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 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1932 printf(" Version : %s\n", get_imsm_version(mpb
));
1933 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1934 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1935 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1936 printf(" Attributes : ");
1937 if (imsm_check_attributes(mpb
->attributes
))
1938 printf("All supported\n");
1940 printf("not supported\n");
1941 getinfo_super_imsm(st
, &info
, NULL
);
1942 fname_from_uuid(st
, &info
, nbuf
, ':');
1943 printf(" UUID : %s\n", nbuf
+ 5);
1944 sum
= __le32_to_cpu(mpb
->check_sum
);
1945 printf(" Checksum : %08x %s\n", sum
,
1946 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1947 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
1948 printf(" Disks : %d\n", mpb
->num_disks
);
1949 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1950 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
1951 super
->disks
->index
, reserved
, super
->sector_size
);
1952 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
1953 struct bbm_log
*log
= super
->bbm_log
;
1956 printf("Bad Block Management Log:\n");
1957 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1958 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1959 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1961 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1963 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1965 super
->current_vol
= i
;
1966 getinfo_super_imsm(st
, &info
, NULL
);
1967 fname_from_uuid(st
, &info
, nbuf
, ':');
1968 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1970 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1971 if (i
== super
->disks
->index
)
1973 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
1974 super
->sector_size
);
1977 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1978 if (dl
->index
== -1)
1979 print_imsm_disk(&dl
->disk
, -1, reserved
,
1980 super
->sector_size
);
1982 examine_migr_rec_imsm(super
);
1985 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1987 /* We just write a generic IMSM ARRAY entry */
1990 struct intel_super
*super
= st
->sb
;
1992 if (!super
->anchor
->num_raid_devs
) {
1993 printf("ARRAY metadata=imsm\n");
1997 getinfo_super_imsm(st
, &info
, NULL
);
1998 fname_from_uuid(st
, &info
, nbuf
, ':');
1999 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
2002 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
2004 /* We just write a generic IMSM ARRAY entry */
2008 struct intel_super
*super
= st
->sb
;
2011 if (!super
->anchor
->num_raid_devs
)
2014 getinfo_super_imsm(st
, &info
, NULL
);
2015 fname_from_uuid(st
, &info
, nbuf
, ':');
2016 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2017 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2019 super
->current_vol
= i
;
2020 getinfo_super_imsm(st
, &info
, NULL
);
2021 fname_from_uuid(st
, &info
, nbuf1
, ':');
2022 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
2023 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
2027 static void export_examine_super_imsm(struct supertype
*st
)
2029 struct intel_super
*super
= st
->sb
;
2030 struct imsm_super
*mpb
= super
->anchor
;
2034 getinfo_super_imsm(st
, &info
, NULL
);
2035 fname_from_uuid(st
, &info
, nbuf
, ':');
2036 printf("MD_METADATA=imsm\n");
2037 printf("MD_LEVEL=container\n");
2038 printf("MD_UUID=%s\n", nbuf
+5);
2039 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
2042 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
2044 /* The second last sector of the device contains
2045 * the "struct imsm_super" metadata.
2046 * This contains mpb_size which is the size in bytes of the
2047 * extended metadata. This is located immediately before
2049 * We want to read all that, plus the last sector which
2050 * may contain a migration record, and write it all
2054 unsigned long long dsize
, offset
;
2056 struct imsm_super
*sb
;
2057 struct intel_super
*super
= st
->sb
;
2058 unsigned int sector_size
= super
->sector_size
;
2059 unsigned int written
= 0;
2061 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
) != 0)
2064 if (!get_dev_size(from
, NULL
, &dsize
))
2067 if (lseek64(from
, dsize
-(2*sector_size
), 0) < 0)
2069 if ((unsigned int)read(from
, buf
, sector_size
) != sector_size
)
2072 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
2075 sectors
= mpb_sectors(sb
, sector_size
) + 2;
2076 offset
= dsize
- sectors
* sector_size
;
2077 if (lseek64(from
, offset
, 0) < 0 ||
2078 lseek64(to
, offset
, 0) < 0)
2080 while (written
< sectors
* sector_size
) {
2081 int n
= sectors
*sector_size
- written
;
2084 if (read(from
, buf
, n
) != n
)
2086 if (write(to
, buf
, n
) != n
)
2097 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
2102 getinfo_super_imsm(st
, &info
, NULL
);
2103 fname_from_uuid(st
, &info
, nbuf
, ':');
2104 printf("\n UUID : %s\n", nbuf
+ 5);
2107 static void brief_detail_super_imsm(struct supertype
*st
)
2111 getinfo_super_imsm(st
, &info
, NULL
);
2112 fname_from_uuid(st
, &info
, nbuf
, ':');
2113 printf(" UUID=%s", nbuf
+ 5);
2116 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
2117 static void fd2devname(int fd
, char *name
);
2119 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2121 /* dump an unsorted list of devices attached to AHCI Intel storage
2122 * controller, as well as non-connected ports
2124 int hba_len
= strlen(hba_path
) + 1;
2129 unsigned long port_mask
= (1 << port_count
) - 1;
2131 if (port_count
> (int)sizeof(port_mask
) * 8) {
2133 pr_err("port_count %d out of range\n", port_count
);
2137 /* scroll through /sys/dev/block looking for devices attached to
2140 dir
= opendir("/sys/dev/block");
2144 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2155 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2157 path
= devt_to_devpath(makedev(major
, minor
));
2160 if (!path_attached_to_hba(path
, hba_path
)) {
2166 /* retrieve the scsi device type */
2167 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2169 pr_err("failed to allocate 'device'\n");
2173 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2174 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2176 pr_err("failed to read device type for %s\n",
2182 type
= strtoul(buf
, NULL
, 10);
2184 /* if it's not a disk print the vendor and model */
2185 if (!(type
== 0 || type
== 7 || type
== 14)) {
2188 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2189 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2190 strncpy(vendor
, buf
, sizeof(vendor
));
2191 vendor
[sizeof(vendor
) - 1] = '\0';
2192 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2193 while (isspace(*c
) || *c
== '\0')
2197 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2198 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2199 strncpy(model
, buf
, sizeof(model
));
2200 model
[sizeof(model
) - 1] = '\0';
2201 c
= (char *) &model
[sizeof(model
) - 1];
2202 while (isspace(*c
) || *c
== '\0')
2206 if (vendor
[0] && model
[0])
2207 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2209 switch (type
) { /* numbers from hald/linux/device.c */
2210 case 1: sprintf(buf
, "tape"); break;
2211 case 2: sprintf(buf
, "printer"); break;
2212 case 3: sprintf(buf
, "processor"); break;
2214 case 5: sprintf(buf
, "cdrom"); break;
2215 case 6: sprintf(buf
, "scanner"); break;
2216 case 8: sprintf(buf
, "media_changer"); break;
2217 case 9: sprintf(buf
, "comm"); break;
2218 case 12: sprintf(buf
, "raid"); break;
2219 default: sprintf(buf
, "unknown");
2225 /* chop device path to 'host%d' and calculate the port number */
2226 c
= strchr(&path
[hba_len
], '/');
2229 pr_err("%s - invalid path name\n", path
+ hba_len
);
2234 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2235 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2239 *c
= '/'; /* repair the full string */
2240 pr_err("failed to determine port number for %s\n",
2247 /* mark this port as used */
2248 port_mask
&= ~(1 << port
);
2250 /* print out the device information */
2252 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2256 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2258 printf(" Port%d : - disk info unavailable -\n", port
);
2260 fd2devname(fd
, buf
);
2261 printf(" Port%d : %s", port
, buf
);
2262 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
2263 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
2278 for (i
= 0; i
< port_count
; i
++)
2279 if (port_mask
& (1 << i
))
2280 printf(" Port%d : - no device attached -\n", i
);
2286 static int print_vmd_attached_devs(struct sys_dev
*hba
)
2294 if (hba
->type
!= SYS_DEV_VMD
)
2297 /* scroll through /sys/dev/block looking for devices attached to
2300 dir
= opendir("/sys/bus/pci/drivers/nvme");
2304 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2307 /* is 'ent' a device? check that the 'subsystem' link exists and
2308 * that its target matches 'bus'
2310 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
2312 n
= readlink(path
, link
, sizeof(link
));
2313 if (n
< 0 || n
>= (int)sizeof(link
))
2316 c
= strrchr(link
, '/');
2319 if (strncmp("pci", c
+1, strlen("pci")) != 0)
2322 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
2324 rp
= realpath(path
, NULL
);
2328 if (path_attached_to_hba(rp
, hba
->path
)) {
2329 printf(" NVMe under VMD : %s\n", rp
);
2338 static void print_found_intel_controllers(struct sys_dev
*elem
)
2340 for (; elem
; elem
= elem
->next
) {
2341 pr_err("found Intel(R) ");
2342 if (elem
->type
== SYS_DEV_SATA
)
2343 fprintf(stderr
, "SATA ");
2344 else if (elem
->type
== SYS_DEV_SAS
)
2345 fprintf(stderr
, "SAS ");
2346 else if (elem
->type
== SYS_DEV_NVME
)
2347 fprintf(stderr
, "NVMe ");
2349 if (elem
->type
== SYS_DEV_VMD
)
2350 fprintf(stderr
, "VMD domain");
2352 fprintf(stderr
, "RAID controller");
2355 fprintf(stderr
, " at %s", elem
->pci_id
);
2356 fprintf(stderr
, ".\n");
2361 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2368 if ((dir
= opendir(hba_path
)) == NULL
)
2371 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2374 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2375 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2377 if (*port_count
== 0)
2379 else if (host
< host_base
)
2382 if (host
+ 1 > *port_count
+ host_base
)
2383 *port_count
= host
+ 1 - host_base
;
2389 static void print_imsm_capability(const struct imsm_orom
*orom
)
2391 printf(" Platform : Intel(R) ");
2392 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2393 printf("Matrix Storage Manager\n");
2394 else if (imsm_orom_is_enterprise(orom
) && orom
->major_ver
>= 6)
2395 printf("Virtual RAID on CPU\n");
2397 printf("Rapid Storage Technology%s\n",
2398 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2399 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2400 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2401 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2402 printf(" RAID Levels :%s%s%s%s%s\n",
2403 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2404 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2405 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2406 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2407 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2408 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2409 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2410 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2411 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2412 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2413 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2414 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2415 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2416 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2417 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2418 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2419 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2420 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2421 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2422 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2423 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2424 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2425 printf(" 2TB volumes :%s supported\n",
2426 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2427 printf(" 2TB disks :%s supported\n",
2428 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2429 printf(" Max Disks : %d\n", orom
->tds
);
2430 printf(" Max Volumes : %d per array, %d per %s\n",
2431 orom
->vpa
, orom
->vphba
,
2432 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2436 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2438 printf("MD_FIRMWARE_TYPE=imsm\n");
2439 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2440 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2441 orom
->hotfix_ver
, orom
->build
);
2442 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2443 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2444 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2445 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2446 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2447 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2448 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2449 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2450 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2451 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2452 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2453 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2454 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2455 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2456 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2457 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2458 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2459 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2460 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2461 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2462 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2463 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2464 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2465 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2466 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2467 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2468 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2469 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2472 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2474 /* There are two components to imsm platform support, the ahci SATA
2475 * controller and the option-rom. To find the SATA controller we
2476 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2477 * controller with the Intel vendor id is present. This approach
2478 * allows mdadm to leverage the kernel's ahci detection logic, with the
2479 * caveat that if ahci.ko is not loaded mdadm will not be able to
2480 * detect platform raid capabilities. The option-rom resides in a
2481 * platform "Adapter ROM". We scan for its signature to retrieve the
2482 * platform capabilities. If raid support is disabled in the BIOS the
2483 * option-rom capability structure will not be available.
2485 struct sys_dev
*list
, *hba
;
2490 if (enumerate_only
) {
2491 if (check_env("IMSM_NO_PLATFORM"))
2493 list
= find_intel_devices();
2496 for (hba
= list
; hba
; hba
= hba
->next
) {
2497 if (find_imsm_capability(hba
)) {
2507 list
= find_intel_devices();
2510 pr_err("no active Intel(R) RAID controller found.\n");
2512 } else if (verbose
> 0)
2513 print_found_intel_controllers(list
);
2515 for (hba
= list
; hba
; hba
= hba
->next
) {
2516 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2518 if (!find_imsm_capability(hba
)) {
2520 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2521 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2522 get_sys_dev_type(hba
->type
));
2528 if (controller_path
&& result
== 1) {
2529 pr_err("no active Intel(R) RAID controller found under %s\n",
2534 const struct orom_entry
*entry
;
2536 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2537 if (entry
->type
== SYS_DEV_VMD
) {
2538 print_imsm_capability(&entry
->orom
);
2539 printf(" 3rd party NVMe :%s supported\n",
2540 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2541 for (hba
= list
; hba
; hba
= hba
->next
) {
2542 if (hba
->type
== SYS_DEV_VMD
) {
2544 printf(" I/O Controller : %s (%s)\n",
2545 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2546 if (print_vmd_attached_devs(hba
)) {
2548 pr_err("failed to get devices attached to VMD domain.\n");
2557 print_imsm_capability(&entry
->orom
);
2558 if (entry
->type
== SYS_DEV_NVME
) {
2559 for (hba
= list
; hba
; hba
= hba
->next
) {
2560 if (hba
->type
== SYS_DEV_NVME
)
2561 printf(" NVMe Device : %s\n", hba
->path
);
2567 struct devid_list
*devid
;
2568 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2569 hba
= device_by_id(devid
->devid
);
2573 printf(" I/O Controller : %s (%s)\n",
2574 hba
->path
, get_sys_dev_type(hba
->type
));
2575 if (hba
->type
== SYS_DEV_SATA
) {
2576 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2577 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2579 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2590 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2592 struct sys_dev
*list
, *hba
;
2595 list
= find_intel_devices();
2598 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2603 for (hba
= list
; hba
; hba
= hba
->next
) {
2604 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2606 if (!find_imsm_capability(hba
) && verbose
> 0) {
2608 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2609 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2615 const struct orom_entry
*entry
;
2617 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2618 if (entry
->type
== SYS_DEV_VMD
) {
2619 for (hba
= list
; hba
; hba
= hba
->next
)
2620 print_imsm_capability_export(&entry
->orom
);
2623 print_imsm_capability_export(&entry
->orom
);
2629 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2631 /* the imsm metadata format does not specify any host
2632 * identification information. We return -1 since we can never
2633 * confirm nor deny whether a given array is "meant" for this
2634 * host. We rely on compare_super and the 'family_num' fields to
2635 * exclude member disks that do not belong, and we rely on
2636 * mdadm.conf to specify the arrays that should be assembled.
2637 * Auto-assembly may still pick up "foreign" arrays.
2643 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2645 /* The uuid returned here is used for:
2646 * uuid to put into bitmap file (Create, Grow)
2647 * uuid for backup header when saving critical section (Grow)
2648 * comparing uuids when re-adding a device into an array
2649 * In these cases the uuid required is that of the data-array,
2650 * not the device-set.
2651 * uuid to recognise same set when adding a missing device back
2652 * to an array. This is a uuid for the device-set.
2654 * For each of these we can make do with a truncated
2655 * or hashed uuid rather than the original, as long as
2657 * In each case the uuid required is that of the data-array,
2658 * not the device-set.
2660 /* imsm does not track uuid's so we synthesis one using sha1 on
2661 * - The signature (Which is constant for all imsm array, but no matter)
2662 * - the orig_family_num of the container
2663 * - the index number of the volume
2664 * - the 'serial' number of the volume.
2665 * Hopefully these are all constant.
2667 struct intel_super
*super
= st
->sb
;
2670 struct sha1_ctx ctx
;
2671 struct imsm_dev
*dev
= NULL
;
2674 /* some mdadm versions failed to set ->orig_family_num, in which
2675 * case fall back to ->family_num. orig_family_num will be
2676 * fixed up with the first metadata update.
2678 family_num
= super
->anchor
->orig_family_num
;
2679 if (family_num
== 0)
2680 family_num
= super
->anchor
->family_num
;
2681 sha1_init_ctx(&ctx
);
2682 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2683 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2684 if (super
->current_vol
>= 0)
2685 dev
= get_imsm_dev(super
, super
->current_vol
);
2687 __u32 vol
= super
->current_vol
;
2688 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2689 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2691 sha1_finish_ctx(&ctx
, buf
);
2692 memcpy(uuid
, buf
, 4*4);
2697 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2699 __u8
*v
= get_imsm_version(mpb
);
2700 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2701 char major
[] = { 0, 0, 0 };
2702 char minor
[] = { 0 ,0, 0 };
2703 char patch
[] = { 0, 0, 0 };
2704 char *ver_parse
[] = { major
, minor
, patch
};
2708 while (*v
!= '\0' && v
< end
) {
2709 if (*v
!= '.' && j
< 2)
2710 ver_parse
[i
][j
++] = *v
;
2718 *m
= strtol(minor
, NULL
, 0);
2719 *p
= strtol(patch
, NULL
, 0);
2723 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2725 /* migr_strip_size when repairing or initializing parity */
2726 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2727 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2729 switch (get_imsm_raid_level(map
)) {
2734 return 128*1024 >> 9;
2738 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2740 /* migr_strip_size when rebuilding a degraded disk, no idea why
2741 * this is different than migr_strip_size_resync(), but it's good
2744 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2745 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2747 switch (get_imsm_raid_level(map
)) {
2750 if (map
->num_members
% map
->num_domains
== 0)
2751 return 128*1024 >> 9;
2755 return max((__u32
) 64*1024 >> 9, chunk
);
2757 return 128*1024 >> 9;
2761 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2763 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2764 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2765 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2766 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2768 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2771 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2773 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2774 int level
= get_imsm_raid_level(lo
);
2776 if (level
== 1 || level
== 10) {
2777 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2779 return hi
->num_domains
;
2781 return num_stripes_per_unit_resync(dev
);
2784 static __u8
imsm_num_data_members(struct imsm_map
*map
)
2786 /* named 'imsm_' because raid0, raid1 and raid10
2787 * counter-intuitively have the same number of data disks
2789 switch (get_imsm_raid_level(map
)) {
2791 return map
->num_members
;
2795 return map
->num_members
/2;
2797 return map
->num_members
- 1;
2799 dprintf("unsupported raid level\n");
2804 static unsigned long long calc_component_size(struct imsm_map
*map
,
2805 struct imsm_dev
*dev
)
2807 unsigned long long component_size
;
2808 unsigned long long dev_size
= imsm_dev_size(dev
);
2809 unsigned long long calc_dev_size
= 0;
2810 unsigned int member_disks
= imsm_num_data_members(map
);
2812 if (member_disks
== 0)
2815 component_size
= per_dev_array_size(map
);
2816 calc_dev_size
= component_size
* member_disks
;
2818 /* Component size is rounded to 1MB so difference between size from
2819 * metadata and size calculated from num_data_stripes equals up to
2820 * 2048 blocks per each device. If the difference is higher it means
2821 * that array size was expanded and num_data_stripes was not updated.
