(super
->bbm_log
);
10176 case update_rwh_policy
: {
10177 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
10185 /* check if we need a larger metadata buffer */
10186 if (super
->next_buf
)
10187 buf_len
= super
->next_len
;
10189 buf_len
= super
->len
;
10191 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
10192 /* ok we need a larger buf than what is currently allocated
10193 * if this allocation fails process_update will notice that
10194 * ->next_len is set and ->next_buf is NULL
10196 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
10197 super
->extra_space
+ len
, sector_size
);
10198 if (super
->next_buf
)
10199 free(super
->next_buf
);
10201 super
->next_len
= buf_len
;
10202 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
10203 memset(super
->next_buf
, 0, buf_len
);
10205 super
->next_buf
= NULL
;
10210 /* must be called while manager is quiesced */
10211 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
10213 struct imsm_super
*mpb
= super
->anchor
;
10215 struct imsm_dev
*dev
;
10216 struct imsm_map
*map
;
10217 unsigned int i
, j
, num_members
;
10218 __u32 ord
, ord_map0
;
10219 struct bbm_log
*log
= super
->bbm_log
;
10221 dprintf("deleting device[%d] from imsm_super\n", index
);
10223 /* shift all indexes down one */
10224 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10225 if (iter
->index
> (int)index
)
10227 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10228 if (iter
->index
> (int)index
)
10231 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10232 dev
= get_imsm_dev(super
, i
);
10233 map
= get_imsm_map(dev
, MAP_0
);
10234 num_members
= map
->num_members
;
10235 for (j
= 0; j
< num_members
; j
++) {
10236 /* update ord entries being careful not to propagate
10237 * ord-flags to the first map
10239 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10240 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10242 if (ord_to_idx(ord
) <= index
)
10245 map
= get_imsm_map(dev
, MAP_0
);
10246 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10247 map
= get_imsm_map(dev
, MAP_1
);
10249 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10253 for (i
= 0; i
< log
->entry_count
; i
++) {
10254 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10256 if (entry
->disk_ordinal
<= index
)
10258 entry
->disk_ordinal
--;
10262 super
->updates_pending
++;
10264 struct dl
*dl
= *dlp
;
10266 *dlp
= (*dlp
)->next
;
10267 __free_imsm_disk(dl
);
10271 static void close_targets(int *targets
, int new_disks
)
10278 for (i
= 0; i
< new_disks
; i
++) {
10279 if (targets
[i
] >= 0) {
10286 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10287 struct intel_super
*super
,
10288 struct imsm_dev
*dev
)
10294 struct imsm_map
*map
;
10297 ret_val
= raid_disks
/2;
10298 /* check map if all disks pairs not failed
10301 map
= get_imsm_map(dev
, MAP_0
);
10302 for (i
= 0; i
< ret_val
; i
++) {
10303 int degradation
= 0;
10304 if (get_imsm_disk(super
, i
) == NULL
)
10306 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10308 if (degradation
== 2)
10311 map
= get_imsm_map(dev
, MAP_1
);
10312 /* if there is no second map
10313 * result can be returned
10317 /* check degradation in second map
10319 for (i
= 0; i
< ret_val
; i
++) {
10320 int degradation
= 0;
10321 if (get_imsm_disk(super
, i
) == NULL
)
10323 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10325 if (degradation
== 2)
10339 /*******************************************************************************
10340 * Function: open_backup_targets
10341 * Description: Function opens file descriptors for all devices given in
10344 * info : general array info
10345 * raid_disks : number of disks
10346 * raid_fds : table of device's file descriptors
10347 * super : intel super for raid10 degradation check
10348 * dev : intel device for raid10 degradation check
10352 ******************************************************************************/
10353 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10354 struct intel_super
*super
, struct imsm_dev
*dev
)
10360 for (i
= 0; i
< raid_disks
; i
++)
10363 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10366 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10367 dprintf("disk is faulty!!\n");
10371 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10374 dn
= map_dev(sd
->disk
.major
,
10375 sd
->disk
.minor
, 1);
10376 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10377 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10378 pr_err("cannot open component\n");
10383 /* check if maximum array degradation level is not exceeded
10385 if ((raid_disks
- opened
) >
10386 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10388 pr_err("Not enough disks can be opened.\n");
10389 close_targets(raid_fds
, raid_disks
);
10395 /*******************************************************************************
10396 * Function: validate_container_imsm
10397 * Description: This routine validates container after assemble,
10398 * eg. if devices in container are under the same controller.
