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)
86 /* Define attributes that are unused but not harmful */
87 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
89 #define MPB_SECTOR_CNT 2210
90 #define IMSM_RESERVED_SECTORS 4096
91 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
92 #define SECT_PER_MB_SHIFT 11
94 /* Disk configuration info. */
95 #define IMSM_MAX_DEVICES 255
97 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
98 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
99 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
100 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
101 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
102 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
103 __u32 status
; /* 0xF0 - 0xF3 */
104 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
105 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
106 #define IMSM_DISK_FILLERS 3
107 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
110 /* map selector for map managment
116 /* RAID map configuration infos. */
118 __u32 pba_of_lba0_lo
; /* start address of partition */
119 __u32 blocks_per_member_lo
;/* blocks per member */
120 __u32 num_data_stripes_lo
; /* number of data stripes */
121 __u16 blocks_per_strip
;
122 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
123 #define IMSM_T_STATE_NORMAL 0
124 #define IMSM_T_STATE_UNINITIALIZED 1
125 #define IMSM_T_STATE_DEGRADED 2
126 #define IMSM_T_STATE_FAILED 3
128 #define IMSM_T_RAID0 0
129 #define IMSM_T_RAID1 1
130 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
131 __u8 num_members
; /* number of member disks */
132 __u8 num_domains
; /* number of parity domains */
133 __u8 failed_disk_num
; /* valid only when state is degraded */
135 __u32 pba_of_lba0_hi
;
136 __u32 blocks_per_member_hi
;
137 __u32 num_data_stripes_hi
;
138 __u32 filler
[4]; /* expansion area */
139 #define IMSM_ORD_REBUILD (1 << 24)
140 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
141 * top byte contains some flags
143 } __attribute__ ((packed
));
146 __u32 curr_migr_unit
;
147 __u32 checkpoint_id
; /* id to access curr_migr_unit */
148 __u8 migr_state
; /* Normal or Migrating */
150 #define MIGR_REBUILD 1
151 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
152 #define MIGR_GEN_MIGR 3
153 #define MIGR_STATE_CHANGE 4
154 #define MIGR_REPAIR 5
155 __u8 migr_type
; /* Initializing, Rebuilding, ... */
157 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
158 __u16 verify_errors
; /* number of mismatches */
159 __u16 bad_blocks
; /* number of bad blocks during verify */
161 struct imsm_map map
[1];
162 /* here comes another one if migr_state */
163 } __attribute__ ((packed
));
166 __u8 volume
[MAX_RAID_SERIAL_LEN
];
169 #define DEV_BOOTABLE __cpu_to_le32(0x01)
170 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
171 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
172 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
173 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
174 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
175 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
176 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
177 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
178 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
179 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
180 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
181 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
182 __u32 status
; /* Persistent RaidDev status */
183 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
187 __u8 cng_master_disk
;
191 #define IMSM_DEV_FILLERS 10
192 __u32 filler
[IMSM_DEV_FILLERS
];
194 } __attribute__ ((packed
));
197 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
198 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
199 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
200 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
201 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
202 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
203 __u32 attributes
; /* 0x34 - 0x37 */
204 __u8 num_disks
; /* 0x38 Number of configured disks */
205 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
206 __u8 error_log_pos
; /* 0x3A */
207 __u8 fill
[1]; /* 0x3B */
208 __u32 cache_size
; /* 0x3c - 0x40 in mb */
209 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
210 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
211 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
212 #define IMSM_FILLERS 35
213 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
214 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
215 /* here comes imsm_dev[num_raid_devs] */
216 /* here comes BBM logs */
217 } __attribute__ ((packed
));
219 #define BBM_LOG_MAX_ENTRIES 254
221 struct bbm_log_entry
{
222 __u64 defective_block_start
;
223 #define UNREADABLE 0xFFFFFFFF
224 __u32 spare_block_offset
;
225 __u16 remapped_marked_count
;
227 } __attribute__ ((__packed__
));
230 __u32 signature
; /* 0xABADB10C */
232 __u32 reserved_spare_block_count
; /* 0 */
233 __u32 reserved
; /* 0xFFFF */
234 __u64 first_spare_lba
;
235 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
236 } __attribute__ ((__packed__
));
239 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
242 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
244 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
246 #define MIGR_REC_BUF_SIZE 512 /* size of migr_record i/o buffer */
247 #define MIGR_REC_POSITION 512 /* migr_record position offset on disk,
248 * MIGR_REC_BUF_SIZE <= MIGR_REC_POSITION
251 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
252 * be recovered using srcMap */
253 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
254 * already been migrated and must
255 * be recovered from checkpoint area */
257 __u32 rec_status
; /* Status used to determine how to restart
258 * migration in case it aborts
260 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
261 __u32 family_num
; /* Family number of MPB
262 * containing the RaidDev
263 * that is migrating */
264 __u32 ascending_migr
; /* True if migrating in increasing
266 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
267 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
269 * advances per unit-of-operation */
270 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
271 __u32 dest_1st_member_lba
; /* First member lba on first
272 * stripe of destination */
273 __u32 num_migr_units
; /* Total num migration units-of-op */
274 __u32 post_migr_vol_cap
; /* Size of volume after
275 * migration completes */
276 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
277 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
278 * migration ckpt record was read from
279 * (for recovered migrations) */
280 } __attribute__ ((__packed__
));
285 * 2: metadata does not match
293 struct md_list
*next
;
296 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
298 static __u8
migr_type(struct imsm_dev
*dev
)
300 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
301 dev
->status
& DEV_VERIFY_AND_FIX
)
304 return dev
->vol
.migr_type
;
307 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
309 /* for compatibility with older oroms convert MIGR_REPAIR, into
310 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
312 if (migr_type
== MIGR_REPAIR
) {
313 dev
->vol
.migr_type
= MIGR_VERIFY
;
314 dev
->status
|= DEV_VERIFY_AND_FIX
;
316 dev
->vol
.migr_type
= migr_type
;
317 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
321 static unsigned int sector_count(__u32 bytes
)
323 return ROUND_UP(bytes
, 512) / 512;
326 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
328 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
332 struct imsm_dev
*dev
;
333 struct intel_dev
*next
;
338 enum sys_dev_type type
;
341 struct intel_hba
*next
;
348 /* internal representation of IMSM metadata */
351 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
352 struct imsm_super
*anchor
; /* immovable parameters */
355 void *migr_rec_buf
; /* buffer for I/O operations */
356 struct migr_record
*migr_rec
; /* migration record */
358 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
359 array, it indicates that mdmon is allowed to clean migration
361 size_t len
; /* size of the 'buf' allocation */
362 void *next_buf
; /* for realloc'ing buf from the manager */
364 int updates_pending
; /* count of pending updates for mdmon */
365 int current_vol
; /* index of raid device undergoing creation */
366 unsigned long long create_offset
; /* common start for 'current_vol' */
367 __u32 random
; /* random data for seeding new family numbers */
368 struct intel_dev
*devlist
;
372 __u8 serial
[MAX_RAID_SERIAL_LEN
];
375 struct imsm_disk disk
;
378 struct extent
*e
; /* for determining freespace @ create */
379 int raiddisk
; /* slot to fill in autolayout */
381 } *disks
, *current_disk
;
382 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
384 struct dl
*missing
; /* disks removed while we weren't looking */
385 struct bbm_log
*bbm_log
;
386 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
387 const struct imsm_orom
*orom
; /* platform firmware support */
388 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
392 struct imsm_disk disk
;
393 #define IMSM_UNKNOWN_OWNER (-1)
395 struct intel_disk
*next
;
399 unsigned long long start
, size
;
402 /* definitions of reshape process types */
403 enum imsm_reshape_type
{
409 /* definition of messages passed to imsm_process_update */
410 enum imsm_update_type
{
411 update_activate_spare
,
415 update_add_remove_disk
,
416 update_reshape_container_disks
,
417 update_reshape_migration
,
419 update_general_migration_checkpoint
,
423 struct imsm_update_activate_spare
{
424 enum imsm_update_type type
;
428 struct imsm_update_activate_spare
*next
;
434 unsigned long long size
;
441 enum takeover_direction
{
445 struct imsm_update_takeover
{
446 enum imsm_update_type type
;
448 enum takeover_direction direction
;
451 struct imsm_update_reshape
{
452 enum imsm_update_type type
;
456 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
459 struct imsm_update_reshape_migration
{
460 enum imsm_update_type type
;
463 /* fields for array migration changes
470 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
473 struct imsm_update_size_change
{
474 enum imsm_update_type type
;
479 struct imsm_update_general_migration_checkpoint
{
480 enum imsm_update_type type
;
481 __u32 curr_migr_unit
;
485 __u8 serial
[MAX_RAID_SERIAL_LEN
];
488 struct imsm_update_create_array
{
489 enum imsm_update_type type
;
494 struct imsm_update_kill_array
{
495 enum imsm_update_type type
;
499 struct imsm_update_rename_array
{
500 enum imsm_update_type type
;
501 __u8 name
[MAX_RAID_SERIAL_LEN
];
505 struct imsm_update_add_remove_disk
{
506 enum imsm_update_type type
;
509 static const char *_sys_dev_type
[] = {
510 [SYS_DEV_UNKNOWN
] = "Unknown",
511 [SYS_DEV_SAS
] = "SAS",
512 [SYS_DEV_SATA
] = "SATA",
513 [SYS_DEV_NVME
] = "NVMe",
514 [SYS_DEV_VMD
] = "VMD"
517 const char *get_sys_dev_type(enum sys_dev_type type
)
519 if (type
>= SYS_DEV_MAX
)
520 type
= SYS_DEV_UNKNOWN
;
522 return _sys_dev_type
[type
];
525 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
527 struct intel_hba
*result
= xmalloc(sizeof(*result
));
529 result
->type
= device
->type
;
530 result
->path
= xstrdup(device
->path
);
532 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
538 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
540 struct intel_hba
*result
;
542 for (result
= hba
; result
; result
= result
->next
) {
543 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
549 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
551 struct intel_hba
*hba
;
553 /* check if disk attached to Intel HBA */
554 hba
= find_intel_hba(super
->hba
, device
);
557 /* Check if HBA is already attached to super */
558 if (super
->hba
== NULL
) {
559 super
->hba
= alloc_intel_hba(device
);
564 /* Intel metadata allows for all disks attached to the same type HBA.
565 * Do not support HBA types mixing
567 if (device
->type
!= hba
->type
)
570 /* Multiple same type HBAs can be used if they share the same OROM */
571 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
573 if (device_orom
!= super
->orom
)
579 hba
->next
= alloc_intel_hba(device
);
583 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
585 struct sys_dev
*list
, *elem
;
588 if ((list
= find_intel_devices()) == NULL
)
592 disk_path
= (char *) devname
;
594 disk_path
= diskfd_to_devpath(fd
);
599 for (elem
= list
; elem
; elem
= elem
->next
)
600 if (path_attached_to_hba(disk_path
, elem
->path
))
603 if (disk_path
!= devname
)
609 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
612 static struct supertype
*match_metadata_desc_imsm(char *arg
)
614 struct supertype
*st
;
616 if (strcmp(arg
, "imsm") != 0 &&
617 strcmp(arg
, "default") != 0
621 st
= xcalloc(1, sizeof(*st
));
622 st
->ss
= &super_imsm
;
623 st
->max_devs
= IMSM_MAX_DEVICES
;
624 st
->minor_version
= 0;
630 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
632 return &mpb
->sig
[MPB_SIG_LEN
];
636 /* retrieve a disk directly from the anchor when the anchor is known to be
637 * up-to-date, currently only at load time
639 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
641 if (index
>= mpb
->num_disks
)
643 return &mpb
->disk
[index
];
646 /* retrieve the disk description based on a index of the disk
649 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
653 for (d
= super
->disks
; d
; d
= d
->next
)
654 if (d
->index
== index
)
659 /* retrieve a disk from the parsed metadata */
660 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
664 dl
= get_imsm_dl_disk(super
, index
);
671 /* generate a checksum directly from the anchor when the anchor is known to be
672 * up-to-date, currently only at load or write_super after coalescing
674 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
676 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
677 __u32
*p
= (__u32
*) mpb
;
681 sum
+= __le32_to_cpu(*p
);
685 return sum
- __le32_to_cpu(mpb
->check_sum
);
688 static size_t sizeof_imsm_map(struct imsm_map
*map
)
690 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
693 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
695 /* A device can have 2 maps if it is in the middle of a migration.
697 * MAP_0 - we return the first map
698 * MAP_1 - we return the second map if it exists, else NULL
699 * MAP_X - we return the second map if it exists, else the first
701 struct imsm_map
*map
= &dev
->vol
.map
[0];
702 struct imsm_map
*map2
= NULL
;
704 if (dev
->vol
.migr_state
)
705 map2
= (void *)map
+ sizeof_imsm_map(map
);
707 switch (second_map
) {
724 /* return the size of the device.
725 * migr_state increases the returned size if map[0] were to be duplicated
727 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
729 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
730 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
732 /* migrating means an additional map */
733 if (dev
->vol
.migr_state
)
734 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
736 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
742 /* retrieve disk serial number list from a metadata update */
743 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
746 struct disk_info
*inf
;
748 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
749 sizeof_imsm_dev(&update
->dev
, 0);
755 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
761 if (index
>= mpb
->num_raid_devs
)
764 /* devices start after all disks */
765 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
767 for (i
= 0; i
<= index
; i
++)
769 return _mpb
+ offset
;
771 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
776 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
778 struct intel_dev
*dv
;
780 if (index
>= super
->anchor
->num_raid_devs
)
782 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
783 if (dv
->index
== index
)
790 * == MAP_0 get first map
791 * == MAP_1 get second map
792 * == MAP_X than get map according to the current migr_state
794 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
798 struct imsm_map
*map
;
800 map
= get_imsm_map(dev
, second_map
);
802 /* top byte identifies disk under rebuild */
803 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
806 #define ord_to_idx(ord) (((ord) << 8) >> 8)
807 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
809 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
811 return ord_to_idx(ord
);
814 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
816 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
819 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
824 for (slot
= 0; slot
< map
->num_members
; slot
++) {
825 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
826 if (ord_to_idx(ord
) == idx
)
833 static int get_imsm_raid_level(struct imsm_map
*map
)
835 if (map
->raid_level
== 1) {
836 if (map
->num_members
== 2)
842 return map
->raid_level
;
845 static int cmp_extent(const void *av
, const void *bv
)
847 const struct extent
*a
= av
;
848 const struct extent
*b
= bv
;
849 if (a
->start
< b
->start
)
851 if (a
->start
> b
->start
)
856 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
861 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
862 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
863 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
865 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
872 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
874 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
876 if (lo
== 0 || hi
== 0)
878 *lo
= __le32_to_cpu((unsigned)n
);
879 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
883 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
885 return (unsigned long long)__le32_to_cpu(lo
) |
886 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
889 static unsigned long long total_blocks(struct imsm_disk
*disk
)
893 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
896 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
900 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
903 static unsigned long long blocks_per_member(struct imsm_map
*map
)
907 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
911 static unsigned long long num_data_stripes(struct imsm_map
*map
)
915 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
918 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
920 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
924 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
926 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
929 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
931 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
934 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
936 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
939 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
941 /* find a list of used extents on the given physical device */
942 struct extent
*rv
, *e
;
944 int memberships
= count_memberships(dl
, super
);
947 /* trim the reserved area for spares, so they can join any array
948 * regardless of whether the OROM has assigned sectors from the
949 * IMSM_RESERVED_SECTORS region
952 reservation
= imsm_min_reserved_sectors(super
);
954 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
956 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
959 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
960 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
961 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
963 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
964 e
->start
= pba_of_lba0(map
);
965 e
->size
= blocks_per_member(map
);
969 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
971 /* determine the start of the metadata
972 * when no raid devices are defined use the default
973 * ...otherwise allow the metadata to truncate the value
974 * as is the case with older versions of imsm
977 struct extent
*last
= &rv
[memberships
- 1];
978 unsigned long long remainder
;
980 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
981 /* round down to 1k block to satisfy precision of the kernel
985 /* make sure remainder is still sane */
986 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
987 remainder
= ROUND_UP(super
->len
, 512) >> 9;
988 if (reservation
> remainder
)
989 reservation
= remainder
;
991 e
->start
= total_blocks(&dl
->disk
) - reservation
;
996 /* try to determine how much space is reserved for metadata from
997 * the last get_extents() entry, otherwise fallback to the
1000 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1006 /* for spares just return a minimal reservation which will grow
1007 * once the spare is picked up by an array
1009 if (dl
->index
== -1)
1010 return MPB_SECTOR_CNT
;
1012 e
= get_extents(super
, dl
);
1014 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1016 /* scroll to last entry */
1017 for (i
= 0; e
[i
].size
; i
++)
1020 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1027 static int is_spare(struct imsm_disk
*disk
)
1029 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1032 static int is_configured(struct imsm_disk
*disk
)
1034 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1037 static int is_failed(struct imsm_disk
*disk
)
1039 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1042 /* try to determine how much space is reserved for metadata from
1043 * the last get_extents() entry on the smallest active disk,
1044 * otherwise fallback to the default
1046 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1050 unsigned long long min_active
;
1052 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1053 struct dl
*dl
, *dl_min
= NULL
;
1059 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1062 unsigned long long blocks
= total_blocks(&dl
->disk
);
1063 if (blocks
< min_active
|| min_active
== 0) {
1065 min_active
= blocks
;
1071 /* find last lba used by subarrays on the smallest active disk */
1072 e
= get_extents(super
, dl_min
);
1075 for (i
= 0; e
[i
].size
; i
++)
1078 remainder
= min_active
- e
[i
].start
;
1081 /* to give priority to recovery we should not require full
1082 IMSM_RESERVED_SECTORS from the spare */
1083 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1085 /* if real reservation is smaller use that value */
1086 return (remainder
< rv
) ? remainder
: rv
;
1089 /* Return minimum size of a spare that can be used in this array*/
1090 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1092 struct intel_super
*super
= st
->sb
;
1096 unsigned long long rv
= 0;
1100 /* find first active disk in array */
1102 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1106 /* find last lba used by subarrays */
1107 e
= get_extents(super
, dl
);
1110 for (i
= 0; e
[i
].size
; i
++)
1113 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1116 /* add the amount of space needed for metadata */
1117 rv
= rv
+ imsm_min_reserved_sectors(super
);
1122 static int is_gen_migration(struct imsm_dev
*dev
);
1125 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1126 struct imsm_dev
*dev
);
1128 static void print_imsm_dev(struct intel_super
*super
,
1129 struct imsm_dev
*dev
,
1135 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1136 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1140 printf("[%.16s]:\n", dev
->volume
);
1141 printf(" UUID : %s\n", uuid
);
1142 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1144 printf(" <-- %d", get_imsm_raid_level(map2
));
1146 printf(" Members : %d", map
->num_members
);
1148 printf(" <-- %d", map2
->num_members
);
1150 printf(" Slots : [");
1151 for (i
= 0; i
< map
->num_members
; i
++) {
1152 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1153 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1158 for (i
= 0; i
< map2
->num_members
; i
++) {
1159 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1160 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1165 printf(" Failed disk : ");
1166 if (map
->failed_disk_num
== 0xff)
1169 printf("%i", map
->failed_disk_num
);
1171 slot
= get_imsm_disk_slot(map
, disk_idx
);
1173 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1174 printf(" This Slot : %d%s\n", slot
,
1175 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1177 printf(" This Slot : ?\n");
1178 sz
= __le32_to_cpu(dev
->size_high
);
1180 sz
+= __le32_to_cpu(dev
->size_low
);
1181 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1182 human_size(sz
* 512));
1183 sz
= blocks_per_member(map
);
1184 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1185 human_size(sz
* 512));
1186 printf(" Sector Offset : %llu\n",
1188 printf(" Num Stripes : %llu\n",
1189 num_data_stripes(map
));
1190 printf(" Chunk Size : %u KiB",
1191 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1193 printf(" <-- %u KiB",
1194 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1196 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1197 printf(" Migrate State : ");
1198 if (dev
->vol
.migr_state
) {
1199 if (migr_type(dev
) == MIGR_INIT
)
1200 printf("initialize\n");
1201 else if (migr_type(dev
) == MIGR_REBUILD
)
1202 printf("rebuild\n");
1203 else if (migr_type(dev
) == MIGR_VERIFY
)
1205 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1206 printf("general migration\n");
1207 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1208 printf("state change\n");
1209 else if (migr_type(dev
) == MIGR_REPAIR
)
1212 printf("<unknown:%d>\n", migr_type(dev
));
1215 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1216 if (dev
->vol
.migr_state
) {
1217 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1219 printf(" <-- %s", map_state_str
[map
->map_state
]);
1220 printf("\n Checkpoint : %u ",
1221 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1222 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1225 printf("(%llu)", (unsigned long long)
1226 blocks_per_migr_unit(super
, dev
));
1229 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1232 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1234 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1237 if (index
< -1 || !disk
)
1241 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1243 printf(" Disk%02d Serial : %s\n", index
, str
);
1245 printf(" Disk Serial : %s\n", str
);
1246 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1247 is_configured(disk
) ? " active" : "",
1248 is_failed(disk
) ? " failed" : "");
1249 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1250 sz
= total_blocks(disk
) - reserved
;
1251 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1252 human_size(sz
* 512));
1255 void examine_migr_rec_imsm(struct intel_super
*super
)
1257 struct migr_record
*migr_rec
= super
->migr_rec
;
1258 struct imsm_super
*mpb
= super
->anchor
;
1261 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1262 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1263 struct imsm_map
*map
;
1266 if (is_gen_migration(dev
) == 0)
1269 printf("\nMigration Record Information:");
1271 /* first map under migration */
1272 map
= get_imsm_map(dev
, MAP_0
);
1274 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1275 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1276 printf(" Empty\n ");
1277 printf("Examine one of first two disks in array\n");
1280 printf("\n Status : ");
1281 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1284 printf("Contains Data\n");
1285 printf(" Current Unit : %u\n",
1286 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1287 printf(" Family : %u\n",
1288 __le32_to_cpu(migr_rec
->family_num
));
1289 printf(" Ascending : %u\n",
1290 __le32_to_cpu(migr_rec
->ascending_migr
));
1291 printf(" Blocks Per Unit : %u\n",
1292 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1293 printf(" Dest. Depth Per Unit : %u\n",
1294 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1295 printf(" Checkpoint Area pba : %u\n",
1296 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1297 printf(" First member lba : %u\n",
1298 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1299 printf(" Total Number of Units : %u\n",
1300 __le32_to_cpu(migr_rec
->num_migr_units
));
1301 printf(" Size of volume : %u\n",
1302 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1303 printf(" Expansion space for LBA64 : %u\n",
1304 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1305 printf(" Record was read from : %u\n",
1306 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1311 #endif /* MDASSEMBLE */
1312 /*******************************************************************************
1313 * function: imsm_check_attributes
1314 * Description: Function checks if features represented by attributes flags
1315 * are supported by mdadm.