2823 if ((unsigned int)abs(calc_dev_size
- dev_size
) >
2824 (1 << SECT_PER_MB_SHIFT
) * member_disks
) {
2825 component_size
= dev_size
/ member_disks
;
2826 dprintf("Invalid num_data_stripes in metadata; expected=%llu, found=%llu\n",
2827 component_size
/ map
->blocks_per_strip
,
2828 num_data_stripes(map
));
2831 return component_size
;
2834 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2836 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2837 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2839 switch(get_imsm_raid_level(map
)) {
2842 return chunk
* map
->num_domains
;
2844 return chunk
* map
->num_members
;
2850 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2852 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2853 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2854 __u32 strip
= block
/ chunk
;
2856 switch (get_imsm_raid_level(map
)) {
2859 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2860 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2862 return vol_stripe
* chunk
+ block
% chunk
;
2864 __u32 stripe
= strip
/ (map
->num_members
- 1);
2866 return stripe
* chunk
+ block
% chunk
;
2873 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2874 struct imsm_dev
*dev
)
2876 /* calculate the conversion factor between per member 'blocks'
2877 * (md/{resync,rebuild}_start) and imsm migration units, return
2878 * 0 for the 'not migrating' and 'unsupported migration' cases
2880 if (!dev
->vol
.migr_state
)
2883 switch (migr_type(dev
)) {
2884 case MIGR_GEN_MIGR
: {
2885 struct migr_record
*migr_rec
= super
->migr_rec
;
2886 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2891 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2892 __u32 stripes_per_unit
;
2893 __u32 blocks_per_unit
;
2902 /* yes, this is really the translation of migr_units to
2903 * per-member blocks in the 'resync' case
2905 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2906 migr_chunk
= migr_strip_blocks_resync(dev
);
2907 disks
= imsm_num_data_members(map
);
2908 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2909 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2910 segment
= blocks_per_unit
/ stripe
;
2911 block_rel
= blocks_per_unit
- segment
* stripe
;
2912 parity_depth
= parity_segment_depth(dev
);
2913 block_map
= map_migr_block(dev
, block_rel
);
2914 return block_map
+ parity_depth
* segment
;
2916 case MIGR_REBUILD
: {
2917 __u32 stripes_per_unit
;
2920 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2921 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2922 return migr_chunk
* stripes_per_unit
;
2924 case MIGR_STATE_CHANGE
:
2930 static int imsm_level_to_layout(int level
)
2938 return ALGORITHM_LEFT_ASYMMETRIC
;
2945 /*******************************************************************************
2946 * Function: read_imsm_migr_rec
2947 * Description: Function reads imsm migration record from last sector of disk
2949 * fd : disk descriptor
2950 * super : metadata info
2954 ******************************************************************************/
2955 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2958 unsigned int sector_size
= super
->sector_size
;
2959 unsigned long long dsize
;
2961 get_dev_size(fd
, NULL
, &dsize
);
2962 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
2964 pr_err("Cannot seek to anchor block: %s\n",
2968 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
2969 MIGR_REC_BUF_SECTORS
*sector_size
) !=
2970 MIGR_REC_BUF_SECTORS
*sector_size
) {
2971 pr_err("Cannot read migr record block: %s\n",
2976 if (sector_size
== 4096)
2977 convert_from_4k_imsm_migr_rec(super
);
2983 static struct imsm_dev
*imsm_get_device_during_migration(
2984 struct intel_super
*super
)
2987 struct intel_dev
*dv
;
2989 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2990 if (is_gen_migration(dv
->dev
))
2996 /*******************************************************************************
2997 * Function: load_imsm_migr_rec
2998 * Description: Function reads imsm migration record (it is stored at the last
3001 * super : imsm internal array info
3002 * info : general array info
3006 * -2 : no migration in progress
3007 ******************************************************************************/
3008 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
3015 struct imsm_dev
*dev
;
3016 struct imsm_map
*map
;
3019 /* find map under migration */
3020 dev
= imsm_get_device_during_migration(super
);
3021 /* nothing to load,no migration in progress?
3027 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
3028 /* read only from one of the first two slots */
3029 if ((sd
->disk
.raid_disk
< 0) ||
3030 (sd
->disk
.raid_disk
> 1))
3033 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3034 fd
= dev_open(nm
, O_RDONLY
);
3040 map
= get_imsm_map(dev
, MAP_0
);
3041 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3042 /* skip spare and failed disks
3046 /* read only from one of the first two slots */
3048 slot
= get_imsm_disk_slot(map
, dl
->index
);
3049 if (map
== NULL
|| slot
> 1 || slot
< 0)
3051 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
3052 fd
= dev_open(nm
, O_RDONLY
);
3059 retval
= read_imsm_migr_rec(fd
, super
);
3067 /*******************************************************************************
3068 * function: imsm_create_metadata_checkpoint_update
3069 * Description: It creates update for checkpoint change.
3071 * super : imsm internal array info
3072 * u : pointer to prepared update
3075 * If length is equal to 0, input pointer u contains no update
3076 ******************************************************************************/
3077 static int imsm_create_metadata_checkpoint_update(
3078 struct intel_super
*super
,
3079 struct imsm_update_general_migration_checkpoint
**u
)
3082 int update_memory_size
= 0;
3084 dprintf("(enter)\n");
3090 /* size of all update data without anchor */
3091 update_memory_size
=
3092 sizeof(struct imsm_update_general_migration_checkpoint
);
3094 *u
= xcalloc(1, update_memory_size
);
3096 dprintf("error: cannot get memory\n");
3099 (*u
)->type
= update_general_migration_checkpoint
;
3100 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
3101 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3103 return update_memory_size
;
3106 static void imsm_update_metadata_locally(struct supertype
*st
,
3107 void *buf
, int len
);
3109 /*******************************************************************************
3110 * Function: write_imsm_migr_rec
3111 * Description: Function writes imsm migration record
3112 * (at the last sector of disk)
3114 * super : imsm internal array info
3118 ******************************************************************************/
3119 static int write_imsm_migr_rec(struct supertype
*st
)
3121 struct intel_super
*super
= st
->sb
;
3122 unsigned int sector_size
= super
->sector_size
;
3123 unsigned long long dsize
;
3129 struct imsm_update_general_migration_checkpoint
*u
;
3130 struct imsm_dev
*dev
;
3131 struct imsm_map
*map
;
3133 /* find map under migration */
3134 dev
= imsm_get_device_during_migration(super
);
3135 /* if no migration, write buffer anyway to clear migr_record
3136 * on disk based on first available device
3139 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3140 super
->current_vol
);
3142 map
= get_imsm_map(dev
, MAP_0
);
3144 if (sector_size
== 4096)
3145 convert_to_4k_imsm_migr_rec(super
);
3146 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3149 /* skip failed and spare devices */
3152 /* write to 2 first slots only */
3154 slot
= get_imsm_disk_slot(map
, sd
->index
);
3155 if (map
== NULL
|| slot
> 1 || slot
< 0)
3158 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3159 fd
= dev_open(nm
, O_RDWR
);
3162 get_dev_size(fd
, NULL
, &dsize
);
3163 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3165 pr_err("Cannot seek to anchor block: %s\n",
3169 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3170 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3171 MIGR_REC_BUF_SECTORS
*sector_size
) {
3172 pr_err("Cannot write migr record block: %s\n",
3179 if (sector_size
== 4096)
3180 convert_from_4k_imsm_migr_rec(super
);
3181 /* update checkpoint information in metadata */
3182 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3184 dprintf("imsm: Cannot prepare update\n");
3187 /* update metadata locally */
3188 imsm_update_metadata_locally(st
, u
, len
);
3189 /* and possibly remotely */
3190 if (st
->update_tail
) {
3191 append_metadata_update(st
, u
, len
);
3192 /* during reshape we do all work inside metadata handler
3193 * manage_reshape(), so metadata update has to be triggered
3196 flush_metadata_updates(st
);
3197 st
->update_tail
= &st
->updates
;
3208 /* spare/missing disks activations are not allowe when
3209 * array/container performs reshape operation, because
3210 * all arrays in container works on the same disks set
3212 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3215 struct intel_dev
*i_dev
;
3216 struct imsm_dev
*dev
;
3218 /* check whole container
3220 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3222 if (is_gen_migration(dev
)) {
3223 /* No repair during any migration in container
3231 static unsigned long long imsm_component_size_alignment_check(int level
,
3233 unsigned int sector_size
,
3234 unsigned long long component_size
)
3236 unsigned int component_size_alignment
;
3238 /* check component size alignment
3240 component_size_alignment
= component_size
% (chunk_size
/sector_size
);
3242 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alignment = %u\n",
3243 level
, chunk_size
, component_size
,
3244 component_size_alignment
);
3246 if (component_size_alignment
&& (level
!= 1) && (level
!= UnSet
)) {
3247 dprintf("imsm: reported component size aligned from %llu ",
3249 component_size
-= component_size_alignment
;
3250 dprintf_cont("to %llu (%i).\n",
3251 component_size
, component_size_alignment
);
3254 return component_size
;
3257 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3259 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3260 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3262 return pba_of_lba0(map
) +
3263 (num_data_stripes(map
) * map
->blocks_per_strip
);
3266 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3268 struct intel_super
*super
= st
->sb
;
3269 struct migr_record
*migr_rec
= super
->migr_rec
;
3270 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3271 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3272 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3273 struct imsm_map
*map_to_analyse
= map
;
3275 int map_disks
= info
->array
.raid_disks
;
3277 memset(info
, 0, sizeof(*info
));
3279 map_to_analyse
= prev_map
;
3281 dl
= super
->current_disk
;
3283 info
->container_member
= super
->current_vol
;
3284 info
->array
.raid_disks
= map
->num_members
;
3285 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3286 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3287 info
->array
.md_minor
= -1;
3288 info
->array
.ctime
= 0;
3289 info
->array
.utime
= 0;
3290 info
->array
.chunk_size
=
3291 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3292 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3293 info
->custom_array_size
= imsm_dev_size(dev
);
3294 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3296 if (is_gen_migration(dev
)) {
3297 info
->reshape_active
= 1;
3298 info
->new_level
= get_imsm_raid_level(map
);
3299 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3300 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3301 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3302 if (info
->delta_disks
) {
3303 /* this needs to be applied to every array
3306 info
->reshape_active
= CONTAINER_RESHAPE
;
3308 /* We shape information that we give to md might have to be
3309 * modify to cope with md's requirement for reshaping arrays.
3310 * For example, when reshaping a RAID0, md requires it to be
3311 * presented as a degraded RAID4.
3312 * Also if a RAID0 is migrating to a RAID5 we need to specify
3313 * the array as already being RAID5, but the 'before' layout
3314 * is a RAID4-like layout.
3316 switch (info
->array
.level
) {
3318 switch(info
->new_level
) {
3320 /* conversion is happening as RAID4 */
3321 info
->array
.level
= 4;
3322 info
->array
.raid_disks
+= 1;
3325 /* conversion is happening as RAID5 */
3326 info
->array
.level
= 5;
3327 info
->array
.layout
= ALGORITHM_PARITY_N
;
3328 info
->delta_disks
-= 1;
3331 /* FIXME error message */
3332 info
->array
.level
= UnSet
;
3338 info
->new_level
= UnSet
;
3339 info
->new_layout
= UnSet
;
3340 info
->new_chunk
= info
->array
.chunk_size
;
3341 info
->delta_disks
= 0;
3345 info
->disk
.major
= dl
->major
;
3346 info
->disk
.minor
= dl
->minor
;
3347 info
->disk
.number
= dl
->index
;
3348 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3352 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3353 info
->component_size
= calc_component_size(map
, dev
);
3354 info
->component_size
= imsm_component_size_alignment_check(
3356 info
->array
.chunk_size
,
3358 info
->component_size
);
3359 info
->bb
.supported
= 1;
3361 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3362 info
->recovery_start
= MaxSector
;
3364 if (info
->array
.level
== 5 &&
3365 (dev
->rwh_policy
== RWH_DISTRIBUTED
||
3366 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)) {
3367 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3368 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3369 if (dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
3370 info
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
3372 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
)
3374 } else if (info
->array
.level
<= 0) {
3375 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3377 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3380 info
->reshape_progress
= 0;
3381 info
->resync_start
= MaxSector
;
3382 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3383 !(info
->array
.state
& 1)) &&
3384 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3385 info
->resync_start
= 0;
3387 if (dev
->vol
.migr_state
) {
3388 switch (migr_type(dev
)) {
3391 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3393 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3395 info
->resync_start
= blocks_per_unit
* units
;
3398 case MIGR_GEN_MIGR
: {
3399 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3401 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
3402 unsigned long long array_blocks
;
3405 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3407 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
3408 (super
->migr_rec
->rec_status
==
3409 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3412 info
->reshape_progress
= blocks_per_unit
* units
;
3414 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3415 (unsigned long long)units
,
3416 (unsigned long long)blocks_per_unit
,
3417 info
->reshape_progress
);
3419 used_disks
= imsm_num_data_members(prev_map
);
3420 if (used_disks
> 0) {
3421 array_blocks
= per_dev_array_size(map
) *
3423 info
->custom_array_size
=
3424 round_size_to_mb(array_blocks
,
3430 /* we could emulate the checkpointing of
3431 * 'sync_action=check' migrations, but for now
3432 * we just immediately complete them
3435 /* this is handled by container_content_imsm() */
3436 case MIGR_STATE_CHANGE
:
3437 /* FIXME handle other migrations */
3439 /* we are not dirty, so... */
3440 info
->resync_start
= MaxSector
;
3444 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3445 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3447 info
->array
.major_version
= -1;
3448 info
->array
.minor_version
= -2;
3449 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3450 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3451 uuid_from_super_imsm(st
, info
->uuid
);
3455 for (i
=0; i
<map_disks
; i
++) {
3457 if (i
< info
->array
.raid_disks
) {
3458 struct imsm_disk
*dsk
;
3459 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3460 dsk
= get_imsm_disk(super
, j
);
3461 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3468 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3469 int failed
, int look_in_map
);
3471 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3474 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3476 if (is_gen_migration(dev
)) {
3479 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3481 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3482 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3483 if (map2
->map_state
!= map_state
) {
3484 map2
->map_state
= map_state
;
3485 super
->updates_pending
++;
3490 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3494 for (d
= super
->missing
; d
; d
= d
->next
)
3495 if (d
->index
== index
)
3500 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3502 struct intel_super
*super
= st
->sb
;
3503 struct imsm_disk
*disk
;
3504 int map_disks
= info
->array
.raid_disks
;
3505 int max_enough
= -1;
3507 struct imsm_super
*mpb
;
3509 if (super
->current_vol
>= 0) {
3510 getinfo_super_imsm_volume(st
, info
, map
);
3513 memset(info
, 0, sizeof(*info
));
3515 /* Set raid_disks to zero so that Assemble will always pull in valid
3518 info
->array
.raid_disks
= 0;
3519 info
->array
.level
= LEVEL_CONTAINER
;
3520 info
->array
.layout
= 0;
3521 info
->array
.md_minor
= -1;
3522 info
->array
.ctime
= 0; /* N/A for imsm */
3523 info
->array
.utime
= 0;
3524 info
->array
.chunk_size
= 0;
3526 info
->disk
.major
= 0;
3527 info
->disk
.minor
= 0;
3528 info
->disk
.raid_disk
= -1;
3529 info
->reshape_active
= 0;
3530 info
->array
.major_version
= -1;
3531 info
->array
.minor_version
= -2;
3532 strcpy(info
->text_version
, "imsm");
3533 info
->safe_mode_delay
= 0;
3534 info
->disk
.number
= -1;
3535 info
->disk
.state
= 0;
3537 info
->recovery_start
= MaxSector
;
3538 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3539 info
->bb
.supported
= 1;
3541 /* do we have the all the insync disks that we expect? */
3542 mpb
= super
->anchor
;
3543 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3545 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3546 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3547 int failed
, enough
, j
, missing
= 0;
3548 struct imsm_map
*map
;
3551 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3552 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3553 map
= get_imsm_map(dev
, MAP_0
);
3555 /* any newly missing disks?
3556 * (catches single-degraded vs double-degraded)
3558 for (j
= 0; j
< map
->num_members
; j
++) {
3559 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3560 __u32 idx
= ord_to_idx(ord
);
3562 if (super
->disks
&& super
->disks
->index
== (int)idx
)
3563 info
->disk
.raid_disk
= j
;
3565 if (!(ord
& IMSM_ORD_REBUILD
) &&
3566 get_imsm_missing(super
, idx
)) {
3572 if (state
== IMSM_T_STATE_FAILED
)
3574 else if (state
== IMSM_T_STATE_DEGRADED
&&
3575 (state
!= map
->map_state
|| missing
))
3577 else /* we're normal, or already degraded */
3579 if (is_gen_migration(dev
) && missing
) {
3580 /* during general migration we need all disks
3581 * that process is running on.
3582 * No new missing disk is allowed.
3586 /* no more checks necessary
3590 /* in the missing/failed disk case check to see
3591 * if at least one array is runnable
3593 max_enough
= max(max_enough
, enough
);
3595 dprintf("enough: %d\n", max_enough
);
3596 info
->container_enough
= max_enough
;
3599 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3601 disk
= &super
->disks
->disk
;
3602 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3603 info
->component_size
= reserved
;
3604 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3605 /* we don't change info->disk.raid_disk here because
3606 * this state will be finalized in mdmon after we have
3607 * found the 'most fresh' version of the metadata
3609 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3610 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3611 0 : (1 << MD_DISK_SYNC
);
3614 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3615 * ->compare_super may have updated the 'num_raid_devs' field for spares
3617 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3618 uuid_from_super_imsm(st
, info
->uuid
);
3620 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3622 /* I don't know how to compute 'map' on imsm, so use safe default */
3625 for (i
= 0; i
< map_disks
; i
++)
3631 /* allocates memory and fills disk in mdinfo structure
3632 * for each disk in array */
3633 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3635 struct mdinfo
*mddev
;
3636 struct intel_super
*super
= st
->sb
;
3637 struct imsm_disk
*disk
;
3640 if (!super
|| !super
->disks
)
3643 mddev
= xcalloc(1, sizeof(*mddev
));
3647 tmp
= xcalloc(1, sizeof(*tmp
));
3649 tmp
->next
= mddev
->devs
;
3651 tmp
->disk
.number
= count
++;
3652 tmp
->disk
.major
= dl
->major
;
3653 tmp
->disk
.minor
= dl
->minor
;
3654 tmp
->disk
.state
= is_configured(disk
) ?
3655 (1 << MD_DISK_ACTIVE
) : 0;
3656 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3657 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3658 tmp
->disk
.raid_disk
= -1;
3664 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3665 char *update
, char *devname
, int verbose
,
3666 int uuid_set
, char *homehost
)
3668 /* For 'assemble' and 'force' we need to return non-zero if any
3669 * change was made. For others, the return value is ignored.
3670 * Update options are:
3671 * force-one : This device looks a bit old but needs to be included,
3672 * update age info appropriately.
3673 * assemble: clear any 'faulty' flag to allow this device to
3675 * force-array: Array is degraded but being forced, mark it clean
3676 * if that will be needed to assemble it.
3678 * newdev: not used ????
3679 * grow: Array has gained a new device - this is currently for
3681 * resync: mark as dirty so a resync will happen.
3682 * name: update the name - preserving the homehost
3683 * uuid: Change the uuid of the array to match watch is given
3685 * Following are not relevant for this imsm:
3686 * sparc2.2 : update from old dodgey metadata
3687 * super-minor: change the preferred_minor number
3688 * summaries: update redundant counters.
3689 * homehost: update the recorded homehost
3690 * _reshape_progress: record new reshape_progress position.
3693 struct intel_super
*super
= st
->sb
;
3694 struct imsm_super
*mpb
;
3696 /* we can only update container info */
3697 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3700 mpb
= super
->anchor
;
3702 if (strcmp(update
, "uuid") == 0) {
3703 /* We take this to mean that the family_num should be updated.
3704 * However that is much smaller than the uuid so we cannot really
3705 * allow an explicit uuid to be given. And it is hard to reliably
3707 * So if !uuid_set we know the current uuid is random and just used
3708 * the first 'int' and copy it to the other 3 positions.
3709 * Otherwise we require the 4 'int's to be the same as would be the
3710 * case if we are using a random uuid. So an explicit uuid will be
3711 * accepted as long as all for ints are the same... which shouldn't hurt
3714 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3717 if (info
->uuid
[0] != info
->uuid
[1] ||
3718 info
->uuid
[1] != info
->uuid
[2] ||
3719 info
->uuid
[2] != info
->uuid
[3])
3725 mpb
->orig_family_num
= info
->uuid
[0];
3726 } else if (strcmp(update
, "assemble") == 0)
3731 /* successful update? recompute checksum */
3733 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3738 static size_t disks_to_mpb_size(int disks
)
3742 size
= sizeof(struct imsm_super
);
3743 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3744 size
+= 2 * sizeof(struct imsm_dev
);
3745 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3746 size
+= (4 - 2) * sizeof(struct imsm_map
);
3747 /* 4 possible disk_ord_tbl's */
3748 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3749 /* maximum bbm log */
3750 size
+= sizeof(struct bbm_log
);
3755 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3756 unsigned long long data_offset
)
3758 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3761 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3764 static void free_devlist(struct intel_super
*super
)
3766 struct intel_dev
*dv
;
3768 while (super
->devlist
) {
3769 dv
= super
->devlist
->next
;
3770 free(super
->devlist
->dev
);
3771 free(super
->devlist
);
3772 super
->devlist
= dv
;
3776 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3778 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3781 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3785 * 0 same, or first was empty, and second was copied
3786 * 1 second had wrong number
3788 * 3 wrong other info
3790 struct intel_super
*first
= st
->sb
;
3791 struct intel_super
*sec
= tst
->sb
;
3798 /* in platform dependent environment test if the disks
3799 * use the same Intel hba
3800 * If not on Intel hba at all, allow anything.