10401 * info : linked list with info about devices used in array
10405 ******************************************************************************/
10406 int validate_container_imsm(struct mdinfo
*info
)
10408 if (check_env("IMSM_NO_PLATFORM"))
10411 struct sys_dev
*idev
;
10412 struct sys_dev
*hba
= NULL
;
10413 struct sys_dev
*intel_devices
= find_intel_devices();
10414 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10415 info
->disk
.minor
));
10417 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10418 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10427 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10428 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10432 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10433 struct mdinfo
*dev
;
10435 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10436 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10438 struct sys_dev
*hba2
= NULL
;
10439 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10440 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10448 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10449 get_orom_by_device_id(hba2
->dev_id
);
10451 if (hba2
&& hba
->type
!= hba2
->type
) {
10452 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10453 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10457 if (orom
!= orom2
) {
10458 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10459 " This operation is not supported and can lead to data loss.\n");
10464 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10465 " This operation is not supported and can lead to data loss.\n");
10473 /*******************************************************************************
10474 * Function: imsm_record_badblock
10475 * Description: This routine stores new bad block record in BBM log
10478 * a : array containing a bad block
10479 * slot : disk number containing a bad block
10480 * sector : bad block sector
10481 * length : bad block sectors range
10485 ******************************************************************************/
10486 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10487 unsigned long long sector
, int length
)
10489 struct intel_super
*super
= a
->container
->sb
;
10493 ord
= imsm_disk_slot_to_ord(a
, slot
);
10497 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10500 super
->updates_pending
++;
10504 /*******************************************************************************
10505 * Function: imsm_clear_badblock
10506 * Description: This routine clears bad block record from BBM log
10509 * a : array containing a bad block
10510 * slot : disk number containing a bad block
10511 * sector : bad block sector
10512 * length : bad block sectors range
10516 ******************************************************************************/
10517 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10518 unsigned long long sector
, int length
)
10520 struct intel_super
*super
= a
->container
->sb
;
10524 ord
= imsm_disk_slot_to_ord(a
, slot
);
10528 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10530 super
->updates_pending
++;
10534 /*******************************************************************************
10535 * Function: imsm_get_badblocks
10536 * Description: This routine get list of bad blocks for an array
10540 * slot : disk number
10542 * bb : structure containing bad blocks
10544 ******************************************************************************/
10545 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10547 int inst
= a
->info
.container_member
;
10548 struct intel_super
*super
= a
->container
->sb
;
10549 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10550 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10553 ord
= imsm_disk_slot_to_ord(a
, slot
);
10557 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10558 per_dev_array_size(map
), &super
->bb
);
10562 /*******************************************************************************
10563 * Function: examine_badblocks_imsm
10564 * Description: Prints list of bad blocks on a disk to the standard output
10567 * st : metadata handler
10568 * fd : open file descriptor for device
10569 * devname : device name
10573 ******************************************************************************/
10574 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10576 struct intel_super
*super
= st
->sb
;
10577 struct bbm_log
*log
= super
->bbm_log
;
10578 struct dl
*d
= NULL
;
10581 for (d
= super
->disks
; d
; d
= d
->next
) {
10582 if (strcmp(d
->devname
, devname
) == 0)
10586 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10587 pr_err("%s doesn't appear to be part of a raid array\n",
10594 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10596 for (i
= 0; i
< log
->entry_count
; i
++) {
10597 if (entry
[i
].disk_ordinal
== d
->index
) {
10598 unsigned long long sector
= __le48_to_cpu(
10599 &entry
[i
].defective_block_start
);
10600 int cnt
= entry
[i
].marked_count
+ 1;
10603 printf("Bad-blocks on %s:\n", devname
);
10607 printf("%20llu for %d sectors\n", sector
, cnt
);
10613 printf("No bad-blocks list configured on %s\n", devname
);
10617 /*******************************************************************************
10618 * Function: init_migr_record_imsm
10619 * Description: Function inits imsm migration record
10621 * super : imsm internal array info
10622 * dev : device under migration
10623 * info : general array info to find the smallest device
10626 ******************************************************************************/
10627 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10628 struct mdinfo
*info
)
10630 struct intel_super
*super
= st
->sb
;
10631 struct migr_record
*migr_rec
= super
->migr_rec
;
10632 int new_data_disks
;
10633 unsigned long long dsize
, dev_sectors
;
10634 long long unsigned min_dev_sectors
= -1LLU;
10638 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10639 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10640 unsigned long long num_migr_units
;
10641 unsigned long long array_blocks
;
10643 memset(migr_rec
, 0, sizeof(struct migr_record
));
10644 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10646 /* only ascending reshape supported now */
10647 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10649 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10650 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10651 migr_rec
->dest_depth_per_unit
*=
10652 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10653 new_data_disks
= imsm_num_data_members(map_dest
);
10654 migr_rec
->blocks_per_unit
=
10655 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10656 migr_rec
->dest_depth_per_unit
=
10657 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10658 array_blocks
= info
->component_size
* new_data_disks
;
10660 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10662 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10664 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
10666 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10667 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10669 /* Find the smallest dev */
10670 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10671 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10672 fd
= dev_open(nm
, O_RDONLY
);
10675 get_dev_size(fd
, NULL
, &dsize
);
10676 dev_sectors
= dsize
/ 512;
10677 if (dev_sectors
< min_dev_sectors
)
10678 min_dev_sectors
= dev_sectors
;
10681 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
10682 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10684 write_imsm_migr_rec(st
);
10689 /*******************************************************************************
10690 * Function: save_backup_imsm
10691 * Description: Function saves critical data stripes to Migration Copy Area
10692 * and updates the current migration unit status.
10693 * Use restore_stripes() to form a destination stripe,
10694 * and to write it to the Copy Area.
10696 * st : supertype information
10697 * dev : imsm device that backup is saved for
10698 * info : general array info
10699 * buf : input buffer
10700 * length : length of data to backup (blocks_per_unit)
10704 ******************************************************************************/
10705 int save_backup_imsm(struct supertype
*st
,
10706 struct imsm_dev
*dev
,
10707 struct mdinfo
*info
,
10712 struct intel_super
*super
= st
->sb
;
10713 unsigned long long *target_offsets
;
10716 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10717 int new_disks
= map_dest
->num_members
;
10718 int dest_layout
= 0;
10720 unsigned long long start
;
10721 int data_disks
= imsm_num_data_members(map_dest
);
10723 targets
= xmalloc(new_disks
* sizeof(int));
10725 for (i
= 0; i
< new_disks
; i
++)
10728 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10730 start
= info
->reshape_progress
* 512;
10731 for (i
= 0; i
< new_disks
; i
++) {
10732 target_offsets
[i
] = (unsigned long long)
10733 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
10734 /* move back copy area adderss, it will be moved forward
10735 * in restore_stripes() using start input variable
10737 target_offsets
[i
] -= start
/data_disks
;
10740 if (open_backup_targets(info
, new_disks
, targets
,
10744 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10745 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10747 if (restore_stripes(targets
, /* list of dest devices */
10748 target_offsets
, /* migration record offsets */
10751 map_dest
->raid_level
,
10753 -1, /* source backup file descriptor */
10754 0, /* input buf offset
10755 * always 0 buf is already offseted */
10759 pr_err("Error restoring stripes\n");
10767 close_targets(targets
, new_disks
);
10770 free(target_offsets
);
10775 /*******************************************************************************
10776 * Function: save_checkpoint_imsm
10777 * Description: Function called for current unit status update
10778 * in the migration record. It writes it to disk.