1317 * attributes - Attributes read from metadata
1319 * 0 - passed attributes contains unsupported features flags
1320 * 1 - all features are supported
1321 ******************************************************************************/
1322 static int imsm_check_attributes(__u32 attributes
)
1325 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1327 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1329 not_supported
&= attributes
;
1330 if (not_supported
) {
1331 pr_err("(IMSM): Unsupported attributes : %x\n",
1332 (unsigned)__le32_to_cpu(not_supported
));
1333 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1334 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1335 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1337 if (not_supported
& MPB_ATTRIB_2TB
) {
1338 dprintf("\t\tMPB_ATTRIB_2TB\n");
1339 not_supported
^= MPB_ATTRIB_2TB
;
1341 if (not_supported
& MPB_ATTRIB_RAID0
) {
1342 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1343 not_supported
^= MPB_ATTRIB_RAID0
;
1345 if (not_supported
& MPB_ATTRIB_RAID1
) {
1346 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1347 not_supported
^= MPB_ATTRIB_RAID1
;
1349 if (not_supported
& MPB_ATTRIB_RAID10
) {
1350 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1351 not_supported
^= MPB_ATTRIB_RAID10
;
1353 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1354 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1355 not_supported
^= MPB_ATTRIB_RAID1E
;
1357 if (not_supported
& MPB_ATTRIB_RAID5
) {
1358 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1359 not_supported
^= MPB_ATTRIB_RAID5
;
1361 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1362 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1363 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1365 if (not_supported
& MPB_ATTRIB_BBM
) {
1366 dprintf("\t\tMPB_ATTRIB_BBM\n");
1367 not_supported
^= MPB_ATTRIB_BBM
;
1369 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1370 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1371 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1373 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1374 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1375 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1377 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1378 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1379 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1381 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1382 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1383 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1385 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1386 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1387 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1391 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1400 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1402 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1404 struct intel_super
*super
= st
->sb
;
1405 struct imsm_super
*mpb
= super
->anchor
;
1406 char str
[MAX_SIGNATURE_LENGTH
];
1411 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1414 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1415 printf(" Magic : %s\n", str
);
1416 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1417 printf(" Version : %s\n", get_imsm_version(mpb
));
1418 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1419 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1420 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1421 printf(" Attributes : ");
1422 if (imsm_check_attributes(mpb
->attributes
))
1423 printf("All supported\n");
1425 printf("not supported\n");
1426 getinfo_super_imsm(st
, &info
, NULL
);
1427 fname_from_uuid(st
, &info
, nbuf
, ':');
1428 printf(" UUID : %s\n", nbuf
+ 5);
1429 sum
= __le32_to_cpu(mpb
->check_sum
);
1430 printf(" Checksum : %08x %s\n", sum
,
1431 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1432 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1433 printf(" Disks : %d\n", mpb
->num_disks
);
1434 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1435 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1436 if (super
->bbm_log
) {
1437 struct bbm_log
*log
= super
->bbm_log
;
1440 printf("Bad Block Management Log:\n");
1441 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1442 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1443 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1444 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1445 printf(" First Spare : %llx\n",
1446 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1448 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1450 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1452 super
->current_vol
= i
;
1453 getinfo_super_imsm(st
, &info
, NULL
);
1454 fname_from_uuid(st
, &info
, nbuf
, ':');
1455 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1457 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1458 if (i
== super
->disks
->index
)
1460 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1463 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1464 if (dl
->index
== -1)
1465 print_imsm_disk(&dl
->disk
, -1, reserved
);
1467 examine_migr_rec_imsm(super
);
1470 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1472 /* We just write a generic IMSM ARRAY entry */
1475 struct intel_super
*super
= st
->sb
;
1477 if (!super
->anchor
->num_raid_devs
) {
1478 printf("ARRAY metadata=imsm\n");
1482 getinfo_super_imsm(st
, &info
, NULL
);
1483 fname_from_uuid(st
, &info
, nbuf
, ':');
1484 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1487 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1489 /* We just write a generic IMSM ARRAY entry */
1493 struct intel_super
*super
= st
->sb
;
1496 if (!super
->anchor
->num_raid_devs
)
1499 getinfo_super_imsm(st
, &info
, NULL
);
1500 fname_from_uuid(st
, &info
, nbuf
, ':');
1501 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1502 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1504 super
->current_vol
= i
;
1505 getinfo_super_imsm(st
, &info
, NULL
);
1506 fname_from_uuid(st
, &info
, nbuf1
, ':');
1507 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1508 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1512 static void export_examine_super_imsm(struct supertype
*st
)
1514 struct intel_super
*super
= st
->sb
;
1515 struct imsm_super
*mpb
= super
->anchor
;
1519 getinfo_super_imsm(st
, &info
, NULL
);
1520 fname_from_uuid(st
, &info
, nbuf
, ':');
1521 printf("MD_METADATA=imsm\n");
1522 printf("MD_LEVEL=container\n");
1523 printf("MD_UUID=%s\n", nbuf
+5);
1524 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1527 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
1529 /* The second last 512byte sector of the device contains
1530 * the "struct imsm_super" metadata.
1531 * This contains mpb_size which is the size in bytes of the
1532 * extended metadata. This is located immediately before
1534 * We want to read all that, plus the last sector which
1535 * may contain a migration record, and write it all
1539 unsigned long long dsize
, offset
;
1541 struct imsm_super
*sb
;
1544 if (posix_memalign(&buf
, 4096, 4096) != 0)
1547 if (!get_dev_size(from
, NULL
, &dsize
))
1550 if (lseek64(from
, dsize
-1024, 0) < 0)
1552 if (read(from
, buf
, 512) != 512)
1555 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
1558 sectors
= mpb_sectors(sb
) + 2;
1559 offset
= dsize
- sectors
* 512;
1560 if (lseek64(from
, offset
, 0) < 0 ||
1561 lseek64(to
, offset
, 0) < 0)
1563 while (written
< sectors
* 512) {
1564 int n
= sectors
*512 - written
;
1567 if (read(from
, buf
, n
) != n
)
1569 if (write(to
, buf
, n
) != n
)
1580 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1585 getinfo_super_imsm(st
, &info
, NULL
);
1586 fname_from_uuid(st
, &info
, nbuf
, ':');
1587 printf("\n UUID : %s\n", nbuf
+ 5);
1590 static void brief_detail_super_imsm(struct supertype
*st
)
1594 getinfo_super_imsm(st
, &info
, NULL
);
1595 fname_from_uuid(st
, &info
, nbuf
, ':');
1596 printf(" UUID=%s", nbuf
+ 5);
1599 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1600 static void fd2devname(int fd
, char *name
);
1602 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1604 /* dump an unsorted list of devices attached to AHCI Intel storage
1605 * controller, as well as non-connected ports
1607 int hba_len
= strlen(hba_path
) + 1;
1612 unsigned long port_mask
= (1 << port_count
) - 1;
1614 if (port_count
> (int)sizeof(port_mask
) * 8) {
1616 pr_err("port_count %d out of range\n", port_count
);
1620 /* scroll through /sys/dev/block looking for devices attached to
1623 dir
= opendir("/sys/dev/block");
1627 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1638 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1640 path
= devt_to_devpath(makedev(major
, minor
));
1643 if (!path_attached_to_hba(path
, hba_path
)) {
1649 /* retrieve the scsi device type */
1650 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1652 pr_err("failed to allocate 'device'\n");
1656 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1657 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
1659 pr_err("failed to read device type for %s\n",
1665 type
= strtoul(buf
, NULL
, 10);
1667 /* if it's not a disk print the vendor and model */
1668 if (!(type
== 0 || type
== 7 || type
== 14)) {
1671 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1672 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
1673 strncpy(vendor
, buf
, sizeof(vendor
));
1674 vendor
[sizeof(vendor
) - 1] = '\0';
1675 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1676 while (isspace(*c
) || *c
== '\0')
1680 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1681 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
1682 strncpy(model
, buf
, sizeof(model
));
1683 model
[sizeof(model
) - 1] = '\0';
1684 c
= (char *) &model
[sizeof(model
) - 1];
1685 while (isspace(*c
) || *c
== '\0')
1689 if (vendor
[0] && model
[0])
1690 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1692 switch (type
) { /* numbers from hald/linux/device.c */
1693 case 1: sprintf(buf
, "tape"); break;
1694 case 2: sprintf(buf
, "printer"); break;
1695 case 3: sprintf(buf
, "processor"); break;
1697 case 5: sprintf(buf
, "cdrom"); break;
1698 case 6: sprintf(buf
, "scanner"); break;
1699 case 8: sprintf(buf
, "media_changer"); break;
1700 case 9: sprintf(buf
, "comm"); break;
1701 case 12: sprintf(buf
, "raid"); break;
1702 default: sprintf(buf
, "unknown");
1708 /* chop device path to 'host%d' and calculate the port number */
1709 c
= strchr(&path
[hba_len
], '/');
1712 pr_err("%s - invalid path name\n", path
+ hba_len
);
1717 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
1718 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
1722 *c
= '/'; /* repair the full string */
1723 pr_err("failed to determine port number for %s\n",
1730 /* mark this port as used */
1731 port_mask
&= ~(1 << port
);
1733 /* print out the device information */
1735 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1739 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1741 printf(" Port%d : - disk info unavailable -\n", port
);
1743 fd2devname(fd
, buf
);
1744 printf(" Port%d : %s", port
, buf
);
1745 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1746 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1761 for (i
= 0; i
< port_count
; i
++)
1762 if (port_mask
& (1 << i
))
1763 printf(" Port%d : - no device attached -\n", i
);
1769 static int print_vmd_attached_devs(struct sys_dev
*hba
)
1777 if (hba
->type
!= SYS_DEV_VMD
)
1780 /* scroll through /sys/dev/block looking for devices attached to
1783 dir
= opendir("/sys/bus/pci/drivers/nvme");
1787 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1790 /* is 'ent' a device? check that the 'subsystem' link exists and
1791 * that its target matches 'bus'
1793 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
1795 n
= readlink(path
, link
, sizeof(link
));
1796 if (n
< 0 || n
>= (int)sizeof(link
))
1799 c
= strrchr(link
, '/');
1802 if (strncmp("pci", c
+1, strlen("pci")) != 0)
1805 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
1806 /* if not a intel NVMe - skip it*/
1807 if (devpath_to_vendor(path
) != 0x8086)
1810 rp
= realpath(path
, NULL
);
1814 if (path_attached_to_hba(rp
, hba
->path
)) {
1815 printf(" NVMe under VMD : %s\n", rp
);
1824 static void print_found_intel_controllers(struct sys_dev
*elem
)
1826 for (; elem
; elem
= elem
->next
) {
1827 pr_err("found Intel(R) ");
1828 if (elem
->type
== SYS_DEV_SATA
)
1829 fprintf(stderr
, "SATA ");
1830 else if (elem
->type
== SYS_DEV_SAS
)
1831 fprintf(stderr
, "SAS ");
1832 else if (elem
->type
== SYS_DEV_NVME
)
1833 fprintf(stderr
, "NVMe ");
1835 if (elem
->type
== SYS_DEV_VMD
)
1836 fprintf(stderr
, "VMD domain");
1838 fprintf(stderr
, "RAID controller");
1841 fprintf(stderr
, " at %s", elem
->pci_id
);
1842 fprintf(stderr
, ".\n");
1847 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1854 if ((dir
= opendir(hba_path
)) == NULL
)
1857 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1860 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
1861 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
1863 if (*port_count
== 0)
1865 else if (host
< host_base
)
1868 if (host
+ 1 > *port_count
+ host_base
)
1869 *port_count
= host
+ 1 - host_base
;
1875 static void print_imsm_capability(const struct imsm_orom
*orom
)
1877 printf(" Platform : Intel(R) ");
1878 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
1879 printf("Matrix Storage Manager\n");
1881 printf("Rapid Storage Technology%s\n",
1882 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
1883 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
1884 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
1885 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
1886 printf(" RAID Levels :%s%s%s%s%s\n",
1887 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1888 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1889 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1890 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1891 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1892 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1893 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1894 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1895 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1896 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1897 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1898 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1899 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1900 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1901 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1902 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1903 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1904 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1905 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1906 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1907 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1908 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1909 printf(" 2TB volumes :%s supported\n",
1910 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
1911 printf(" 2TB disks :%s supported\n",
1912 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
1913 printf(" Max Disks : %d\n", orom
->tds
);
1914 printf(" Max Volumes : %d per array, %d per %s\n",
1915 orom
->vpa
, orom
->vphba
,
1916 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
1920 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
1922 printf("MD_FIRMWARE_TYPE=imsm\n");
1923 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
1924 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1925 orom
->hotfix_ver
, orom
->build
);
1926 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
1927 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
1928 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
1929 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
1930 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
1931 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
1932 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1933 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
1934 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
1935 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
1936 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
1937 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
1938 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
1939 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
1940 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
1941 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
1942 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
1943 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
1944 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
1945 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
1946 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
1947 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
1948 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
1949 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
1950 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
1951 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
1952 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
1953 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
1956 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
1958 /* There are two components to imsm platform support, the ahci SATA
1959 * controller and the option-rom. To find the SATA controller we
1960 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1961 * controller with the Intel vendor id is present. This approach
1962 * allows mdadm to leverage the kernel's ahci detection logic, with the
1963 * caveat that if ahci.ko is not loaded mdadm will not be able to
1964 * detect platform raid capabilities. The option-rom resides in a
1965 * platform "Adapter ROM". We scan for its signature to retrieve the
1966 * platform capabilities. If raid support is disabled in the BIOS the
1967 * option-rom capability structure will not be available.
1969 struct sys_dev
*list
, *hba
;
1974 if (enumerate_only
) {
1975 if (check_env("IMSM_NO_PLATFORM"))
1977 list
= find_intel_devices();
1980 for (hba
= list
; hba
; hba
= hba
->next
) {
1981 if (find_imsm_capability(hba
)) {
1991 list
= find_intel_devices();
1994 pr_err("no active Intel(R) RAID controller found.\n");
1996 } else if (verbose
> 0)
1997 print_found_intel_controllers(list
);
1999 for (hba
= list
; hba
; hba
= hba
->next
) {
2000 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2002 if (!find_imsm_capability(hba
)) {
2004 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2005 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2006 get_sys_dev_type(hba
->type
));
2012 if (controller_path
&& result
== 1) {
2013 pr_err("no active Intel(R) RAID controller found under %s\n",
2018 const struct orom_entry
*entry
;
2020 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2021 if (entry
->type
== SYS_DEV_VMD
) {
2022 print_imsm_capability(&entry
->orom
);
2023 for (hba
= list
; hba
; hba
= hba
->next
) {
2024 if (hba
->type
== SYS_DEV_VMD
) {
2026 printf(" I/O Controller : %s (%s)\n",
2027 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2028 if (print_vmd_attached_devs(hba
)) {
2030 pr_err("failed to get devices attached to VMD domain.\n");
2039 print_imsm_capability(&entry
->orom
);
2040 if (entry
->type
== SYS_DEV_NVME
) {
2041 for (hba
= list
; hba
; hba
= hba
->next
) {
2042 if (hba
->type
== SYS_DEV_NVME
)
2043 printf(" NVMe Device : %s\n", hba
->path
);
2049 struct devid_list
*devid
;
2050 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2051 hba
= device_by_id(devid
->devid
);
2055 printf(" I/O Controller : %s (%s)\n",
2056 hba
->path
, get_sys_dev_type(hba
->type
));
2057 if (hba
->type
== SYS_DEV_SATA
) {
2058 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2059 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2061 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2072 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2074 struct sys_dev
*list
, *hba
;
2077 list
= find_intel_devices();
2080 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2085 for (hba
= list
; hba
; hba
= hba
->next
) {
2086 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2088 if (!find_imsm_capability(hba
) && verbose
> 0) {
2090 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2091 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2097 const struct orom_entry
*entry
;
2099 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2100 if (entry
->type
== SYS_DEV_VMD
) {
2101 for (hba
= list
; hba
; hba
= hba
->next
)
2102 print_imsm_capability_export(&entry
->orom
);
2105 print_imsm_capability_export(&entry
->orom
);
2113 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2115 /* the imsm metadata format does not specify any host
2116 * identification information. We return -1 since we can never
2117 * confirm nor deny whether a given array is "meant" for this
2118 * host. We rely on compare_super and the 'family_num' fields to
2119 * exclude member disks that do not belong, and we rely on
2120 * mdadm.conf to specify the arrays that should be assembled.
2121 * Auto-assembly may still pick up "foreign" arrays.
2127 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2129 /* The uuid returned here is used for:
2130 * uuid to put into bitmap file (Create, Grow)
2131 * uuid for backup header when saving critical section (Grow)
2132 * comparing uuids when re-adding a device into an array
2133 * In these cases the uuid required is that of the data-array,
2134 * not the device-set.
2135 * uuid to recognise same set when adding a missing device back
2136 * to an array. This is a uuid for the device-set.
2138 * For each of these we can make do with a truncated
2139 * or hashed uuid rather than the original, as long as
2141 * In each case the uuid required is that of the data-array,
2142 * not the device-set.
2144 /* imsm does not track uuid's so we synthesis one using sha1 on
2145 * - The signature (Which is constant for all imsm array, but no matter)
2146 * - the orig_family_num of the container
2147 * - the index number of the volume
2148 * - the 'serial' number of the volume.
2149 * Hopefully these are all constant.
2151 struct intel_super
*super
= st
->sb
;
2154 struct sha1_ctx ctx
;
2155 struct imsm_dev
*dev
= NULL
;
2158 /* some mdadm versions failed to set ->orig_family_num, in which
2159 * case fall back to ->family_num. orig_family_num will be
2160 * fixed up with the first metadata update.