3802 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3803 if (first
->hba
->type
!= sec
->hba
->type
) {
3805 "HBAs of devices do not match %s != %s\n",
3806 get_sys_dev_type(first
->hba
->type
),
3807 get_sys_dev_type(sec
->hba
->type
));
3810 if (first
->orom
!= sec
->orom
) {
3812 "HBAs of devices do not match %s != %s\n",
3813 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3818 /* if an anchor does not have num_raid_devs set then it is a free
3821 if (first
->anchor
->num_raid_devs
> 0 &&
3822 sec
->anchor
->num_raid_devs
> 0) {
3823 /* Determine if these disks might ever have been
3824 * related. Further disambiguation can only take place
3825 * in load_super_imsm_all
3827 __u32 first_family
= first
->anchor
->orig_family_num
;
3828 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3830 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3831 MAX_SIGNATURE_LENGTH
) != 0)
3834 if (first_family
== 0)
3835 first_family
= first
->anchor
->family_num
;
3836 if (sec_family
== 0)
3837 sec_family
= sec
->anchor
->family_num
;
3839 if (first_family
!= sec_family
)
3844 /* if 'first' is a spare promote it to a populated mpb with sec's
3847 if (first
->anchor
->num_raid_devs
== 0 &&
3848 sec
->anchor
->num_raid_devs
> 0) {
3850 struct intel_dev
*dv
;
3851 struct imsm_dev
*dev
;
3853 /* we need to copy raid device info from sec if an allocation
3854 * fails here we don't associate the spare
3856 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3857 dv
= xmalloc(sizeof(*dv
));
3858 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3861 dv
->next
= first
->devlist
;
3862 first
->devlist
= dv
;
3864 if (i
< sec
->anchor
->num_raid_devs
) {
3865 /* allocation failure */
3866 free_devlist(first
);
3867 pr_err("imsm: failed to associate spare\n");
3870 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3871 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3872 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3873 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3874 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3875 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3881 static void fd2devname(int fd
, char *name
)
3885 char dname
[PATH_MAX
];
3890 if (fstat(fd
, &st
) != 0)
3892 sprintf(path
, "/sys/dev/block/%d:%d",
3893 major(st
.st_rdev
), minor(st
.st_rdev
));
3895 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3900 nm
= strrchr(dname
, '/');
3903 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3907 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3910 char *name
= fd2kname(fd
);
3915 if (strncmp(name
, "nvme", 4) != 0)
3918 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3920 return load_sys(path
, buf
, buf_len
);
3923 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3925 static int imsm_read_serial(int fd
, char *devname
,
3926 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3935 memset(buf
, 0, sizeof(buf
));
3937 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3940 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3942 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3943 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3944 fd2devname(fd
, (char *) serial
);
3950 pr_err("Failed to retrieve serial for %s\n",
3955 /* trim all whitespace and non-printable characters and convert
3958 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
3961 /* ':' is reserved for use in placeholder serial
3962 * numbers for missing disks
3973 /* truncate leading characters */
3974 if (len
> MAX_RAID_SERIAL_LEN
) {
3975 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3976 len
= MAX_RAID_SERIAL_LEN
;
3979 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3980 memcpy(serial
, dest
, len
);
3985 static int serialcmp(__u8
*s1
, __u8
*s2
)
3987 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3990 static void serialcpy(__u8
*dest
, __u8
*src
)
3992 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3995 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3999 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4000 if (serialcmp(dl
->serial
, serial
) == 0)
4006 static struct imsm_disk
*
4007 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
4011 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4012 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4014 if (serialcmp(disk
->serial
, serial
) == 0) {
4025 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4027 struct imsm_disk
*disk
;
4032 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4034 rv
= imsm_read_serial(fd
, devname
, serial
);
4039 dl
= xcalloc(1, sizeof(*dl
));
4042 dl
->major
= major(stb
.st_rdev
);
4043 dl
->minor
= minor(stb
.st_rdev
);
4044 dl
->next
= super
->disks
;
4045 dl
->fd
= keep_fd
? fd
: -1;
4046 assert(super
->disks
== NULL
);
4048 serialcpy(dl
->serial
, serial
);
4051 fd2devname(fd
, name
);
4053 dl
->devname
= xstrdup(devname
);
4055 dl
->devname
= xstrdup(name
);
4057 /* look up this disk's index in the current anchor */
4058 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
4061 /* only set index on disks that are a member of a
4062 * populated contianer, i.e. one with raid_devs
4064 if (is_failed(&dl
->disk
))
4066 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
4073 /* When migrating map0 contains the 'destination' state while map1
4074 * contains the current state. When not migrating map0 contains the
4075 * current state. This routine assumes that map[0].map_state is set to
4076 * the current array state before being called.
4078 * Migration is indicated by one of the following states
4079 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4080 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4081 * map1state=unitialized)
4082 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4084 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4085 * map1state=degraded)
4086 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4089 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4090 __u8 to_state
, int migr_type
)
4092 struct imsm_map
*dest
;
4093 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4095 dev
->vol
.migr_state
= 1;
4096 set_migr_type(dev
, migr_type
);
4097 dev
->vol
.curr_migr_unit
= 0;
4098 dest
= get_imsm_map(dev
, MAP_1
);
4100 /* duplicate and then set the target end state in map[0] */
4101 memcpy(dest
, src
, sizeof_imsm_map(src
));
4102 if (migr_type
== MIGR_GEN_MIGR
) {
4106 for (i
= 0; i
< src
->num_members
; i
++) {
4107 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4108 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4112 if (migr_type
== MIGR_GEN_MIGR
)
4113 /* Clear migration record */
4114 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4116 src
->map_state
= to_state
;
4119 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4122 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4123 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4127 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4128 * completed in the last migration.
4130 * FIXME add support for raid-level-migration
4132 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4133 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4134 /* when final map state is other than expected
4135 * merge maps (not for migration)
4139 for (i
= 0; i
< prev
->num_members
; i
++)
4140 for (j
= 0; j
< map
->num_members
; j
++)
4141 /* during online capacity expansion
4142 * disks position can be changed
4143 * if takeover is used
4145 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4146 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4147 map
->disk_ord_tbl
[j
] |=
4148 prev
->disk_ord_tbl
[i
];
4151 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4152 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4155 dev
->vol
.migr_state
= 0;
4156 set_migr_type(dev
, 0);
4157 dev
->vol
.curr_migr_unit
= 0;
4158 map
->map_state
= map_state
;
4161 static int parse_raid_devices(struct intel_super
*super
)
4164 struct imsm_dev
*dev_new
;
4165 size_t len
, len_migr
;
4167 size_t space_needed
= 0;
4168 struct imsm_super
*mpb
= super
->anchor
;
4170 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4171 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4172 struct intel_dev
*dv
;
4174 len
= sizeof_imsm_dev(dev_iter
, 0);
4175 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4177 space_needed
+= len_migr
- len
;
4179 dv
= xmalloc(sizeof(*dv
));
4180 if (max_len
< len_migr
)
4182 if (max_len
> len_migr
)
4183 space_needed
+= max_len
- len_migr
;
4184 dev_new
= xmalloc(max_len
);
4185 imsm_copy_dev(dev_new
, dev_iter
);
4188 dv
->next
= super
->devlist
;
4189 super
->devlist
= dv
;
4192 /* ensure that super->buf is large enough when all raid devices
4195 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4198 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4199 super
->sector_size
);
4200 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4203 memcpy(buf
, super
->buf
, super
->len
);
4204 memset(buf
+ super
->len
, 0, len
- super
->len
);
4210 super
->extra_space
+= space_needed
;
4215 /*******************************************************************************
4216 * Function: check_mpb_migr_compatibility
4217 * Description: Function checks for unsupported migration features:
4218 * - migration optimization area (pba_of_lba0)
4219 * - descending reshape (ascending_migr)
4221 * super : imsm metadata information
4223 * 0 : migration is compatible
4224 * -1 : migration is not compatible
4225 ******************************************************************************/
4226 int check_mpb_migr_compatibility(struct intel_super
*super
)
4228 struct imsm_map
*map0
, *map1
;
4229 struct migr_record
*migr_rec
= super
->migr_rec
;
4232 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4233 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4236 dev_iter
->vol
.migr_state
== 1 &&
4237 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4238 /* This device is migrating */
4239 map0
= get_imsm_map(dev_iter
, MAP_0
);
4240 map1
= get_imsm_map(dev_iter
, MAP_1
);
4241 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4242 /* migration optimization area was used */
4244 if (migr_rec
->ascending_migr
== 0 &&
4245 migr_rec
->dest_depth_per_unit
> 0)
4246 /* descending reshape not supported yet */
4253 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4255 /* load_imsm_mpb - read matrix metadata
4256 * allocates super->mpb to be freed by free_imsm
4258 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4260 unsigned long long dsize
;
4261 unsigned long long sectors
;
4262 unsigned int sector_size
= super
->sector_size
;
4264 struct imsm_super
*anchor
;
4267 get_dev_size(fd
, NULL
, &dsize
);
4268 if (dsize
< 2*sector_size
) {
4270 pr_err("%s: device to small for imsm\n",
4275 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4277 pr_err("Cannot seek to anchor block on %s: %s\n",
4278 devname
, strerror(errno
));
4282 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4284 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4287 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4289 pr_err("Cannot read anchor block on %s: %s\n",
4290 devname
, strerror(errno
));
4295 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4297 pr_err("no IMSM anchor on %s\n", devname
);
4302 __free_imsm(super
, 0);
4303 /* reload capability and hba */
4305 /* capability and hba must be updated with new super allocation */
4306 find_intel_hba_capability(fd
, super
, devname
);
4307 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4308 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4310 pr_err("unable to allocate %zu byte mpb buffer\n",
4315 memcpy(super
->buf
, anchor
, sector_size
);
4317 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4320 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4321 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4322 pr_err("could not allocate migr_rec buffer\n");
4326 super
->clean_migration_record_by_mdmon
= 0;
4329 check_sum
= __gen_imsm_checksum(super
->anchor
);
4330 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4332 pr_err("IMSM checksum %x != %x on %s\n",
4334 __le32_to_cpu(super
->anchor
->check_sum
),
4342 /* read the extended mpb */
4343 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4345 pr_err("Cannot seek to extended mpb on %s: %s\n",
4346 devname
, strerror(errno
));
4350 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4351 super
->len
- sector_size
) != super
->len
- sector_size
) {
4353 pr_err("Cannot read extended mpb on %s: %s\n",
4354 devname
, strerror(errno
));
4358 check_sum
= __gen_imsm_checksum(super
->anchor
);
4359 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4361 pr_err("IMSM checksum %x != %x on %s\n",
4362 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4370 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4372 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4373 static void clear_hi(struct intel_super
*super
)
4375 struct imsm_super
*mpb
= super
->anchor
;
4377 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4379 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4380 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4381 disk
->total_blocks_hi
= 0;
4383 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4384 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4387 for (n
= 0; n
< 2; ++n
) {
4388 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4391 map
->pba_of_lba0_hi
= 0;
4392 map
->blocks_per_member_hi
= 0;
4393 map
->num_data_stripes_hi
= 0;
4399 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4403 err
= load_imsm_mpb(fd
, super
, devname
);
4406 if (super
->sector_size
== 4096)
4407 convert_from_4k(super
);
4408 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4411 err
= parse_raid_devices(super
);
4414 err
= load_bbm_log(super
);
4419 static void __free_imsm_disk(struct dl
*d
)
4431 static void free_imsm_disks(struct intel_super
*super
)
4435 while (super
->disks
) {
4437 super
->disks
= d
->next
;
4438 __free_imsm_disk(d
);
4440 while (super
->disk_mgmt_list
) {
4441 d
= super
->disk_mgmt_list
;
4442 super
->disk_mgmt_list
= d
->next
;
4443 __free_imsm_disk(d
);
4445 while (super
->missing
) {
4447 super
->missing
= d
->next
;
4448 __free_imsm_disk(d
);
4453 /* free all the pieces hanging off of a super pointer */
4454 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4456 struct intel_hba
*elem
, *next
;
4462 /* unlink capability description */
4464 if (super
->migr_rec_buf
) {
4465 free(super
->migr_rec_buf
);
4466 super
->migr_rec_buf
= NULL
;
4469 free_imsm_disks(super
);
4470 free_devlist(super
);
4474 free((void *)elem
->path
);
4480 free(super
->bbm_log
);
4484 static void free_imsm(struct intel_super
*super
)
4486 __free_imsm(super
, 1);
4487 free(super
->bb
.entries
);
4491 static void free_super_imsm(struct supertype
*st
)
4493 struct intel_super
*super
= st
->sb
;
4502 static struct intel_super
*alloc_super(void)
4504 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4506 super
->current_vol
= -1;
4507 super
->create_offset
= ~((unsigned long long) 0);
4509 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4510 sizeof(struct md_bb_entry
));
4511 if (!super
->bb
.entries
) {
4520 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4522 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4524 struct sys_dev
*hba_name
;
4527 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4532 hba_name
= find_disk_attached_hba(fd
, NULL
);
4535 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4539 rv
= attach_hba_to_super(super
, hba_name
);
4542 struct intel_hba
*hba
= super
->hba
;
4544 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4545 " but the container is assigned to Intel(R) %s %s (",
4547 get_sys_dev_type(hba_name
->type
),
4548 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4549 hba_name
->pci_id
? : "Err!",
4550 get_sys_dev_type(super
->hba
->type
),
4551 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4554 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4556 fprintf(stderr
, ", ");
4559 fprintf(stderr
, ").\n"
4560 " Mixing devices attached to different controllers is not allowed.\n");
4564 super
->orom
= find_imsm_capability(hba_name
);
4571 /* find_missing - helper routine for load_super_imsm_all that identifies
4572 * disks that have disappeared from the system. This routine relies on
4573 * the mpb being uptodate, which it is at load time.
4575 static int find_missing(struct intel_super
*super
)
4578 struct imsm_super
*mpb
= super
->anchor
;
4580 struct imsm_disk
*disk
;
4582 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4583 disk
= __get_imsm_disk(mpb
, i
);
4584 dl
= serial_to_dl(disk
->serial
, super
);
4588 dl
= xmalloc(sizeof(*dl
));
4592 dl
->devname
= xstrdup("missing");
4594 serialcpy(dl
->serial
, disk
->serial
);
4597 dl
->next
= super
->missing
;
4598 super
->missing
= dl
;
4604 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4606 struct intel_disk
*idisk
= disk_list
;
4609 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4611 idisk
= idisk
->next
;
4617 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4618 struct intel_super
*super
,
4619 struct intel_disk
**disk_list
)
4621 struct imsm_disk
*d
= &super
->disks
->disk
;
4622 struct imsm_super
*mpb
= super
->anchor
;
4625 for (i
= 0; i
< tbl_size
; i
++) {
4626 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4627 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4629 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4630 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4631 dprintf("mpb from %d:%d matches %d:%d\n",
4632 super
->disks
->major
,
4633 super
->disks
->minor
,
4634 table
[i
]->disks
->major
,
4635 table
[i
]->disks
->minor
);
4639 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4640 is_configured(d
) == is_configured(tbl_d
)) &&
4641 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4642 /* current version of the mpb is a
4643 * better candidate than the one in
4644 * super_table, but copy over "cross
4645 * generational" status
4647 struct intel_disk
*idisk
;
4649 dprintf("mpb from %d:%d replaces %d:%d\n",
4650 super
->disks
->major
,
4651 super
->disks
->minor
,
4652 table
[i
]->disks
->major
,
4653 table
[i
]->disks
->minor
);
4655 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4656 if (idisk
&& is_failed(&idisk
->disk
))
4657 tbl_d
->status
|= FAILED_DISK
;
4660 struct intel_disk
*idisk
;
4661 struct imsm_disk
*disk
;
4663 /* tbl_mpb is more up to date, but copy
4664 * over cross generational status before
4667 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4668 if (disk
&& is_failed(disk
))
4669 d
->status
|= FAILED_DISK
;
4671 idisk
= disk_list_get(d
->serial
, *disk_list
);
4674 if (disk
&& is_configured(disk
))
4675 idisk
->disk
.status
|= CONFIGURED_DISK
;
4678 dprintf("mpb from %d:%d prefer %d:%d\n",
4679 super
->disks
->major
,
4680 super
->disks
->minor
,
4681 table
[i
]->disks
->major
,
4682 table
[i
]->disks
->minor
);
4690 table
[tbl_size
++] = super
;
4694 /* update/extend the merged list of imsm_disk records */
4695 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4696 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4697 struct intel_disk
*idisk
;
4699 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4701 idisk
->disk
.status
|= disk
->status
;
4702 if (is_configured(&idisk
->disk
) ||
4703 is_failed(&idisk
->disk
))
4704 idisk
->disk
.status
&= ~(SPARE_DISK
);
4706 idisk
= xcalloc(1, sizeof(*idisk
));
4707 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4708 idisk
->disk
= *disk
;
4709 idisk
->next
= *disk_list
;
4713 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4720 static struct intel_super
*
4721 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4724 struct imsm_super
*mpb
= super
->anchor
;
4728 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4729 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4730 struct intel_disk
*idisk
;
4732 idisk
= disk_list_get(disk
->serial
, disk_list
);
4734 if (idisk
->owner
== owner
||
4735 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4738 dprintf("'%.16s' owner %d != %d\n",
4739 disk
->serial
, idisk
->owner
,
4742 dprintf("unknown disk %x [%d]: %.16s\n",
4743 __le32_to_cpu(mpb
->family_num
), i
,
4749 if (ok_count
== mpb
->num_disks
)
4754 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4756 struct intel_super
*s
;
4758 for (s
= super_list
; s
; s
= s
->next
) {
4759 if (family_num
!= s
->anchor
->family_num
)
4761 pr_err("Conflict, offlining family %#x on '%s'\n",
4762 __le32_to_cpu(family_num
), s
->disks
->devname
);
4766 static struct intel_super
*
4767 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4769 struct intel_super
*super_table
[len
];
4770 struct intel_disk
*disk_list
= NULL
;
4771 struct intel_super
*champion
, *spare
;
4772 struct intel_super
*s
, **del
;
4777 memset(super_table
, 0, sizeof(super_table
));
4778 for (s
= *super_list
; s
; s
= s
->next
)
4779 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4781 for (i
= 0; i
< tbl_size
; i
++) {
4782 struct imsm_disk
*d
;
4783 struct intel_disk
*idisk
;
4784 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4787 d
= &s
->disks
->disk
;
4789 /* 'd' must appear in merged disk list for its
4790 * configuration to be valid
4792 idisk
= disk_list_get(d
->serial
, disk_list
);
4793 if (idisk
&& idisk
->owner
== i
)
4794 s
= validate_members(s
, disk_list
, i
);
4799 dprintf("marking family: %#x from %d:%d offline\n",
4801 super_table
[i
]->disks
->major
,
4802 super_table
[i
]->disks
->minor
);
4806 /* This is where the mdadm implementation differs from the Windows
4807 * driver which has no strict concept of a container. We can only
4808 * assemble one family from a container, so when returning a prodigal
4809 * array member to this system the code will not be able to disambiguate
4810 * the container contents that should be assembled ("foreign" versus
4811 * "local"). It requires user intervention to set the orig_family_num
4812 * to a new value to establish a new container. The Windows driver in
4813 * this situation fixes up the volume name in place and manages the
4814 * foreign array as an independent entity.
4819 for (i
= 0; i
< tbl_size
; i
++) {
4820 struct intel_super
*tbl_ent
= super_table
[i
];
4826 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4831 if (s
&& !is_spare
) {
4832 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4834 } else if (!s
&& !is_spare
)
4847 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4848 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4850 /* collect all dl's onto 'champion', and update them to
4851 * champion's version of the status
4853 for (s
= *super_list
; s
; s
= s
->next
) {
4854 struct imsm_super
*mpb
= champion
->anchor
;
4855 struct dl
*dl
= s
->disks
;
4860 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4862 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4863 struct imsm_disk
*disk
;
4865 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4868 /* only set index on disks that are a member of
4869 * a populated contianer, i.e. one with
4872 if (is_failed(&dl
->disk
))
4874 else if (is_spare(&dl
->disk
))
4880 if (i
>= mpb
->num_disks
) {
4881 struct intel_disk
*idisk
;
4883 idisk
= disk_list_get(dl
->serial
, disk_list
);
4884 if (idisk
&& is_spare(&idisk
->disk
) &&
4885 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4893 dl
->next
= champion
->disks
;
4894 champion
->disks
= dl
;
4898 /* delete 'champion' from super_list */
4899 for (del
= super_list
; *del
; ) {
4900 if (*del
== champion
) {
4901 *del
= (*del
)->next
;
4904 del
= &(*del
)->next
;
4906 champion
->next
= NULL
;
4910 struct intel_disk
*idisk
= disk_list
;
4912 disk_list
= disk_list
->next
;
4920 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4921 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4922 int major
, int minor
, int keep_fd
);
4924 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4925 int *max
, int keep_fd
);
4927 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4928 char *devname
, struct md_list
*devlist
,
4931 struct intel_super
*super_list
= NULL
;
4932 struct intel_super
*super
= NULL
;
4937 /* 'fd' is an opened container */
4938 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4940 /* get super block from devlist devices */
4941 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4944 /* all mpbs enter, maybe one leaves */
4945 super
= imsm_thunderdome(&super_list
, i
);
4951 if (find_missing(super
) != 0) {
4957 /* load migration record */
4958 err
= load_imsm_migr_rec(super
, NULL
);
4960 /* migration is in progress,
4961 * but migr_rec cannot be loaded,
4967 /* Check migration compatibility */
4968 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
4969 pr_err("Unsupported migration detected");
4971 fprintf(stderr
, " on %s\n", devname
);
4973 fprintf(stderr
, " (IMSM).\n");
4982 while (super_list
) {
4983 struct intel_super
*s
= super_list
;
4985 super_list
= super_list
->next
;
4994 strcpy(st
->container_devnm
, fd2devnm(fd
));
4996 st
->container_devnm
[0] = 0;
4997 if (err
== 0 && st
->ss
== NULL
) {
4998 st
->ss
= &super_imsm
;
4999 st
->minor_version
= 0;
5000 st
->max_devs
= IMSM_MAX_DEVICES
;
5006 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
5007 int *max
, int keep_fd
)
5009 struct md_list
*tmpdev
;
5013 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5014 if (tmpdev
->used
!= 1)
5016 if (tmpdev
->container
== 1) {
5018 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
5020 pr_err("cannot open device %s: %s\n",
5021 tmpdev
->devname
, strerror(errno
));
5025 err
= get_sra_super_block(fd
, super_list
,
5026 tmpdev
->devname
, &lmax
,
5035 int major
= major(tmpdev
->st_rdev
);
5036 int minor
= minor(tmpdev
->st_rdev
);
5037 err
= get_super_block(super_list
,
5054 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
5055 int major
, int minor
, int keep_fd
)
5057 struct intel_super
*s
;
5069 sprintf(nm
, "%d:%d", major
, minor
);
5070 dfd
= dev_open(nm
, O_RDWR
);
5076 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5077 find_intel_hba_capability(dfd
, s
, devname
);
5078 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5080 /* retry the load if we might have raced against mdmon */
5081 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5082 for (retry
= 0; retry
< 3; retry
++) {
5084 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5090 s
->next
= *super_list
;
5098 if (dfd
>= 0 && !keep_fd
)
5105 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5112 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5116 if (sra
->array
.major_version
!= -1 ||
5117 sra
->array
.minor_version
!= -2 ||
5118 strcmp(sra
->text_version
, "imsm") != 0) {
5123 devnm
= fd2devnm(fd
);
5124 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5125 if (get_super_block(super_list
, devnm
, devname
,
5126 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5137 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5139 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5142 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5144 struct intel_super
*super
;
5148 if (test_partition(fd
))
5149 /* IMSM not allowed on partitions */
5152 free_super_imsm(st
);
5154 super
= alloc_super();
5155 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5158 /* Load hba and capabilities if they exist.