10780 * super : imsm internal array info
10781 * info : general array info
10785 * 2: failure, means no valid migration record
10786 * / no general migration in progress /
10787 ******************************************************************************/
10788 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10790 struct intel_super
*super
= st
->sb
;
10791 unsigned long long blocks_per_unit
;
10792 unsigned long long curr_migr_unit
;
10794 if (load_imsm_migr_rec(super
, info
) != 0) {
10795 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10799 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10800 if (blocks_per_unit
== 0) {
10801 dprintf("imsm: no migration in progress.\n");
10804 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10805 /* check if array is alligned to copy area
10806 * if it is not alligned, add one to current migration unit value
10807 * this can happend on array reshape finish only
10809 if (info
->reshape_progress
% blocks_per_unit
)
10812 super
->migr_rec
->curr_migr_unit
=
10813 __cpu_to_le32(curr_migr_unit
);
10814 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10815 super
->migr_rec
->dest_1st_member_lba
=
10816 __cpu_to_le32(curr_migr_unit
*
10817 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
10818 if (write_imsm_migr_rec(st
) < 0) {
10819 dprintf("imsm: Cannot write migration record outside backup area\n");
10826 /*******************************************************************************
10827 * Function: recover_backup_imsm
10828 * Description: Function recovers critical data from the Migration Copy Area
10829 * while assembling an array.
10831 * super : imsm internal array info
10832 * info : general array info
10834 * 0 : success (or there is no data to recover)
10836 ******************************************************************************/
10837 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10839 struct intel_super
*super
= st
->sb
;
10840 struct migr_record
*migr_rec
= super
->migr_rec
;
10841 struct imsm_map
*map_dest
;
10842 struct intel_dev
*id
= NULL
;
10843 unsigned long long read_offset
;
10844 unsigned long long write_offset
;
10846 int *targets
= NULL
;
10847 int new_disks
, i
, err
;
10850 unsigned int sector_size
= super
->sector_size
;
10851 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
10852 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
10854 int skipped_disks
= 0;
10856 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10860 /* recover data only during assemblation */
10861 if (strncmp(buffer
, "inactive", 8) != 0)
10863 /* no data to recover */
10864 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10866 if (curr_migr_unit
>= num_migr_units
)
10869 /* find device during reshape */
10870 for (id
= super
->devlist
; id
; id
= id
->next
)
10871 if (is_gen_migration(id
->dev
))
10876 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10877 new_disks
= map_dest
->num_members
;
10879 read_offset
= (unsigned long long)
10880 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
10882 write_offset
= ((unsigned long long)
10883 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
10884 pba_of_lba0(map_dest
)) * 512;
10886 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10887 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10889 targets
= xcalloc(new_disks
, sizeof(int));
10891 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10892 pr_err("Cannot open some devices belonging to array.\n");
10896 for (i
= 0; i
< new_disks
; i
++) {
10897 if (targets
[i
] < 0) {
10901 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
10902 pr_err("Cannot seek to block: %s\n",
10907 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
10908 pr_err("Cannot read copy area block: %s\n",
10913 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
10914 pr_err("Cannot seek to block: %s\n",
10919 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
10920 pr_err("Cannot restore block: %s\n",
10927 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
10931 pr_err("Cannot restore data from backup. Too many failed disks\n");
10935 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
10936 /* ignore error == 2, this can mean end of reshape here
10938 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
10944 for (i
= 0; i
< new_disks
; i
++)
10953 static char disk_by_path
[] = "/dev/disk/by-path/";
10955 static const char *imsm_get_disk_controller_domain(const char *path
)
10957 char disk_path
[PATH_MAX
];
10961 strcpy(disk_path
, disk_by_path
);
10962 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
10963 if (stat(disk_path
, &st
) == 0) {
10964 struct sys_dev
* hba
;
10967 path
= devt_to_devpath(st
.st_rdev
);
10970 hba
= find_disk_attached_hba(-1, path
);
10971 if (hba
&& hba
->type
== SYS_DEV_SAS
)
10973 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
10975 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
10977 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
10981 dprintf("path: %s hba: %s attached: %s\n",
10982 path
, (hba
) ? hba
->path
: "NULL", drv
);
10988 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
10990 static char devnm
[32];
10991 char subdev_name
[20];
10992 struct mdstat_ent
*mdstat
;
10994 sprintf(subdev_name
, "%d", subdev
);
10995 mdstat
= mdstat_by_subdev(subdev_name
, container
);
10999 strcpy(devnm
, mdstat
->devnm
);
11000 free_mdstat(mdstat
);
11004 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
11005 struct geo_params
*geo
,
11006 int *old_raid_disks
,
11009 /* currently we only support increasing the number of devices
11010 * for a container. This increases the number of device for each
11011 * member array. They must all be RAID0 or RAID5.
11014 struct mdinfo
*info
, *member
;
11015 int devices_that_can_grow
= 0;
11017 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
11019 if (geo
->size
> 0 ||
11020 geo
->level
!= UnSet
||
11021 geo
->layout
!= UnSet
||
11022 geo
->chunksize
!= 0 ||
11023 geo
->raid_disks
== UnSet
) {
11024 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
11028 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11029 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
11033 info
= container_content_imsm(st
, NULL
);
11034 for (member
= info
; member
; member
= member
->next
) {
11037 dprintf("imsm: checking device_num: %i\n",
11038 member
->container_member
);
11040 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
11041 /* we work on container for Online Capacity Expansion
11042 * only so raid_disks has to grow
11044 dprintf("imsm: for container operation raid disks increase is required\n");
11048 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
11049 /* we cannot use this container with other raid level
11051 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
11052 info
->array
.level
);
11055 /* check for platform support
11056 * for this raid level configuration
11058 struct intel_super
*super
= st
->sb
;
11059 if (!is_raid_level_supported(super
->orom
,
11060 member
->array
.level
,
11061 geo
->raid_disks
)) {
11062 dprintf("platform does not support raid%d with %d disk%s\n",
11065 geo
->raid_disks
> 1 ? "s" : "");
11068 /* check if component size is aligned to chunk size
11070 if (info
->component_size
%
11071 (info
->array
.chunk_size
/512)) {
11072 dprintf("Component size is not aligned to chunk size\n");
11077 if (*old_raid_disks
&&
11078 info
->array
.raid_disks
!= *old_raid_disks
)
11080 *old_raid_disks
= info
->array
.raid_disks
;
11082 /* All raid5 and raid0 volumes in container
11083 * have to be ready for Online Capacity Expansion
11084 * so they need to be assembled. We have already
11085 * checked that no recovery etc is happening.