2162 family_num
= super
->anchor
->orig_family_num
;
2163 if (family_num
== 0)
2164 family_num
= super
->anchor
->family_num
;
2165 sha1_init_ctx(&ctx
);
2166 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2167 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2168 if (super
->current_vol
>= 0)
2169 dev
= get_imsm_dev(super
, super
->current_vol
);
2171 __u32 vol
= super
->current_vol
;
2172 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2173 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2175 sha1_finish_ctx(&ctx
, buf
);
2176 memcpy(uuid
, buf
, 4*4);
2181 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2183 __u8
*v
= get_imsm_version(mpb
);
2184 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2185 char major
[] = { 0, 0, 0 };
2186 char minor
[] = { 0 ,0, 0 };
2187 char patch
[] = { 0, 0, 0 };
2188 char *ver_parse
[] = { major
, minor
, patch
};
2192 while (*v
!= '\0' && v
< end
) {
2193 if (*v
!= '.' && j
< 2)
2194 ver_parse
[i
][j
++] = *v
;
2202 *m
= strtol(minor
, NULL
, 0);
2203 *p
= strtol(patch
, NULL
, 0);
2207 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2209 /* migr_strip_size when repairing or initializing parity */
2210 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2211 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2213 switch (get_imsm_raid_level(map
)) {
2218 return 128*1024 >> 9;
2222 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2224 /* migr_strip_size when rebuilding a degraded disk, no idea why
2225 * this is different than migr_strip_size_resync(), but it's good
2228 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2229 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2231 switch (get_imsm_raid_level(map
)) {
2234 if (map
->num_members
% map
->num_domains
== 0)
2235 return 128*1024 >> 9;
2239 return max((__u32
) 64*1024 >> 9, chunk
);
2241 return 128*1024 >> 9;
2245 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2247 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2248 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2249 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2250 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2252 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2255 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2257 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2258 int level
= get_imsm_raid_level(lo
);
2260 if (level
== 1 || level
== 10) {
2261 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2263 return hi
->num_domains
;
2265 return num_stripes_per_unit_resync(dev
);
2268 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2270 /* named 'imsm_' because raid0, raid1 and raid10
2271 * counter-intuitively have the same number of data disks
2273 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2275 switch (get_imsm_raid_level(map
)) {
2277 return map
->num_members
;
2281 return map
->num_members
/2;
2283 return map
->num_members
- 1;
2285 dprintf("unsupported raid level\n");
2290 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2292 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2293 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2295 switch(get_imsm_raid_level(map
)) {
2298 return chunk
* map
->num_domains
;
2300 return chunk
* map
->num_members
;
2306 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2308 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2309 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2310 __u32 strip
= block
/ chunk
;
2312 switch (get_imsm_raid_level(map
)) {
2315 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2316 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2318 return vol_stripe
* chunk
+ block
% chunk
;
2320 __u32 stripe
= strip
/ (map
->num_members
- 1);
2322 return stripe
* chunk
+ block
% chunk
;
2329 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2330 struct imsm_dev
*dev
)
2332 /* calculate the conversion factor between per member 'blocks'
2333 * (md/{resync,rebuild}_start) and imsm migration units, return
2334 * 0 for the 'not migrating' and 'unsupported migration' cases
2336 if (!dev
->vol
.migr_state
)
2339 switch (migr_type(dev
)) {
2340 case MIGR_GEN_MIGR
: {
2341 struct migr_record
*migr_rec
= super
->migr_rec
;
2342 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2347 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2348 __u32 stripes_per_unit
;
2349 __u32 blocks_per_unit
;
2358 /* yes, this is really the translation of migr_units to
2359 * per-member blocks in the 'resync' case
2361 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2362 migr_chunk
= migr_strip_blocks_resync(dev
);
2363 disks
= imsm_num_data_members(dev
, MAP_0
);
2364 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2365 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2366 segment
= blocks_per_unit
/ stripe
;
2367 block_rel
= blocks_per_unit
- segment
* stripe
;
2368 parity_depth
= parity_segment_depth(dev
);
2369 block_map
= map_migr_block(dev
, block_rel
);
2370 return block_map
+ parity_depth
* segment
;
2372 case MIGR_REBUILD
: {
2373 __u32 stripes_per_unit
;
2376 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2377 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2378 return migr_chunk
* stripes_per_unit
;
2380 case MIGR_STATE_CHANGE
:
2386 static int imsm_level_to_layout(int level
)
2394 return ALGORITHM_LEFT_ASYMMETRIC
;
2401 /*******************************************************************************
2402 * Function: read_imsm_migr_rec
2403 * Description: Function reads imsm migration record from last sector of disk
2405 * fd : disk descriptor
2406 * super : metadata info
2410 ******************************************************************************/
2411 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2414 unsigned long long dsize
;
2416 get_dev_size(fd
, NULL
, &dsize
);
2417 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2418 pr_err("Cannot seek to anchor block: %s\n",
2422 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2423 MIGR_REC_BUF_SIZE
) {
2424 pr_err("Cannot read migr record block: %s\n",
2434 static struct imsm_dev
*imsm_get_device_during_migration(
2435 struct intel_super
*super
)
2438 struct intel_dev
*dv
;
2440 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2441 if (is_gen_migration(dv
->dev
))
2447 /*******************************************************************************
2448 * Function: load_imsm_migr_rec
2449 * Description: Function reads imsm migration record (it is stored at the last
2452 * super : imsm internal array info
2453 * info : general array info
2457 * -2 : no migration in progress
2458 ******************************************************************************/
2459 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2466 struct imsm_dev
*dev
;
2467 struct imsm_map
*map
;
2470 /* find map under migration */
2471 dev
= imsm_get_device_during_migration(super
);
2472 /* nothing to load,no migration in progress?
2478 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2479 /* read only from one of the first two slots */
2480 if ((sd
->disk
.raid_disk
< 0) ||
2481 (sd
->disk
.raid_disk
> 1))
2484 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2485 fd
= dev_open(nm
, O_RDONLY
);
2491 map
= get_imsm_map(dev
, MAP_0
);
2492 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2493 /* skip spare and failed disks
2497 /* read only from one of the first two slots */
2499 slot
= get_imsm_disk_slot(map
, dl
->index
);
2500 if (map
== NULL
|| slot
> 1 || slot
< 0)
2502 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2503 fd
= dev_open(nm
, O_RDONLY
);
2510 retval
= read_imsm_migr_rec(fd
, super
);
2519 /*******************************************************************************
2520 * function: imsm_create_metadata_checkpoint_update
2521 * Description: It creates update for checkpoint change.
2523 * super : imsm internal array info
2524 * u : pointer to prepared update
2527 * If length is equal to 0, input pointer u contains no update
2528 ******************************************************************************/
2529 static int imsm_create_metadata_checkpoint_update(
2530 struct intel_super
*super
,
2531 struct imsm_update_general_migration_checkpoint
**u
)
2534 int update_memory_size
= 0;
2536 dprintf("(enter)\n");
2542 /* size of all update data without anchor */
2543 update_memory_size
=
2544 sizeof(struct imsm_update_general_migration_checkpoint
);
2546 *u
= xcalloc(1, update_memory_size
);
2548 dprintf("error: cannot get memory\n");
2551 (*u
)->type
= update_general_migration_checkpoint
;
2552 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2553 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
2555 return update_memory_size
;
2558 static void imsm_update_metadata_locally(struct supertype
*st
,
2559 void *buf
, int len
);
2561 /*******************************************************************************
2562 * Function: write_imsm_migr_rec
2563 * Description: Function writes imsm migration record
2564 * (at the last sector of disk)
2566 * super : imsm internal array info
2570 ******************************************************************************/
2571 static int write_imsm_migr_rec(struct supertype
*st
)
2573 struct intel_super
*super
= st
->sb
;
2574 unsigned long long dsize
;
2580 struct imsm_update_general_migration_checkpoint
*u
;
2581 struct imsm_dev
*dev
;
2582 struct imsm_map
*map
;
2584 /* find map under migration */
2585 dev
= imsm_get_device_during_migration(super
);
2586 /* if no migration, write buffer anyway to clear migr_record
2587 * on disk based on first available device
2590 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2591 super
->current_vol
);
2593 map
= get_imsm_map(dev
, MAP_0
);
2595 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2598 /* skip failed and spare devices */
2601 /* write to 2 first slots only */
2603 slot
= get_imsm_disk_slot(map
, sd
->index
);
2604 if (map
== NULL
|| slot
> 1 || slot
< 0)
2607 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2608 fd
= dev_open(nm
, O_RDWR
);
2611 get_dev_size(fd
, NULL
, &dsize
);
2612 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2613 pr_err("Cannot seek to anchor block: %s\n",
2617 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2618 MIGR_REC_BUF_SIZE
) {
2619 pr_err("Cannot write migr record block: %s\n",
2626 /* update checkpoint information in metadata */
2627 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2630 dprintf("imsm: Cannot prepare update\n");
2633 /* update metadata locally */
2634 imsm_update_metadata_locally(st
, u
, len
);
2635 /* and possibly remotely */
2636 if (st
->update_tail
) {
2637 append_metadata_update(st
, u
, len
);
2638 /* during reshape we do all work inside metadata handler
2639 * manage_reshape(), so metadata update has to be triggered
2642 flush_metadata_updates(st
);
2643 st
->update_tail
= &st
->updates
;
2653 #endif /* MDASSEMBLE */
2655 /* spare/missing disks activations are not allowe when
2656 * array/container performs reshape operation, because
2657 * all arrays in container works on the same disks set
2659 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2662 struct intel_dev
*i_dev
;
2663 struct imsm_dev
*dev
;
2665 /* check whole container
2667 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2669 if (is_gen_migration(dev
)) {
2670 /* No repair during any migration in container
2678 static unsigned long long imsm_component_size_aligment_check(int level
,
2680 unsigned long long component_size
)
2682 unsigned int component_size_alligment
;
2684 /* check component size aligment
2686 component_size_alligment
= component_size
% (chunk_size
/512);
2688 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
2689 level
, chunk_size
, component_size
,
2690 component_size_alligment
);
2692 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
2693 dprintf("imsm: reported component size alligned from %llu ",
2695 component_size
-= component_size_alligment
;
2696 dprintf_cont("to %llu (%i).\n",
2697 component_size
, component_size_alligment
);
2700 return component_size
;
2703 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2705 struct intel_super
*super
= st
->sb
;
2706 struct migr_record
*migr_rec
= super
->migr_rec
;
2707 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2708 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2709 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2710 struct imsm_map
*map_to_analyse
= map
;
2712 int map_disks
= info
->array
.raid_disks
;
2714 memset(info
, 0, sizeof(*info
));
2716 map_to_analyse
= prev_map
;
2718 dl
= super
->current_disk
;
2720 info
->container_member
= super
->current_vol
;
2721 info
->array
.raid_disks
= map
->num_members
;
2722 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2723 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2724 info
->array
.md_minor
= -1;
2725 info
->array
.ctime
= 0;
2726 info
->array
.utime
= 0;
2727 info
->array
.chunk_size
=
2728 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2729 info
->array
.state
= !dev
->vol
.dirty
;
2730 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2731 info
->custom_array_size
<<= 32;
2732 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2733 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2735 if (is_gen_migration(dev
)) {
2736 info
->reshape_active
= 1;
2737 info
->new_level
= get_imsm_raid_level(map
);
2738 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2739 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2740 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2741 if (info
->delta_disks
) {
2742 /* this needs to be applied to every array
2745 info
->reshape_active
= CONTAINER_RESHAPE
;
2747 /* We shape information that we give to md might have to be
2748 * modify to cope with md's requirement for reshaping arrays.
2749 * For example, when reshaping a RAID0, md requires it to be
2750 * presented as a degraded RAID4.
2751 * Also if a RAID0 is migrating to a RAID5 we need to specify
2752 * the array as already being RAID5, but the 'before' layout
2753 * is a RAID4-like layout.
2755 switch (info
->array
.level
) {
2757 switch(info
->new_level
) {
2759 /* conversion is happening as RAID4 */
2760 info
->array
.level
= 4;
2761 info
->array
.raid_disks
+= 1;
2764 /* conversion is happening as RAID5 */
2765 info
->array
.level
= 5;
2766 info
->array
.layout
= ALGORITHM_PARITY_N
;
2767 info
->delta_disks
-= 1;
2770 /* FIXME error message */
2771 info
->array
.level
= UnSet
;
2777 info
->new_level
= UnSet
;
2778 info
->new_layout
= UnSet
;
2779 info
->new_chunk
= info
->array
.chunk_size
;
2780 info
->delta_disks
= 0;
2784 info
->disk
.major
= dl
->major
;
2785 info
->disk
.minor
= dl
->minor
;
2786 info
->disk
.number
= dl
->index
;
2787 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2791 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2792 info
->component_size
= blocks_per_member(map_to_analyse
);
2794 info
->component_size
= imsm_component_size_aligment_check(
2796 info
->array
.chunk_size
,
2797 info
->component_size
);
2799 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2800 info
->recovery_start
= MaxSector
;
2802 info
->reshape_progress
= 0;
2803 info
->resync_start
= MaxSector
;
2804 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2806 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2807 info
->resync_start
= 0;
2809 if (dev
->vol
.migr_state
) {
2810 switch (migr_type(dev
)) {
2813 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2815 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2817 info
->resync_start
= blocks_per_unit
* units
;
2820 case MIGR_GEN_MIGR
: {
2821 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2823 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2824 unsigned long long array_blocks
;
2827 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2829 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2830 (super
->migr_rec
->rec_status
==
2831 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2834 info
->reshape_progress
= blocks_per_unit
* units
;
2836 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
2837 (unsigned long long)units
,
2838 (unsigned long long)blocks_per_unit
,
2839 info
->reshape_progress
);
2841 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2842 if (used_disks
> 0) {
2843 array_blocks
= blocks_per_member(map
) *
2845 /* round array size down to closest MB
2847 info
->custom_array_size
= (array_blocks
2848 >> SECT_PER_MB_SHIFT
)
2849 << SECT_PER_MB_SHIFT
;
2853 /* we could emulate the checkpointing of
2854 * 'sync_action=check' migrations, but for now
2855 * we just immediately complete them
2858 /* this is handled by container_content_imsm() */
2859 case MIGR_STATE_CHANGE
:
2860 /* FIXME handle other migrations */
2862 /* we are not dirty, so... */
2863 info
->resync_start
= MaxSector
;
2867 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2868 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2870 info
->array
.major_version
= -1;
2871 info
->array
.minor_version
= -2;
2872 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
2873 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2874 uuid_from_super_imsm(st
, info
->uuid
);
2878 for (i
=0; i
<map_disks
; i
++) {
2880 if (i
< info
->array
.raid_disks
) {
2881 struct imsm_disk
*dsk
;
2882 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2883 dsk
= get_imsm_disk(super
, j
);
2884 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2891 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2892 int failed
, int look_in_map
);
2894 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2898 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2900 if (is_gen_migration(dev
)) {
2903 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2905 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2906 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2907 if (map2
->map_state
!= map_state
) {
2908 map2
->map_state
= map_state
;
2909 super
->updates_pending
++;
2915 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2919 for (d
= super
->missing
; d
; d
= d
->next
)
2920 if (d
->index
== index
)
2925 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2927 struct intel_super
*super
= st
->sb
;
2928 struct imsm_disk
*disk
;
2929 int map_disks
= info
->array
.raid_disks
;
2930 int max_enough
= -1;
2932 struct imsm_super
*mpb
;
2934 if (super
->current_vol
>= 0) {
2935 getinfo_super_imsm_volume(st
, info
, map
);
2938 memset(info
, 0, sizeof(*info
));
2940 /* Set raid_disks to zero so that Assemble will always pull in valid
2943 info
->array
.raid_disks
= 0;
2944 info
->array
.level
= LEVEL_CONTAINER
;
2945 info
->array
.layout
= 0;
2946 info
->array
.md_minor
= -1;
2947 info
->array
.ctime
= 0; /* N/A for imsm */
2948 info
->array
.utime
= 0;
2949 info
->array
.chunk_size
= 0;
2951 info
->disk
.major
= 0;
2952 info
->disk
.minor
= 0;
2953 info
->disk
.raid_disk
= -1;
2954 info
->reshape_active
= 0;
2955 info
->array
.major_version
= -1;
2956 info
->array
.minor_version
= -2;
2957 strcpy(info
->text_version
, "imsm");
2958 info
->safe_mode_delay
= 0;
2959 info
->disk
.number
= -1;
2960 info
->disk
.state
= 0;
2962 info
->recovery_start
= MaxSector
;
2963 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2965 /* do we have the all the insync disks that we expect? */
2966 mpb
= super
->anchor
;
2968 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2969 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2970 int failed
, enough
, j
, missing
= 0;
2971 struct imsm_map
*map
;
2974 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2975 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2976 map
= get_imsm_map(dev
, MAP_0
);
2978 /* any newly missing disks?
2979 * (catches single-degraded vs double-degraded)
2981 for (j
= 0; j
< map
->num_members
; j
++) {
2982 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2983 __u32 idx
= ord_to_idx(ord
);
2985 if (!(ord
& IMSM_ORD_REBUILD
) &&
2986 get_imsm_missing(super
, idx
)) {
2992 if (state
== IMSM_T_STATE_FAILED
)
2994 else if (state
== IMSM_T_STATE_DEGRADED
&&
2995 (state
!= map
->map_state
|| missing
))
2997 else /* we're normal, or already degraded */
2999 if (is_gen_migration(dev
) && missing
) {
3000 /* during general migration we need all disks
3001 * that process is running on.
3002 * No new missing disk is allowed.
3006 /* no more checks necessary
3010 /* in the missing/failed disk case check to see
3011 * if at least one array is runnable
3013 max_enough
= max(max_enough
, enough
);
3015 dprintf("enough: %d\n", max_enough
);
3016 info
->container_enough
= max_enough
;
3019 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3021 disk
= &super
->disks
->disk
;
3022 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3023 info
->component_size
= reserved
;
3024 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3025 /* we don't change info->disk.raid_disk here because
3026 * this state will be finalized in mdmon after we have
3027 * found the 'most fresh' version of the metadata
3029 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3030 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3033 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3034 * ->compare_super may have updated the 'num_raid_devs' field for spares
3036 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3037 uuid_from_super_imsm(st
, info
->uuid
);
3039 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3041 /* I don't know how to compute 'map' on imsm, so use safe default */
3044 for (i
= 0; i
< map_disks
; i
++)
3050 /* allocates memory and fills disk in mdinfo structure
3051 * for each disk in array */
3052 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3054 struct mdinfo
*mddev
;
3055 struct intel_super
*super
= st
->sb
;
3056 struct imsm_disk
*disk
;
3059 if (!super
|| !super
->disks
)
3062 mddev
= xcalloc(1, sizeof(*mddev
));
3066 tmp
= xcalloc(1, sizeof(*tmp
));
3068 tmp
->next
= mddev
->devs
;
3070 tmp
->disk
.number
= count
++;
3071 tmp
->disk
.major
= dl
->major
;
3072 tmp
->disk
.minor
= dl
->minor
;
3073 tmp
->disk
.state
= is_configured(disk
) ?
3074 (1 << MD_DISK_ACTIVE
) : 0;
3075 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3076 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3077 tmp
->disk
.raid_disk
= -1;
3083 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3084 char *update
, char *devname
, int verbose
,
3085 int uuid_set
, char *homehost
)
3087 /* For 'assemble' and 'force' we need to return non-zero if any
3088 * change was made. For others, the return value is ignored.
3089 * Update options are:
3090 * force-one : This device looks a bit old but needs to be included,
3091 * update age info appropriately.
3092 * assemble: clear any 'faulty' flag to allow this device to
3094 * force-array: Array is degraded but being forced, mark it clean
3095 * if that will be needed to assemble it.
3097 * newdev: not used ????
3098 * grow: Array has gained a new device - this is currently for
3100 * resync: mark as dirty so a resync will happen.
3101 * name: update the name - preserving the homehost
3102 * uuid: Change the uuid of the array to match watch is given
3104 * Following are not relevant for this imsm:
3105 * sparc2.2 : update from old dodgey metadata
3106 * super-minor: change the preferred_minor number
3107 * summaries: update redundant counters.
3108 * homehost: update the recorded homehost
3109 * _reshape_progress: record new reshape_progress position.
3112 struct intel_super
*super
= st
->sb
;
3113 struct imsm_super
*mpb
;
3115 /* we can only update container info */
3116 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3119 mpb
= super
->anchor
;
3121 if (strcmp(update
, "uuid") == 0) {
3122 /* We take this to mean that the family_num should be updated.
3123 * However that is much smaller than the uuid so we cannot really
3124 * allow an explicit uuid to be given. And it is hard to reliably
3126 * So if !uuid_set we know the current uuid is random and just used
3127 * the first 'int' and copy it to the other 3 positions.