5159 * But do not preclude loading metadata in case capabilities or hba are
5160 * non-compliant and ignore_hw_compat is set.
5162 rv
= find_intel_hba_capability(fd
, super
, devname
);
5163 /* no orom/efi or non-intel hba of the disk */
5164 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5166 pr_err("No OROM/EFI properties for %s\n", devname
);
5170 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5172 /* retry the load if we might have raced against mdmon */
5174 struct mdstat_ent
*mdstat
= NULL
;
5175 char *name
= fd2kname(fd
);
5178 mdstat
= mdstat_by_component(name
);
5180 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5181 for (retry
= 0; retry
< 3; retry
++) {
5183 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5189 free_mdstat(mdstat
);
5194 pr_err("Failed to load all information sections on %s\n", devname
);
5200 if (st
->ss
== NULL
) {
5201 st
->ss
= &super_imsm
;
5202 st
->minor_version
= 0;
5203 st
->max_devs
= IMSM_MAX_DEVICES
;
5206 /* load migration record */
5207 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5208 /* Check for unsupported migration features */
5209 if (check_mpb_migr_compatibility(super
) != 0) {
5210 pr_err("Unsupported migration detected");
5212 fprintf(stderr
, " on %s\n", devname
);
5214 fprintf(stderr
, " (IMSM).\n");
5222 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5224 if (info
->level
== 1)
5226 return info
->chunk_size
>> 9;
5229 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5230 unsigned long long size
)
5232 if (info
->level
== 1)
5235 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5238 static void imsm_update_version_info(struct intel_super
*super
)
5240 /* update the version and attributes */
5241 struct imsm_super
*mpb
= super
->anchor
;
5243 struct imsm_dev
*dev
;
5244 struct imsm_map
*map
;
5247 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5248 dev
= get_imsm_dev(super
, i
);
5249 map
= get_imsm_map(dev
, MAP_0
);
5250 if (__le32_to_cpu(dev
->size_high
) > 0)
5251 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5253 /* FIXME detect when an array spans a port multiplier */
5255 mpb
->attributes
|= MPB_ATTRIB_PM
;
5258 if (mpb
->num_raid_devs
> 1 ||
5259 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5260 version
= MPB_VERSION_ATTRIBS
;
5261 switch (get_imsm_raid_level(map
)) {
5262 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5263 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5264 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5265 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5268 if (map
->num_members
>= 5)
5269 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5270 else if (dev
->status
== DEV_CLONE_N_GO
)
5271 version
= MPB_VERSION_CNG
;
5272 else if (get_imsm_raid_level(map
) == 5)
5273 version
= MPB_VERSION_RAID5
;
5274 else if (map
->num_members
>= 3)
5275 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5276 else if (get_imsm_raid_level(map
) == 1)
5277 version
= MPB_VERSION_RAID1
;
5279 version
= MPB_VERSION_RAID0
;
5281 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5285 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5287 struct imsm_super
*mpb
= super
->anchor
;
5288 char *reason
= NULL
;
5291 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
5292 reason
= "must be 16 characters or less";
5294 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5295 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5297 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5298 reason
= "already exists";
5303 if (reason
&& !quiet
)
5304 pr_err("imsm volume name %s\n", reason
);
5309 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5310 struct shape
*s
, char *name
,
5311 char *homehost
, int *uuid
,
5312 long long data_offset
)
5314 /* We are creating a volume inside a pre-existing container.
5315 * so st->sb is already set.
5317 struct intel_super
*super
= st
->sb
;
5318 unsigned int sector_size
= super
->sector_size
;
5319 struct imsm_super
*mpb
= super
->anchor
;
5320 struct intel_dev
*dv
;
5321 struct imsm_dev
*dev
;
5322 struct imsm_vol
*vol
;
5323 struct imsm_map
*map
;
5324 int idx
= mpb
->num_raid_devs
;
5327 unsigned long long array_blocks
;
5328 size_t size_old
, size_new
;
5329 unsigned long long num_data_stripes
;
5330 unsigned int data_disks
;
5331 unsigned long long size_per_member
;
5333 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5334 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5338 /* ensure the mpb is large enough for the new data */
5339 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5340 size_new
= disks_to_mpb_size(info
->nr_disks
);
5341 if (size_new
> size_old
) {
5343 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5345 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5346 pr_err("could not allocate new mpb\n");
5349 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5350 MIGR_REC_BUF_SECTORS
*
5351 MAX_SECTOR_SIZE
) != 0) {
5352 pr_err("could not allocate migr_rec buffer\n");
5358 memcpy(mpb_new
, mpb
, size_old
);
5361 super
->anchor
= mpb_new
;
5362 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5363 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5364 super
->len
= size_round
;
5366 super
->current_vol
= idx
;
5368 /* handle 'failed_disks' by either:
5369 * a) create dummy disk entries in the table if this the first
5370 * volume in the array. We add them here as this is the only
5371 * opportunity to add them. add_to_super_imsm_volume()
5372 * handles the non-failed disks and continues incrementing
5374 * b) validate that 'failed_disks' matches the current number
5375 * of missing disks if the container is populated
5377 if (super
->current_vol
== 0) {
5379 for (i
= 0; i
< info
->failed_disks
; i
++) {
5380 struct imsm_disk
*disk
;
5383 disk
= __get_imsm_disk(mpb
, i
);
5384 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5385 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5386 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5387 "missing:%d", (__u8
)i
);
5389 find_missing(super
);
5394 for (d
= super
->missing
; d
; d
= d
->next
)
5396 if (info
->failed_disks
> missing
) {
5397 pr_err("unable to add 'missing' disk to container\n");
5402 if (!check_name(super
, name
, 0))
5404 dv
= xmalloc(sizeof(*dv
));
5405 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5407 * Explicitly allow truncating to not confuse gcc's
5408 * -Werror=stringop-truncation
5410 namelen
= min((int) strlen(name
), MAX_RAID_SERIAL_LEN
);
5411 memcpy(dev
->volume
, name
, namelen
);
5412 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5413 info
->layout
, info
->chunk_size
,
5414 s
->size
* BLOCKS_PER_KB
);
5415 data_disks
= get_data_disks(info
->level
, info
->layout
,
5417 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5418 size_per_member
= array_blocks
/ data_disks
;
5420 set_imsm_dev_size(dev
, array_blocks
);
5421 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5423 vol
->migr_state
= 0;
5424 set_migr_type(dev
, MIGR_INIT
);
5425 vol
->dirty
= !info
->state
;
5426 vol
->curr_migr_unit
= 0;
5427 map
= get_imsm_map(dev
, MAP_0
);
5428 set_pba_of_lba0(map
, super
->create_offset
);
5429 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5430 map
->failed_disk_num
= ~0;
5431 if (info
->level
> 0)
5432 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5433 : IMSM_T_STATE_UNINITIALIZED
);
5435 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5436 IMSM_T_STATE_NORMAL
;
5439 if (info
->level
== 1 && info
->raid_disks
> 2) {
5442 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5446 map
->raid_level
= info
->level
;
5447 if (info
->level
== 10) {
5448 map
->raid_level
= 1;
5449 map
->num_domains
= info
->raid_disks
/ 2;
5450 } else if (info
->level
== 1)
5451 map
->num_domains
= info
->raid_disks
;
5453 map
->num_domains
= 1;
5455 /* info->size is only int so use the 'size' parameter instead */
5456 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5457 num_data_stripes
/= map
->num_domains
;
5458 set_num_data_stripes(map
, num_data_stripes
);
5460 size_per_member
+= NUM_BLOCKS_DIRTY_STRIPE_REGION
;
5461 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5465 map
->num_members
= info
->raid_disks
;
5466 for (i
= 0; i
< map
->num_members
; i
++) {
5467 /* initialized in add_to_super */
5468 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5470 mpb
->num_raid_devs
++;
5471 mpb
->num_raid_devs_created
++;
5472 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5474 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5475 dev
->rwh_policy
= RWH_MULTIPLE_OFF
;
5476 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5477 dev
->rwh_policy
= RWH_MULTIPLE_DISTRIBUTED
;
5481 pr_err("imsm does not support consistency policy %s\n",
5482 map_num(consistency_policies
, s
->consistency_policy
));
5487 dv
->index
= super
->current_vol
;
5488 dv
->next
= super
->devlist
;
5489 super
->devlist
= dv
;
5491 imsm_update_version_info(super
);
5496 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5497 struct shape
*s
, char *name
,
5498 char *homehost
, int *uuid
,
5499 unsigned long long data_offset
)
5501 /* This is primarily called by Create when creating a new array.
5502 * We will then get add_to_super called for each component, and then
5503 * write_init_super called to write it out to each device.
5504 * For IMSM, Create can create on fresh devices or on a pre-existing
5506 * To create on a pre-existing array a different method will be called.
5507 * This one is just for fresh drives.
5509 struct intel_super
*super
;
5510 struct imsm_super
*mpb
;
5514 if (data_offset
!= INVALID_SECTORS
) {
5515 pr_err("data-offset not supported by imsm\n");
5520 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5524 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5526 mpb_size
= MAX_SECTOR_SIZE
;
5528 super
= alloc_super();
5530 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5535 pr_err("could not allocate superblock\n");
5538 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5539 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5540 pr_err("could not allocate migr_rec buffer\n");
5545 memset(super
->buf
, 0, mpb_size
);
5547 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5551 /* zeroing superblock */
5555 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5557 version
= (char *) mpb
->sig
;
5558 strcpy(version
, MPB_SIGNATURE
);
5559 version
+= strlen(MPB_SIGNATURE
);
5560 strcpy(version
, MPB_VERSION_RAID0
);
5565 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5567 unsigned int member_sector_size
;
5570 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5574 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5576 if (member_sector_size
!= super
->sector_size
)
5581 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5582 int fd
, char *devname
)
5584 struct intel_super
*super
= st
->sb
;
5585 struct imsm_super
*mpb
= super
->anchor
;
5586 struct imsm_disk
*_disk
;
5587 struct imsm_dev
*dev
;
5588 struct imsm_map
*map
;
5592 dev
= get_imsm_dev(super
, super
->current_vol
);
5593 map
= get_imsm_map(dev
, MAP_0
);
5595 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5596 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5602 /* we're doing autolayout so grab the pre-marked (in
5603 * validate_geometry) raid_disk
5605 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5606 if (dl
->raiddisk
== dk
->raid_disk
)
5609 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5610 if (dl
->major
== dk
->major
&&
5611 dl
->minor
== dk
->minor
)
5616 pr_err("%s is not a member of the same container\n", devname
);
5620 if (!drive_validate_sector_size(super
, dl
)) {
5621 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5625 /* add a pristine spare to the metadata */
5626 if (dl
->index
< 0) {
5627 dl
->index
= super
->anchor
->num_disks
;
5628 super
->anchor
->num_disks
++;
5630 /* Check the device has not already been added */
5631 slot
= get_imsm_disk_slot(map
, dl
->index
);
5633 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5634 pr_err("%s has been included in this array twice\n",
5638 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5639 dl
->disk
.status
= CONFIGURED_DISK
;
5641 /* update size of 'missing' disks to be at least as large as the
5642 * largest acitve member (we only have dummy missing disks when
5643 * creating the first volume)
5645 if (super
->current_vol
== 0) {
5646 for (df
= super
->missing
; df
; df
= df
->next
) {
5647 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5648 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5649 _disk
= __get_imsm_disk(mpb
, df
->index
);
5654 /* refresh unset/failed slots to point to valid 'missing' entries */
5655 for (df
= super
->missing
; df
; df
= df
->next
)
5656 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5657 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5659 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5661 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5662 if (is_gen_migration(dev
)) {
5663 struct imsm_map
*map2
= get_imsm_map(dev
,
5665 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5666 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5667 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5670 if ((unsigned)df
->index
==
5672 set_imsm_ord_tbl_ent(map2
,
5678 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5682 /* if we are creating the first raid device update the family number */
5683 if (super
->current_vol
== 0) {
5685 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5687 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5688 if (!_dev
|| !_disk
) {
5689 pr_err("BUG mpb setup error\n");
5695 sum
+= __gen_imsm_checksum(mpb
);
5696 mpb
->family_num
= __cpu_to_le32(sum
);
5697 mpb
->orig_family_num
= mpb
->family_num
;
5699 super
->current_disk
= dl
;
5704 * Function marks disk as spare and restores disk serial
5705 * in case it was previously marked as failed by takeover operation
5707 * -1 : critical error
5708 * 0 : disk is marked as spare but serial is not set
5711 int mark_spare(struct dl
*disk
)
5713 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5720 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5721 /* Restore disk serial number, because takeover marks disk
5722 * as failed and adds to serial ':0' before it becomes
5725 serialcpy(disk
->serial
, serial
);
5726 serialcpy(disk
->disk
.serial
, serial
);
5729 disk
->disk
.status
= SPARE_DISK
;
5735 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5736 int fd
, char *devname
,
5737 unsigned long long data_offset
)
5739 struct intel_super
*super
= st
->sb
;
5741 unsigned long long size
;
5742 unsigned int member_sector_size
;
5747 /* If we are on an RAID enabled platform check that the disk is
5748 * attached to the raid controller.
5749 * We do not need to test disks attachment for container based additions,
5750 * they shall be already tested when container was created/assembled.
5752 rv
= find_intel_hba_capability(fd
, super
, devname
);
5753 /* no orom/efi or non-intel hba of the disk */
5755 dprintf("capability: %p fd: %d ret: %d\n",
5756 super
->orom
, fd
, rv
);
5760 if (super
->current_vol
>= 0)
5761 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5764 dd
= xcalloc(sizeof(*dd
), 1);
5765 dd
->major
= major(stb
.st_rdev
);
5766 dd
->minor
= minor(stb
.st_rdev
);
5767 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5770 dd
->action
= DISK_ADD
;
5771 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5773 pr_err("failed to retrieve scsi serial, aborting\n");
5779 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5780 (super
->hba
->type
== SYS_DEV_VMD
))) {
5782 char *devpath
= diskfd_to_devpath(fd
);
5783 char controller_path
[PATH_MAX
];
5786 pr_err("failed to get devpath, aborting\n");
5793 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5796 if (devpath_to_vendor(controller_path
) == 0x8086) {
5798 * If Intel's NVMe drive has serial ended with
5799 * "-A","-B","-1" or "-2" it means that this is "x8"
5800 * device (double drive on single PCIe card).
5801 * User should be warned about potential data loss.