11087 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
11088 st
->container_devnm
);
11089 if (result
== NULL
) {
11090 dprintf("imsm: cannot find array\n");
11093 devices_that_can_grow
++;
11096 if (!member
&& devices_that_can_grow
)
11100 dprintf("Container operation allowed\n");
11102 dprintf("Error: %i\n", ret_val
);
11107 /* Function: get_spares_for_grow
11108 * Description: Allocates memory and creates list of spare devices
11109 * avaliable in container. Checks if spare drive size is acceptable.
11110 * Parameters: Pointer to the supertype structure
11111 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
11114 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
11116 struct spare_criteria sc
;
11118 get_spare_criteria_imsm(st
, &sc
);
11119 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
11122 /******************************************************************************
11123 * function: imsm_create_metadata_update_for_reshape
11124 * Function creates update for whole IMSM container.
11126 ******************************************************************************/
11127 static int imsm_create_metadata_update_for_reshape(
11128 struct supertype
*st
,
11129 struct geo_params
*geo
,
11130 int old_raid_disks
,
11131 struct imsm_update_reshape
**updatep
)
11133 struct intel_super
*super
= st
->sb
;
11134 struct imsm_super
*mpb
= super
->anchor
;
11135 int update_memory_size
;
11136 struct imsm_update_reshape
*u
;
11137 struct mdinfo
*spares
;
11140 struct mdinfo
*dev
;
11142 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
11144 delta_disks
= geo
->raid_disks
- old_raid_disks
;
11146 /* size of all update data without anchor */
11147 update_memory_size
= sizeof(struct imsm_update_reshape
);
11149 /* now add space for spare disks that we need to add. */
11150 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
11152 u
= xcalloc(1, update_memory_size
);
11153 u
->type
= update_reshape_container_disks
;
11154 u
->old_raid_disks
= old_raid_disks
;
11155 u
->new_raid_disks
= geo
->raid_disks
;
11157 /* now get spare disks list
11159 spares
= get_spares_for_grow(st
);
11161 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
11162 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
11167 /* we have got spares
11168 * update disk list in imsm_disk list table in anchor
11170 dprintf("imsm: %i spares are available.\n\n",
11171 spares
->array
.spare_disks
);
11173 dev
= spares
->devs
;
11174 for (i
= 0; i
< delta_disks
; i
++) {
11179 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
11181 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
11182 dl
->index
= mpb
->num_disks
;
11190 sysfs_free(spares
);
11192 dprintf("imsm: reshape update preparation :");
11193 if (i
== delta_disks
) {
11194 dprintf_cont(" OK\n");
11196 return update_memory_size
;
11199 dprintf_cont(" Error\n");
11204 /******************************************************************************
11205 * function: imsm_create_metadata_update_for_size_change()
11206 * Creates update for IMSM array for array size change.
11208 ******************************************************************************/
11209 static int imsm_create_metadata_update_for_size_change(
11210 struct supertype
*st
,
11211 struct geo_params
*geo
,
11212 struct imsm_update_size_change
**updatep
)
11214 struct intel_super
*super
= st
->sb
;
11215 int update_memory_size
;
11216 struct imsm_update_size_change
*u
;
11218 dprintf("(enter) New size = %llu\n", geo
->size
);
11220 /* size of all update data without anchor */
11221 update_memory_size
= sizeof(struct imsm_update_size_change
);
11223 u
= xcalloc(1, update_memory_size
);
11224 u
->type
= update_size_change
;
11225 u
->subdev
= super
->current_vol
;
11226 u
->new_size
= geo
->size
;
11228 dprintf("imsm: reshape update preparation : OK\n");
11231 return update_memory_size
;
11234 /******************************************************************************
11235 * function: imsm_create_metadata_update_for_migration()
11236 * Creates update for IMSM array.