3128 * Otherwise we require the 4 'int's to be the same as would be the
3129 * case if we are using a random uuid. So an explicit uuid will be
3130 * accepted as long as all for ints are the same... which shouldn't hurt
3133 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3136 if (info
->uuid
[0] != info
->uuid
[1] ||
3137 info
->uuid
[1] != info
->uuid
[2] ||
3138 info
->uuid
[2] != info
->uuid
[3])
3144 mpb
->orig_family_num
= info
->uuid
[0];
3145 } else if (strcmp(update
, "assemble") == 0)
3150 /* successful update? recompute checksum */
3152 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3157 static size_t disks_to_mpb_size(int disks
)
3161 size
= sizeof(struct imsm_super
);
3162 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3163 size
+= 2 * sizeof(struct imsm_dev
);
3164 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3165 size
+= (4 - 2) * sizeof(struct imsm_map
);
3166 /* 4 possible disk_ord_tbl's */
3167 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3172 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3173 unsigned long long data_offset
)
3175 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3178 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3181 static void free_devlist(struct intel_super
*super
)
3183 struct intel_dev
*dv
;
3185 while (super
->devlist
) {
3186 dv
= super
->devlist
->next
;
3187 free(super
->devlist
->dev
);
3188 free(super
->devlist
);
3189 super
->devlist
= dv
;
3193 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3195 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3198 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3202 * 0 same, or first was empty, and second was copied
3203 * 1 second had wrong number
3205 * 3 wrong other info
3207 struct intel_super
*first
= st
->sb
;
3208 struct intel_super
*sec
= tst
->sb
;
3215 /* in platform dependent environment test if the disks
3216 * use the same Intel hba
3217 * If not on Intel hba at all, allow anything.
3219 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3220 if (first
->hba
->type
!= sec
->hba
->type
) {
3222 "HBAs of devices do not match %s != %s\n",
3223 get_sys_dev_type(first
->hba
->type
),
3224 get_sys_dev_type(sec
->hba
->type
));
3227 if (first
->orom
!= sec
->orom
) {
3229 "HBAs of devices do not match %s != %s\n",
3230 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3235 /* if an anchor does not have num_raid_devs set then it is a free
3238 if (first
->anchor
->num_raid_devs
> 0 &&
3239 sec
->anchor
->num_raid_devs
> 0) {
3240 /* Determine if these disks might ever have been
3241 * related. Further disambiguation can only take place
3242 * in load_super_imsm_all
3244 __u32 first_family
= first
->anchor
->orig_family_num
;
3245 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3247 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3248 MAX_SIGNATURE_LENGTH
) != 0)
3251 if (first_family
== 0)
3252 first_family
= first
->anchor
->family_num
;
3253 if (sec_family
== 0)
3254 sec_family
= sec
->anchor
->family_num
;
3256 if (first_family
!= sec_family
)
3261 /* if 'first' is a spare promote it to a populated mpb with sec's
3264 if (first
->anchor
->num_raid_devs
== 0 &&
3265 sec
->anchor
->num_raid_devs
> 0) {
3267 struct intel_dev
*dv
;
3268 struct imsm_dev
*dev
;
3270 /* we need to copy raid device info from sec if an allocation
3271 * fails here we don't associate the spare
3273 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3274 dv
= xmalloc(sizeof(*dv
));
3275 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3278 dv
->next
= first
->devlist
;
3279 first
->devlist
= dv
;
3281 if (i
< sec
->anchor
->num_raid_devs
) {
3282 /* allocation failure */
3283 free_devlist(first
);
3284 pr_err("imsm: failed to associate spare\n");
3287 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3288 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3289 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3290 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3291 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3292 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3298 static void fd2devname(int fd
, char *name
)
3302 char dname
[PATH_MAX
];
3307 if (fstat(fd
, &st
) != 0)
3309 sprintf(path
, "/sys/dev/block/%d:%d",
3310 major(st
.st_rdev
), minor(st
.st_rdev
));
3312 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3317 nm
= strrchr(dname
, '/');
3320 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3324 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3327 char *name
= fd2kname(fd
);
3332 if (strncmp(name
, "nvme", 4) != 0)
3335 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3337 return load_sys(path
, buf
, buf_len
);
3340 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3342 static int imsm_read_serial(int fd
, char *devname
,
3343 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3352 memset(buf
, 0, sizeof(buf
));
3354 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3357 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3359 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3360 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3361 fd2devname(fd
, (char *) serial
);
3367 pr_err("Failed to retrieve serial for %s\n",
3372 /* trim all whitespace and non-printable characters and convert
3375 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
3378 /* ':' is reserved for use in placeholder serial
3379 * numbers for missing disks
3390 /* truncate leading characters */
3391 if (len
> MAX_RAID_SERIAL_LEN
) {
3392 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3393 len
= MAX_RAID_SERIAL_LEN
;
3396 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3397 memcpy(serial
, dest
, len
);
3402 static int serialcmp(__u8
*s1
, __u8
*s2
)
3404 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3407 static void serialcpy(__u8
*dest
, __u8
*src
)
3409 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3412 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3416 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3417 if (serialcmp(dl
->serial
, serial
) == 0)
3423 static struct imsm_disk
*
3424 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3428 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3429 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3431 if (serialcmp(disk
->serial
, serial
) == 0) {
3442 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3444 struct imsm_disk
*disk
;
3449 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3451 rv
= imsm_read_serial(fd
, devname
, serial
);
3456 dl
= xcalloc(1, sizeof(*dl
));
3459 dl
->major
= major(stb
.st_rdev
);
3460 dl
->minor
= minor(stb
.st_rdev
);
3461 dl
->next
= super
->disks
;
3462 dl
->fd
= keep_fd
? fd
: -1;
3463 assert(super
->disks
== NULL
);
3465 serialcpy(dl
->serial
, serial
);
3468 fd2devname(fd
, name
);
3470 dl
->devname
= xstrdup(devname
);
3472 dl
->devname
= xstrdup(name
);
3474 /* look up this disk's index in the current anchor */
3475 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3478 /* only set index on disks that are a member of a
3479 * populated contianer, i.e. one with raid_devs
3481 if (is_failed(&dl
->disk
))
3483 else if (is_spare(&dl
->disk
))
3491 /* When migrating map0 contains the 'destination' state while map1
3492 * contains the current state. When not migrating map0 contains the
3493 * current state. This routine assumes that map[0].map_state is set to
3494 * the current array state before being called.
3496 * Migration is indicated by one of the following states
3497 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3498 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3499 * map1state=unitialized)
3500 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3502 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3503 * map1state=degraded)
3504 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3507 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3508 __u8 to_state
, int migr_type
)
3510 struct imsm_map
*dest
;
3511 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3513 dev
->vol
.migr_state
= 1;
3514 set_migr_type(dev
, migr_type
);
3515 dev
->vol
.curr_migr_unit
= 0;
3516 dest
= get_imsm_map(dev
, MAP_1
);
3518 /* duplicate and then set the target end state in map[0] */
3519 memcpy(dest
, src
, sizeof_imsm_map(src
));
3520 if (migr_type
== MIGR_REBUILD
|| migr_type
== MIGR_GEN_MIGR
) {
3524 for (i
= 0; i
< src
->num_members
; i
++) {
3525 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3526 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3530 if (migr_type
== MIGR_GEN_MIGR
)
3531 /* Clear migration record */
3532 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3534 src
->map_state
= to_state
;
3537 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3540 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3541 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3545 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3546 * completed in the last migration.
3548 * FIXME add support for raid-level-migration
3550 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
3551 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
3552 /* when final map state is other than expected
3553 * merge maps (not for migration)
3557 for (i
= 0; i
< prev
->num_members
; i
++)
3558 for (j
= 0; j
< map
->num_members
; j
++)
3559 /* during online capacity expansion
3560 * disks position can be changed
3561 * if takeover is used
3563 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3564 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3565 map
->disk_ord_tbl
[j
] |=
3566 prev
->disk_ord_tbl
[i
];
3569 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3570 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3573 dev
->vol
.migr_state
= 0;
3574 set_migr_type(dev
, 0);
3575 dev
->vol
.curr_migr_unit
= 0;
3576 map
->map_state
= map_state
;
3580 static int parse_raid_devices(struct intel_super
*super
)
3583 struct imsm_dev
*dev_new
;
3584 size_t len
, len_migr
;
3586 size_t space_needed
= 0;
3587 struct imsm_super
*mpb
= super
->anchor
;
3589 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3590 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3591 struct intel_dev
*dv
;
3593 len
= sizeof_imsm_dev(dev_iter
, 0);
3594 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3596 space_needed
+= len_migr
- len
;
3598 dv
= xmalloc(sizeof(*dv
));
3599 if (max_len
< len_migr
)
3601 if (max_len
> len_migr
)
3602 space_needed
+= max_len
- len_migr
;
3603 dev_new
= xmalloc(max_len
);
3604 imsm_copy_dev(dev_new
, dev_iter
);
3607 dv
->next
= super
->devlist
;
3608 super
->devlist
= dv
;
3611 /* ensure that super->buf is large enough when all raid devices
3614 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3617 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3618 if (posix_memalign(&buf
, 512, len
) != 0)
3621 memcpy(buf
, super
->buf
, super
->len
);
3622 memset(buf
+ super
->len
, 0, len
- super
->len
);
3631 /* retrieve a pointer to the bbm log which starts after all raid devices */
3632 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3636 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3638 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3644 /*******************************************************************************
3645 * Function: check_mpb_migr_compatibility
3646 * Description: Function checks for unsupported migration features:
3647 * - migration optimization area (pba_of_lba0)
3648 * - descending reshape (ascending_migr)
3650 * super : imsm metadata information
3652 * 0 : migration is compatible
3653 * -1 : migration is not compatible
3654 ******************************************************************************/
3655 int check_mpb_migr_compatibility(struct intel_super
*super
)
3657 struct imsm_map
*map0
, *map1
;
3658 struct migr_record
*migr_rec
= super
->migr_rec
;
3661 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3662 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3665 dev_iter
->vol
.migr_state
== 1 &&
3666 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3667 /* This device is migrating */
3668 map0
= get_imsm_map(dev_iter
, MAP_0
);
3669 map1
= get_imsm_map(dev_iter
, MAP_1
);
3670 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3671 /* migration optimization area was used */
3673 if (migr_rec
->ascending_migr
== 0
3674 && migr_rec
->dest_depth_per_unit
> 0)
3675 /* descending reshape not supported yet */
3682 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3684 /* load_imsm_mpb - read matrix metadata
3685 * allocates super->mpb to be freed by free_imsm
3687 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3689 unsigned long long dsize
;
3690 unsigned long long sectors
;
3692 struct imsm_super
*anchor
;
3695 get_dev_size(fd
, NULL
, &dsize
);
3698 pr_err("%s: device to small for imsm\n",
3703 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3705 pr_err("Cannot seek to anchor block on %s: %s\n",
3706 devname
, strerror(errno
));
3710 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3712 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
3715 if (read(fd
, anchor
, 512) != 512) {
3717 pr_err("Cannot read anchor block on %s: %s\n",
3718 devname
, strerror(errno
));
3723 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3725 pr_err("no IMSM anchor on %s\n", devname
);
3730 __free_imsm(super
, 0);
3731 /* reload capability and hba */
3733 /* capability and hba must be updated with new super allocation */
3734 find_intel_hba_capability(fd
, super
, devname
);
3735 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3736 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3738 pr_err("unable to allocate %zu byte mpb buffer\n",
3743 memcpy(super
->buf
, anchor
, 512);
3745 sectors
= mpb_sectors(anchor
) - 1;
3748 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3749 pr_err("could not allocate migr_rec buffer\n");
3753 super
->clean_migration_record_by_mdmon
= 0;
3756 check_sum
= __gen_imsm_checksum(super
->anchor
);
3757 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3759 pr_err("IMSM checksum %x != %x on %s\n",
3761 __le32_to_cpu(super
->anchor
->check_sum
),
3769 /* read the extended mpb */
3770 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3772 pr_err("Cannot seek to extended mpb on %s: %s\n",
3773 devname
, strerror(errno
));
3777 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3779 pr_err("Cannot read extended mpb on %s: %s\n",
3780 devname
, strerror(errno
));
3784 check_sum
= __gen_imsm_checksum(super
->anchor
);
3785 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3787 pr_err("IMSM checksum %x != %x on %s\n",
3788 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3793 /* FIXME the BBM log is disk specific so we cannot use this global
3794 * buffer for all disks. Ok for now since we only look at the global
3795 * bbm_log_size parameter to gate assembly
3797 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3802 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3804 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3805 static void clear_hi(struct intel_super
*super
)
3807 struct imsm_super
*mpb
= super
->anchor
;
3809 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3811 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3812 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3813 disk
->total_blocks_hi
= 0;
3815 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3816 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3819 for (n
= 0; n
< 2; ++n
) {
3820 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3823 map
->pba_of_lba0_hi
= 0;
3824 map
->blocks_per_member_hi
= 0;
3825 map
->num_data_stripes_hi
= 0;
3831 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3835 err
= load_imsm_mpb(fd
, super
, devname
);
3838 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3841 err
= parse_raid_devices(super
);
3846 static void __free_imsm_disk(struct dl
*d
)
3858 static void free_imsm_disks(struct intel_super
*super
)
3862 while (super
->disks
) {
3864 super
->disks
= d
->next
;
3865 __free_imsm_disk(d
);
3867 while (super
->disk_mgmt_list
) {
3868 d
= super
->disk_mgmt_list
;
3869 super
->disk_mgmt_list
= d
->next
;
3870 __free_imsm_disk(d
);
3872 while (super
->missing
) {
3874 super
->missing
= d
->next
;
3875 __free_imsm_disk(d
);
3880 /* free all the pieces hanging off of a super pointer */
3881 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3883 struct intel_hba
*elem
, *next
;
3889 /* unlink capability description */
3891 if (super
->migr_rec_buf
) {
3892 free(super
->migr_rec_buf
);
3893 super
->migr_rec_buf
= NULL
;
3896 free_imsm_disks(super
);
3897 free_devlist(super
);
3901 free((void *)elem
->path
);
3909 static void free_imsm(struct intel_super
*super
)
3911 __free_imsm(super
, 1);
3915 static void free_super_imsm(struct supertype
*st
)
3917 struct intel_super
*super
= st
->sb
;
3926 static struct intel_super
*alloc_super(void)
3928 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
3930 super
->current_vol
= -1;
3931 super
->create_offset
= ~((unsigned long long) 0);
3936 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3938 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3940 struct sys_dev
*hba_name
;
3943 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
3948 hba_name
= find_disk_attached_hba(fd
, NULL
);
3951 pr_err("%s is not attached to Intel(R) RAID controller.\n",
3955 rv
= attach_hba_to_super(super
, hba_name
);
3958 struct intel_hba
*hba
= super
->hba
;
3960 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
3961 " but the container is assigned to Intel(R) %s %s (",
3963 get_sys_dev_type(hba_name
->type
),
3964 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
3965 hba_name
->pci_id
? : "Err!",
3966 get_sys_dev_type(super
->hba
->type
),
3967 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
3970 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3972 fprintf(stderr
, ", ");
3975 fprintf(stderr
, ").\n"
3976 " Mixing devices attached to different %s is not allowed.\n",
3977 hba_name
->type
== SYS_DEV_VMD
? "VMD domains" : "controllers");
3981 super
->orom
= find_imsm_capability(hba_name
);
3988 /* find_missing - helper routine for load_super_imsm_all that identifies
3989 * disks that have disappeared from the system. This routine relies on
3990 * the mpb being uptodate, which it is at load time.
3992 static int find_missing(struct intel_super
*super
)
3995 struct imsm_super
*mpb
= super
->anchor
;
3997 struct imsm_disk
*disk
;
3999 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4000 disk
= __get_imsm_disk(mpb
, i
);
4001 dl
= serial_to_dl(disk
->serial
, super
);
4005 dl
= xmalloc(sizeof(*dl
));
4009 dl
->devname
= xstrdup("missing");
4011 serialcpy(dl
->serial
, disk
->serial
);
4014 dl
->next
= super
->missing
;
4015 super
->missing
= dl
;
4022 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4024 struct intel_disk
*idisk
= disk_list
;
4027 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4029 idisk
= idisk
->next
;
4035 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4036 struct intel_super
*super
,
4037 struct intel_disk
**disk_list
)
4039 struct imsm_disk
*d
= &super
->disks
->disk
;
4040 struct imsm_super
*mpb
= super
->anchor
;
4043 for (i
= 0; i
< tbl_size
; i
++) {
4044 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4045 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4047 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4048 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4049 dprintf("mpb from %d:%d matches %d:%d\n",
4050 super
->disks
->major
,
4051 super
->disks
->minor
,
4052 table
[i
]->disks
->major
,
4053 table
[i
]->disks
->minor
);
4057 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4058 is_configured(d
) == is_configured(tbl_d
)) &&
4059 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4060 /* current version of the mpb is a
4061 * better candidate than the one in
4062 * super_table, but copy over "cross
4063 * generational" status
4065 struct intel_disk
*idisk
;
4067 dprintf("mpb from %d:%d replaces %d:%d\n",
4068 super
->disks
->major
,
4069 super
->disks
->minor
,
4070 table
[i
]->disks
->major
,
4071 table
[i
]->disks
->minor
);
4073 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4074 if (idisk
&& is_failed(&idisk
->disk
))
4075 tbl_d
->status
|= FAILED_DISK
;
4078 struct intel_disk
*idisk
;
4079 struct imsm_disk
*disk
;
4081 /* tbl_mpb is more up to date, but copy
4082 * over cross generational status before
4085 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4086 if (disk
&& is_failed(disk
))
4087 d
->status
|= FAILED_DISK
;
4089 idisk
= disk_list_get(d
->serial
, *disk_list
);
4092 if (disk
&& is_configured(disk
))
4093 idisk
->disk
.status
|= CONFIGURED_DISK
;
4096 dprintf("mpb from %d:%d prefer %d:%d\n",
4097 super
->disks
->major
,
4098 super
->disks
->minor
,
4099 table
[i
]->disks
->major
,
4100 table
[i
]->disks
->minor
);
4108 table
[tbl_size
++] = super
;
4112 /* update/extend the merged list of imsm_disk records */
4113 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4114 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4115 struct intel_disk
*idisk
;
4117 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4119 idisk
->disk
.status
|= disk
->status
;
4120 if (is_configured(&idisk
->disk
) ||
4121 is_failed(&idisk
->disk
))
4122 idisk
->disk
.status
&= ~(SPARE_DISK
);
4124 idisk
= xcalloc(1, sizeof(*idisk
));
4125 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4126 idisk
->disk
= *disk
;
4127 idisk
->next
= *disk_list
;
4131 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4138 static struct intel_super
*
4139 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4142 struct imsm_super
*mpb
= super
->anchor
;
4146 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4147 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4148 struct intel_disk
*idisk
;
4150 idisk
= disk_list_get(disk
->serial
, disk_list
);
4152 if (idisk
->owner
== owner
||
4153 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4156 dprintf("'%.16s' owner %d != %d\n",
4157 disk
->serial
, idisk
->owner
,
4160 dprintf("unknown disk %x [%d]: %.16s\n",
4161 __le32_to_cpu(mpb
->family_num
), i
,
4167 if (ok_count
== mpb
->num_disks
)
4172 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4174 struct intel_super
*s
;
4176 for (s
= super_list
; s
; s
= s
->next
) {
4177 if (family_num
!= s
->anchor
->family_num
)
4179 pr_err("Conflict, offlining family %#x on '%s'\n",
4180 __le32_to_cpu(family_num
), s
->disks
->devname
);
4184 static struct intel_super
*
4185 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4187 struct intel_super
*super_table
[len
];
4188 struct intel_disk
*disk_list
= NULL
;
4189 struct intel_super
*champion
, *spare
;
4190 struct intel_super
*s
, **del
;
4195 memset(super_table
, 0, sizeof(super_table
));
4196 for (s
= *super_list
; s
; s
= s
->next
)
4197 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4199 for (i
= 0; i
< tbl_size
; i
++) {
4200 struct imsm_disk
*d
;
4201 struct intel_disk
*idisk
;
4202 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4205 d
= &s
->disks
->disk
;
4207 /* 'd' must appear in merged disk list for its
4208 * configuration to be valid
4210 idisk
= disk_list_get(d
->serial
, disk_list
);
4211 if (idisk
&& idisk
->owner
== i
)
4212 s
= validate_members(s
, disk_list
, i
);
4217 dprintf("marking family: %#x from %d:%d offline\n",
4219 super_table
[i
]->disks
->major
,
4220 super_table
[i
]->disks
->minor
);
4224 /* This is where the mdadm implementation differs from the Windows
4225 * driver which has no strict concept of a container. We can only
4226 * assemble one family from a container, so when returning a prodigal
4227 * array member to this system the code will not be able to disambiguate
4228 * the container contents that should be assembled ("foreign" versus
4229 * "local"). It requires user intervention to set the orig_family_num
4230 * to a new value to establish a new container. The Windows driver in
4231 * this situation fixes up the volume name in place and manages the
4232 * foreign array as an independent entity.