5803 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5804 /* Skip empty character at the end */
5805 if (dd
->serial
[i
] == 0)
5808 if (((dd
->serial
[i
] == 'A') ||
5809 (dd
->serial
[i
] == 'B') ||
5810 (dd
->serial
[i
] == '1') ||
5811 (dd
->serial
[i
] == '2')) &&
5812 (dd
->serial
[i
-1] == '-'))
5813 pr_err("\tThe action you are about to take may put your data at risk.\n"
5814 "\tPlease note that x8 devices may consist of two separate x4 devices "
5815 "located on a single PCIe port.\n"
5816 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5819 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5820 !imsm_orom_has_tpv_support(super
->orom
)) {
5821 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5822 "\tPlease refer to Intel(R) RSTe/VROC user guide.\n");
5829 get_dev_size(fd
, NULL
, &size
);
5830 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5832 if (super
->sector_size
== 0) {
5833 /* this a first device, so sector_size is not set yet */
5834 super
->sector_size
= member_sector_size
;
5837 /* clear migr_rec when adding disk to container */
5838 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5839 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5841 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5842 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5843 MIGR_REC_BUF_SECTORS
*member_sector_size
)
5844 perror("Write migr_rec failed");
5848 serialcpy(dd
->disk
.serial
, dd
->serial
);
5849 set_total_blocks(&dd
->disk
, size
);
5850 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5851 struct imsm_super
*mpb
= super
->anchor
;
5852 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5855 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5856 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5858 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5860 if (st
->update_tail
) {
5861 dd
->next
= super
->disk_mgmt_list
;
5862 super
->disk_mgmt_list
= dd
;
5864 dd
->next
= super
->disks
;
5866 super
->updates_pending
++;
5872 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5874 struct intel_super
*super
= st
->sb
;
5877 /* remove from super works only in mdmon - for communication
5878 * manager - monitor. Check if communication memory buffer
5881 if (!st
->update_tail
) {
5882 pr_err("shall be used in mdmon context only\n");
5885 dd
= xcalloc(1, sizeof(*dd
));
5886 dd
->major
= dk
->major
;
5887 dd
->minor
= dk
->minor
;
5890 dd
->action
= DISK_REMOVE
;
5892 dd
->next
= super
->disk_mgmt_list
;
5893 super
->disk_mgmt_list
= dd
;
5898 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5901 char buf
[MAX_SECTOR_SIZE
];
5902 struct imsm_super anchor
;
5903 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5905 /* spare records have their own family number and do not have any defined raid
5908 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5910 struct imsm_super
*mpb
= super
->anchor
;
5911 struct imsm_super
*spare
= &spare_record
.anchor
;
5915 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5916 spare
->generation_num
= __cpu_to_le32(1UL);
5917 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5918 spare
->num_disks
= 1;
5919 spare
->num_raid_devs
= 0;
5920 spare
->cache_size
= mpb
->cache_size
;
5921 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5923 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5924 MPB_SIGNATURE MPB_VERSION_RAID0
);
5926 for (d
= super
->disks
; d
; d
= d
->next
) {
5930 spare
->disk
[0] = d
->disk
;
5931 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5932 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5934 if (super
->sector_size
== 4096)
5935 convert_to_4k_imsm_disk(&spare
->disk
[0]);
5937 sum
= __gen_imsm_checksum(spare
);
5938 spare
->family_num
= __cpu_to_le32(sum
);
5939 spare
->orig_family_num
= 0;
5940 sum
= __gen_imsm_checksum(spare
);
5941 spare
->check_sum
= __cpu_to_le32(sum
);
5943 if (store_imsm_mpb(d
->fd
, spare
)) {
5944 pr_err("failed for device %d:%d %s\n",
5945 d
->major
, d
->minor
, strerror(errno
));
5957 static int write_super_imsm(struct supertype
*st
, int doclose
)
5959 struct intel_super
*super
= st
->sb
;
5960 unsigned int sector_size
= super
->sector_size
;
5961 struct imsm_super
*mpb
= super
->anchor
;
5967 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5969 int clear_migration_record
= 1;
5972 /* 'generation' is incremented everytime the metadata is written */
5973 generation
= __le32_to_cpu(mpb
->generation_num
);
5975 mpb
->generation_num
= __cpu_to_le32(generation
);
5977 /* fix up cases where previous mdadm releases failed to set
5980 if (mpb
->orig_family_num
== 0)
5981 mpb
->orig_family_num
= mpb
->family_num
;
5983 for (d
= super
->disks
; d
; d
= d
->next
) {
5987 mpb
->disk
[d
->index
] = d
->disk
;
5991 for (d
= super
->missing
; d
; d
= d
->next
) {
5992 mpb
->disk
[d
->index
] = d
->disk
;
5995 mpb
->num_disks
= num_disks
;
5996 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5998 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5999 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
6000 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
6002 imsm_copy_dev(dev
, dev2
);
6003 mpb_size
+= sizeof_imsm_dev(dev
, 0);
6005 if (is_gen_migration(dev2
))
6006 clear_migration_record
= 0;
6009 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
6012 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
6013 mpb
->attributes
|= MPB_ATTRIB_BBM
;
6015 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
6017 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
6018 mpb_size
+= bbm_log_size
;
6019 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
6022 assert(super
->len
== 0 || mpb_size
<= super
->len
);
6025 /* recalculate checksum */
6026 sum
= __gen_imsm_checksum(mpb
);
6027 mpb
->check_sum
= __cpu_to_le32(sum
);
6029 if (super
->clean_migration_record_by_mdmon
) {
6030 clear_migration_record
= 1;
6031 super
->clean_migration_record_by_mdmon
= 0;
6033 if (clear_migration_record
)
6034 memset(super
->migr_rec_buf
, 0,
6035 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
6037 if (sector_size
== 4096)
6038 convert_to_4k(super
);
6040 /* write the mpb for disks that compose raid devices */
6041 for (d
= super
->disks
; d
; d
= d
->next
) {
6042 if (d
->index
< 0 || is_failed(&d
->disk
))
6045 if (clear_migration_record
) {
6046 unsigned long long dsize
;
6048 get_dev_size(d
->fd
, NULL
, &dsize
);
6049 if (lseek64(d
->fd
, dsize
- sector_size
,
6051 if ((unsigned int)write(d
->fd
,
6052 super
->migr_rec_buf
,
6053 MIGR_REC_BUF_SECTORS
*sector_size
) !=
6054 MIGR_REC_BUF_SECTORS
*sector_size
)
6055 perror("Write migr_rec failed");
6059 if (store_imsm_mpb(d
->fd
, mpb
))
6061 "failed for device %d:%d (fd: %d)%s\n",
6063 d
->fd
, strerror(errno
));
6072 return write_super_imsm_spares(super
, doclose
);
6077 static int create_array(struct supertype
*st
, int dev_idx
)
6080 struct imsm_update_create_array
*u
;
6081 struct intel_super
*super
= st
->sb
;
6082 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6083 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6084 struct disk_info
*inf
;
6085 struct imsm_disk
*disk
;
6088 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6089 sizeof(*inf
) * map
->num_members
;
6091 u
->type
= update_create_array
;
6092 u
->dev_idx
= dev_idx
;
6093 imsm_copy_dev(&u
->dev
, dev
);
6094 inf
= get_disk_info(u
);
6095 for (i
= 0; i
< map
->num_members
; i
++) {
6096 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6098 disk
= get_imsm_disk(super
, idx
);
6100 disk
= get_imsm_missing(super
, idx
);
6101 serialcpy(inf
[i
].serial
, disk
->serial
);
6103 append_metadata_update(st
, u
, len
);
6108 static int mgmt_disk(struct supertype
*st
)
6110 struct intel_super
*super
= st
->sb
;
6112 struct imsm_update_add_remove_disk
*u
;
6114 if (!super
->disk_mgmt_list
)
6119 u
->type
= update_add_remove_disk
;
6120 append_metadata_update(st
, u
, len
);
6125 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6127 static int write_ppl_header(unsigned long long ppl_sector
, int fd
, void *buf
)
6129 struct ppl_header
*ppl_hdr
= buf
;
6132 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6134 if (lseek64(fd
, ppl_sector
* 512, SEEK_SET
) < 0) {
6136 perror("Failed to seek to PPL header location");
6140 if (write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6142 perror("Write PPL header failed");
6151 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6153 struct intel_super
*super
= st
->sb
;
6155 struct ppl_header
*ppl_hdr
;
6158 /* first clear entire ppl space */
6159 ret
= zero_disk_range(fd
, info
->ppl_sector
, info
->ppl_size
);
6163 ret
= posix_memalign(&buf
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
);
6165 pr_err("Failed to allocate PPL header buffer\n");
6169 memset(buf
, 0, PPL_HEADER_SIZE
);
6171 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6172 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6174 if (info
->mismatch_cnt
) {
6176 * We are overwriting an invalid ppl. Make one entry with wrong
6177 * checksum to prevent the kernel from skipping resync.
6179 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6180 ppl_hdr
->entries
[0].checksum
= ~0;
6183 ret
= write_ppl_header(info
->ppl_sector
, fd
, buf
);
6189 static int is_rebuilding(struct imsm_dev
*dev
);
6191 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6192 struct mdinfo
*disk
)
6194 struct intel_super
*super
= st
->sb
;
6196 void *buf_orig
, *buf
, *buf_prev
= NULL
;
6198 struct ppl_header
*ppl_hdr
= NULL
;
6200 struct imsm_dev
*dev
;
6203 unsigned long long ppl_offset
= 0;
6204 unsigned long long prev_gen_num
= 0;
6206 if (disk
->disk
.raid_disk
< 0)
6209 dev
= get_imsm_dev(super
, info
->container_member
);
6210 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_0
);
6211 d
= get_imsm_dl_disk(super
, idx
);
6213 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6216 if (posix_memalign(&buf_orig
, MAX_SECTOR_SIZE
, PPL_HEADER_SIZE
* 2)) {
6217 pr_err("Failed to allocate PPL header buffer\n");
6223 while (ppl_offset
< MULTIPLE_PPL_AREA_SIZE_IMSM
) {
6226 dprintf("Checking potential PPL at offset: %llu\n", ppl_offset
);
6228 if (lseek64(d
->fd
, info
->ppl_sector
* 512 + ppl_offset
,
6230 perror("Failed to seek to PPL header location");
6235 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6236 perror("Read PPL header failed");
6243 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6244 ppl_hdr
->checksum
= 0;
6246 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6247 dprintf("Wrong PPL header checksum on %s\n",
6252 if (prev_gen_num
> __le64_to_cpu(ppl_hdr
->generation
)) {
6253 /* previous was newest, it was already checked */
6257 if ((__le32_to_cpu(ppl_hdr
->signature
) !=
6258 super
->anchor
->orig_family_num
)) {
6259 dprintf("Wrong PPL header signature on %s\n",
6266 prev_gen_num
= __le64_to_cpu(ppl_hdr
->generation
);
6268 ppl_offset
+= PPL_HEADER_SIZE
;
6269 for (i
= 0; i
< __le32_to_cpu(ppl_hdr
->entries_count
); i
++)
6271 __le32_to_cpu(ppl_hdr
->entries
[i
].pp_size
);
6274 buf_prev
= buf
+ PPL_HEADER_SIZE
;
6286 * Update metadata to use mutliple PPLs area (1MB).
6287 * This is done once for all RAID members
6289 if (info
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
6290 info
->ppl_size
!= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9)) {
6292 struct mdinfo
*member_dev
;
6294 sprintf(subarray
, "%d", info
->container_member
);
6296 if (mdmon_running(st
->container_devnm
))
6297 st
->update_tail
= &st
->updates
;
6299 if (st
->ss
->update_subarray(st
, subarray
, "ppl", NULL
)) {
6300 pr_err("Failed to update subarray %s\n",
6303 if (st
->update_tail
)
6304 flush_metadata_updates(st
);
6306 st
->ss
->sync_metadata(st
);
6307 info
->ppl_size
= (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6308 for (member_dev
= info
->devs
; member_dev
;
6309 member_dev
= member_dev
->next
)
6310 member_dev
->ppl_size
=
6311 (MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9);
6316 struct imsm_map
*map
= get_imsm_map(dev
, MAP_X
);
6318 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
6319 (map
->map_state
== IMSM_T_STATE_NORMAL
&&
6320 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) ||
6321 (is_rebuilding(dev
) &&
6322 dev
->vol
.curr_migr_unit
== 0 &&
6323 get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_1
) != idx
))
6324 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6326 info
->mismatch_cnt
++;
6327 } else if (ret
== 0 &&
6328 ppl_hdr
->entries_count
== 0 &&
6329 is_rebuilding(dev
) &&
6330 info
->resync_start
== 0) {
6332 * The header has no entries - add a single empty entry and
6333 * rewrite the header to prevent the kernel from going into
6334 * resync after an interrupted rebuild.
6336 ppl_hdr
->entries_count
= __cpu_to_le32(1);
6337 ret
= write_ppl_header(info
->ppl_sector
, d
->fd
, buf
);
6345 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6347 struct intel_super
*super
= st
->sb
;
6351 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6352 info
->array
.level
!= 5)
6355 for (d
= super
->disks
; d
; d
= d
->next
) {
6356 if (d
->index
< 0 || is_failed(&d
->disk
))
6359 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6367 static int write_init_super_imsm(struct supertype
*st
)
6369 struct intel_super
*super
= st
->sb
;
6370 int current_vol
= super
->current_vol
;
6374 getinfo_super_imsm(st
, &info
, NULL
);
6376 /* we are done with current_vol reset it to point st at the container */
6377 super
->current_vol
= -1;
6379 if (st
->update_tail
) {
6380 /* queue the recently created array / added disk
6381 * as a metadata update */
6383 /* determine if we are creating a volume or adding a disk */
6384 if (current_vol
< 0) {
6385 /* in the mgmt (add/remove) disk case we are running
6386 * in mdmon context, so don't close fd's
6390 rv
= write_init_ppl_imsm_all(st
, &info
);
6392 rv
= create_array(st
, current_vol
);
6396 for (d
= super
->disks
; d
; d
= d
->next
)
6397 Kill(d
->devname
, NULL
, 0, -1, 1);
6398 if (current_vol
>= 0)
6399 rv
= write_init_ppl_imsm_all(st
, &info
);
6401 rv
= write_super_imsm(st
, 1);
6407 static int store_super_imsm(struct supertype
*st
, int fd
)
6409 struct intel_super
*super
= st
->sb
;
6410 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6415 if (super
->sector_size
== 4096)
6416 convert_to_4k(super
);
6417 return store_imsm_mpb(fd
, mpb
);
6420 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6421 int layout
, int raiddisks
, int chunk
,
6422 unsigned long long size
,
6423 unsigned long long data_offset
,
6425 unsigned long long *freesize
,
6429 unsigned long long ldsize
;
6430 struct intel_super
*super
;
6433 if (level
!= LEVEL_CONTAINER
)
6438 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6441 pr_err("imsm: Cannot open %s: %s\n",
6442 dev
, strerror(errno
));
6445 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6450 /* capabilities retrieve could be possible
6451 * note that there is no fd for the disks in array.
6453 super
= alloc_super();
6458 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6464 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6468 fd2devname(fd
, str
);
6469 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6470 fd
, str
, super
->orom
, rv
, raiddisks
);
6472 /* no orom/efi or non-intel hba of the disk */
6479 if (raiddisks
> super
->orom
->tds
) {
6481 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6482 raiddisks
, super
->orom
->tds
);
6486 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6487 (ldsize
>> 9) >> 32 > 0) {
6489 pr_err("%s exceeds maximum platform supported size\n", dev
);
6495 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6501 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6503 const unsigned long long base_start
= e
[*idx
].start
;
6504 unsigned long long end
= base_start
+ e
[*idx
].size
;
6507 if (base_start
== end
)
6511 for (i
= *idx
; i
< num_extents
; i
++) {
6512 /* extend overlapping extents */
6513 if (e
[i
].start
>= base_start
&&
6514 e
[i
].start
<= end
) {
6517 if (e
[i
].start
+ e
[i
].size
> end
)
6518 end
= e
[i
].start
+ e
[i
].size
;
6519 } else if (e
[i
].start
> end
) {
6525 return end
- base_start
;
6528 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6530 /* build a composite disk with all known extents and generate a new
6531 * 'maxsize' given the "all disks in an array must share a common start
6532 * offset" constraint
6534 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6538 unsigned long long pos
;
6539 unsigned long long start
= 0;
6540 unsigned long long maxsize
;
6541 unsigned long reserve
;
6543 /* coalesce and sort all extents. also, check to see if we need to
6544 * reserve space between member arrays
6547 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6550 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6553 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6558 while (i
< sum_extents
) {
6559 e
[j
].start
= e
[i
].start
;
6560 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6562 if (e
[j
-1].size
== 0)
6571 unsigned long long esize
;
6573 esize
= e
[i
].start
- pos
;
6574 if (esize
>= maxsize
) {
6579 pos
= e
[i
].start
+ e
[i
].size
;
6581 } while (e
[i
-1].size
);
6587 /* FIXME assumes volume at offset 0 is the first volume in a
6590 if (start_extent
> 0)
6591 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6595 if (maxsize
< reserve
)
6598 super
->create_offset
= ~((unsigned long long) 0);
6599 if (start
+ reserve
> super
->create_offset
)
6600 return 0; /* start overflows create_offset */
6601 super
->create_offset
= start
+ reserve
;
6603 return maxsize
- reserve
;
6606 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6608 if (level
< 0 || level
== 6 || level
== 4)
6611 /* if we have an orom prevent invalid raid levels */
6614 case 0: return imsm_orom_has_raid0(orom
);
6617 return imsm_orom_has_raid1e(orom
);
6618 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6619 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6620 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6623 return 1; /* not on an Intel RAID platform so anything goes */
6629 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6630 int dpa
, int verbose
)
6632 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6633 struct mdstat_ent
*memb
;
6639 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6640 if (memb
->metadata_version
&&
6641 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6642 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6643 !is_subarray(memb
->metadata_version
+9) &&
6645 struct dev_member
*dev
= memb
->members
;
6647 while(dev
&& (fd
< 0)) {
6648 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6649 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6651 fd
= open(path
, O_RDONLY
, 0);
6652 if (num
<= 0 || fd
< 0) {
6653 pr_vrb("Cannot open %s: %s\n",
6654 dev
->name
, strerror(errno
));
6660 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6661 struct mdstat_ent
*vol
;
6662 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6663 if (vol
->active
> 0 &&
6664 vol
->metadata_version
&&
6665 is_container_member(vol
, memb
->devnm
)) {
6670 if (*devlist
&& (found
< dpa
)) {
6671 dv
= xcalloc(1, sizeof(*dv
));
6672 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6673 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6676 dv
->next
= *devlist
;
6684 free_mdstat(mdstat
);
6689 static struct md_list
*
6690 get_loop_devices(void)
6693 struct md_list
*devlist
= NULL
;
6696 for(i
= 0; i
< 12; i
++) {
6697 dv
= xcalloc(1, sizeof(*dv
));
6698 dv
->devname
= xmalloc(40);
6699 sprintf(dv
->devname
, "/dev/loop%d", i
);
6707 static struct md_list
*
6708 get_devices(const char *hba_path
)
6710 struct md_list
*devlist
= NULL
;
6717 devlist
= get_loop_devices();
6720 /* scroll through /sys/dev/block looking for devices attached to
6723 dir
= opendir("/sys/dev/block");
6724 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6729 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6731 path
= devt_to_devpath(makedev(major
, minor
));
6734 if (!path_attached_to_hba(path
, hba_path
)) {
6741 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6743 fd2devname(fd
, buf
);
6746 pr_err("cannot open device: %s\n",
6751 dv
= xcalloc(1, sizeof(*dv
));
6752 dv
->devname
= xstrdup(buf
);
6759 devlist
= devlist
->next
;
6769 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6770 int verbose
, int *found
)
6772 struct md_list
*tmpdev
;
6774 struct supertype
*st
;
6776 /* first walk the list of devices to find a consistent set
6777 * that match the criterea, if that is possible.
6778 * We flag the ones we like with 'used'.
6781 st
= match_metadata_desc_imsm("imsm");
6783 pr_vrb("cannot allocate memory for imsm supertype\n");
6787 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6788 char *devname
= tmpdev
->devname
;
6790 struct supertype
*tst
;
6792 if (tmpdev
->used
> 1)
6794 tst
= dup_super(st
);
6796 pr_vrb("cannot allocate memory for imsm supertype\n");
6799 tmpdev
->container
= 0;
6800 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6802 dprintf("cannot open device %s: %s\n",
6803 devname
, strerror(errno
));
6805 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
6807 } else if (must_be_container(dfd
)) {
6808 struct supertype
*cst
;
6809 cst
= super_by_fd(dfd
, NULL
);
6811 dprintf("cannot recognize container type %s\n",
6814 } else if (tst
->ss
!= st
->ss
) {
6815 dprintf("non-imsm container - ignore it: %s\n",
6818 } else if (!tst
->ss
->load_container
||
6819 tst
->ss
->load_container(tst
, dfd
, NULL
))
6822 tmpdev
->container
= 1;
6825 cst
->ss
->free_super(cst
);
6827 tmpdev
->st_rdev
= rdev
;
6828 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6829 dprintf("no RAID superblock on %s\n",
6832 } else if (tst
->ss
->compare_super
== NULL
) {
6833 dprintf("Cannot assemble %s metadata on %s\n",
6834 tst
->ss
->name
, devname
);
6840 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6841 /* Ignore unrecognised devices during auto-assembly */
6846 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6848 if (st
->minor_version
== -1)
6849 st
->minor_version
= tst
->minor_version
;
6851 if (memcmp(info
.uuid
, uuid_zero
,
6852 sizeof(int[4])) == 0) {
6853 /* this is a floating spare. It cannot define
6854 * an array unless there are no more arrays of
6855 * this type to be found. It can be included
6856 * in an array of this type though.
6862 if (st
->ss
!= tst
->ss
||
6863 st
->minor_version
!= tst
->minor_version
||
6864 st
->ss
->compare_super(st
, tst
) != 0) {
6865 /* Some mismatch. If exactly one array matches this host,
6866 * we can resolve on that one.
6867 * Or, if we are auto assembling, we just ignore the second
6870 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6876 dprintf("found: devname: %s\n", devname
);
6880 tst
->ss
->free_super(tst
);
6884 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6885 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6886 for (iter
= head
; iter
; iter
= iter
->next
) {
6887 dprintf("content->text_version: %s vol\n",
6888 iter
->text_version
);
6889 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6890 /* do not assemble arrays with unsupported
6892 dprintf("Cannot activate member %s.\n",
6893 iter
->text_version
);
6900 dprintf("No valid super block on device list: err: %d %p\n",
6904 dprintf("no more devices to examine\n");
6907 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6908 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6910 if (count
< tmpdev
->found
)
6913 count
-= tmpdev
->found
;
6916 if (tmpdev
->used
== 1)
6921 st
->ss
->free_super(st
);
6925 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
6928 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6930 const struct orom_entry
*entry
;
6931 struct devid_list
*dv
, *devid_list
;
6936 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
6937 if (strstr(idev
->path
, hba_path
))
6941 if (!idev
|| !idev
->dev_id
)
6944 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
6946 if (!entry
|| !entry
->devid_list
)
6949 devid_list
= entry
->devid_list
;
6950 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
6951 struct md_list
*devlist
;
6952 struct sys_dev
*device
= NULL
;
6957 device
= device_by_id_and_path(dv
->devid
, hba_path
);
6959 device
= device_by_id(dv
->devid
);
6962 hpath
= device
->path
;
6966 devlist
= get_devices(hpath
);
6967 /* if no intel devices return zero volumes */
6968 if (devlist
== NULL
)
6971 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
6973 dprintf("path: %s active arrays: %d\n", hpath
, count
);
6974 if (devlist
== NULL
)
6978 count
+= count_volumes_list(devlist
,
6982 dprintf("found %d count: %d\n", found
, count
);
6985 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
6988 struct md_list
*dv
= devlist
;
6989 devlist
= devlist
->next
;
6997 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
7001 if (hba
->type
== SYS_DEV_VMD
) {
7002 struct sys_dev
*dev
;
7005 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
7006 if (dev
->type
== SYS_DEV_VMD
)
7007 count
+= __count_volumes(dev
->path
, dpa
,
7012 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
7015 static int imsm_default_chunk(const struct imsm_orom
*orom
)
7017 /* up to 512 if the plaform supports it, otherwise the platform max.