11238 ******************************************************************************/
11239 static int imsm_create_metadata_update_for_migration(
11240 struct supertype
*st
,
11241 struct geo_params
*geo
,
11242 struct imsm_update_reshape_migration
**updatep
)
11244 struct intel_super
*super
= st
->sb
;
11245 int update_memory_size
;
11246 struct imsm_update_reshape_migration
*u
;
11247 struct imsm_dev
*dev
;
11248 int previous_level
= -1;
11250 dprintf("(enter) New Level = %i\n", geo
->level
);
11252 /* size of all update data without anchor */
11253 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11255 u
= xcalloc(1, update_memory_size
);
11256 u
->type
= update_reshape_migration
;
11257 u
->subdev
= super
->current_vol
;
11258 u
->new_level
= geo
->level
;
11259 u
->new_layout
= geo
->layout
;
11260 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11261 u
->new_disks
[0] = -1;
11262 u
->new_chunksize
= -1;
11264 dev
= get_imsm_dev(super
, u
->subdev
);
11266 struct imsm_map
*map
;
11268 map
= get_imsm_map(dev
, MAP_0
);
11270 int current_chunk_size
=
11271 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11273 if (geo
->chunksize
!= current_chunk_size
) {
11274 u
->new_chunksize
= geo
->chunksize
/ 1024;
11275 dprintf("imsm: chunk size change from %i to %i\n",
11276 current_chunk_size
, u
->new_chunksize
);
11278 previous_level
= map
->raid_level
;
11281 if (geo
->level
== 5 && previous_level
== 0) {
11282 struct mdinfo
*spares
= NULL
;
11284 u
->new_raid_disks
++;
11285 spares
= get_spares_for_grow(st
);
11286 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11288 sysfs_free(spares
);
11289 update_memory_size
= 0;
11290 pr_err("cannot get spare device for requested migration\n");
11293 sysfs_free(spares
);
11295 dprintf("imsm: reshape update preparation : OK\n");
11298 return update_memory_size
;
11301 static void imsm_update_metadata_locally(struct supertype
*st
,
11302 void *buf
, int len
)
11304 struct metadata_update mu
;
11309 mu
.space_list
= NULL
;
11311 if (imsm_prepare_update(st
, &mu
))
11312 imsm_process_update(st
, &mu
);
11314 while (mu
.space_list
) {
11315 void **space
= mu
.space_list
;
11316 mu
.space_list
= *space
;
11321 /***************************************************************************
11322 * Function: imsm_analyze_change
11323 * Description: Function analyze change for single volume
11324 * and validate if transition is supported
11325 * Parameters: Geometry parameters, supertype structure,
11326 * metadata change direction (apply/rollback)
11327 * Returns: Operation type code on success, -1 if fail
11328 ****************************************************************************/
11329 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11330 struct geo_params
*geo
,
11333 struct mdinfo info
;
11335 int check_devs
= 0;
11337 /* number of added/removed disks in operation result */
11338 int devNumChange
= 0;
11339 /* imsm compatible layout value for array geometry verification */
11340 int imsm_layout
= -1;
11342 struct imsm_dev
*dev
;
11343 struct imsm_map
*map
;
11344 struct intel_super
*super
;
11345 unsigned long long current_size
;
11346 unsigned long long free_size
;
11347 unsigned long long max_size
;
11350 getinfo_super_imsm_volume(st
, &info
, NULL
);
11351 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11352 geo
->level
!= UnSet
) {
11353 switch (info
.array
.level
) {
11355 if (geo
->level
== 5) {
11356 change
= CH_MIGRATION
;
11357 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11358 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11360 goto analyse_change_exit
;
11362 imsm_layout
= geo
->layout
;
11364 devNumChange
= 1; /* parity disk added */
11365 } else if (geo
->level
== 10) {
11366 change
= CH_TAKEOVER
;
11368 devNumChange
= 2; /* two mirrors added */
11369 imsm_layout
= 0x102; /* imsm supported layout */
11374 if (geo
->level
== 0) {
11375 change
= CH_TAKEOVER
;
11377 devNumChange
= -(geo
->raid_disks
/2);
11378 imsm_layout
= 0; /* imsm raid0 layout */
11382 if (change
== -1) {
11383 pr_err("Error. Level Migration from %d to %d not supported!\n",
11384 info
.array
.level
, geo
->level
);
11385 goto analyse_change_exit
;
11388 geo
->level
= info
.array
.level
;
11390 if (geo
->layout
!= info
.array
.layout
&&
11391 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11392 change
= CH_MIGRATION
;
11393 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11394 geo
->layout
== 5) {
11395 /* reshape 5 -> 4 */
11396 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11397 geo
->layout
== 0) {
11398 /* reshape 4 -> 5 */
11402 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11403 info
.array
.layout
, geo
->layout
);
11405 goto analyse_change_exit
;
11408 geo
->layout
= info
.array
.layout
;
11409 if (imsm_layout
== -1)
11410 imsm_layout
= info
.array
.layout
;
11413 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11414 geo
->chunksize
!= info
.array
.chunk_size
) {
11415 if (info
.array
.level
== 10) {
11416 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11418 goto analyse_change_exit
;
11419 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11420 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11421 geo
->chunksize
/1024, info
.component_size
/2);
11423 goto analyse_change_exit
;
11425 change
= CH_MIGRATION
;
11427 geo
->chunksize
= info
.array
.chunk_size
;
11430 chunk
= geo
->chunksize
/ 1024;
11433 dev
= get_imsm_dev(super
, super
->current_vol
);
11434 map
= get_imsm_map(dev
, MAP_0
);
11435 data_disks
= imsm_num_data_members(map
);
11436 /* compute current size per disk member
11438 current_size
= info
.custom_array_size
/ data_disks
;
11440 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11441 /* align component size
11443 geo
->size
= imsm_component_size_alignment_check(
11444 get_imsm_raid_level(dev
->vol
.map
),
11445 chunk
* 1024, super
->sector_size
,
11447 if (geo
->size
== 0) {
11448 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11450 goto analyse_change_exit
;
11454 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11455 if (change
!= -1) {
11456 pr_err("Error. Size change should be the only one at a time.\n");
11458 goto analyse_change_exit
;
11460 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11461 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11462 super
->current_vol
, st
->devnm
);
11463 goto analyse_change_exit
;
11465 /* check the maximum available size
11467 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11468 0, chunk
, &free_size
);
11470 /* Cannot find maximum available space
11474 max_size
= free_size
+ current_size
;
11475 /* align component size
11477 max_size
= imsm_component_size_alignment_check(
11478 get_imsm_raid_level(dev
->vol
.map
),
11479 chunk
* 1024, super
->sector_size
,
11482 if (geo
->size
== MAX_SIZE
) {
11483 /* requested size change to the maximum available size
11485 if (max_size
== 0) {
11486 pr_err("Error. Cannot find maximum available space.\n");
11488 goto analyse_change_exit
;
11490 geo
->size
= max_size
;
11493 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11494 /* accept size for rollback only
11497 /* round size due to metadata compatibility
11499 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11500 << SECT_PER_MB_SHIFT
;