4237 for (i
= 0; i
< tbl_size
; i
++) {
4238 struct intel_super
*tbl_ent
= super_table
[i
];
4244 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4249 if (s
&& !is_spare
) {
4250 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4252 } else if (!s
&& !is_spare
)
4265 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4266 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4268 /* collect all dl's onto 'champion', and update them to
4269 * champion's version of the status
4271 for (s
= *super_list
; s
; s
= s
->next
) {
4272 struct imsm_super
*mpb
= champion
->anchor
;
4273 struct dl
*dl
= s
->disks
;
4278 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4280 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4281 struct imsm_disk
*disk
;
4283 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4286 /* only set index on disks that are a member of
4287 * a populated contianer, i.e. one with
4290 if (is_failed(&dl
->disk
))
4292 else if (is_spare(&dl
->disk
))
4298 if (i
>= mpb
->num_disks
) {
4299 struct intel_disk
*idisk
;
4301 idisk
= disk_list_get(dl
->serial
, disk_list
);
4302 if (idisk
&& is_spare(&idisk
->disk
) &&
4303 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4311 dl
->next
= champion
->disks
;
4312 champion
->disks
= dl
;
4316 /* delete 'champion' from super_list */
4317 for (del
= super_list
; *del
; ) {
4318 if (*del
== champion
) {
4319 *del
= (*del
)->next
;
4322 del
= &(*del
)->next
;
4324 champion
->next
= NULL
;
4328 struct intel_disk
*idisk
= disk_list
;
4330 disk_list
= disk_list
->next
;
4338 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4339 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4340 int major
, int minor
, int keep_fd
);
4342 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4343 int *max
, int keep_fd
);
4345 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4346 char *devname
, struct md_list
*devlist
,
4349 struct intel_super
*super_list
= NULL
;
4350 struct intel_super
*super
= NULL
;
4355 /* 'fd' is an opened container */
4356 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4358 /* get super block from devlist devices */
4359 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4362 /* all mpbs enter, maybe one leaves */
4363 super
= imsm_thunderdome(&super_list
, i
);
4369 if (find_missing(super
) != 0) {
4375 /* load migration record */
4376 err
= load_imsm_migr_rec(super
, NULL
);
4378 /* migration is in progress,
4379 * but migr_rec cannot be loaded,
4385 /* Check migration compatibility */
4386 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
4387 pr_err("Unsupported migration detected");
4389 fprintf(stderr
, " on %s\n", devname
);
4391 fprintf(stderr
, " (IMSM).\n");
4400 while (super_list
) {
4401 struct intel_super
*s
= super_list
;
4403 super_list
= super_list
->next
;
4412 strcpy(st
->container_devnm
, fd2devnm(fd
));
4414 st
->container_devnm
[0] = 0;
4415 if (err
== 0 && st
->ss
== NULL
) {
4416 st
->ss
= &super_imsm
;
4417 st
->minor_version
= 0;
4418 st
->max_devs
= IMSM_MAX_DEVICES
;
4424 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4425 int *max
, int keep_fd
)
4427 struct md_list
*tmpdev
;
4431 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4432 if (tmpdev
->used
!= 1)
4434 if (tmpdev
->container
== 1) {
4436 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4438 pr_err("cannot open device %s: %s\n",
4439 tmpdev
->devname
, strerror(errno
));
4443 err
= get_sra_super_block(fd
, super_list
,
4444 tmpdev
->devname
, &lmax
,
4453 int major
= major(tmpdev
->st_rdev
);
4454 int minor
= minor(tmpdev
->st_rdev
);
4455 err
= get_super_block(super_list
,
4472 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4473 int major
, int minor
, int keep_fd
)
4475 struct intel_super
*s
;
4487 sprintf(nm
, "%d:%d", major
, minor
);
4488 dfd
= dev_open(nm
, O_RDWR
);
4494 find_intel_hba_capability(dfd
, s
, devname
);
4495 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4497 /* retry the load if we might have raced against mdmon */
4498 if (err
== 3 && devnm
&& mdmon_running(devnm
))
4499 for (retry
= 0; retry
< 3; retry
++) {
4501 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4507 s
->next
= *super_list
;
4515 if (dfd
>= 0 && !keep_fd
)
4522 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4529 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4533 if (sra
->array
.major_version
!= -1 ||
4534 sra
->array
.minor_version
!= -2 ||
4535 strcmp(sra
->text_version
, "imsm") != 0) {
4540 devnm
= fd2devnm(fd
);
4541 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4542 if (get_super_block(super_list
, devnm
, devname
,
4543 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4554 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4556 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4560 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4562 struct intel_super
*super
;
4566 if (test_partition(fd
))
4567 /* IMSM not allowed on partitions */
4570 free_super_imsm(st
);
4572 super
= alloc_super();
4573 /* Load hba and capabilities if they exist.
4574 * But do not preclude loading metadata in case capabilities or hba are
4575 * non-compliant and ignore_hw_compat is set.
4577 rv
= find_intel_hba_capability(fd
, super
, devname
);
4578 /* no orom/efi or non-intel hba of the disk */
4579 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
4581 pr_err("No OROM/EFI properties for %s\n", devname
);
4585 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4587 /* retry the load if we might have raced against mdmon */
4589 struct mdstat_ent
*mdstat
= NULL
;
4590 char *name
= fd2kname(fd
);
4593 mdstat
= mdstat_by_component(name
);
4595 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
4596 for (retry
= 0; retry
< 3; retry
++) {
4598 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4604 free_mdstat(mdstat
);
4609 pr_err("Failed to load all information sections on %s\n", devname
);
4615 if (st
->ss
== NULL
) {
4616 st
->ss
= &super_imsm
;
4617 st
->minor_version
= 0;
4618 st
->max_devs
= IMSM_MAX_DEVICES
;
4621 /* load migration record */
4622 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4623 /* Check for unsupported migration features */
4624 if (check_mpb_migr_compatibility(super
) != 0) {
4625 pr_err("Unsupported migration detected");
4627 fprintf(stderr
, " on %s\n", devname
);
4629 fprintf(stderr
, " (IMSM).\n");
4637 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4639 if (info
->level
== 1)
4641 return info
->chunk_size
>> 9;
4644 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4645 unsigned long long size
)
4647 if (info
->level
== 1)
4650 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4653 static void imsm_update_version_info(struct intel_super
*super
)
4655 /* update the version and attributes */
4656 struct imsm_super
*mpb
= super
->anchor
;
4658 struct imsm_dev
*dev
;
4659 struct imsm_map
*map
;
4662 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4663 dev
= get_imsm_dev(super
, i
);
4664 map
= get_imsm_map(dev
, MAP_0
);
4665 if (__le32_to_cpu(dev
->size_high
) > 0)
4666 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4668 /* FIXME detect when an array spans a port multiplier */
4670 mpb
->attributes
|= MPB_ATTRIB_PM
;
4673 if (mpb
->num_raid_devs
> 1 ||
4674 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4675 version
= MPB_VERSION_ATTRIBS
;
4676 switch (get_imsm_raid_level(map
)) {
4677 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4678 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4679 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4680 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4683 if (map
->num_members
>= 5)
4684 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4685 else if (dev
->status
== DEV_CLONE_N_GO
)
4686 version
= MPB_VERSION_CNG
;
4687 else if (get_imsm_raid_level(map
) == 5)
4688 version
= MPB_VERSION_RAID5
;
4689 else if (map
->num_members
>= 3)
4690 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4691 else if (get_imsm_raid_level(map
) == 1)
4692 version
= MPB_VERSION_RAID1
;
4694 version
= MPB_VERSION_RAID0
;
4696 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4700 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4702 struct imsm_super
*mpb
= super
->anchor
;
4703 char *reason
= NULL
;
4706 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4707 reason
= "must be 16 characters or less";
4709 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4710 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4712 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4713 reason
= "already exists";
4718 if (reason
&& !quiet
)
4719 pr_err("imsm volume name %s\n", reason
);
4724 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4725 unsigned long long size
, char *name
,
4726 char *homehost
, int *uuid
,
4727 long long data_offset
)
4729 /* We are creating a volume inside a pre-existing container.
4730 * so st->sb is already set.
4732 struct intel_super
*super
= st
->sb
;
4733 struct imsm_super
*mpb
= super
->anchor
;
4734 struct intel_dev
*dv
;
4735 struct imsm_dev
*dev
;
4736 struct imsm_vol
*vol
;
4737 struct imsm_map
*map
;
4738 int idx
= mpb
->num_raid_devs
;
4740 unsigned long long array_blocks
;
4741 size_t size_old
, size_new
;
4742 unsigned long long num_data_stripes
;
4744 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4745 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
4749 /* ensure the mpb is large enough for the new data */
4750 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4751 size_new
= disks_to_mpb_size(info
->nr_disks
);
4752 if (size_new
> size_old
) {
4754 size_t size_round
= ROUND_UP(size_new
, 512);
4756 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4757 pr_err("could not allocate new mpb\n");
4760 if (posix_memalign(&super
->migr_rec_buf
, 512,
4761 MIGR_REC_BUF_SIZE
) != 0) {
4762 pr_err("could not allocate migr_rec buffer\n");
4768 memcpy(mpb_new
, mpb
, size_old
);
4771 super
->anchor
= mpb_new
;
4772 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4773 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4775 super
->current_vol
= idx
;
4777 /* handle 'failed_disks' by either:
4778 * a) create dummy disk entries in the table if this the first
4779 * volume in the array. We add them here as this is the only
4780 * opportunity to add them. add_to_super_imsm_volume()
4781 * handles the non-failed disks and continues incrementing
4783 * b) validate that 'failed_disks' matches the current number
4784 * of missing disks if the container is populated
4786 if (super
->current_vol
== 0) {
4788 for (i
= 0; i
< info
->failed_disks
; i
++) {
4789 struct imsm_disk
*disk
;
4792 disk
= __get_imsm_disk(mpb
, i
);
4793 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4794 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4795 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4798 find_missing(super
);
4803 for (d
= super
->missing
; d
; d
= d
->next
)
4805 if (info
->failed_disks
> missing
) {
4806 pr_err("unable to add 'missing' disk to container\n");
4811 if (!check_name(super
, name
, 0))
4813 dv
= xmalloc(sizeof(*dv
));
4814 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4815 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4816 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4817 info
->layout
, info
->chunk_size
,
4819 /* round array size down to closest MB */
4820 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4822 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4823 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4824 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4826 vol
->migr_state
= 0;
4827 set_migr_type(dev
, MIGR_INIT
);
4828 vol
->dirty
= !info
->state
;
4829 vol
->curr_migr_unit
= 0;
4830 map
= get_imsm_map(dev
, MAP_0
);
4831 set_pba_of_lba0(map
, super
->create_offset
);
4832 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4833 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4834 map
->failed_disk_num
= ~0;
4835 if (info
->level
> 0)
4836 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
4837 : IMSM_T_STATE_UNINITIALIZED
);
4839 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4840 IMSM_T_STATE_NORMAL
;
4843 if (info
->level
== 1 && info
->raid_disks
> 2) {
4846 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
4850 map
->raid_level
= info
->level
;
4851 if (info
->level
== 10) {
4852 map
->raid_level
= 1;
4853 map
->num_domains
= info
->raid_disks
/ 2;
4854 } else if (info
->level
== 1)
4855 map
->num_domains
= info
->raid_disks
;
4857 map
->num_domains
= 1;
4859 /* info->size is only int so use the 'size' parameter instead */
4860 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4861 num_data_stripes
/= map
->num_domains
;
4862 set_num_data_stripes(map
, num_data_stripes
);
4864 map
->num_members
= info
->raid_disks
;
4865 for (i
= 0; i
< map
->num_members
; i
++) {
4866 /* initialized in add_to_super */
4867 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4869 mpb
->num_raid_devs
++;
4872 dv
->index
= super
->current_vol
;
4873 dv
->next
= super
->devlist
;
4874 super
->devlist
= dv
;
4876 imsm_update_version_info(super
);
4881 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4882 unsigned long long size
, char *name
,
4883 char *homehost
, int *uuid
,
4884 unsigned long long data_offset
)
4886 /* This is primarily called by Create when creating a new array.
4887 * We will then get add_to_super called for each component, and then
4888 * write_init_super called to write it out to each device.
4889 * For IMSM, Create can create on fresh devices or on a pre-existing
4891 * To create on a pre-existing array a different method will be called.
4892 * This one is just for fresh drives.
4894 struct intel_super
*super
;
4895 struct imsm_super
*mpb
;
4899 if (data_offset
!= INVALID_SECTORS
) {
4900 pr_err("data-offset not supported by imsm\n");
4905 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
,
4909 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4913 super
= alloc_super();
4914 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4919 pr_err("could not allocate superblock\n");
4922 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4923 pr_err("could not allocate migr_rec buffer\n");
4928 memset(super
->buf
, 0, mpb_size
);
4930 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4934 /* zeroing superblock */
4938 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4940 version
= (char *) mpb
->sig
;
4941 strcpy(version
, MPB_SIGNATURE
);
4942 version
+= strlen(MPB_SIGNATURE
);
4943 strcpy(version
, MPB_VERSION_RAID0
);
4949 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4950 int fd
, char *devname
)
4952 struct intel_super
*super
= st
->sb
;
4953 struct imsm_super
*mpb
= super
->anchor
;
4954 struct imsm_disk
*_disk
;
4955 struct imsm_dev
*dev
;
4956 struct imsm_map
*map
;
4960 dev
= get_imsm_dev(super
, super
->current_vol
);
4961 map
= get_imsm_map(dev
, MAP_0
);
4963 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4964 pr_err("%s: Cannot add spare devices to IMSM volume\n",
4970 /* we're doing autolayout so grab the pre-marked (in
4971 * validate_geometry) raid_disk
4973 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4974 if (dl
->raiddisk
== dk
->raid_disk
)
4977 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4978 if (dl
->major
== dk
->major
&&
4979 dl
->minor
== dk
->minor
)
4984 pr_err("%s is not a member of the same container\n", devname
);
4988 /* add a pristine spare to the metadata */
4989 if (dl
->index
< 0) {
4990 dl
->index
= super
->anchor
->num_disks
;
4991 super
->anchor
->num_disks
++;
4993 /* Check the device has not already been added */
4994 slot
= get_imsm_disk_slot(map
, dl
->index
);
4996 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4997 pr_err("%s has been included in this array twice\n",
5001 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5002 dl
->disk
.status
= CONFIGURED_DISK
;
5004 /* update size of 'missing' disks to be at least as large as the
5005 * largest acitve member (we only have dummy missing disks when
5006 * creating the first volume)
5008 if (super
->current_vol
== 0) {
5009 for (df
= super
->missing
; df
; df
= df
->next
) {
5010 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5011 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5012 _disk
= __get_imsm_disk(mpb
, df
->index
);
5017 /* refresh unset/failed slots to point to valid 'missing' entries */
5018 for (df
= super
->missing
; df
; df
= df
->next
)
5019 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5020 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5022 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5024 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5025 if (is_gen_migration(dev
)) {
5026 struct imsm_map
*map2
= get_imsm_map(dev
,
5028 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5029 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5030 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5033 if ((unsigned)df
->index
==
5035 set_imsm_ord_tbl_ent(map2
,
5041 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5045 /* if we are creating the first raid device update the family number */
5046 if (super
->current_vol
== 0) {
5048 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5050 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5051 if (!_dev
|| !_disk
) {
5052 pr_err("BUG mpb setup error\n");
5058 sum
+= __gen_imsm_checksum(mpb
);
5059 mpb
->family_num
= __cpu_to_le32(sum
);
5060 mpb
->orig_family_num
= mpb
->family_num
;
5062 super
->current_disk
= dl
;
5067 * Function marks disk as spare and restores disk serial
5068 * in case it was previously marked as failed by takeover operation
5070 * -1 : critical error
5071 * 0 : disk is marked as spare but serial is not set
5074 int mark_spare(struct dl
*disk
)
5076 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5083 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5084 /* Restore disk serial number, because takeover marks disk
5085 * as failed and adds to serial ':0' before it becomes
5088 serialcpy(disk
->serial
, serial
);
5089 serialcpy(disk
->disk
.serial
, serial
);
5092 disk
->disk
.status
= SPARE_DISK
;
5098 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5099 int fd
, char *devname
,
5100 unsigned long long data_offset
)
5102 struct intel_super
*super
= st
->sb
;
5104 unsigned long long size
;
5109 /* If we are on an RAID enabled platform check that the disk is
5110 * attached to the raid controller.
5111 * We do not need to test disks attachment for container based additions,
5112 * they shall be already tested when container was created/assembled.
5114 rv
= find_intel_hba_capability(fd
, super
, devname
);
5115 /* no orom/efi or non-intel hba of the disk */
5117 dprintf("capability: %p fd: %d ret: %d\n",
5118 super
->orom
, fd
, rv
);
5122 if (super
->current_vol
>= 0)
5123 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5126 dd
= xcalloc(sizeof(*dd
), 1);
5127 dd
->major
= major(stb
.st_rdev
);
5128 dd
->minor
= minor(stb
.st_rdev
);
5129 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5132 dd
->action
= DISK_ADD
;
5133 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5135 pr_err("failed to retrieve scsi serial, aborting\n");
5141 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5142 (super
->hba
->type
== SYS_DEV_VMD
))) {
5144 char *devpath
= diskfd_to_devpath(fd
);
5145 char controller_path
[PATH_MAX
];
5148 pr_err("failed to get devpath, aborting\n");
5155 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5158 if (devpath_to_vendor(controller_path
) == 0x8086) {
5160 * If Intel's NVMe drive has serial ended with
5161 * "-A","-B","-1" or "-2" it means that this is "x8"
5162 * device (double drive on single PCIe card).
5163 * User should be warned about potential data loss.