7018 * 128 if no platform detected
7020 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
7022 return min(512, (1 << fs
));
7026 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
7027 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
7029 /* check/set platform and metadata limits/defaults */
7030 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
7031 pr_vrb("platform supports a maximum of %d disks per array\n",
7036 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
7037 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
7038 pr_vrb("platform does not support raid%d with %d disk%s\n",
7039 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
7043 if (*chunk
== 0 || *chunk
== UnSet
)
7044 *chunk
= imsm_default_chunk(super
->orom
);
7046 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
7047 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
7051 if (layout
!= imsm_level_to_layout(level
)) {
7053 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
7054 else if (level
== 10)
7055 pr_vrb("imsm raid 10 only supports the n2 layout\n");
7057 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
7062 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
7063 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
7064 pr_vrb("platform does not support a volume size over 2TB\n");
7071 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
7072 * FIX ME add ahci details
7074 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
7075 int layout
, int raiddisks
, int *chunk
,
7076 unsigned long long size
,
7077 unsigned long long data_offset
,
7079 unsigned long long *freesize
,
7083 struct intel_super
*super
= st
->sb
;
7084 struct imsm_super
*mpb
;
7086 unsigned long long pos
= 0;
7087 unsigned long long maxsize
;
7091 /* We must have the container info already read in. */
7095 mpb
= super
->anchor
;
7097 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
7098 pr_err("RAID geometry validation failed. Cannot proceed with the action(s).\n");
7102 /* General test: make sure there is space for
7103 * 'raiddisks' device extents of size 'size' at a given
7106 unsigned long long minsize
= size
;
7107 unsigned long long start_offset
= MaxSector
;
7110 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
7111 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7116 e
= get_extents(super
, dl
);
7119 unsigned long long esize
;
7120 esize
= e
[i
].start
- pos
;
7121 if (esize
>= minsize
)
7123 if (found
&& start_offset
== MaxSector
) {
7126 } else if (found
&& pos
!= start_offset
) {
7130 pos
= e
[i
].start
+ e
[i
].size
;
7132 } while (e
[i
-1].size
);
7137 if (dcnt
< raiddisks
) {
7139 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
7146 /* This device must be a member of the set */
7147 if (!stat_is_blkdev(dev
, &rdev
))
7149 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7150 if (dl
->major
== (int)major(rdev
) &&
7151 dl
->minor
== (int)minor(rdev
))
7156 pr_err("%s is not in the same imsm set\n", dev
);
7158 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
7159 /* If a volume is present then the current creation attempt
7160 * cannot incorporate new spares because the orom may not
7161 * understand this configuration (all member disks must be
7162 * members of each array in the container).
7164 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
7165 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7167 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
7168 mpb
->num_disks
!= raiddisks
) {
7169 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
7173 /* retrieve the largest free space block */
7174 e
= get_extents(super
, dl
);
7179 unsigned long long esize
;
7181 esize
= e
[i
].start
- pos
;
7182 if (esize
>= maxsize
)
7184 pos
= e
[i
].start
+ e
[i
].size
;
7186 } while (e
[i
-1].size
);
7191 pr_err("unable to determine free space for: %s\n",
7195 if (maxsize
< size
) {
7197 pr_err("%s not enough space (%llu < %llu)\n",
7198 dev
, maxsize
, size
);
7202 /* count total number of extents for merge */
7204 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7206 i
+= dl
->extent_cnt
;
7208 maxsize
= merge_extents(super
, i
);
7210 if (!check_env("IMSM_NO_PLATFORM") &&
7211 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7212 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7216 if (maxsize
< size
|| maxsize
== 0) {
7219 pr_err("no free space left on device. Aborting...\n");
7221 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7227 *freesize
= maxsize
;
7230 int count
= count_volumes(super
->hba
,
7231 super
->orom
->dpa
, verbose
);
7232 if (super
->orom
->vphba
<= count
) {
7233 pr_vrb("platform does not support more than %d raid volumes.\n",
7234 super
->orom
->vphba
);
7241 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7242 unsigned long long size
, int chunk
,
7243 unsigned long long *freesize
)
7245 struct intel_super
*super
= st
->sb
;
7246 struct imsm_super
*mpb
= super
->anchor
;
7251 unsigned long long maxsize
;
7252 unsigned long long minsize
;
7256 /* find the largest common start free region of the possible disks */
7260 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7266 /* don't activate new spares if we are orom constrained
7267 * and there is already a volume active in the container
7269 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7272 e
= get_extents(super
, dl
);
7275 for (i
= 1; e
[i
-1].size
; i
++)
7283 maxsize
= merge_extents(super
, extent_cnt
);
7287 minsize
= chunk
* 2;
7289 if (cnt
< raiddisks
||
7290 (super
->orom
&& used
&& used
!= raiddisks
) ||
7291 maxsize
< minsize
||
7293 pr_err("not enough devices with space to create array.\n");
7294 return 0; /* No enough free spaces large enough */
7305 if (!check_env("IMSM_NO_PLATFORM") &&
7306 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7307 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7311 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7313 dl
->raiddisk
= cnt
++;
7317 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7322 static int reserve_space(struct supertype
*st
, int raiddisks
,
7323 unsigned long long size
, int chunk
,
7324 unsigned long long *freesize
)
7326 struct intel_super
*super
= st
->sb
;
7331 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7334 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7336 dl
->raiddisk
= cnt
++;
7343 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7344 int raiddisks
, int *chunk
, unsigned long long size
,
7345 unsigned long long data_offset
,
7346 char *dev
, unsigned long long *freesize
,
7347 int consistency_policy
, int verbose
)
7354 * if given unused devices create a container
7355 * if given given devices in a container create a member volume
7357 if (level
== LEVEL_CONTAINER
) {
7358 /* Must be a fresh device to add to a container */
7359 return validate_geometry_imsm_container(st
, level
, layout
,
7367 if (size
&& ((size
< 1024) || (*chunk
!= UnSet
&&
7368 size
< (unsigned long long) *chunk
))) {
7369 pr_err("Given size must be greater than 1M and chunk size.\n");
7370 /* Depends on algorithm in Create.c :
7371 * if container was given (dev == NULL) return -1,
7372 * if block device was given ( dev != NULL) return 0.
7374 return dev
? -1 : 0;
7379 struct intel_super
*super
= st
->sb
;
7380 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7381 raiddisks
, chunk
, size
,
7384 /* we are being asked to automatically layout a
7385 * new volume based on the current contents of
7386 * the container. If the the parameters can be
7387 * satisfied reserve_space will record the disks,
7388 * start offset, and size of the volume to be
7389 * created. add_to_super and getinfo_super
7390 * detect when autolayout is in progress.
7392 /* assuming that freesize is always given when array is
7394 if (super
->orom
&& freesize
) {
7396 count
= count_volumes(super
->hba
,
7397 super
->orom
->dpa
, verbose
);
7398 if (super
->orom
->vphba
<= count
) {
7399 pr_vrb("platform does not support more than %d raid volumes.\n",
7400 super
->orom
->vphba
);
7405 return reserve_space(st
, raiddisks
, size
,
7411 /* creating in a given container */
7412 return validate_geometry_imsm_volume(st
, level
, layout
,
7413 raiddisks
, chunk
, size
,
7415 dev
, freesize
, verbose
);
7418 /* This device needs to be a device in an 'imsm' container */
7419 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7422 pr_err("Cannot create this array on device %s\n",
7427 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7429 pr_err("Cannot open %s: %s\n",
7430 dev
, strerror(errno
));
7433 /* Well, it is in use by someone, maybe an 'imsm' container. */
7434 cfd
= open_container(fd
);
7438 pr_err("Cannot use %s: It is busy\n",
7442 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7443 if (sra
&& sra
->array
.major_version
== -1 &&
7444 strcmp(sra
->text_version
, "imsm") == 0)
7448 /* This is a member of a imsm container. Load the container
7449 * and try to create a volume
7451 struct intel_super
*super
;
7453 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7455 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7457 return validate_geometry_imsm_volume(st
, level
, layout
,
7459 size
, data_offset
, dev
,
7466 pr_err("failed container membership check\n");
7472 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7474 struct intel_super
*super
= st
->sb
;
7476 if (level
&& *level
== UnSet
)
7477 *level
= LEVEL_CONTAINER
;
7479 if (level
&& layout
&& *layout
== UnSet
)
7480 *layout
= imsm_level_to_layout(*level
);
7482 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7483 *chunk
= imsm_default_chunk(super
->orom
);
7486 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7488 static int kill_subarray_imsm(struct supertype
*st
)
7490 /* remove the subarray currently referenced by ->current_vol */
7492 struct intel_dev
**dp
;
7493 struct intel_super
*super
= st
->sb
;
7494 __u8 current_vol
= super
->current_vol
;
7495 struct imsm_super
*mpb
= super
->anchor
;
7497 if (super
->current_vol
< 0)
7499 super
->current_vol
= -1; /* invalidate subarray cursor */
7501 /* block deletions that would change the uuid of active subarrays
7503 * FIXME when immutable ids are available, but note that we'll
7504 * also need to fixup the invalidated/active subarray indexes in
7507 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7510 if (i
< current_vol
)
7512 sprintf(subarray
, "%u", i
);
7513 if (is_subarray_active(subarray
, st
->devnm
)) {
7514 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7521 if (st
->update_tail
) {
7522 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7524 u
->type
= update_kill_array
;
7525 u
->dev_idx
= current_vol
;
7526 append_metadata_update(st
, u
, sizeof(*u
));
7531 for (dp
= &super
->devlist
; *dp
;)
7532 if ((*dp
)->index
== current_vol
) {
7535 handle_missing(super
, (*dp
)->dev
);
7536 if ((*dp
)->index
> current_vol
)
7541 /* no more raid devices, all active components are now spares,
7542 * but of course failed are still failed
7544 if (--mpb
->num_raid_devs
== 0) {
7547 for (d
= super
->disks
; d
; d
= d
->next
)
7552 super
->updates_pending
++;
7557 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7558 char *update
, struct mddev_ident
*ident
)
7560 /* update the subarray currently referenced by ->current_vol */
7561 struct intel_super
*super
= st
->sb
;
7562 struct imsm_super
*mpb
= super
->anchor
;
7564 if (strcmp(update
, "name") == 0) {
7565 char *name
= ident
->name
;
7569 if (is_subarray_active(subarray
, st
->devnm
)) {
7570 pr_err("Unable to update name of active subarray\n");
7574 if (!check_name(super
, name
, 0))
7577 vol
= strtoul(subarray
, &ep
, 10);
7578 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7581 if (st
->update_tail
) {
7582 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7584 u
->type
= update_rename_array
;
7586 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7587 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7588 append_metadata_update(st
, u
, sizeof(*u
));
7590 struct imsm_dev
*dev
;
7593 dev
= get_imsm_dev(super
, vol
);
7594 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
7595 dev
->volume
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7596 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7597 dev
= get_imsm_dev(super
, i
);
7598 handle_missing(super
, dev
);
7600 super
->updates_pending
++;
7602 } else if (strcmp(update
, "ppl") == 0 ||
7603 strcmp(update
, "no-ppl") == 0) {
7606 int vol
= strtoul(subarray
, &ep
, 10);
7608 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7611 if (strcmp(update
, "ppl") == 0)
7612 new_policy
= RWH_MULTIPLE_DISTRIBUTED
;
7614 new_policy
= RWH_MULTIPLE_OFF
;
7616 if (st
->update_tail
) {
7617 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7619 u
->type
= update_rwh_policy
;
7621 u
->new_policy
= new_policy
;
7622 append_metadata_update(st
, u
, sizeof(*u
));
7624 struct imsm_dev
*dev
;
7626 dev
= get_imsm_dev(super
, vol
);
7627 dev
->rwh_policy
= new_policy
;
7628 super
->updates_pending
++;
7636 static int is_gen_migration(struct imsm_dev
*dev
)
7641 if (!dev
->vol
.migr_state
)
7644 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7650 static int is_rebuilding(struct imsm_dev
*dev
)
7652 struct imsm_map
*migr_map
;
7654 if (!dev
->vol
.migr_state
)
7657 if (migr_type(dev
) != MIGR_REBUILD
)
7660 migr_map
= get_imsm_map(dev
, MAP_1
);
7662 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7668 static int is_initializing(struct imsm_dev
*dev
)
7670 struct imsm_map
*migr_map
;
7672 if (!dev
->vol
.migr_state
)
7675 if (migr_type(dev
) != MIGR_INIT
)
7678 migr_map
= get_imsm_map(dev
, MAP_1
);
7680 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7686 static void update_recovery_start(struct intel_super
*super
,
7687 struct imsm_dev
*dev
,
7688 struct mdinfo
*array
)
7690 struct mdinfo
*rebuild
= NULL
;
7694 if (!is_rebuilding(dev
))
7697 /* Find the rebuild target, but punt on the dual rebuild case */
7698 for (d
= array
->devs
; d
; d
= d
->next
)
7699 if (d
->recovery_start
== 0) {
7706 /* (?) none of the disks are marked with
7707 * IMSM_ORD_REBUILD, so assume they are missing and the
7708 * disk_ord_tbl was not correctly updated
7710 dprintf("failed to locate out-of-sync disk\n");
7714 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7715 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7718 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7720 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7722 /* Given a container loaded by load_super_imsm_all,
7723 * extract information about all the arrays into
7725 * If 'subarray' is given, just extract info about that array.
7727 * For each imsm_dev create an mdinfo, fill it in,
7728 * then look for matching devices in super->disks
7729 * and create appropriate device mdinfo.
7731 struct intel_super
*super
= st
->sb
;
7732 struct imsm_super
*mpb
= super
->anchor
;
7733 struct mdinfo
*rest
= NULL
;
7737 int spare_disks
= 0;
7739 /* do not assemble arrays when not all attributes are supported */
7740 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7742 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7745 /* count spare devices, not used in maps
7747 for (d
= super
->disks
; d
; d
= d
->next
)
7751 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7752 struct imsm_dev
*dev
;
7753 struct imsm_map
*map
;
7754 struct imsm_map
*map2
;
7755 struct mdinfo
*this;
7762 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7765 dev
= get_imsm_dev(super
, i
);
7766 map
= get_imsm_map(dev
, MAP_0
);
7767 map2
= get_imsm_map(dev
, MAP_1
);
7768 level
= get_imsm_raid_level(map
);
7770 /* do not publish arrays that are in the middle of an
7771 * unsupported migration
7773 if (dev
->vol
.migr_state
&&
7774 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7775 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7779 /* do not publish arrays that are not support by controller's
7783 this = xmalloc(sizeof(*this));
7785 super
->current_vol
= i
;
7786 getinfo_super_imsm_volume(st
, this, NULL
);
7788 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7789 /* mdadm does not support all metadata features- set the bit in all arrays state */
7790 if (!validate_geometry_imsm_orom(super
,
7791 level
, /* RAID level */
7792 imsm_level_to_layout(level
),
7793 map
->num_members
, /* raid disks */
7794 &chunk
, imsm_dev_size(dev
),
7796 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7798 this->array
.state
|=
7799 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7800 (1<<MD_SB_BLOCK_VOLUME
);
7803 /* if array has bad blocks, set suitable bit in all arrays state */
7805 this->array
.state
|=
7806 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7807 (1<<MD_SB_BLOCK_VOLUME
);
7809 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7810 unsigned long long recovery_start
;
7811 struct mdinfo
*info_d
;
7819 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7820 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7821 for (d
= super
->disks
; d
; d
= d
->next
)
7822 if (d
->index
== idx
)
7825 recovery_start
= MaxSector
;
7828 if (d
&& is_failed(&d
->disk
))
7830 if (!skip
&& (ord
& IMSM_ORD_REBUILD
))
7834 * if we skip some disks the array will be assmebled degraded;
7835 * reset resync start to avoid a dirty-degraded
7836 * situation when performing the intial sync
7841 if (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
)) {
7842 if ((!able_to_resync(level
, missing
) ||
7843 recovery_start
== 0))
7844 this->resync_start
= MaxSector
;
7847 * FIXME handle dirty degraded
7854 info_d
= xcalloc(1, sizeof(*info_d
));
7855 info_d
->next
= this->devs
;
7856 this->devs
= info_d
;
7858 info_d
->disk
.number
= d
->index
;
7859 info_d
->disk
.major
= d
->major
;
7860 info_d
->disk
.minor
= d
->minor
;
7861 info_d
->disk
.raid_disk
= slot
;
7862 info_d
->recovery_start
= recovery_start
;
7864 if (slot
< map2
->num_members
)
7865 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7867 this->array
.spare_disks
++;
7869 if (slot
< map
->num_members
)
7870 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7872 this->array
.spare_disks
++;
7874 if (info_d
->recovery_start
== MaxSector
)
7875 this->array
.working_disks
++;
7877 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7878 info_d
->data_offset
= pba_of_lba0(map
);
7879 info_d
->component_size
= calc_component_size(map
, dev
);
7881 if (map
->raid_level
== 5) {
7882 info_d
->ppl_sector
= this->ppl_sector
;
7883 info_d
->ppl_size
= this->ppl_size
;
7884 if (this->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
7885 recovery_start
== 0)
7886 this->resync_start
= 0;
7889 info_d
->bb
.supported
= 1;
7890 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7891 info_d
->data_offset
,
7892 info_d
->component_size
,
7895 /* now that the disk list is up-to-date fixup recovery_start */
7896 update_recovery_start(super
, dev
, this);
7897 this->array
.spare_disks
+= spare_disks
;
7899 /* check for reshape */
7900 if (this->reshape_active
== 1)
7901 recover_backup_imsm(st
, this);
7908 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7909 int failed
, int look_in_map
)
7911 struct imsm_map
*map
;
7913 map
= get_imsm_map(dev
, look_in_map
);
7916 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
7917 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
7919 switch (get_imsm_raid_level(map
)) {
7921 return IMSM_T_STATE_FAILED
;
7924 if (failed
< map
->num_members
)
7925 return IMSM_T_STATE_DEGRADED
;
7927 return IMSM_T_STATE_FAILED
;
7932 * check to see if any mirrors have failed, otherwise we
7933 * are degraded. Even numbered slots are mirrored on
7937 /* gcc -Os complains that this is unused */
7938 int insync
= insync
;
7940 for (i
= 0; i
< map
->num_members
; i
++) {
7941 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
7942 int idx
= ord_to_idx(ord
);
7943 struct imsm_disk
*disk
;
7945 /* reset the potential in-sync count on even-numbered
7946 * slots. num_copies is always 2 for imsm raid10
7951 disk
= get_imsm_disk(super
, idx
);
7952 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7955 /* no in-sync disks left in this mirror the
7959 return IMSM_T_STATE_FAILED
;
7962 return IMSM_T_STATE_DEGRADED
;
7966 return IMSM_T_STATE_DEGRADED
;
7968 return IMSM_T_STATE_FAILED
;
7974 return map
->map_state
;
7977 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
7982 struct imsm_disk
*disk
;
7983 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7984 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
7985 struct imsm_map
*map_for_loop
;
7990 /* at the beginning of migration we set IMSM_ORD_REBUILD on
7991 * disks that are being rebuilt. New failures are recorded to
7992 * map[0]. So we look through all the disks we started with and
7993 * see if any failures are still present, or if any new ones
7997 if (prev
&& (map
->num_members
< prev
->num_members
))
7998 map_for_loop
= prev
;
8000 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
8002 /* when MAP_X is passed both maps failures are counted
8005 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
8006 i
< prev
->num_members
) {
8007 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
8008 idx_1
= ord_to_idx(ord
);
8010 disk
= get_imsm_disk(super
, idx_1
);
8011 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
8014 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
8015 i
< map
->num_members
) {
8016 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
8017 idx
= ord_to_idx(ord
);
8020 disk
= get_imsm_disk(super
, idx
);
8021 if (!disk
|| is_failed(disk
) ||
8022 ord
& IMSM_ORD_REBUILD
)
8031 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
8034 struct intel_super
*super
= c
->sb
;
8035 struct imsm_super
*mpb
= super
->anchor
;
8036 struct imsm_update_prealloc_bb_mem u
;
8038 if (atoi(inst
) >= mpb
->num_raid_devs
) {
8039 pr_err("subarry index %d, out of range\n", atoi(inst
));
8043 dprintf("imsm: open_new %s\n", inst
);
8044 a
->info
.container_member
= atoi(inst
);
8046 u
.type
= update_prealloc_badblocks_mem
;
8047 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
8052 static int is_resyncing(struct imsm_dev
*dev
)
8054 struct imsm_map
*migr_map
;
8056 if (!dev
->vol
.migr_state
)
8059 if (migr_type(dev
) == MIGR_INIT
||
8060 migr_type(dev
) == MIGR_REPAIR
)
8063 if (migr_type(dev
) == MIGR_GEN_MIGR
)
8066 migr_map
= get_imsm_map(dev
, MAP_1
);
8068 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
8069 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
8075 /* return true if we recorded new information */
8076 static int mark_failure(struct intel_super
*super
,
8077 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8081 struct imsm_map
*map
;
8082 char buf
[MAX_RAID_SERIAL_LEN
+3];
8083 unsigned int len
, shift
= 0;
8085 /* new failures are always set in map[0] */
8086 map
= get_imsm_map(dev
, MAP_0
);
8088 slot
= get_imsm_disk_slot(map
, idx
);
8092 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
8093 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
8096 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
8097 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
8099 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
8100 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
8101 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
8103 disk
->status
|= FAILED_DISK
;
8104 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
8105 /* mark failures in second map if second map exists and this disk
8107 * This is valid for migration, initialization and rebuild
8109 if (dev
->vol
.migr_state
) {
8110 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
8111 int slot2
= get_imsm_disk_slot(map2
, idx
);
8113 if (slot2
< map2
->num_members
&& slot2
>= 0)
8114 set_imsm_ord_tbl_ent(map2
, slot2
,
8115 idx
| IMSM_ORD_REBUILD
);
8117 if (map
->failed_disk_num
== 0xff)
8118 map
->failed_disk_num
= slot
;
8120 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
8125 static void mark_missing(struct intel_super
*super
,
8126 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
8128 mark_failure(super
, dev
, disk
, idx
);
8130 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
8133 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
8134 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
8137 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
8141 if (!super
->missing
)
8144 /* When orom adds replacement for missing disk it does
8145 * not remove entry of missing disk, but just updates map with
8146 * new added disk. So it is not enough just to test if there is
8147 * any missing disk, we have to look if there are any failed disks
8148 * in map to stop migration */
8150 dprintf("imsm: mark missing\n");
8151 /* end process for initialization and rebuild only
8153 if (is_gen_migration(dev
) == 0) {
8154 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8158 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8159 struct imsm_map
*map1
;
8160 int i
, ord
, ord_map1
;
8163 for (i
= 0; i
< map
->num_members
; i
++) {
8164 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8165 if (!(ord
& IMSM_ORD_REBUILD
))
8168 map1
= get_imsm_map(dev
, MAP_1
);
8172 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
8173 if (ord_map1
& IMSM_ORD_REBUILD
)
8178 map_state
= imsm_check_degraded(super
, dev
,
8180 end_migration(dev
, super
, map_state
);
8184 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
8185 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
8186 super
->updates_pending
++;
8189 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
8192 unsigned long long array_blocks
;
8193 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8194 int used_disks
= imsm_num_data_members(map
);
8196 if (used_disks
== 0) {
8197 /* when problems occures
8198 * return current array_blocks value
8200 array_blocks
= imsm_dev_size(dev
);
8202 return array_blocks
;
8205 /* set array size in metadata
8208 /* OLCE size change is caused by added disks
8210 array_blocks
= per_dev_array_size(map
) * used_disks
;
8212 /* Online Volume Size Change
8213 * Using available free space
8215 array_blocks
= new_size
;
8217 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8218 set_imsm_dev_size(dev
, array_blocks
);
8220 return array_blocks
;
8223 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8225 static void imsm_progress_container_reshape(struct intel_super
*super
)
8227 /* if no device has a migr_state, but some device has a
8228 * different number of members than the previous device, start
8229 * changing the number of devices in this device to match
8232 struct imsm_super
*mpb
= super
->anchor
;
8233 int prev_disks
= -1;
8237 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8238 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8239 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8240 struct imsm_map
*map2
;
8241 int prev_num_members
;
8243 if (dev
->vol
.migr_state
)
8246 if (prev_disks
== -1)
8247 prev_disks
= map
->num_members
;
8248 if (prev_disks
== map
->num_members
)
8251 /* OK, this array needs to enter reshape mode.