11501 dprintf("Prepare update for size change to %llu\n",
11503 if (current_size
>= geo
->size
) {
11504 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11505 current_size
, geo
->size
);
11506 goto analyse_change_exit
;
11508 if (max_size
&& geo
->size
> max_size
) {
11509 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11510 max_size
, geo
->size
);
11511 goto analyse_change_exit
;
11514 geo
->size
*= data_disks
;
11515 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11516 change
= CH_ARRAY_SIZE
;
11518 if (!validate_geometry_imsm(st
,
11521 geo
->raid_disks
+ devNumChange
,
11523 geo
->size
, INVALID_SECTORS
,
11524 0, 0, info
.consistency_policy
, 1))
11528 struct intel_super
*super
= st
->sb
;
11529 struct imsm_super
*mpb
= super
->anchor
;
11531 if (mpb
->num_raid_devs
> 1) {
11532 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11538 analyse_change_exit
:
11539 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11540 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11541 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11547 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11549 struct intel_super
*super
= st
->sb
;
11550 struct imsm_update_takeover
*u
;
11552 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11554 u
->type
= update_takeover
;
11555 u
->subarray
= super
->current_vol
;
11557 /* 10->0 transition */
11558 if (geo
->level
== 0)
11559 u
->direction
= R10_TO_R0
;
11561 /* 0->10 transition */
11562 if (geo
->level
== 10)
11563 u
->direction
= R0_TO_R10
;
11565 /* update metadata locally */
11566 imsm_update_metadata_locally(st
, u
,
11567 sizeof(struct imsm_update_takeover
));
11568 /* and possibly remotely */
11569 if (st
->update_tail
)
11570 append_metadata_update(st
, u
,
11571 sizeof(struct imsm_update_takeover
));
11578 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11580 int layout
, int chunksize
, int raid_disks
,
11581 int delta_disks
, char *backup
, char *dev
,
11582 int direction
, int verbose
)
11585 struct geo_params geo
;
11587 dprintf("(enter)\n");
11589 memset(&geo
, 0, sizeof(struct geo_params
));
11591 geo
.dev_name
= dev
;
11592 strcpy(geo
.devnm
, st
->devnm
);
11595 geo
.layout
= layout
;
11596 geo
.chunksize
= chunksize
;
11597 geo
.raid_disks
= raid_disks
;
11598 if (delta_disks
!= UnSet
)
11599 geo
.raid_disks
+= delta_disks
;
11601 dprintf("for level : %i\n", geo
.level
);
11602 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11604 if (experimental() == 0)
11607 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11608 /* On container level we can only increase number of devices. */
11609 dprintf("imsm: info: Container operation\n");
11610 int old_raid_disks
= 0;
11612 if (imsm_reshape_is_allowed_on_container(
11613 st
, &geo
, &old_raid_disks
, direction
)) {
11614 struct imsm_update_reshape
*u
= NULL
;
11617 len
= imsm_create_metadata_update_for_reshape(
11618 st
, &geo
, old_raid_disks
, &u
);
11621 dprintf("imsm: Cannot prepare update\n");
11622 goto exit_imsm_reshape_super
;
11626 /* update metadata locally */
11627 imsm_update_metadata_locally(st
, u
, len
);
11628 /* and possibly remotely */
11629 if (st
->update_tail
)
11630 append_metadata_update(st
, u
, len
);
11635 pr_err("(imsm) Operation is not allowed on this container\n");
11638 /* On volume level we support following operations
11639 * - takeover: raid10 -> raid0; raid0 -> raid10
11640 * - chunk size migration
11641 * - migration: raid5 -> raid0; raid0 -> raid5
11643 struct intel_super
*super
= st
->sb
;
11644 struct intel_dev
*dev
= super
->devlist
;
11646 dprintf("imsm: info: Volume operation\n");
11647 /* find requested device */
11650 imsm_find_array_devnm_by_subdev(
11651 dev
->index
, st
->container_devnm
);
11652 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11657 pr_err("Cannot find %s (%s) subarray\n",
11658 geo
.dev_name
, geo
.devnm
);
11659 goto exit_imsm_reshape_super
;
11661 super
->current_vol
= dev
->index
;
11662 change
= imsm_analyze_change(st
, &geo
, direction
);
11665 ret_val
= imsm_takeover(st
, &geo
);
11667 case CH_MIGRATION
: {
11668 struct imsm_update_reshape_migration
*u
= NULL
;
11670 imsm_create_metadata_update_for_migration(
11673 dprintf("imsm: Cannot prepare update\n");
11677 /* update metadata locally */
11678 imsm_update_metadata_locally(st
, u
, len
);
11679 /* and possibly remotely */
11680 if (st
->update_tail
)
11681 append_metadata_update(st
, u
, len
);
11686 case CH_ARRAY_SIZE
: {
11687 struct imsm_update_size_change
*u
= NULL
;
11689 imsm_create_metadata_update_for_size_change(
11692 dprintf("imsm: Cannot prepare update\n");
11696 /* update metadata locally */
11697 imsm_update_metadata_locally(st
, u
, len
);
11698 /* and possibly remotely */
11699 if (st
->update_tail
)
11700 append_metadata_update(st
, u
, len
);
11710 exit_imsm_reshape_super
:
11711 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11715 #define COMPLETED_OK 0
11716 #define COMPLETED_NONE 1
11717 #define COMPLETED_DELAYED 2
11719 static int read_completed(int fd
, unsigned long long *val
)
11724 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11728 ret
= COMPLETED_OK
;
11729 if (strncmp(buf
, "none", 4) == 0) {
11730 ret
= COMPLETED_NONE
;
11731 } else if (strncmp(buf
, "delayed", 7) == 0) {
11732 ret
= COMPLETED_DELAYED
;
11735 *val
= strtoull(buf
, &ep
, 0);
11736 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11742 /*******************************************************************************
11743 * Function: wait_for_reshape_imsm
11744 * Description: Function writes new sync_max value and waits until
11745 * reshape process reach new position
11747 * sra : general array info
11748 * ndata : number of disks in new array's layout
11751 * 1 : there is no reshape in progress,
11753 ******************************************************************************/
11754 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11756 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11758 unsigned long long completed
;
11759 /* to_complete : new sync_max position */
11760 unsigned long long to_complete
= sra
->reshape_progress
;
11761 unsigned long long position_to_set
= to_complete
/ ndata
;
11764 dprintf("cannot open reshape_position\n");
11769 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11771 dprintf("cannot read reshape_position (no reshape in progres)\n");
11780 if (completed
> position_to_set
) {
11781 dprintf("wrong next position to set %llu (%llu)\n",
11782 to_complete
, position_to_set
);
11786 dprintf("Position set: %llu\n", position_to_set
);
11787 if (sysfs_set_num(sra
, NULL
, "sync_max",
11788 position_to_set
) != 0) {
11789 dprintf("cannot set reshape position to %llu\n",
11798 int timeout
= 3000;
11800 sysfs_wait(fd
, &timeout
);
11801 if (sysfs_get_str(sra
, NULL
, "sync_action",
11803 strncmp(action
, "reshape", 7) != 0) {
11804 if (strncmp(action
, "idle", 4) == 0)
11810 rc
= read_completed(fd
, &completed
);
11812 dprintf("cannot read reshape_position (in loop)\n");
11815 } else if (rc
== COMPLETED_NONE
)
11817 } while (completed
< position_to_set
);
11823 /*******************************************************************************
11824 * Function: check_degradation_change
11825 * Description: Check that array hasn't become failed.