5165 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5166 /* Skip empty character at the end */
5167 if (dd
->serial
[i
] == 0)
5170 if (((dd
->serial
[i
] == 'A') ||
5171 (dd
->serial
[i
] == 'B') ||
5172 (dd
->serial
[i
] == '1') ||
5173 (dd
->serial
[i
] == '2')) &&
5174 (dd
->serial
[i
-1] == '-'))
5175 pr_err("\tThe action you are about to take may put your data at risk.\n"
5176 "\tPlease note that x8 devices may consist of two separate x4 devices "
5177 "located on a single PCIe port.\n"
5178 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5184 get_dev_size(fd
, NULL
, &size
);
5185 /* clear migr_rec when adding disk to container */
5186 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5187 if (lseek64(fd
, size
- MIGR_REC_POSITION
, SEEK_SET
) >= 0) {
5188 if (write(fd
, super
->migr_rec_buf
,
5189 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5190 perror("Write migr_rec failed");
5194 serialcpy(dd
->disk
.serial
, dd
->serial
);
5195 set_total_blocks(&dd
->disk
, size
);
5196 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5197 struct imsm_super
*mpb
= super
->anchor
;
5198 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5201 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5202 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5204 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5206 if (st
->update_tail
) {
5207 dd
->next
= super
->disk_mgmt_list
;
5208 super
->disk_mgmt_list
= dd
;
5210 dd
->next
= super
->disks
;
5212 super
->updates_pending
++;
5218 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5220 struct intel_super
*super
= st
->sb
;
5223 /* remove from super works only in mdmon - for communication
5224 * manager - monitor. Check if communication memory buffer
5227 if (!st
->update_tail
) {
5228 pr_err("shall be used in mdmon context only\n");
5231 dd
= xcalloc(1, sizeof(*dd
));
5232 dd
->major
= dk
->major
;
5233 dd
->minor
= dk
->minor
;
5236 dd
->action
= DISK_REMOVE
;
5238 dd
->next
= super
->disk_mgmt_list
;
5239 super
->disk_mgmt_list
= dd
;
5244 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5248 struct imsm_super anchor
;
5249 } spare_record
__attribute__ ((aligned(512)));
5251 /* spare records have their own family number and do not have any defined raid
5254 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5256 struct imsm_super
*mpb
= super
->anchor
;
5257 struct imsm_super
*spare
= &spare_record
.anchor
;
5261 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5262 spare
->generation_num
= __cpu_to_le32(1UL);
5263 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5264 spare
->num_disks
= 1;
5265 spare
->num_raid_devs
= 0;
5266 spare
->cache_size
= mpb
->cache_size
;
5267 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5269 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5270 MPB_SIGNATURE MPB_VERSION_RAID0
);
5272 for (d
= super
->disks
; d
; d
= d
->next
) {
5276 spare
->disk
[0] = d
->disk
;
5277 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5278 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5280 sum
= __gen_imsm_checksum(spare
);
5281 spare
->family_num
= __cpu_to_le32(sum
);
5282 spare
->orig_family_num
= 0;
5283 sum
= __gen_imsm_checksum(spare
);
5284 spare
->check_sum
= __cpu_to_le32(sum
);
5286 if (store_imsm_mpb(d
->fd
, spare
)) {
5287 pr_err("failed for device %d:%d %s\n",
5288 d
->major
, d
->minor
, strerror(errno
));
5300 static int write_super_imsm(struct supertype
*st
, int doclose
)
5302 struct intel_super
*super
= st
->sb
;
5303 struct imsm_super
*mpb
= super
->anchor
;
5309 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5311 int clear_migration_record
= 1;
5313 /* 'generation' is incremented everytime the metadata is written */
5314 generation
= __le32_to_cpu(mpb
->generation_num
);
5316 mpb
->generation_num
= __cpu_to_le32(generation
);
5318 /* fix up cases where previous mdadm releases failed to set
5321 if (mpb
->orig_family_num
== 0)
5322 mpb
->orig_family_num
= mpb
->family_num
;
5324 for (d
= super
->disks
; d
; d
= d
->next
) {
5328 mpb
->disk
[d
->index
] = d
->disk
;
5332 for (d
= super
->missing
; d
; d
= d
->next
) {
5333 mpb
->disk
[d
->index
] = d
->disk
;
5336 mpb
->num_disks
= num_disks
;
5337 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5339 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5340 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5341 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5343 imsm_copy_dev(dev
, dev2
);
5344 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5346 if (is_gen_migration(dev2
))
5347 clear_migration_record
= 0;
5349 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5350 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5352 /* recalculate checksum */
5353 sum
= __gen_imsm_checksum(mpb
);
5354 mpb
->check_sum
= __cpu_to_le32(sum
);
5356 if (super
->clean_migration_record_by_mdmon
) {
5357 clear_migration_record
= 1;
5358 super
->clean_migration_record_by_mdmon
= 0;
5360 if (clear_migration_record
)
5361 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5363 /* write the mpb for disks that compose raid devices */
5364 for (d
= super
->disks
; d
; d
= d
->next
) {
5365 if (d
->index
< 0 || is_failed(&d
->disk
))
5368 if (clear_migration_record
) {
5369 unsigned long long dsize
;
5371 get_dev_size(d
->fd
, NULL
, &dsize
);
5372 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5373 if (write(d
->fd
, super
->migr_rec_buf
,
5374 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5375 perror("Write migr_rec failed");
5379 if (store_imsm_mpb(d
->fd
, mpb
))
5381 "failed for device %d:%d (fd: %d)%s\n",
5383 d
->fd
, strerror(errno
));
5392 return write_super_imsm_spares(super
, doclose
);
5397 static int create_array(struct supertype
*st
, int dev_idx
)
5400 struct imsm_update_create_array
*u
;
5401 struct intel_super
*super
= st
->sb
;
5402 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5403 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5404 struct disk_info
*inf
;
5405 struct imsm_disk
*disk
;
5408 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5409 sizeof(*inf
) * map
->num_members
;
5411 u
->type
= update_create_array
;
5412 u
->dev_idx
= dev_idx
;
5413 imsm_copy_dev(&u
->dev
, dev
);
5414 inf
= get_disk_info(u
);
5415 for (i
= 0; i
< map
->num_members
; i
++) {
5416 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5418 disk
= get_imsm_disk(super
, idx
);
5420 disk
= get_imsm_missing(super
, idx
);
5421 serialcpy(inf
[i
].serial
, disk
->serial
);
5423 append_metadata_update(st
, u
, len
);
5428 static int mgmt_disk(struct supertype
*st
)
5430 struct intel_super
*super
= st
->sb
;
5432 struct imsm_update_add_remove_disk
*u
;
5434 if (!super
->disk_mgmt_list
)
5439 u
->type
= update_add_remove_disk
;
5440 append_metadata_update(st
, u
, len
);
5445 static int write_init_super_imsm(struct supertype
*st
)
5447 struct intel_super
*super
= st
->sb
;
5448 int current_vol
= super
->current_vol
;
5450 /* we are done with current_vol reset it to point st at the container */
5451 super
->current_vol
= -1;
5453 if (st
->update_tail
) {
5454 /* queue the recently created array / added disk
5455 * as a metadata update */
5458 /* determine if we are creating a volume or adding a disk */
5459 if (current_vol
< 0) {
5460 /* in the mgmt (add/remove) disk case we are running
5461 * in mdmon context, so don't close fd's
5463 return mgmt_disk(st
);
5465 rv
= create_array(st
, current_vol
);
5470 for (d
= super
->disks
; d
; d
= d
->next
)
5471 Kill(d
->devname
, NULL
, 0, -1, 1);
5472 return write_super_imsm(st
, 1);
5477 static int store_super_imsm(struct supertype
*st
, int fd
)
5479 struct intel_super
*super
= st
->sb
;
5480 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5486 return store_imsm_mpb(fd
, mpb
);
5492 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5494 return __le32_to_cpu(mpb
->bbm_log_size
);
5498 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5499 int layout
, int raiddisks
, int chunk
,
5500 unsigned long long size
,
5501 unsigned long long data_offset
,
5503 unsigned long long *freesize
,
5507 unsigned long long ldsize
;
5508 struct intel_super
*super
;
5511 if (level
!= LEVEL_CONTAINER
)
5516 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5519 pr_err("imsm: Cannot open %s: %s\n",
5520 dev
, strerror(errno
));
5523 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5528 /* capabilities retrieve could be possible
5529 * note that there is no fd for the disks in array.
5531 super
= alloc_super();
5532 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
5536 fd2devname(fd
, str
);
5537 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5538 fd
, str
, super
->orom
, rv
, raiddisks
);
5540 /* no orom/efi or non-intel hba of the disk */
5547 if (raiddisks
> super
->orom
->tds
) {
5549 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
5550 raiddisks
, super
->orom
->tds
);
5554 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5555 (ldsize
>> 9) >> 32 > 0) {
5557 pr_err("%s exceeds maximum platform supported size\n", dev
);
5563 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
5569 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5571 const unsigned long long base_start
= e
[*idx
].start
;
5572 unsigned long long end
= base_start
+ e
[*idx
].size
;
5575 if (base_start
== end
)
5579 for (i
= *idx
; i
< num_extents
; i
++) {
5580 /* extend overlapping extents */
5581 if (e
[i
].start
>= base_start
&&
5582 e
[i
].start
<= end
) {
5585 if (e
[i
].start
+ e
[i
].size
> end
)
5586 end
= e
[i
].start
+ e
[i
].size
;
5587 } else if (e
[i
].start
> end
) {
5593 return end
- base_start
;
5596 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5598 /* build a composite disk with all known extents and generate a new
5599 * 'maxsize' given the "all disks in an array must share a common start
5600 * offset" constraint
5602 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
5606 unsigned long long pos
;
5607 unsigned long long start
= 0;
5608 unsigned long long maxsize
;
5609 unsigned long reserve
;
5611 /* coalesce and sort all extents. also, check to see if we need to
5612 * reserve space between member arrays
5615 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5618 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5621 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5626 while (i
< sum_extents
) {
5627 e
[j
].start
= e
[i
].start
;
5628 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5630 if (e
[j
-1].size
== 0)
5639 unsigned long long esize
;
5641 esize
= e
[i
].start
- pos
;
5642 if (esize
>= maxsize
) {
5647 pos
= e
[i
].start
+ e
[i
].size
;
5649 } while (e
[i
-1].size
);
5655 /* FIXME assumes volume at offset 0 is the first volume in a
5658 if (start_extent
> 0)
5659 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5663 if (maxsize
< reserve
)
5666 super
->create_offset
= ~((unsigned long long) 0);
5667 if (start
+ reserve
> super
->create_offset
)
5668 return 0; /* start overflows create_offset */
5669 super
->create_offset
= start
+ reserve
;
5671 return maxsize
- reserve
;
5674 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5676 if (level
< 0 || level
== 6 || level
== 4)
5679 /* if we have an orom prevent invalid raid levels */
5682 case 0: return imsm_orom_has_raid0(orom
);
5685 return imsm_orom_has_raid1e(orom
);
5686 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5687 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5688 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5691 return 1; /* not on an Intel RAID platform so anything goes */
5697 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5698 int dpa
, int verbose
)
5700 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5701 struct mdstat_ent
*memb
;
5707 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5708 if (memb
->metadata_version
&&
5709 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5710 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5711 !is_subarray(memb
->metadata_version
+9) &&
5713 struct dev_member
*dev
= memb
->members
;
5715 while(dev
&& (fd
< 0)) {
5716 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5717 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5719 fd
= open(path
, O_RDONLY
, 0);
5720 if (num
<= 0 || fd
< 0) {
5721 pr_vrb("Cannot open %s: %s\n",
5722 dev
->name
, strerror(errno
));
5728 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
5729 struct mdstat_ent
*vol
;
5730 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5731 if (vol
->active
> 0 &&
5732 vol
->metadata_version
&&
5733 is_container_member(vol
, memb
->devnm
)) {
5738 if (*devlist
&& (found
< dpa
)) {
5739 dv
= xcalloc(1, sizeof(*dv
));
5740 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
5741 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
5744 dv
->next
= *devlist
;
5752 free_mdstat(mdstat
);
5757 static struct md_list
*
5758 get_loop_devices(void)
5761 struct md_list
*devlist
= NULL
;
5764 for(i
= 0; i
< 12; i
++) {
5765 dv
= xcalloc(1, sizeof(*dv
));
5766 dv
->devname
= xmalloc(40);
5767 sprintf(dv
->devname
, "/dev/loop%d", i
);
5775 static struct md_list
*
5776 get_devices(const char *hba_path
)
5778 struct md_list
*devlist
= NULL
;
5785 devlist
= get_loop_devices();
5788 /* scroll through /sys/dev/block looking for devices attached to
5791 dir
= opendir("/sys/dev/block");
5792 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5797 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5799 path
= devt_to_devpath(makedev(major
, minor
));
5802 if (!path_attached_to_hba(path
, hba_path
)) {
5809 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5811 fd2devname(fd
, buf
);
5814 pr_err("cannot open device: %s\n",
5819 dv
= xcalloc(1, sizeof(*dv
));
5820 dv
->devname
= xstrdup(buf
);
5827 devlist
= devlist
->next
;
5837 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5838 int verbose
, int *found
)
5840 struct md_list
*tmpdev
;
5842 struct supertype
*st
;
5844 /* first walk the list of devices to find a consistent set
5845 * that match the criterea, if that is possible.
5846 * We flag the ones we like with 'used'.
5849 st
= match_metadata_desc_imsm("imsm");
5851 pr_vrb("cannot allocate memory for imsm supertype\n");
5855 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5856 char *devname
= tmpdev
->devname
;
5858 struct supertype
*tst
;
5860 if (tmpdev
->used
> 1)
5862 tst
= dup_super(st
);
5864 pr_vrb("cannot allocate memory for imsm supertype\n");
5867 tmpdev
->container
= 0;
5868 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5870 dprintf("cannot open device %s: %s\n",
5871 devname
, strerror(errno
));
5873 } else if (fstat(dfd
, &stb
)< 0) {
5875 dprintf("fstat failed for %s: %s\n",
5876 devname
, strerror(errno
));
5878 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5879 dprintf("%s is not a block device.\n",
5882 } else if (must_be_container(dfd
)) {
5883 struct supertype
*cst
;
5884 cst
= super_by_fd(dfd
, NULL
);
5886 dprintf("cannot recognize container type %s\n",
5889 } else if (tst
->ss
!= st
->ss
) {
5890 dprintf("non-imsm container - ignore it: %s\n",
5893 } else if (!tst
->ss
->load_container
||
5894 tst
->ss
->load_container(tst
, dfd
, NULL
))
5897 tmpdev
->container
= 1;
5900 cst
->ss
->free_super(cst
);
5902 tmpdev
->st_rdev
= stb
.st_rdev
;
5903 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5904 dprintf("no RAID superblock on %s\n",
5907 } else if (tst
->ss
->compare_super
== NULL
) {
5908 dprintf("Cannot assemble %s metadata on %s\n",
5909 tst
->ss
->name
, devname
);
5915 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5916 /* Ignore unrecognised devices during auto-assembly */
5921 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5923 if (st
->minor_version
== -1)
5924 st
->minor_version
= tst
->minor_version
;
5926 if (memcmp(info
.uuid
, uuid_zero
,
5927 sizeof(int[4])) == 0) {
5928 /* this is a floating spare. It cannot define
5929 * an array unless there are no more arrays of
5930 * this type to be found. It can be included
5931 * in an array of this type though.
5937 if (st
->ss
!= tst
->ss
||
5938 st
->minor_version
!= tst
->minor_version
||
5939 st
->ss
->compare_super(st
, tst
) != 0) {
5940 /* Some mismatch. If exactly one array matches this host,
5941 * we can resolve on that one.
5942 * Or, if we are auto assembling, we just ignore the second
5945 dprintf("superblock on %s doesn't match others - assembly aborted\n",
5951 dprintf("found: devname: %s\n", devname
);
5955 tst
->ss
->free_super(tst
);
5959 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5960 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5961 for (iter
= head
; iter
; iter
= iter
->next
) {
5962 dprintf("content->text_version: %s vol\n",
5963 iter
->text_version
);
5964 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5965 /* do not assemble arrays with unsupported
5967 dprintf("Cannot activate member %s.\n",
5968 iter
->text_version
);
5975 dprintf("No valid super block on device list: err: %d %p\n",
5979 dprintf("no more devices to examine\n");
5982 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5983 if (tmpdev
->used
== 1 && tmpdev
->found
) {
5985 if (count
< tmpdev
->found
)
5988 count
-= tmpdev
->found
;
5991 if (tmpdev
->used
== 1)
5996 st
->ss
->free_super(st
);
6001 count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
6003 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6005 const struct orom_entry
*entry
;
6006 struct devid_list
*dv
, *devid_list
;
6008 if (!hba
|| !hba
->path
)
6011 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
6012 if (strstr(idev
->path
, hba
->path
))
6016 if (!idev
|| !idev
->dev_id
)
6019 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
6021 if (!entry
|| !entry
->devid_list
)
6024 devid_list
= entry
->devid_list
;
6025 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
6026 struct md_list
*devlist
;
6027 struct sys_dev
*device
= device_by_id(dv
->devid
);
6032 hba_path
= device
->path
;
6036 devlist
= get_devices(hba_path
);
6037 /* if no intel devices return zero volumes */
6038 if (devlist
== NULL
)
6041 count
+= active_arrays_by_format("imsm", hba_path
, &devlist
, dpa
, verbose
);
6042 dprintf("path: %s active arrays: %d\n", hba_path
, count
);
6043 if (devlist
== NULL
)
6047 count
+= count_volumes_list(devlist
,
6051 dprintf("found %d count: %d\n", found
, count
);
6054 dprintf("path: %s total number of volumes: %d\n", hba_path
, count
);
6057 struct md_list
*dv
= devlist
;
6058 devlist
= devlist
->next
;
6066 static int imsm_default_chunk(const struct imsm_orom
*orom
)
6068 /* up to 512 if the plaform supports it, otherwise the platform max.
6069 * 128 if no platform detected
6071 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
6073 return min(512, (1 << fs
));
6077 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
6078 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
6080 /* check/set platform and metadata limits/defaults */
6081 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
6082 pr_vrb("platform supports a maximum of %d disks per array\n",
6087 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6088 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
6089 pr_vrb("platform does not support raid%d with %d disk%s\n",
6090 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
6094 if (*chunk
== 0 || *chunk
== UnSet
)
6095 *chunk
= imsm_default_chunk(super
->orom
);
6097 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
6098 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
6102 if (layout
!= imsm_level_to_layout(level
)) {
6104 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
6105 else if (level
== 10)
6106 pr_vrb("imsm raid 10 only supports the n2 layout\n");
6108 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
6113 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
6114 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
6115 pr_vrb("platform does not support a volume size over 2TB\n");
6122 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
6123 * FIX ME add ahci details
6125 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
6126 int layout
, int raiddisks
, int *chunk
,
6127 unsigned long long size
,
6128 unsigned long long data_offset
,
6130 unsigned long long *freesize
,
6134 struct intel_super
*super
= st
->sb
;
6135 struct imsm_super
*mpb
;
6137 unsigned long long pos
= 0;
6138 unsigned long long maxsize
;
6142 /* We must have the container info already read in. */
6146 mpb
= super
->anchor
;
6148 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
6149 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6153 /* General test: make sure there is space for
6154 * 'raiddisks' device extents of size 'size' at a given
6157 unsigned long long minsize
= size
;
6158 unsigned long long start_offset
= MaxSector
;
6161 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
6162 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6167 e
= get_extents(super
, dl
);
6170 unsigned long long esize
;
6171 esize
= e
[i
].start
- pos
;
6172 if (esize
>= minsize
)
6174 if (found
&& start_offset
== MaxSector
) {
6177 } else if (found
&& pos
!= start_offset
) {
6181 pos
= e
[i
].start
+ e
[i
].size
;
6183 } while (e
[i
-1].size
);
6188 if (dcnt
< raiddisks
) {
6190 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6197 /* This device must be a member of the set */
6198 if (stat(dev
, &stb
) < 0)
6200 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6202 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6203 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6204 dl
->minor
== (int)minor(stb
.st_rdev
))
6209 pr_err("%s is not in the same imsm set\n", dev
);
6211 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6212 /* If a volume is present then the current creation attempt
6213 * cannot incorporate new spares because the orom may not
6214 * understand this configuration (all member disks must be
6215 * members of each array in the container).
6217 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6218 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6220 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6221 mpb
->num_disks
!= raiddisks
) {
6222 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6226 /* retrieve the largest free space block */
6227 e
= get_extents(super
, dl
);
6232 unsigned long long esize
;
6234 esize
= e
[i
].start
- pos
;
6235 if (esize
>= maxsize
)
6237 pos
= e
[i
].start
+ e
[i
].size
;
6239 } while (e
[i
-1].size
);
6244 pr_err("unable to determine free space for: %s\n",
6248 if (maxsize
< size
) {
6250 pr_err("%s not enough space (%llu < %llu)\n",
6251 dev
, maxsize
, size
);
6255 /* count total number of extents for merge */
6257 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6259 i
+= dl
->extent_cnt
;
6261 maxsize
= merge_extents(super
, i
);
6263 if (!check_env("IMSM_NO_PLATFORM") &&
6264 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6265 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6269 if (maxsize
< size
|| maxsize
== 0) {
6272 pr_err("no free space left on device. Aborting...\n");
6274 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
6280 *freesize
= maxsize
;
6283 int count
= count_volumes(super
->hba
,
6284 super
->orom
->dpa
, verbose
);
6285 if (super
->orom
->vphba
<= count
) {
6286 pr_vrb("platform does not support more than %d raid volumes.\n",
6287 super
->orom
->vphba
);
6294 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6295 unsigned long long size
, int chunk
,
6296 unsigned long long *freesize
)
6298 struct intel_super
*super
= st
->sb
;
6299 struct imsm_super
*mpb
= super
->anchor
;
6304 unsigned long long maxsize
;
6305 unsigned long long minsize
;
6309 /* find the largest common start free region of the possible disks */
6313 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6319 /* don't activate new spares if we are orom constrained
6320 * and there is already a volume active in the container
6322 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6325 e
= get_extents(super
, dl
);
6328 for (i
= 1; e
[i
-1].size
; i
++)
6336 maxsize
= merge_extents(super
, extent_cnt
);
6340 minsize
= chunk
* 2;
6342 if (cnt
< raiddisks
||
6343 (super
->orom
&& used
&& used
!= raiddisks
) ||
6344 maxsize
< minsize
||
6346 pr_err("not enough devices with space to create array.\n");
6347 return 0; /* No enough free spaces large enough */
6358 if (!check_env("IMSM_NO_PLATFORM") &&
6359 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6360 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6364 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6366 dl
->raiddisk
= cnt
++;
6370 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6375 static int reserve_space(struct supertype
*st
, int raiddisks
,
6376 unsigned long long size
, int chunk
,
6377 unsigned long long *freesize
)
6379 struct intel_super
*super
= st
->sb
;
6384 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6387 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6389 dl
->raiddisk
= cnt
++;
6396 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6397 int raiddisks
, int *chunk
, unsigned long long size
,
6398 unsigned long long data_offset
,
6399 char *dev
, unsigned long long *freesize
,
6407 * if given unused devices create a container
6408 * if given given devices in a container create a member volume
6410 if (level
== LEVEL_CONTAINER
) {
6411 /* Must be a fresh device to add to a container */
6412 return validate_geometry_imsm_container(st
, level
, layout
,
6422 struct intel_super
*super
= st
->sb
;
6423 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6424 raiddisks
, chunk
, size
,
6427 /* we are being asked to automatically layout a
6428 * new volume based on the current contents of
6429 * the container. If the the parameters can be
6430 * satisfied reserve_space will record the disks,
6431 * start offset, and size of the volume to be
6432 * created. add_to_super and getinfo_super
6433 * detect when autolayout is in progress.