8252 * i.e it needs a migr_state
8255 copy_map_size
= sizeof_imsm_map(map
);
8256 prev_num_members
= map
->num_members
;
8257 map
->num_members
= prev_disks
;
8258 dev
->vol
.migr_state
= 1;
8259 dev
->vol
.curr_migr_unit
= 0;
8260 set_migr_type(dev
, MIGR_GEN_MIGR
);
8261 for (i
= prev_num_members
;
8262 i
< map
->num_members
; i
++)
8263 set_imsm_ord_tbl_ent(map
, i
, i
);
8264 map2
= get_imsm_map(dev
, MAP_1
);
8265 /* Copy the current map */
8266 memcpy(map2
, map
, copy_map_size
);
8267 map2
->num_members
= prev_num_members
;
8269 imsm_set_array_size(dev
, -1);
8270 super
->clean_migration_record_by_mdmon
= 1;
8271 super
->updates_pending
++;
8275 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8276 * states are handled in imsm_set_disk() with one exception, when a
8277 * resync is stopped due to a new failure this routine will set the
8278 * 'degraded' state for the array.
8280 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8282 int inst
= a
->info
.container_member
;
8283 struct intel_super
*super
= a
->container
->sb
;
8284 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8285 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8286 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8287 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8288 __u32 blocks_per_unit
;
8290 if (dev
->vol
.migr_state
&&
8291 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8292 /* array state change is blocked due to reshape action
8294 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8295 * - finish the reshape (if last_checkpoint is big and action != reshape)
8296 * - update curr_migr_unit
8298 if (a
->curr_action
== reshape
) {
8299 /* still reshaping, maybe update curr_migr_unit */
8300 goto mark_checkpoint
;
8302 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8303 /* for some reason we aborted the reshape.
8305 * disable automatic metadata rollback
8306 * user action is required to recover process
8309 struct imsm_map
*map2
=
8310 get_imsm_map(dev
, MAP_1
);
8311 dev
->vol
.migr_state
= 0;
8312 set_migr_type(dev
, 0);
8313 dev
->vol
.curr_migr_unit
= 0;
8315 sizeof_imsm_map(map2
));
8316 super
->updates_pending
++;
8319 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8320 unsigned long long array_blocks
;
8324 used_disks
= imsm_num_data_members(map
);
8325 if (used_disks
> 0) {
8327 per_dev_array_size(map
) *
8330 round_size_to_mb(array_blocks
,
8332 a
->info
.custom_array_size
= array_blocks
;
8333 /* encourage manager to update array
8337 a
->check_reshape
= 1;
8339 /* finalize online capacity expansion/reshape */
8340 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8342 mdi
->disk
.raid_disk
,
8345 imsm_progress_container_reshape(super
);
8350 /* before we activate this array handle any missing disks */
8351 if (consistent
== 2)
8352 handle_missing(super
, dev
);
8354 if (consistent
== 2 &&
8355 (!is_resync_complete(&a
->info
) ||
8356 map_state
!= IMSM_T_STATE_NORMAL
||
8357 dev
->vol
.migr_state
))
8360 if (is_resync_complete(&a
->info
)) {
8361 /* complete intialization / resync,
8362 * recovery and interrupted recovery is completed in
8365 if (is_resyncing(dev
)) {
8366 dprintf("imsm: mark resync done\n");
8367 end_migration(dev
, super
, map_state
);
8368 super
->updates_pending
++;
8369 a
->last_checkpoint
= 0;
8371 } else if ((!is_resyncing(dev
) && !failed
) &&
8372 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8373 /* mark the start of the init process if nothing is failed */
8374 dprintf("imsm: mark resync start\n");
8375 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8376 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8378 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8379 super
->updates_pending
++;
8383 /* skip checkpointing for general migration,
8384 * it is controlled in mdadm
8386 if (is_gen_migration(dev
))
8387 goto skip_mark_checkpoint
;
8389 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8390 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8391 if (blocks_per_unit
) {
8395 units
= a
->last_checkpoint
/ blocks_per_unit
;
8398 /* check that we did not overflow 32-bits, and that
8399 * curr_migr_unit needs updating
8401 if (units32
== units
&&
8403 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8404 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8405 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8406 super
->updates_pending
++;
8410 skip_mark_checkpoint
:
8411 /* mark dirty / clean */
8412 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8413 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8414 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8416 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8418 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8419 if (dev
->rwh_policy
== RWH_DISTRIBUTED
||
8420 dev
->rwh_policy
== RWH_MULTIPLE_DISTRIBUTED
)
8421 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8423 super
->updates_pending
++;
8429 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8431 int inst
= a
->info
.container_member
;
8432 struct intel_super
*super
= a
->container
->sb
;
8433 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8434 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8436 if (slot
> map
->num_members
) {
8437 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8438 slot
, map
->num_members
- 1);
8445 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8448 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8450 int inst
= a
->info
.container_member
;
8451 struct intel_super
*super
= a
->container
->sb
;
8452 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8453 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8454 struct imsm_disk
*disk
;
8456 int recovery_not_finished
= 0;
8460 int rebuild_done
= 0;
8463 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8467 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8468 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8470 /* check for new failures */
8471 if (state
& DS_FAULTY
) {
8472 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8473 super
->updates_pending
++;
8476 /* check if in_sync */
8477 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8478 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8480 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8482 super
->updates_pending
++;
8485 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8486 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8488 /* check if recovery complete, newly degraded, or failed */
8489 dprintf("imsm: Detected transition to state ");
8490 switch (map_state
) {
8491 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8492 dprintf("normal: ");
8493 if (is_rebuilding(dev
)) {
8494 dprintf_cont("while rebuilding");
8495 /* check if recovery is really finished */
8496 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8497 if (mdi
->recovery_start
!= MaxSector
) {
8498 recovery_not_finished
= 1;
8501 if (recovery_not_finished
) {
8503 dprintf("Rebuild has not finished yet, state not changed");
8504 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8505 a
->last_checkpoint
= mdi
->recovery_start
;
8506 super
->updates_pending
++;
8510 end_migration(dev
, super
, map_state
);
8511 map
= get_imsm_map(dev
, MAP_0
);
8512 map
->failed_disk_num
= ~0;
8513 super
->updates_pending
++;
8514 a
->last_checkpoint
= 0;
8517 if (is_gen_migration(dev
)) {
8518 dprintf_cont("while general migration");
8519 if (a
->last_checkpoint
>= a
->info
.component_size
)
8520 end_migration(dev
, super
, map_state
);
8522 map
->map_state
= map_state
;
8523 map
= get_imsm_map(dev
, MAP_0
);
8524 map
->failed_disk_num
= ~0;
8525 super
->updates_pending
++;
8529 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8530 dprintf_cont("degraded: ");
8531 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8532 dprintf_cont("mark degraded");
8533 map
->map_state
= map_state
;
8534 super
->updates_pending
++;
8535 a
->last_checkpoint
= 0;
8538 if (is_rebuilding(dev
)) {
8539 dprintf_cont("while rebuilding.");
8540 if (map
->map_state
!= map_state
) {
8541 dprintf_cont(" Map state change");
8542 end_migration(dev
, super
, map_state
);
8543 super
->updates_pending
++;
8544 } else if (!rebuild_done
) {
8548 /* check if recovery is really finished */
8549 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8550 if (mdi
->recovery_start
!= MaxSector
) {
8551 recovery_not_finished
= 1;
8554 if (recovery_not_finished
) {
8556 dprintf("Rebuild has not finished yet, state not changed");
8557 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8558 a
->last_checkpoint
=
8559 mdi
->recovery_start
;
8560 super
->updates_pending
++;
8565 dprintf_cont(" Rebuild done, still degraded");
8566 dev
->vol
.migr_state
= 0;
8567 set_migr_type(dev
, 0);
8568 dev
->vol
.curr_migr_unit
= 0;
8570 for (i
= 0; i
< map
->num_members
; i
++) {
8571 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8573 if (idx
& IMSM_ORD_REBUILD
)
8574 map
->failed_disk_num
= i
;
8576 super
->updates_pending
++;
8579 if (is_gen_migration(dev
)) {
8580 dprintf_cont("while general migration");
8581 if (a
->last_checkpoint
>= a
->info
.component_size
)
8582 end_migration(dev
, super
, map_state
);
8584 map
->map_state
= map_state
;
8585 manage_second_map(super
, dev
);
8587 super
->updates_pending
++;
8590 if (is_initializing(dev
)) {
8591 dprintf_cont("while initialization.");
8592 map
->map_state
= map_state
;
8593 super
->updates_pending
++;
8597 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8598 dprintf_cont("failed: ");
8599 if (is_gen_migration(dev
)) {
8600 dprintf_cont("while general migration");
8601 map
->map_state
= map_state
;
8602 super
->updates_pending
++;
8605 if (map
->map_state
!= map_state
) {
8606 dprintf_cont("mark failed");
8607 end_migration(dev
, super
, map_state
);
8608 super
->updates_pending
++;
8609 a
->last_checkpoint
= 0;
8614 dprintf_cont("state %i\n", map_state
);
8619 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8622 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8623 unsigned long long dsize
;
8624 unsigned long long sectors
;
8625 unsigned int sector_size
;
8627 get_dev_sector_size(fd
, NULL
, §or_size
);
8628 get_dev_size(fd
, NULL
, &dsize
);
8630 if (mpb_size
> sector_size
) {
8631 /* -1 to account for anchor */
8632 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8634 /* write the extended mpb to the sectors preceeding the anchor */
8635 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8639 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8640 sector_size
* sectors
) != sector_size
* sectors
)
8644 /* first block is stored on second to last sector of the disk */
8645 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8648 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8654 static void imsm_sync_metadata(struct supertype
*container
)
8656 struct intel_super
*super
= container
->sb
;
8658 dprintf("sync metadata: %d\n", super
->updates_pending
);
8659 if (!super
->updates_pending
)
8662 write_super_imsm(container
, 0);
8664 super
->updates_pending
= 0;
8667 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8669 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8670 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8673 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8677 if (dl
&& is_failed(&dl
->disk
))
8681 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8686 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8687 struct active_array
*a
, int activate_new
,
8688 struct mdinfo
*additional_test_list
)
8690 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8691 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8692 struct imsm_super
*mpb
= super
->anchor
;
8693 struct imsm_map
*map
;
8694 unsigned long long pos
;
8699 __u32 array_start
= 0;
8700 __u32 array_end
= 0;
8702 struct mdinfo
*test_list
;
8704 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8705 /* If in this array, skip */
8706 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8707 if (d
->state_fd
>= 0 &&
8708 d
->disk
.major
== dl
->major
&&
8709 d
->disk
.minor
== dl
->minor
) {
8710 dprintf("%x:%x already in array\n",
8711 dl
->major
, dl
->minor
);
8716 test_list
= additional_test_list
;
8718 if (test_list
->disk
.major
== dl
->major
&&
8719 test_list
->disk
.minor
== dl
->minor
) {
8720 dprintf("%x:%x already in additional test list\n",
8721 dl
->major
, dl
->minor
);
8724 test_list
= test_list
->next
;
8729 /* skip in use or failed drives */
8730 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8732 dprintf("%x:%x status (failed: %d index: %d)\n",
8733 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8737 /* skip pure spares when we are looking for partially
8738 * assimilated drives
8740 if (dl
->index
== -1 && !activate_new
)
8743 if (!drive_validate_sector_size(super
, dl
))
8746 /* Does this unused device have the requisite free space?
8747 * It needs to be able to cover all member volumes
8749 ex
= get_extents(super
, dl
);
8751 dprintf("cannot get extents\n");
8754 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8755 dev
= get_imsm_dev(super
, i
);
8756 map
= get_imsm_map(dev
, MAP_0
);
8758 /* check if this disk is already a member of
8761 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8767 array_start
= pba_of_lba0(map
);
8768 array_end
= array_start
+
8769 per_dev_array_size(map
) - 1;
8772 /* check that we can start at pba_of_lba0 with
8773 * num_data_stripes*blocks_per_stripe of space
8775 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8779 pos
= ex
[j
].start
+ ex
[j
].size
;
8781 } while (ex
[j
-1].size
);
8788 if (i
< mpb
->num_raid_devs
) {
8789 dprintf("%x:%x does not have %u to %u available\n",
8790 dl
->major
, dl
->minor
, array_start
, array_end
);
8800 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8802 struct imsm_dev
*dev2
;
8803 struct imsm_map
*map
;
8809 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8811 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8812 if (state
== IMSM_T_STATE_FAILED
) {
8813 map
= get_imsm_map(dev2
, MAP_0
);
8816 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8818 * Check if failed disks are deleted from intel
8819 * disk list or are marked to be deleted
8821 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8822 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8824 * Do not rebuild the array if failed disks
8825 * from failed sub-array are not removed from
8829 is_failed(&idisk
->disk
) &&
8830 (idisk
->action
!= DISK_REMOVE
))
8838 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8839 struct metadata_update
**updates
)
8842 * Find a device with unused free space and use it to replace a
8843 * failed/vacant region in an array. We replace failed regions one a
8844 * array at a time. The result is that a new spare disk will be added
8845 * to the first failed array and after the monitor has finished
8846 * propagating failures the remainder will be consumed.
8848 * FIXME add a capability for mdmon to request spares from another
8852 struct intel_super
*super
= a
->container
->sb
;
8853 int inst
= a
->info
.container_member
;
8854 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8855 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8856 int failed
= a
->info
.array
.raid_disks
;
8857 struct mdinfo
*rv
= NULL
;
8860 struct metadata_update
*mu
;
8862 struct imsm_update_activate_spare
*u
;
8867 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8868 if ((d
->curr_state
& DS_FAULTY
) &&
8870 /* wait for Removal to happen */
8872 if (d
->state_fd
>= 0)
8876 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8877 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8879 if (imsm_reshape_blocks_arrays_changes(super
))
8882 /* Cannot activate another spare if rebuild is in progress already
8884 if (is_rebuilding(dev
)) {
8885 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8889 if (a
->info
.array
.level
== 4)
8890 /* No repair for takeovered array
8891 * imsm doesn't support raid4
8895 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8896 IMSM_T_STATE_DEGRADED
)
8899 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8900 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8905 * If there are any failed disks check state of the other volume.
8906 * Block rebuild if the another one is failed until failed disks
8907 * are removed from container.
8910 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8911 MAX_RAID_SERIAL_LEN
, dev
->volume
);
8912 /* check if states of the other volumes allow for rebuild */
8913 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
8915 allowed
= imsm_rebuild_allowed(a
->container
,
8923 /* For each slot, if it is not working, find a spare */
8924 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
8925 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8926 if (d
->disk
.raid_disk
== i
)
8928 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
8929 if (d
&& (d
->state_fd
>= 0))
8933 * OK, this device needs recovery. Try to re-add the
8934 * previous occupant of this slot, if this fails see if
8935 * we can continue the assimilation of a spare that was
8936 * partially assimilated, finally try to activate a new
8939 dl
= imsm_readd(super
, i
, a
);
8941 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
8943 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
8947 /* found a usable disk with enough space */
8948 di
= xcalloc(1, sizeof(*di
));
8950 /* dl->index will be -1 in the case we are activating a
8951 * pristine spare. imsm_process_update() will create a
8952 * new index in this case. Once a disk is found to be
8953 * failed in all member arrays it is kicked from the
8956 di
->disk
.number
= dl
->index
;
8958 /* (ab)use di->devs to store a pointer to the device
8961 di
->devs
= (struct mdinfo
*) dl
;
8963 di
->disk
.raid_disk
= i
;
8964 di
->disk
.major
= dl
->major
;
8965 di
->disk
.minor
= dl
->minor
;
8967 di
->recovery_start
= 0;
8968 di
->data_offset
= pba_of_lba0(map
);
8969 di
->component_size
= a
->info
.component_size
;
8970 di
->container_member
= inst
;
8971 di
->bb
.supported
= 1;
8972 if (a
->info
.consistency_policy
== CONSISTENCY_POLICY_PPL
) {
8973 di
->ppl_sector
= get_ppl_sector(super
, inst
);
8974 di
->ppl_size
= MULTIPLE_PPL_AREA_SIZE_IMSM
>> 9;
8976 super
->random
= random32();
8980 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
8981 i
, di
->data_offset
);
8985 /* No spares found */
8987 /* Now 'rv' has a list of devices to return.