11827 * info : for sysfs access
11828 * sources : source disks descriptors
11829 * degraded: previous degradation level
11831 * degradation level
11832 ******************************************************************************/
11833 int check_degradation_change(struct mdinfo
*info
,
11837 unsigned long long new_degraded
;
11840 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11841 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11842 /* check each device to ensure it is still working */
11845 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11846 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11848 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11851 if (sysfs_get_str(info
,
11852 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11853 strstr(sbuf
, "faulty") ||
11854 strstr(sbuf
, "in_sync") == NULL
) {
11855 /* this device is dead */
11856 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11857 if (sd
->disk
.raid_disk
>= 0 &&
11858 sources
[sd
->disk
.raid_disk
] >= 0) {
11860 sd
->disk
.raid_disk
]);
11861 sources
[sd
->disk
.raid_disk
] =
11870 return new_degraded
;
11873 /*******************************************************************************
11874 * Function: imsm_manage_reshape
11875 * Description: Function finds array under reshape and it manages reshape
11876 * process. It creates stripes backups (if required) and sets
11879 * afd : Backup handle (nattive) - not used
11880 * sra : general array info
11881 * reshape : reshape parameters - not used
11882 * st : supertype structure
11883 * blocks : size of critical section [blocks]
11884 * fds : table of source device descriptor
11885 * offsets : start of array (offest per devices)
11887 * destfd : table of destination device descriptor
11888 * destoffsets : table of destination offsets (per device)
11890 * 1 : success, reshape is done
11892 ******************************************************************************/
11893 static int imsm_manage_reshape(
11894 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
11895 struct supertype
*st
, unsigned long backup_blocks
,
11896 int *fds
, unsigned long long *offsets
,
11897 int dests
, int *destfd
, unsigned long long *destoffsets
)
11900 struct intel_super
*super
= st
->sb
;
11901 struct intel_dev
*dv
;
11902 unsigned int sector_size
= super
->sector_size
;
11903 struct imsm_dev
*dev
= NULL
;
11904 struct imsm_map
*map_src
, *map_dest
;
11905 int migr_vol_qan
= 0;
11906 int ndata
, odata
; /* [bytes] */
11907 int chunk
; /* [bytes] */
11908 struct migr_record
*migr_rec
;
11910 unsigned int buf_size
; /* [bytes] */
11911 unsigned long long max_position
; /* array size [bytes] */
11912 unsigned long long next_step
; /* [blocks]/[bytes] */
11913 unsigned long long old_data_stripe_length
;
11914 unsigned long long start_src
; /* [bytes] */
11915 unsigned long long start
; /* [bytes] */
11916 unsigned long long start_buf_shift
; /* [bytes] */
11918 int source_layout
= 0;
11923 if (!fds
|| !offsets
)
11926 /* Find volume during the reshape */
11927 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
11928 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
11929 dv
->dev
->vol
.migr_state
== 1) {
11934 /* Only one volume can migrate at the same time */
11935 if (migr_vol_qan
!= 1) {
11936 pr_err("%s", migr_vol_qan
?
11937 "Number of migrating volumes greater than 1\n" :
11938 "There is no volume during migrationg\n");
11942 map_dest
= get_imsm_map(dev
, MAP_0
);
11943 map_src
= get_imsm_map(dev
, MAP_1
);
11944 if (map_src
== NULL
)
11947 ndata
= imsm_num_data_members(map_dest
);
11948 odata
= imsm_num_data_members(map_src
);
11950 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
11951 old_data_stripe_length
= odata
* chunk
;
11953 migr_rec
= super
->migr_rec
;
11955 /* initialize migration record for start condition */
11956 if (sra
->reshape_progress
== 0)
11957 init_migr_record_imsm(st
, dev
, sra
);
11959 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
11960 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
11963 /* Save checkpoint to update migration record for current
11964 * reshape position (in md). It can be farther than current
11965 * reshape position in metadata.
11967 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11968 /* ignore error == 2, this can mean end of reshape here
11970 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
11975 /* size for data */
11976 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
11977 /* extend buffer size for parity disk */
11978 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11979 /* add space for stripe alignment */
11980 buf_size
+= old_data_stripe_length
;
11981 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
11982 dprintf("imsm: Cannot allocate checkpoint buffer\n");
11986 max_position
= sra
->component_size
* ndata
;
11987 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
11989 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
11990 __le32_to_cpu(migr_rec
->num_migr_units
)) {
11991 /* current reshape position [blocks] */
11992 unsigned long long current_position
=
11993 __le32_to_cpu(migr_rec
->blocks_per_unit
)
11994 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
11995 unsigned long long border
;
11997 /* Check that array hasn't become failed.