6435 /* assuming that freesize is always given when array is
6437 if (super
->orom
&& freesize
) {
6439 count
= count_volumes(super
->hba
,
6440 super
->orom
->dpa
, verbose
);
6441 if (super
->orom
->vphba
<= count
) {
6442 pr_vrb("platform does not support more than %d raid volumes.\n",
6443 super
->orom
->vphba
);
6448 return reserve_space(st
, raiddisks
, size
,
6454 /* creating in a given container */
6455 return validate_geometry_imsm_volume(st
, level
, layout
,
6456 raiddisks
, chunk
, size
,
6458 dev
, freesize
, verbose
);
6461 /* This device needs to be a device in an 'imsm' container */
6462 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6465 pr_err("Cannot create this array on device %s\n",
6470 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6472 pr_err("Cannot open %s: %s\n",
6473 dev
, strerror(errno
));
6476 /* Well, it is in use by someone, maybe an 'imsm' container. */
6477 cfd
= open_container(fd
);
6481 pr_err("Cannot use %s: It is busy\n",
6485 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
6486 if (sra
&& sra
->array
.major_version
== -1 &&
6487 strcmp(sra
->text_version
, "imsm") == 0)
6491 /* This is a member of a imsm container. Load the container
6492 * and try to create a volume
6494 struct intel_super
*super
;
6496 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6498 strcpy(st
->container_devnm
, fd2devnm(cfd
));
6500 return validate_geometry_imsm_volume(st
, level
, layout
,
6502 size
, data_offset
, dev
,
6509 pr_err("failed container membership check\n");
6515 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6517 struct intel_super
*super
= st
->sb
;
6519 if (level
&& *level
== UnSet
)
6520 *level
= LEVEL_CONTAINER
;
6522 if (level
&& layout
&& *layout
== UnSet
)
6523 *layout
= imsm_level_to_layout(*level
);
6525 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6526 *chunk
= imsm_default_chunk(super
->orom
);
6529 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6531 static int kill_subarray_imsm(struct supertype
*st
)
6533 /* remove the subarray currently referenced by ->current_vol */
6535 struct intel_dev
**dp
;
6536 struct intel_super
*super
= st
->sb
;
6537 __u8 current_vol
= super
->current_vol
;
6538 struct imsm_super
*mpb
= super
->anchor
;
6540 if (super
->current_vol
< 0)
6542 super
->current_vol
= -1; /* invalidate subarray cursor */
6544 /* block deletions that would change the uuid of active subarrays
6546 * FIXME when immutable ids are available, but note that we'll
6547 * also need to fixup the invalidated/active subarray indexes in
6550 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6553 if (i
< current_vol
)
6555 sprintf(subarray
, "%u", i
);
6556 if (is_subarray_active(subarray
, st
->devnm
)) {
6557 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6564 if (st
->update_tail
) {
6565 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
6567 u
->type
= update_kill_array
;
6568 u
->dev_idx
= current_vol
;
6569 append_metadata_update(st
, u
, sizeof(*u
));
6574 for (dp
= &super
->devlist
; *dp
;)
6575 if ((*dp
)->index
== current_vol
) {
6578 handle_missing(super
, (*dp
)->dev
);
6579 if ((*dp
)->index
> current_vol
)
6584 /* no more raid devices, all active components are now spares,
6585 * but of course failed are still failed
6587 if (--mpb
->num_raid_devs
== 0) {
6590 for (d
= super
->disks
; d
; d
= d
->next
)
6595 super
->updates_pending
++;
6600 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6601 char *update
, struct mddev_ident
*ident
)
6603 /* update the subarray currently referenced by ->current_vol */
6604 struct intel_super
*super
= st
->sb
;
6605 struct imsm_super
*mpb
= super
->anchor
;
6607 if (strcmp(update
, "name") == 0) {
6608 char *name
= ident
->name
;
6612 if (is_subarray_active(subarray
, st
->devnm
)) {
6613 pr_err("Unable to update name of active subarray\n");
6617 if (!check_name(super
, name
, 0))
6620 vol
= strtoul(subarray
, &ep
, 10);
6621 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6624 if (st
->update_tail
) {
6625 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
6627 u
->type
= update_rename_array
;
6629 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6630 append_metadata_update(st
, u
, sizeof(*u
));
6632 struct imsm_dev
*dev
;
6635 dev
= get_imsm_dev(super
, vol
);
6636 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6637 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6638 dev
= get_imsm_dev(super
, i
);
6639 handle_missing(super
, dev
);
6641 super
->updates_pending
++;
6648 #endif /* MDASSEMBLE */
6650 static int is_gen_migration(struct imsm_dev
*dev
)
6655 if (!dev
->vol
.migr_state
)
6658 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6664 static int is_rebuilding(struct imsm_dev
*dev
)
6666 struct imsm_map
*migr_map
;
6668 if (!dev
->vol
.migr_state
)
6671 if (migr_type(dev
) != MIGR_REBUILD
)
6674 migr_map
= get_imsm_map(dev
, MAP_1
);
6676 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6683 static int is_initializing(struct imsm_dev
*dev
)
6685 struct imsm_map
*migr_map
;
6687 if (!dev
->vol
.migr_state
)
6690 if (migr_type(dev
) != MIGR_INIT
)
6693 migr_map
= get_imsm_map(dev
, MAP_1
);
6695 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6702 static void update_recovery_start(struct intel_super
*super
,
6703 struct imsm_dev
*dev
,
6704 struct mdinfo
*array
)
6706 struct mdinfo
*rebuild
= NULL
;
6710 if (!is_rebuilding(dev
))
6713 /* Find the rebuild target, but punt on the dual rebuild case */
6714 for (d
= array
->devs
; d
; d
= d
->next
)
6715 if (d
->recovery_start
== 0) {
6722 /* (?) none of the disks are marked with
6723 * IMSM_ORD_REBUILD, so assume they are missing and the
6724 * disk_ord_tbl was not correctly updated
6726 dprintf("failed to locate out-of-sync disk\n");
6730 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6731 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6735 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6738 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6740 /* Given a container loaded by load_super_imsm_all,
6741 * extract information about all the arrays into
6743 * If 'subarray' is given, just extract info about that array.
6745 * For each imsm_dev create an mdinfo, fill it in,
6746 * then look for matching devices in super->disks
6747 * and create appropriate device mdinfo.
6749 struct intel_super
*super
= st
->sb
;
6750 struct imsm_super
*mpb
= super
->anchor
;
6751 struct mdinfo
*rest
= NULL
;
6755 int spare_disks
= 0;
6757 /* do not assemble arrays when not all attributes are supported */
6758 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6760 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
6763 /* check for bad blocks */
6764 if (imsm_bbm_log_size(super
->anchor
)) {
6765 pr_err("BBM log found in IMSM metadata.Arrays activation is blocked.\n");
6769 /* count spare devices, not used in maps
6771 for (d
= super
->disks
; d
; d
= d
->next
)
6775 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6776 struct imsm_dev
*dev
;
6777 struct imsm_map
*map
;
6778 struct imsm_map
*map2
;
6779 struct mdinfo
*this;
6787 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6790 dev
= get_imsm_dev(super
, i
);
6791 map
= get_imsm_map(dev
, MAP_0
);
6792 map2
= get_imsm_map(dev
, MAP_1
);
6794 /* do not publish arrays that are in the middle of an
6795 * unsupported migration
6797 if (dev
->vol
.migr_state
&&
6798 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6799 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
6803 /* do not publish arrays that are not support by controller's
6807 this = xmalloc(sizeof(*this));
6809 super
->current_vol
= i
;
6810 getinfo_super_imsm_volume(st
, this, NULL
);
6813 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6814 /* mdadm does not support all metadata features- set the bit in all arrays state */
6815 if (!validate_geometry_imsm_orom(super
,
6816 get_imsm_raid_level(map
), /* RAID level */
6817 imsm_level_to_layout(get_imsm_raid_level(map
)),
6818 map
->num_members
, /* raid disks */
6819 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6821 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
6823 this->array
.state
|=
6824 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6825 (1<<MD_SB_BLOCK_VOLUME
);
6829 /* if array has bad blocks, set suitable bit in all arrays state */
6831 this->array
.state
|=
6832 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6833 (1<<MD_SB_BLOCK_VOLUME
);
6835 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6836 unsigned long long recovery_start
;
6837 struct mdinfo
*info_d
;
6844 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6845 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6846 for (d
= super
->disks
; d
; d
= d
->next
)
6847 if (d
->index
== idx
)
6850 recovery_start
= MaxSector
;
6853 if (d
&& is_failed(&d
->disk
))
6855 if (ord
& IMSM_ORD_REBUILD
)
6859 * if we skip some disks the array will be assmebled degraded;
6860 * reset resync start to avoid a dirty-degraded
6861 * situation when performing the intial sync
6863 * FIXME handle dirty degraded
6865 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6866 this->resync_start
= MaxSector
;
6870 info_d
= xcalloc(1, sizeof(*info_d
));
6871 info_d
->next
= this->devs
;
6872 this->devs
= info_d
;
6874 info_d
->disk
.number
= d
->index
;
6875 info_d
->disk
.major
= d
->major
;
6876 info_d
->disk
.minor
= d
->minor
;
6877 info_d
->disk
.raid_disk
= slot
;
6878 info_d
->recovery_start
= recovery_start
;
6880 if (slot
< map2
->num_members
)
6881 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6883 this->array
.spare_disks
++;
6885 if (slot
< map
->num_members
)
6886 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6888 this->array
.spare_disks
++;
6890 if (info_d
->recovery_start
== MaxSector
)
6891 this->array
.working_disks
++;
6893 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6894 info_d
->data_offset
= pba_of_lba0(map
);
6895 info_d
->component_size
= blocks_per_member(map
);
6897 /* now that the disk list is up-to-date fixup recovery_start */
6898 update_recovery_start(super
, dev
, this);
6899 this->array
.spare_disks
+= spare_disks
;
6902 /* check for reshape */
6903 if (this->reshape_active
== 1)
6904 recover_backup_imsm(st
, this);
6912 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6913 int failed
, int look_in_map
)
6915 struct imsm_map
*map
;
6917 map
= get_imsm_map(dev
, look_in_map
);
6920 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6921 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6923 switch (get_imsm_raid_level(map
)) {
6925 return IMSM_T_STATE_FAILED
;
6928 if (failed
< map
->num_members
)
6929 return IMSM_T_STATE_DEGRADED
;
6931 return IMSM_T_STATE_FAILED
;
6936 * check to see if any mirrors have failed, otherwise we
6937 * are degraded. Even numbered slots are mirrored on
6941 /* gcc -Os complains that this is unused */
6942 int insync
= insync
;
6944 for (i
= 0; i
< map
->num_members
; i
++) {
6945 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6946 int idx
= ord_to_idx(ord
);
6947 struct imsm_disk
*disk
;
6949 /* reset the potential in-sync count on even-numbered
6950 * slots. num_copies is always 2 for imsm raid10
6955 disk
= get_imsm_disk(super
, idx
);
6956 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6959 /* no in-sync disks left in this mirror the
6963 return IMSM_T_STATE_FAILED
;
6966 return IMSM_T_STATE_DEGRADED
;
6970 return IMSM_T_STATE_DEGRADED
;
6972 return IMSM_T_STATE_FAILED
;
6978 return map
->map_state
;
6981 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6986 struct imsm_disk
*disk
;
6987 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6988 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6989 struct imsm_map
*map_for_loop
;
6994 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6995 * disks that are being rebuilt. New failures are recorded to
6996 * map[0]. So we look through all the disks we started with and
6997 * see if any failures are still present, or if any new ones
7001 if (prev
&& (map
->num_members
< prev
->num_members
))
7002 map_for_loop
= prev
;
7004 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
7006 /* when MAP_X is passed both maps failures are counted
7009 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
7010 i
< prev
->num_members
) {
7011 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
7012 idx_1
= ord_to_idx(ord
);
7014 disk
= get_imsm_disk(super
, idx_1
);
7015 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7018 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
7019 i
< map
->num_members
) {
7020 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
7021 idx
= ord_to_idx(ord
);
7024 disk
= get_imsm_disk(super
, idx
);
7025 if (!disk
|| is_failed(disk
) ||
7026 ord
& IMSM_ORD_REBUILD
)
7036 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
7039 struct intel_super
*super
= c
->sb
;
7040 struct imsm_super
*mpb
= super
->anchor
;
7042 if (atoi(inst
) >= mpb
->num_raid_devs
) {
7043 pr_err("subarry index %d, out of range\n", atoi(inst
));
7047 dprintf("imsm: open_new %s\n", inst
);
7048 a
->info
.container_member
= atoi(inst
);
7052 static int is_resyncing(struct imsm_dev
*dev
)
7054 struct imsm_map
*migr_map
;
7056 if (!dev
->vol
.migr_state
)
7059 if (migr_type(dev
) == MIGR_INIT
||
7060 migr_type(dev
) == MIGR_REPAIR
)
7063 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7066 migr_map
= get_imsm_map(dev
, MAP_1
);
7068 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
7069 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
7075 /* return true if we recorded new information */
7076 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7080 struct imsm_map
*map
;
7081 char buf
[MAX_RAID_SERIAL_LEN
+3];
7082 unsigned int len
, shift
= 0;
7084 /* new failures are always set in map[0] */
7085 map
= get_imsm_map(dev
, MAP_0
);
7087 slot
= get_imsm_disk_slot(map
, idx
);
7091 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
7092 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
7095 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
7096 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
7098 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
7099 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
7100 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
7102 disk
->status
|= FAILED_DISK
;
7103 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
7104 /* mark failures in second map if second map exists and this disk
7106 * This is valid for migration, initialization and rebuild
7108 if (dev
->vol
.migr_state
) {
7109 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
7110 int slot2
= get_imsm_disk_slot(map2
, idx
);
7112 if (slot2
< map2
->num_members
&& slot2
>= 0)
7113 set_imsm_ord_tbl_ent(map2
, slot2
,
7114 idx
| IMSM_ORD_REBUILD
);
7116 if (map
->failed_disk_num
== 0xff)
7117 map
->failed_disk_num
= slot
;
7121 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7123 mark_failure(dev
, disk
, idx
);
7125 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
7128 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
7129 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
7132 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7136 if (!super
->missing
)
7139 /* When orom adds replacement for missing disk it does
7140 * not remove entry of missing disk, but just updates map with
7141 * new added disk. So it is not enough just to test if there is
7142 * any missing disk, we have to look if there are any failed disks
7143 * in map to stop migration */
7145 dprintf("imsm: mark missing\n");
7146 /* end process for initialization and rebuild only
7148 if (is_gen_migration(dev
) == 0) {
7152 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7153 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7156 end_migration(dev
, super
, map_state
);
7158 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7159 mark_missing(dev
, &dl
->disk
, dl
->index
);
7160 super
->updates_pending
++;
7163 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7166 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7167 unsigned long long array_blocks
;
7168 struct imsm_map
*map
;
7170 if (used_disks
== 0) {
7171 /* when problems occures
7172 * return current array_blocks value
7174 array_blocks
= __le32_to_cpu(dev
->size_high
);
7175 array_blocks
= array_blocks
<< 32;
7176 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7178 return array_blocks
;
7181 /* set array size in metadata
7183 if (new_size
<= 0) {
7184 /* OLCE size change is caused by added disks
7186 map
= get_imsm_map(dev
, MAP_0
);
7187 array_blocks
= blocks_per_member(map
) * used_disks
;
7189 /* Online Volume Size Change
7190 * Using available free space
7192 array_blocks
= new_size
;
7195 /* round array size down to closest MB
7197 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7198 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7199 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7201 return array_blocks
;
7204 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7206 static void imsm_progress_container_reshape(struct intel_super
*super
)
7208 /* if no device has a migr_state, but some device has a
7209 * different number of members than the previous device, start
7210 * changing the number of devices in this device to match
7213 struct imsm_super
*mpb
= super
->anchor
;
7214 int prev_disks
= -1;
7218 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7219 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7220 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7221 struct imsm_map
*map2
;
7222 int prev_num_members
;
7224 if (dev
->vol
.migr_state
)
7227 if (prev_disks
== -1)
7228 prev_disks
= map
->num_members
;
7229 if (prev_disks
== map
->num_members
)
7232 /* OK, this array needs to enter reshape mode.
7233 * i.e it needs a migr_state
7236 copy_map_size
= sizeof_imsm_map(map
);
7237 prev_num_members
= map
->num_members
;
7238 map
->num_members
= prev_disks
;
7239 dev
->vol
.migr_state
= 1;
7240 dev
->vol
.curr_migr_unit
= 0;
7241 set_migr_type(dev
, MIGR_GEN_MIGR
);
7242 for (i
= prev_num_members
;
7243 i
< map
->num_members
; i
++)
7244 set_imsm_ord_tbl_ent(map
, i
, i
);
7245 map2
= get_imsm_map(dev
, MAP_1
);
7246 /* Copy the current map */
7247 memcpy(map2
, map
, copy_map_size
);
7248 map2
->num_members
= prev_num_members
;
7250 imsm_set_array_size(dev
, -1);
7251 super
->clean_migration_record_by_mdmon
= 1;
7252 super
->updates_pending
++;
7256 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7257 * states are handled in imsm_set_disk() with one exception, when a
7258 * resync is stopped due to a new failure this routine will set the
7259 * 'degraded' state for the array.
7261 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7263 int inst
= a
->info
.container_member
;
7264 struct intel_super
*super
= a
->container
->sb
;
7265 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7266 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7267 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7268 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7269 __u32 blocks_per_unit
;
7271 if (dev
->vol
.migr_state
&&
7272 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7273 /* array state change is blocked due to reshape action
7275 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7276 * - finish the reshape (if last_checkpoint is big and action != reshape)
7277 * - update curr_migr_unit
7279 if (a
->curr_action
== reshape
) {
7280 /* still reshaping, maybe update curr_migr_unit */
7281 goto mark_checkpoint
;
7283 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7284 /* for some reason we aborted the reshape.