8988 * Create a metadata_update record to update the
8989 * disk_ord_tbl for the array
8991 mu
= xmalloc(sizeof(*mu
));
8992 mu
->buf
= xcalloc(num_spares
,
8993 sizeof(struct imsm_update_activate_spare
));
8995 mu
->space_list
= NULL
;
8996 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
8997 mu
->next
= *updates
;
8998 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
9000 for (di
= rv
; di
; di
= di
->next
) {
9001 u
->type
= update_activate_spare
;
9002 u
->dl
= (struct dl
*) di
->devs
;
9004 u
->slot
= di
->disk
.raid_disk
;
9015 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
9017 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
9018 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9019 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9020 struct disk_info
*inf
= get_disk_info(u
);
9021 struct imsm_disk
*disk
;
9025 for (i
= 0; i
< map
->num_members
; i
++) {
9026 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
9027 for (j
= 0; j
< new_map
->num_members
; j
++)
9028 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
9035 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
9039 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9040 if (dl
->major
== major
&& dl
->minor
== minor
)
9045 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
9051 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
9052 if (dl
->major
== major
&& dl
->minor
== minor
) {
9055 prev
->next
= dl
->next
;
9057 super
->disks
= dl
->next
;
9059 __free_imsm_disk(dl
);
9060 dprintf("removed %x:%x\n", major
, minor
);
9068 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
9070 static int add_remove_disk_update(struct intel_super
*super
)
9072 int check_degraded
= 0;
9075 /* add/remove some spares to/from the metadata/contrainer */
9076 while (super
->disk_mgmt_list
) {
9077 struct dl
*disk_cfg
;
9079 disk_cfg
= super
->disk_mgmt_list
;
9080 super
->disk_mgmt_list
= disk_cfg
->next
;
9081 disk_cfg
->next
= NULL
;
9083 if (disk_cfg
->action
== DISK_ADD
) {
9084 disk_cfg
->next
= super
->disks
;
9085 super
->disks
= disk_cfg
;
9087 dprintf("added %x:%x\n",
9088 disk_cfg
->major
, disk_cfg
->minor
);
9089 } else if (disk_cfg
->action
== DISK_REMOVE
) {
9090 dprintf("Disk remove action processed: %x.%x\n",
9091 disk_cfg
->major
, disk_cfg
->minor
);
9092 disk
= get_disk_super(super
,
9096 /* store action status */
9097 disk
->action
= DISK_REMOVE
;
9098 /* remove spare disks only */
9099 if (disk
->index
== -1) {
9100 remove_disk_super(super
,
9105 /* release allocate disk structure */
9106 __free_imsm_disk(disk_cfg
);
9109 return check_degraded
;
9112 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
9113 struct intel_super
*super
,
9116 struct intel_dev
*id
;
9117 void **tofree
= NULL
;
9120 dprintf("(enter)\n");
9121 if (u
->subdev
< 0 || u
->subdev
> 1) {
9122 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9125 if (space_list
== NULL
|| *space_list
== NULL
) {
9126 dprintf("imsm: Error: Memory is not allocated\n");
9130 for (id
= super
->devlist
; id
; id
= id
->next
) {
9131 if (id
->index
== (unsigned)u
->subdev
) {
9132 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9133 struct imsm_map
*map
;
9134 struct imsm_dev
*new_dev
=
9135 (struct imsm_dev
*)*space_list
;
9136 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
9138 struct dl
*new_disk
;
9140 if (new_dev
== NULL
)
9142 *space_list
= **space_list
;
9143 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
9144 map
= get_imsm_map(new_dev
, MAP_0
);
9146 dprintf("imsm: Error: migration in progress");
9150 to_state
= map
->map_state
;
9151 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
9153 /* this should not happen */
9154 if (u
->new_disks
[0] < 0) {
9155 map
->failed_disk_num
=
9156 map
->num_members
- 1;
9157 to_state
= IMSM_T_STATE_DEGRADED
;
9159 to_state
= IMSM_T_STATE_NORMAL
;
9161 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
9162 if (u
->new_level
> -1)
9163 map
->raid_level
= u
->new_level
;
9164 migr_map
= get_imsm_map(new_dev
, MAP_1
);
9165 if ((u
->new_level
== 5) &&
9166 (migr_map
->raid_level
== 0)) {
9167 int ord
= map
->num_members
- 1;
9168 migr_map
->num_members
--;
9169 if (u
->new_disks
[0] < 0)
9170 ord
|= IMSM_ORD_REBUILD
;
9171 set_imsm_ord_tbl_ent(map
,
9172 map
->num_members
- 1,
9176 tofree
= (void **)dev
;
9178 /* update chunk size
9180 if (u
->new_chunksize
> 0) {
9181 unsigned long long num_data_stripes
;
9182 struct imsm_map
*dest_map
=
9183 get_imsm_map(dev
, MAP_0
);
9185 imsm_num_data_members(dest_map
);
9187 if (used_disks
== 0)
9190 map
->blocks_per_strip
=
9191 __cpu_to_le16(u
->new_chunksize
* 2);
9193 imsm_dev_size(dev
) / used_disks
;
9194 num_data_stripes
/= map
->blocks_per_strip
;
9195 num_data_stripes
/= map
->num_domains
;
9196 set_num_data_stripes(map
, num_data_stripes
);
9199 /* ensure blocks_per_member has valid value
9201 set_blocks_per_member(map
,
9202 per_dev_array_size(map
) +
9203 NUM_BLOCKS_DIRTY_STRIPE_REGION
);
9207 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
9208 migr_map
->raid_level
== map
->raid_level
)
9211 if (u
->new_disks
[0] >= 0) {
9214 new_disk
= get_disk_super(super
,
9215 major(u
->new_disks
[0]),
9216 minor(u
->new_disks
[0]));
9217 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9218 major(u
->new_disks
[0]),
9219 minor(u
->new_disks
[0]),
9220 new_disk
, new_disk
->index
);
9221 if (new_disk
== NULL
)
9222 goto error_disk_add
;
9224 new_disk
->index
= map
->num_members
- 1;
9225 /* slot to fill in autolayout
9227 new_disk
->raiddisk
= new_disk
->index
;
9228 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9229 new_disk
->disk
.status
&= ~SPARE_DISK
;
9231 goto error_disk_add
;
9234 *tofree
= *space_list
;
9235 /* calculate new size
9237 imsm_set_array_size(new_dev
, -1);
9244 *space_list
= tofree
;
9248 dprintf("Error: imsm: Cannot find disk.\n");
9252 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9253 struct intel_super
*super
)
9255 struct intel_dev
*id
;
9258 dprintf("(enter)\n");
9259 if (u
->subdev
< 0 || u
->subdev
> 1) {
9260 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9264 for (id
= super
->devlist
; id
; id
= id
->next
) {
9265 if (id
->index
== (unsigned)u
->subdev
) {
9266 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9267 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9268 int used_disks
= imsm_num_data_members(map
);
9269 unsigned long long blocks_per_member
;
9270 unsigned long long num_data_stripes
;
9271 unsigned long long new_size_per_disk
;
9273 if (used_disks
== 0)
9276 /* calculate new size
9278 new_size_per_disk
= u
->new_size
/ used_disks
;
9279 blocks_per_member
= new_size_per_disk
+
9280 NUM_BLOCKS_DIRTY_STRIPE_REGION
;
9281 num_data_stripes
= new_size_per_disk
/
9282 map
->blocks_per_strip
;
9283 num_data_stripes
/= map
->num_domains
;
9284 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9285 u
->new_size
, new_size_per_disk
,
9287 set_blocks_per_member(map
, blocks_per_member
);
9288 set_num_data_stripes(map
, num_data_stripes
);
9289 imsm_set_array_size(dev
, u
->new_size
);
9299 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9300 struct intel_super
*super
,
9301 struct active_array
*active_array
)
9303 struct imsm_super
*mpb
= super
->anchor
;
9304 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9305 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9306 struct imsm_map
*migr_map
;
9307 struct active_array
*a
;
9308 struct imsm_disk
*disk
;
9315 int second_map_created
= 0;
9317 for (; u
; u
= u
->next
) {
9318 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9323 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9328 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9333 /* count failures (excluding rebuilds and the victim)
9334 * to determine map[0] state
9337 for (i
= 0; i
< map
->num_members
; i
++) {
9340 disk
= get_imsm_disk(super
,
9341 get_imsm_disk_idx(dev
, i
, MAP_X
));
9342 if (!disk
|| is_failed(disk
))
9346 /* adding a pristine spare, assign a new index */
9347 if (dl
->index
< 0) {
9348 dl
->index
= super
->anchor
->num_disks
;
9349 super
->anchor
->num_disks
++;
9352 disk
->status
|= CONFIGURED_DISK
;
9353 disk
->status
&= ~SPARE_DISK
;
9356 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9357 if (!second_map_created
) {
9358 second_map_created
= 1;
9359 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9360 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9362 map
->map_state
= to_state
;
9363 migr_map
= get_imsm_map(dev
, MAP_1
);
9364 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9365 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9366 dl
->index
| IMSM_ORD_REBUILD
);
9368 /* update the family_num to mark a new container
9369 * generation, being careful to record the existing
9370 * family_num in orig_family_num to clean up after
9371 * earlier mdadm versions that neglected to set it.
9373 if (mpb
->orig_family_num
== 0)
9374 mpb
->orig_family_num
= mpb
->family_num
;
9375 mpb
->family_num
+= super
->random
;
9377 /* count arrays using the victim in the metadata */
9379 for (a
= active_array
; a
; a
= a
->next
) {
9380 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9381 map
= get_imsm_map(dev
, MAP_0
);
9383 if (get_imsm_disk_slot(map
, victim
) >= 0)
9387 /* delete the victim if it is no longer being
9393 /* We know that 'manager' isn't touching anything,
9394 * so it is safe to delete
9396 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9397 if ((*dlp
)->index
== victim
)
9400 /* victim may be on the missing list */
9402 for (dlp
= &super
->missing
; *dlp
;
9403 dlp
= &(*dlp
)->next
)
9404 if ((*dlp
)->index
== victim
)
9406 imsm_delete(super
, dlp
, victim
);
9413 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9414 struct intel_super
*super
,
9417 struct dl
*new_disk
;
9418 struct intel_dev
*id
;
9420 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9421 int disk_count
= u
->old_raid_disks
;
9422 void **tofree
= NULL
;
9423 int devices_to_reshape
= 1;
9424 struct imsm_super
*mpb
= super
->anchor
;
9426 unsigned int dev_id
;
9428 dprintf("(enter)\n");
9430 /* enable spares to use in array */
9431 for (i
= 0; i
< delta_disks
; i
++) {
9432 new_disk
= get_disk_super(super
,
9433 major(u
->new_disks
[i
]),
9434 minor(u
->new_disks
[i
]));
9435 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9436 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9437 new_disk
, new_disk
->index
);
9438 if (new_disk
== NULL
||
9439 (new_disk
->index
>= 0 &&
9440 new_disk
->index
< u
->old_raid_disks
))
9441 goto update_reshape_exit
;
9442 new_disk
->index
= disk_count
++;
9443 /* slot to fill in autolayout
9445 new_disk
->raiddisk
= new_disk
->index
;
9446 new_disk
->disk
.status
|=
9448 new_disk
->disk
.status
&= ~SPARE_DISK
;
9451 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9452 mpb
->num_raid_devs
);
9453 /* manage changes in volume
9455 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9456 void **sp
= *space_list
;
9457 struct imsm_dev
*newdev
;
9458 struct imsm_map
*newmap
, *oldmap
;
9460 for (id
= super
->devlist
; id
; id
= id
->next
) {
9461 if (id
->index
== dev_id
)
9470 /* Copy the dev, but not (all of) the map */
9471 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9472 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9473 newmap
= get_imsm_map(newdev
, MAP_0
);
9474 /* Copy the current map */
9475 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9476 /* update one device only
9478 if (devices_to_reshape
) {
9479 dprintf("imsm: modifying subdev: %i\n",
9481 devices_to_reshape
--;
9482 newdev
->vol
.migr_state
= 1;
9483 newdev
->vol
.curr_migr_unit
= 0;
9484 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9485 newmap
->num_members
= u
->new_raid_disks
;
9486 for (i
= 0; i
< delta_disks
; i
++) {
9487 set_imsm_ord_tbl_ent(newmap
,
9488 u
->old_raid_disks
+ i
,
9489 u
->old_raid_disks
+ i
);
9491 /* New map is correct, now need to save old map
9493 newmap
= get_imsm_map(newdev
, MAP_1
);
9494 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9496 imsm_set_array_size(newdev
, -1);
9499 sp
= (void **)id
->dev
;
9504 /* Clear migration record */
9505 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9508 *space_list
= tofree
;
9511 update_reshape_exit
:
9516 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9517 struct intel_super
*super
,
9520 struct imsm_dev
*dev
= NULL
;
9521 struct intel_dev
*dv
;
9522 struct imsm_dev
*dev_new
;
9523 struct imsm_map
*map
;
9527 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9528 if (dv
->index
== (unsigned int)u
->subarray
) {
9536 map
= get_imsm_map(dev
, MAP_0
);
9538 if (u
->direction
== R10_TO_R0
) {
9539 unsigned long long num_data_stripes
;
9541 map
->num_domains
= 1;
9542 num_data_stripes
= imsm_dev_size(dev
) / 2;
9543 num_data_stripes
/= map
->blocks_per_strip
;
9544 num_data_stripes
/= map
->num_domains
;
9545 set_num_data_stripes(map
, num_data_stripes
);
9547 /* Number of failed disks must be half of initial disk number */
9548 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9549 (map
->num_members
/ 2))
9552 /* iterate through devices to mark removed disks as spare */
9553 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9554 if (dm
->disk
.status
& FAILED_DISK
) {
9555 int idx
= dm
->index
;
9556 /* update indexes on the disk list */
9557 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9558 the index values will end up being correct.... NB */
9559 for (du
= super
->disks
; du
; du
= du
->next
)
9560 if (du
->index
> idx
)
9562 /* mark as spare disk */
9567 map
->num_members
= map
->num_members
/ 2;
9568 map
->map_state
= IMSM_T_STATE_NORMAL
;
9569 map
->num_domains
= 1;
9570 map
->raid_level
= 0;
9571 map
->failed_disk_num
= -1;
9574 if (u
->direction
== R0_TO_R10
) {
9576 /* update slots in current disk list */
9577 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9581 /* create new *missing* disks */
9582 for (i
= 0; i
< map
->num_members
; i
++) {
9583 space
= *space_list
;
9586 *space_list
= *space
;
9588 memcpy(du
, super
->disks
, sizeof(*du
));
9592 du
->index
= (i
* 2) + 1;
9593 sprintf((char *)du
->disk
.serial
,
9594 " MISSING_%d", du
->index
);
9595 sprintf((char *)du
->serial
,
9596 "MISSING_%d", du
->index
);
9597 du
->next
= super
->missing
;
9598 super
->missing
= du
;
9600 /* create new dev and map */
9601 space
= *space_list
;
9604 *space_list
= *space
;
9605 dev_new
= (void *)space
;
9606 memcpy(dev_new
, dev
, sizeof(*dev
));
9607 /* update new map */
9608 map
= get_imsm_map(dev_new
, MAP_0
);
9609 map
->num_members
= map
->num_members
* 2;
9610 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9611 map
->num_domains
= 2;
9612 map
->raid_level
= 1;
9613 /* replace dev<->dev_new */
9616 /* update disk order table */
9617 for (du
= super
->disks
; du
; du
= du
->next
)
9619 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9620 for (du
= super
->missing
; du
; du
= du
->next
)
9621 if (du
->index
>= 0) {
9622 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9623 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9629 static void imsm_process_update(struct supertype
*st
,
9630 struct metadata_update
*update
)
9633 * crack open the metadata_update envelope to find the update record
9634 * update can be one of:
9635 * update_reshape_container_disks - all the arrays in the container
9636 * are being reshaped to have more devices. We need to mark
9637 * the arrays for general migration and convert selected spares
9638 * into active devices.
9639 * update_activate_spare - a spare device has replaced a failed
9640 * device in an array, update the disk_ord_tbl. If this disk is
9641 * present in all member arrays then also clear the SPARE_DISK
9643 * update_create_array
9645 * update_rename_array
9646 * update_add_remove_disk
9648 struct intel_super
*super
= st
->sb
;
9649 struct imsm_super
*mpb
;
9650 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9652 /* update requires a larger buf but the allocation failed */
9653 if (super
->next_len
&& !super
->next_buf
) {
9654 super
->next_len
= 0;
9658 if (super
->next_buf
) {
9659 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9661 super
->len
= super
->next_len
;
9662 super
->buf
= super
->next_buf
;
9664 super
->next_len
= 0;
9665 super
->next_buf
= NULL
;
9668 mpb
= super
->anchor
;
9671 case update_general_migration_checkpoint
: {
9672 struct intel_dev
*id
;
9673 struct imsm_update_general_migration_checkpoint
*u
=
9674 (void *)update
->buf
;
9676 dprintf("called for update_general_migration_checkpoint\n");
9678 /* find device under general migration */
9679 for (id
= super
->devlist
; id
; id
= id
->next
) {
9680 if (is_gen_migration(id
->dev
)) {
9681 id
->dev
->vol
.curr_migr_unit
=
9682 __cpu_to_le32(u
->curr_migr_unit
);
9683 super
->updates_pending
++;
9688 case update_takeover
: {
9689 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9690 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9691 imsm_update_version_info(super
);
9692 super
->updates_pending
++;
9697 case update_reshape_container_disks
: {
9698 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9699 if (apply_reshape_container_disks_update(
9700 u
, super
, &update
->space_list
))
9701 super
->updates_pending
++;
9704 case update_reshape_migration
: {
9705 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9706 if (apply_reshape_migration_update(
9707 u
, super
, &update
->space_list
))
9708 super
->updates_pending
++;
9711 case update_size_change
: {
9712 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9713 if (apply_size_change_update(u
, super
))
9714 super
->updates_pending
++;
9717 case update_activate_spare
: {
9718 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9719 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9720 super
->updates_pending
++;
9723 case update_create_array
: {
9724 /* someone wants to create a new array, we need to be aware of
9725 * a few races/collisions:
9726 * 1/ 'Create' called by two separate instances of mdadm
9727 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9728 * devices that have since been assimilated via
9730 * In the event this update can not be carried out mdadm will
9731 * (FIX ME) notice that its update did not take hold.
9733 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9734 struct intel_dev
*dv
;
9735 struct imsm_dev
*dev
;
9736 struct imsm_map
*map
, *new_map
;
9737 unsigned long long start
, end
;
9738 unsigned long long new_start
, new_end
;
9740 struct disk_info
*inf
;
9743 /* handle racing creates: first come first serve */
9744 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9745 dprintf("subarray %d already defined\n", u
->dev_idx
);
9749 /* check update is next in sequence */
9750 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9751 dprintf("can not create array %d expected index %d\n",
9752 u
->dev_idx
, mpb
->num_raid_devs
);
9756 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9757 new_start
= pba_of_lba0(new_map
);
9758 new_end
= new_start
+ per_dev_array_size(new_map
);
9759 inf
= get_disk_info(u
);
9761 /* handle activate_spare versus create race:
9762 * check to make sure that overlapping arrays do not include
9765 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9766 dev
= get_imsm_dev(super
, i
);
9767 map
= get_imsm_map(dev
, MAP_0
);
9768 start
= pba_of_lba0(map
);
9769 end
= start
+ per_dev_array_size(map
);
9770 if ((new_start
>= start
&& new_start
<= end
) ||
9771 (start
>= new_start
&& start
<= new_end
))
9776 if (disks_overlap(super
, i
, u
)) {
9777 dprintf("arrays overlap\n");
9782 /* check that prepare update was successful */
9783 if (!update
->space
) {
9784 dprintf("prepare update failed\n");
9788 /* check that all disks are still active before committing
9789 * changes. FIXME: could we instead handle this by creating a
9790 * degraded array? That's probably not what the user expects,
9791 * so better to drop this update on the floor.
9793 for (i
= 0; i
< new_map
->num_members
; i
++) {
9794 dl
= serial_to_dl(inf
[i
].serial
, super
);
9796 dprintf("disk disappeared\n");
9801 super
->updates_pending
++;
9803 /* convert spares to members and fixup ord_tbl */
9804 for (i
= 0; i
< new_map
->num_members
; i
++) {
9805 dl
= serial_to_dl(inf
[i
].serial
, super
);
9806 if (dl
->index
== -1) {
9807 dl
->index
= mpb
->num_disks
;
9809 dl
->disk
.status
|= CONFIGURED_DISK
;
9810 dl
->disk
.status
&= ~SPARE_DISK
;
9812 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9817 update
->space
= NULL
;
9818 imsm_copy_dev(dev
, &u
->dev
);
9819 dv
->index
= u
->dev_idx
;
9820 dv
->next
= super
->devlist
;
9821 super
->devlist
= dv
;
9822 mpb
->num_raid_devs
++;
9824 imsm_update_version_info(super
);
9827 /* mdmon knows how to release update->space, but not
9828 * ((struct intel_dev *) update->space)->dev
9830 if (update
->space
) {
9836 case update_kill_array
: {
9837 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9838 int victim
= u
->dev_idx
;
9839 struct active_array
*a
;
9840 struct intel_dev
**dp
;
9841 struct imsm_dev
*dev
;
9843 /* sanity check that we are not affecting the uuid of
9844 * active arrays, or deleting an active array
9846 * FIXME when immutable ids are available, but note that
9847 * we'll also need to fixup the invalidated/active
9848 * subarray indexes in mdstat
9850 for (a
= st
->arrays
; a
; a
= a
->next
)
9851 if (a
->info
.container_member
>= victim
)
9853 /* by definition if mdmon is running at least one array
9854 * is active in the container, so checking
9855 * mpb->num_raid_devs is just extra paranoia
9857 dev
= get_imsm_dev(super
, victim
);
9858 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9859 dprintf("failed to delete subarray-%d\n", victim
);
9863 for (dp
= &super
->devlist
; *dp
;)
9864 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9867 if ((*dp
)->index
> (unsigned)victim
)
9871 mpb
->num_raid_devs
--;
9872 super
->updates_pending
++;
9875 case update_rename_array
: {
9876 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9877 char name
[MAX_RAID_SERIAL_LEN
+1];
9878 int target
= u
->dev_idx
;
9879 struct active_array
*a
;
9880 struct imsm_dev
*dev
;
9882 /* sanity check that we are not affecting the uuid of
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 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
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
,