11999 degraded
= check_degradation_change(sra
, fds
, degraded
);
12000 if (degraded
> 1) {
12001 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
12005 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
12007 if ((current_position
+ next_step
) > max_position
)
12008 next_step
= max_position
- current_position
;
12010 start
= current_position
* 512;
12012 /* align reading start to old geometry */
12013 start_buf_shift
= start
% old_data_stripe_length
;
12014 start_src
= start
- start_buf_shift
;
12016 border
= (start_src
/ odata
) - (start
/ ndata
);
12018 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
12019 /* save critical stripes to buf
12020 * start - start address of current unit
12021 * to backup [bytes]
12022 * start_src - start address of current unit
12023 * to backup alligned to source array
12026 unsigned long long next_step_filler
;
12027 unsigned long long copy_length
= next_step
* 512;
12029 /* allign copy area length to stripe in old geometry */
12030 next_step_filler
= ((copy_length
+ start_buf_shift
)
12031 % old_data_stripe_length
);
12032 if (next_step_filler
)
12033 next_step_filler
= (old_data_stripe_length
12034 - next_step_filler
);
12035 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
12036 start
, start_src
, copy_length
,
12037 start_buf_shift
, next_step_filler
);
12039 if (save_stripes(fds
, offsets
, map_src
->num_members
,
12040 chunk
, map_src
->raid_level
,
12041 source_layout
, 0, NULL
, start_src
,
12043 next_step_filler
+ start_buf_shift
,
12045 dprintf("imsm: Cannot save stripes to buffer\n");
12048 /* Convert data to destination format and store it
12049 * in backup general migration area
12051 if (save_backup_imsm(st
, dev
, sra
,
12052 buf
+ start_buf_shift
, copy_length
)) {
12053 dprintf("imsm: Cannot save stripes to target devices\n");
12056 if (save_checkpoint_imsm(st
, sra
,
12057 UNIT_SRC_IN_CP_AREA
)) {
12058 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
12062 /* set next step to use whole border area */
12063 border
/= next_step
;
12065 next_step
*= border
;
12067 /* When data backed up, checkpoint stored,
12068 * kick the kernel to reshape unit of data
12070 next_step
= next_step
+ sra
->reshape_progress
;
12071 /* limit next step to array max position */
12072 if (next_step
> max_position
)
12073 next_step
= max_position
;
12074 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
12075 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
12076 sra
->reshape_progress
= next_step
;
12078 /* wait until reshape finish */
12079 if (wait_for_reshape_imsm(sra
, ndata
)) {
12080 dprintf("wait_for_reshape_imsm returned error!\n");
12086 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
12087 /* ignore error == 2, this can mean end of reshape here
12089 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
12095 /* clear migr_rec on disks after successful migration */
12098 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
12099 for (d
= super
->disks
; d
; d
= d
->next
) {
12100 if (d
->index
< 0 || is_failed(&d
->disk
))
12102 unsigned long long dsize
;
12104 get_dev_size(d
->fd
, NULL
, &dsize
);
12105 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
12107 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
12108 MIGR_REC_BUF_SECTORS
*sector_size
) !=
12109 MIGR_REC_BUF_SECTORS
*sector_size
)
12110 perror("Write migr_rec failed");
12114 /* return '1' if done */
12118 /* See Grow.c: abort_reshape() for further explanation */
12119 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
12120 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
12121 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
12126 struct superswitch super_imsm
= {
12127 .examine_super
= examine_super_imsm
,
12128 .brief_examine_super
= brief_examine_super_imsm
,
12129 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
12130 .export_examine_super
= export_examine_super_imsm
,
12131 .detail_super
= detail_super_imsm
,
12132 .brief_detail_super
= brief_detail_super_imsm
,
12133 .write_init_super
= write_init_super_imsm
,
12134 .validate_geometry
= validate_geometry_imsm
,
12135 .add_to_super
= add_to_super_imsm
,
12136 .remove_from_super
= remove_from_super_imsm
,
12137 .detail_platform
= detail_platform_imsm
,
12138 .export_detail_platform
= export_detail_platform_imsm
,
12139 .kill_subarray
= kill_subarray_imsm
,
12140 .update_subarray
= update_subarray_imsm
,
12141 .load_container
= load_container_imsm
,
12142 .default_geometry
= default_geometry_imsm
,
12143 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
12144 .reshape_super
= imsm_reshape_super
,
12145 .manage_reshape
= imsm_manage_reshape
,
12146 .recover_backup
= recover_backup_imsm
,
12147 .copy_metadata
= copy_metadata_imsm
,
12148 .examine_badblocks
= examine_badblocks_imsm
,
12149 .match_home
= match_home_imsm
,
12150 .uuid_from_super
= uuid_from_super_imsm
,
12151 .getinfo_super
= getinfo_super_imsm
,
12152 .getinfo_super_disks
= getinfo_super_disks_imsm
,
12153 .update_super
= update_super_imsm
,
12155 .avail_size
= avail_size_imsm
,
12156 .get_spare_criteria
= get_spare_criteria_imsm
,
12158 .compare_super
= compare_super_imsm
,
12160 .load_super
= load_super_imsm
,
12161 .init_super
= init_super_imsm
,
12162 .store_super
= store_super_imsm
,
12163 .free_super
= free_super_imsm
,
12164 .match_metadata_desc
= match_metadata_desc_imsm
,
12165 .container_content
= container_content_imsm
,
12166 .validate_container
= validate_container_imsm
,
12168 .write_init_ppl
= write_init_ppl_imsm
,
12169 .validate_ppl
= validate_ppl_imsm
,
12175 .open_new
= imsm_open_new
,
12176 .set_array_state
= imsm_set_array_state
,
12177 .set_disk
= imsm_set_disk
,
12178 .sync_metadata
= imsm_sync_metadata
,
12179 .activate_spare
= imsm_activate_spare
,
12180 .process_update
= imsm_process_update
,
12181 .prepare_update
= imsm_prepare_update
,
12182 .record_bad_block
= imsm_record_badblock
,
12183 .clear_bad_block
= imsm_clear_badblock
,
12184 .get_bad_blocks
= imsm_get_badblocks
,