7286 * disable automatic metadata rollback
7287 * user action is required to recover process
7290 struct imsm_map
*map2
=
7291 get_imsm_map(dev
, MAP_1
);
7292 dev
->vol
.migr_state
= 0;
7293 set_migr_type(dev
, 0);
7294 dev
->vol
.curr_migr_unit
= 0;
7296 sizeof_imsm_map(map2
));
7297 super
->updates_pending
++;
7300 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7301 unsigned long long array_blocks
;
7305 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7306 if (used_disks
> 0) {
7308 blocks_per_member(map
) *
7310 /* round array size down to closest MB
7312 array_blocks
= (array_blocks
7313 >> SECT_PER_MB_SHIFT
)
7314 << SECT_PER_MB_SHIFT
;
7315 a
->info
.custom_array_size
= array_blocks
;
7316 /* encourage manager to update array
7320 a
->check_reshape
= 1;
7322 /* finalize online capacity expansion/reshape */
7323 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7325 mdi
->disk
.raid_disk
,
7328 imsm_progress_container_reshape(super
);
7333 /* before we activate this array handle any missing disks */
7334 if (consistent
== 2)
7335 handle_missing(super
, dev
);
7337 if (consistent
== 2 &&
7338 (!is_resync_complete(&a
->info
) ||
7339 map_state
!= IMSM_T_STATE_NORMAL
||
7340 dev
->vol
.migr_state
))
7343 if (is_resync_complete(&a
->info
)) {
7344 /* complete intialization / resync,
7345 * recovery and interrupted recovery is completed in
7348 if (is_resyncing(dev
)) {
7349 dprintf("imsm: mark resync done\n");
7350 end_migration(dev
, super
, map_state
);
7351 super
->updates_pending
++;
7352 a
->last_checkpoint
= 0;
7354 } else if ((!is_resyncing(dev
) && !failed
) &&
7355 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7356 /* mark the start of the init process if nothing is failed */
7357 dprintf("imsm: mark resync start\n");
7358 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7359 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7361 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7362 super
->updates_pending
++;
7366 /* skip checkpointing for general migration,
7367 * it is controlled in mdadm
7369 if (is_gen_migration(dev
))
7370 goto skip_mark_checkpoint
;
7372 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7373 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7374 if (blocks_per_unit
) {
7378 units
= a
->last_checkpoint
/ blocks_per_unit
;
7381 /* check that we did not overflow 32-bits, and that
7382 * curr_migr_unit needs updating
7384 if (units32
== units
&&
7386 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7387 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7388 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7389 super
->updates_pending
++;
7393 skip_mark_checkpoint
:
7394 /* mark dirty / clean */
7395 if (dev
->vol
.dirty
!= !consistent
) {
7396 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7401 super
->updates_pending
++;
7407 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7409 int inst
= a
->info
.container_member
;
7410 struct intel_super
*super
= a
->container
->sb
;
7411 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7412 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7413 struct imsm_disk
*disk
;
7415 int recovery_not_finished
= 0;
7420 if (n
> map
->num_members
)
7421 pr_err("imsm: set_disk %d out of range 0..%d\n",
7422 n
, map
->num_members
- 1);
7427 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7429 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7430 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7432 /* check for new failures */
7433 if (state
& DS_FAULTY
) {
7434 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7435 super
->updates_pending
++;
7438 /* check if in_sync */
7439 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7440 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7442 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7443 super
->updates_pending
++;
7446 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7447 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7449 /* check if recovery complete, newly degraded, or failed */
7450 dprintf("imsm: Detected transition to state ");
7451 switch (map_state
) {
7452 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7453 dprintf("normal: ");
7454 if (is_rebuilding(dev
)) {
7455 dprintf_cont("while rebuilding");
7456 /* check if recovery is really finished */
7457 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7458 if (mdi
->recovery_start
!= MaxSector
) {
7459 recovery_not_finished
= 1;
7462 if (recovery_not_finished
) {
7464 dprintf("Rebuild has not finished yet, state not changed");
7465 if (a
->last_checkpoint
< mdi
->recovery_start
) {
7466 a
->last_checkpoint
= mdi
->recovery_start
;
7467 super
->updates_pending
++;
7471 end_migration(dev
, super
, map_state
);
7472 map
= get_imsm_map(dev
, MAP_0
);
7473 map
->failed_disk_num
= ~0;
7474 super
->updates_pending
++;
7475 a
->last_checkpoint
= 0;
7478 if (is_gen_migration(dev
)) {
7479 dprintf_cont("while general migration");
7480 if (a
->last_checkpoint
>= a
->info
.component_size
)
7481 end_migration(dev
, super
, map_state
);
7483 map
->map_state
= map_state
;
7484 map
= get_imsm_map(dev
, MAP_0
);
7485 map
->failed_disk_num
= ~0;
7486 super
->updates_pending
++;
7490 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7491 dprintf_cont("degraded: ");
7492 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
7493 dprintf_cont("mark degraded");
7494 map
->map_state
= map_state
;
7495 super
->updates_pending
++;
7496 a
->last_checkpoint
= 0;
7499 if (is_rebuilding(dev
)) {
7500 dprintf_cont("while rebuilding.");
7501 if (map
->map_state
!= map_state
) {
7502 dprintf_cont(" Map state change");
7503 end_migration(dev
, super
, map_state
);
7504 super
->updates_pending
++;
7508 if (is_gen_migration(dev
)) {
7509 dprintf_cont("while general migration");
7510 if (a
->last_checkpoint
>= a
->info
.component_size
)
7511 end_migration(dev
, super
, map_state
);
7513 map
->map_state
= map_state
;
7514 manage_second_map(super
, dev
);
7516 super
->updates_pending
++;
7519 if (is_initializing(dev
)) {
7520 dprintf_cont("while initialization.");
7521 map
->map_state
= map_state
;
7522 super
->updates_pending
++;
7526 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7527 dprintf_cont("failed: ");
7528 if (is_gen_migration(dev
)) {
7529 dprintf_cont("while general migration");
7530 map
->map_state
= map_state
;
7531 super
->updates_pending
++;
7534 if (map
->map_state
!= map_state
) {
7535 dprintf_cont("mark failed");
7536 end_migration(dev
, super
, map_state
);
7537 super
->updates_pending
++;
7538 a
->last_checkpoint
= 0;
7543 dprintf_cont("state %i\n", map_state
);
7548 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7551 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7552 unsigned long long dsize
;
7553 unsigned long long sectors
;
7555 get_dev_size(fd
, NULL
, &dsize
);
7557 if (mpb_size
> 512) {
7558 /* -1 to account for anchor */
7559 sectors
= mpb_sectors(mpb
) - 1;
7561 /* write the extended mpb to the sectors preceeding the anchor */
7562 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7565 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7570 /* first block is stored on second to last sector of the disk */
7571 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7574 if (write(fd
, buf
, 512) != 512)
7580 static void imsm_sync_metadata(struct supertype
*container
)
7582 struct intel_super
*super
= container
->sb
;
7584 dprintf("sync metadata: %d\n", super
->updates_pending
);
7585 if (!super
->updates_pending
)
7588 write_super_imsm(container
, 0);
7590 super
->updates_pending
= 0;
7593 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7595 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7596 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7599 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7603 if (dl
&& is_failed(&dl
->disk
))
7607 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
7612 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7613 struct active_array
*a
, int activate_new
,
7614 struct mdinfo
*additional_test_list
)
7616 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7617 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7618 struct imsm_super
*mpb
= super
->anchor
;
7619 struct imsm_map
*map
;
7620 unsigned long long pos
;
7625 __u32 array_start
= 0;
7626 __u32 array_end
= 0;
7628 struct mdinfo
*test_list
;
7630 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7631 /* If in this array, skip */
7632 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7633 if (d
->state_fd
>= 0 &&
7634 d
->disk
.major
== dl
->major
&&
7635 d
->disk
.minor
== dl
->minor
) {
7636 dprintf("%x:%x already in array\n",
7637 dl
->major
, dl
->minor
);
7642 test_list
= additional_test_list
;
7644 if (test_list
->disk
.major
== dl
->major
&&
7645 test_list
->disk
.minor
== dl
->minor
) {
7646 dprintf("%x:%x already in additional test list\n",
7647 dl
->major
, dl
->minor
);
7650 test_list
= test_list
->next
;
7655 /* skip in use or failed drives */
7656 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7658 dprintf("%x:%x status (failed: %d index: %d)\n",
7659 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7663 /* skip pure spares when we are looking for partially
7664 * assimilated drives
7666 if (dl
->index
== -1 && !activate_new
)
7669 /* Does this unused device have the requisite free space?
7670 * It needs to be able to cover all member volumes
7672 ex
= get_extents(super
, dl
);
7674 dprintf("cannot get extents\n");
7677 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7678 dev
= get_imsm_dev(super
, i
);
7679 map
= get_imsm_map(dev
, MAP_0
);
7681 /* check if this disk is already a member of
7684 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7690 array_start
= pba_of_lba0(map
);
7691 array_end
= array_start
+
7692 blocks_per_member(map
) - 1;
7695 /* check that we can start at pba_of_lba0 with
7696 * blocks_per_member of space
7698 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7702 pos
= ex
[j
].start
+ ex
[j
].size
;
7704 } while (ex
[j
-1].size
);
7711 if (i
< mpb
->num_raid_devs
) {
7712 dprintf("%x:%x does not have %u to %u available\n",
7713 dl
->major
, dl
->minor
, array_start
, array_end
);
7723 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7725 struct imsm_dev
*dev2
;
7726 struct imsm_map
*map
;
7732 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7734 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7735 if (state
== IMSM_T_STATE_FAILED
) {
7736 map
= get_imsm_map(dev2
, MAP_0
);
7739 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7741 * Check if failed disks are deleted from intel
7742 * disk list or are marked to be deleted
7744 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7745 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7747 * Do not rebuild the array if failed disks
7748 * from failed sub-array are not removed from
7752 is_failed(&idisk
->disk
) &&
7753 (idisk
->action
!= DISK_REMOVE
))
7761 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7762 struct metadata_update
**updates
)
7765 * Find a device with unused free space and use it to replace a
7766 * failed/vacant region in an array. We replace failed regions one a
7767 * array at a time. The result is that a new spare disk will be added
7768 * to the first failed array and after the monitor has finished
7769 * propagating failures the remainder will be consumed.
7771 * FIXME add a capability for mdmon to request spares from another
7775 struct intel_super
*super
= a
->container
->sb
;
7776 int inst
= a
->info
.container_member
;
7777 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7778 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7779 int failed
= a
->info
.array
.raid_disks
;
7780 struct mdinfo
*rv
= NULL
;
7783 struct metadata_update
*mu
;
7785 struct imsm_update_activate_spare
*u
;
7790 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7791 if ((d
->curr_state
& DS_FAULTY
) &&
7793 /* wait for Removal to happen */
7795 if (d
->state_fd
>= 0)
7799 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7800 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7802 if (imsm_reshape_blocks_arrays_changes(super
))
7805 /* Cannot activate another spare if rebuild is in progress already
7807 if (is_rebuilding(dev
)) {
7808 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
7812 if (a
->info
.array
.level
== 4)
7813 /* No repair for takeovered array
7814 * imsm doesn't support raid4
7818 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7819 IMSM_T_STATE_DEGRADED
)
7822 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
7823 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
7828 * If there are any failed disks check state of the other volume.
7829 * Block rebuild if the another one is failed until failed disks
7830 * are removed from container.
7833 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
7834 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7835 /* check if states of the other volumes allow for rebuild */
7836 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7838 allowed
= imsm_rebuild_allowed(a
->container
,
7846 /* For each slot, if it is not working, find a spare */
7847 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7848 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7849 if (d
->disk
.raid_disk
== i
)
7851 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7852 if (d
&& (d
->state_fd
>= 0))
7856 * OK, this device needs recovery. Try to re-add the
7857 * previous occupant of this slot, if this fails see if
7858 * we can continue the assimilation of a spare that was
7859 * partially assimilated, finally try to activate a new
7862 dl
= imsm_readd(super
, i
, a
);
7864 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7866 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7870 /* found a usable disk with enough space */
7871 di
= xcalloc(1, sizeof(*di
));
7873 /* dl->index will be -1 in the case we are activating a
7874 * pristine spare. imsm_process_update() will create a
7875 * new index in this case. Once a disk is found to be
7876 * failed in all member arrays it is kicked from the
7879 di
->disk
.number
= dl
->index
;
7881 /* (ab)use di->devs to store a pointer to the device
7884 di
->devs
= (struct mdinfo
*) dl
;
7886 di
->disk
.raid_disk
= i
;
7887 di
->disk
.major
= dl
->major
;
7888 di
->disk
.minor
= dl
->minor
;
7890 di
->recovery_start
= 0;
7891 di
->data_offset
= pba_of_lba0(map
);
7892 di
->component_size
= a
->info
.component_size
;
7893 di
->container_member
= inst
;
7894 super
->random
= random32();
7898 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7899 i
, di
->data_offset
);
7903 /* No spares found */
7905 /* Now 'rv' has a list of devices to return.
7906 * Create a metadata_update record to update the
7907 * disk_ord_tbl for the array
7909 mu
= xmalloc(sizeof(*mu
));
7910 mu
->buf
= xcalloc(num_spares
,
7911 sizeof(struct imsm_update_activate_spare
));
7913 mu
->space_list
= NULL
;
7914 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7915 mu
->next
= *updates
;
7916 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7918 for (di
= rv
; di
; di
= di
->next
) {
7919 u
->type
= update_activate_spare
;
7920 u
->dl
= (struct dl
*) di
->devs
;
7922 u
->slot
= di
->disk
.raid_disk
;
7933 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7935 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7936 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7937 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7938 struct disk_info
*inf
= get_disk_info(u
);
7939 struct imsm_disk
*disk
;
7943 for (i
= 0; i
< map
->num_members
; i
++) {
7944 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7945 for (j
= 0; j
< new_map
->num_members
; j
++)
7946 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7953 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7957 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7958 if (dl
->major
== major
&& dl
->minor
== minor
)
7963 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7969 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7970 if (dl
->major
== major
&& dl
->minor
== minor
) {
7973 prev
->next
= dl
->next
;
7975 super
->disks
= dl
->next
;
7977 __free_imsm_disk(dl
);
7978 dprintf("removed %x:%x\n", major
, minor
);
7986 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7988 static int add_remove_disk_update(struct intel_super
*super
)
7990 int check_degraded
= 0;
7993 /* add/remove some spares to/from the metadata/contrainer */
7994 while (super
->disk_mgmt_list
) {
7995 struct dl
*disk_cfg
;
7997 disk_cfg
= super
->disk_mgmt_list
;
7998 super
->disk_mgmt_list
= disk_cfg
->next
;
7999 disk_cfg
->next
= NULL
;
8001 if (disk_cfg
->action
== DISK_ADD
) {
8002 disk_cfg
->next
= super
->disks
;
8003 super
->disks
= disk_cfg
;
8005 dprintf("added %x:%x\n",
8006 disk_cfg
->major
, disk_cfg
->minor
);
8007 } else if (disk_cfg
->action
== DISK_REMOVE
) {
8008 dprintf("Disk remove action processed: %x.%x\n",
8009 disk_cfg
->major
, disk_cfg
->minor
);
8010 disk
= get_disk_super(super
,
8014 /* store action status */
8015 disk
->action
= DISK_REMOVE
;
8016 /* remove spare disks only */
8017 if (disk
->index
== -1) {
8018 remove_disk_super(super
,
8023 /* release allocate disk structure */
8024 __free_imsm_disk(disk_cfg
);
8027 return check_degraded
;
8030 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
8031 struct intel_super
*super
,
8034 struct intel_dev
*id
;
8035 void **tofree
= NULL
;
8038 dprintf("(enter)\n");
8039 if (u
->subdev
< 0 || u
->subdev
> 1) {
8040 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8043 if (space_list
== NULL
|| *space_list
== NULL
) {
8044 dprintf("imsm: Error: Memory is not allocated\n");
8048 for (id
= super
->devlist
; id
; id
= id
->next
) {
8049 if (id
->index
== (unsigned)u
->subdev
) {
8050 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8051 struct imsm_map
*map
;
8052 struct imsm_dev
*new_dev
=
8053 (struct imsm_dev
*)*space_list
;
8054 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8056 struct dl
*new_disk
;
8058 if (new_dev
== NULL
)
8060 *space_list
= **space_list
;
8061 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
8062 map
= get_imsm_map(new_dev
, MAP_0
);
8064 dprintf("imsm: Error: migration in progress");
8068 to_state
= map
->map_state
;
8069 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
8071 /* this should not happen */
8072 if (u
->new_disks
[0] < 0) {
8073 map
->failed_disk_num
=
8074 map
->num_members
- 1;
8075 to_state
= IMSM_T_STATE_DEGRADED
;
8077 to_state
= IMSM_T_STATE_NORMAL
;
8079 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
8080 if (u
->new_level
> -1)
8081 map
->raid_level
= u
->new_level
;
8082 migr_map
= get_imsm_map(new_dev
, MAP_1
);
8083 if ((u
->new_level
== 5) &&
8084 (migr_map
->raid_level
== 0)) {
8085 int ord
= map
->num_members
- 1;
8086 migr_map
->num_members
--;
8087 if (u
->new_disks
[0] < 0)
8088 ord
|= IMSM_ORD_REBUILD
;
8089 set_imsm_ord_tbl_ent(map
,
8090 map
->num_members
- 1,
8094 tofree
= (void **)dev
;
8096 /* update chunk size
8098 if (u
->new_chunksize
> 0)
8099 map
->blocks_per_strip
=
8100 __cpu_to_le16(u
->new_chunksize
* 2);
8104 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
8105 migr_map
->raid_level
== map
->raid_level
)
8108 if (u
->new_disks
[0] >= 0) {
8111 new_disk
= get_disk_super(super
,
8112 major(u
->new_disks
[0]),
8113 minor(u
->new_disks
[0]));
8114 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8115 major(u
->new_disks
[0]),
8116 minor(u
->new_disks
[0]),
8117 new_disk
, new_disk
->index
);
8118 if (new_disk
== NULL
)
8119 goto error_disk_add
;
8121 new_disk
->index
= map
->num_members
- 1;
8122 /* slot to fill in autolayout
8124 new_disk
->raiddisk
= new_disk
->index
;
8125 new_disk
->disk
.status
|= CONFIGURED_DISK
;
8126 new_disk
->disk
.status
&= ~SPARE_DISK
;
8128 goto error_disk_add
;
8131 *tofree
= *space_list
;
8132 /* calculate new size
8134 imsm_set_array_size(new_dev
, -1);
8141 *space_list
= tofree
;
8145 dprintf("Error: imsm: Cannot find disk.\n");
8149 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8150 struct intel_super
*super
)
8152 struct intel_dev
*id
;
8155 dprintf("(enter)\n");
8156 if (u
->subdev
< 0 || u
->subdev
> 1) {
8157 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8161 for (id
= super
->devlist
; id
; id
= id
->next
) {
8162 if (id
->index
== (unsigned)u
->subdev
) {
8163 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8164 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8165 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8166 unsigned long long blocks_per_member
;
8168 /* calculate new size
8170 blocks_per_member
= u
->new_size
/ used_disks
;
8171 dprintf("(size: %llu, blocks per member: %llu)\n",
8172 u
->new_size
, blocks_per_member
);
8173 set_blocks_per_member(map
, blocks_per_member
);
8174 imsm_set_array_size(dev
, u
->new_size
);
8184 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8185 struct intel_super
*super
,
8186 struct active_array
*active_array
)
8188 struct imsm_super
*mpb
= super
->anchor
;
8189 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8190 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8191 struct imsm_map
*migr_map
;
8192 struct active_array
*a
;
8193 struct imsm_disk
*disk
;
8200 int second_map_created
= 0;
8202 for (; u
; u
= u
->next
) {
8203 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8208 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8213 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
8218 /* count failures (excluding rebuilds and the victim)
8219 * to determine map[0] state
8222 for (i
= 0; i
< map
->num_members
; i
++) {
8225 disk
= get_imsm_disk(super
,
8226 get_imsm_disk_idx(dev
, i
, MAP_X
));
8227 if (!disk
|| is_failed(disk
))
8231 /* adding a pristine spare, assign a new index */
8232 if (dl
->index
< 0) {
8233 dl
->index
= super
->anchor
->num_disks
;
8234 super
->anchor
->num_disks
++;
8237 disk
->status
|= CONFIGURED_DISK
;
8238 disk
->status
&= ~SPARE_DISK
;
8241 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8242 if (!second_map_created
) {
8243 second_map_created
= 1;
8244 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8245 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8247 map
->map_state
= to_state
;
8248 migr_map
= get_imsm_map(dev
, MAP_1
);
8249 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8250 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8251 dl
->index
| IMSM_ORD_REBUILD
);
8253 /* update the family_num to mark a new container
8254 * generation, being careful to record the existing
8255 * family_num in orig_family_num to clean up after
8256 * earlier mdadm versions that neglected to set it.
8258 if (mpb
->orig_family_num
== 0)
8259 mpb
->orig_family_num
= mpb
->family_num
;
8260 mpb
->family_num
+= super
->random
;
8262 /* count arrays using the victim in the metadata */
8264 for (a
= active_array
; a
; a
= a
->next
) {
8265 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8266 map
= get_imsm_map(dev
, MAP_0
);
8268 if (get_imsm_disk_slot(map
, victim
) >= 0)
8272 /* delete the victim if it is no longer being
8278 /* We know that 'manager' isn't touching anything,
8279 * so it is safe to delete
8281 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8282 if ((*dlp
)->index
== victim
)
8285 /* victim may be on the missing list */
8287 for (dlp
= &super
->missing
; *dlp
;
8288 dlp
= &(*dlp
)->next
)
8289 if ((*dlp
)->index
== victim
)
8291 imsm_delete(super
, dlp
, victim
);
8298 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8299 struct intel_super
*super
,
8302 struct dl
*new_disk
;
8303 struct intel_dev
*id
;
8305 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8306 int disk_count
= u
->old_raid_disks
;
8307 void **tofree
= NULL
;
8308 int devices_to_reshape
= 1;
8309 struct imsm_super
*mpb
= super
->anchor
;
8311 unsigned int dev_id
;
8313 dprintf("(enter)\n");
8315 /* enable spares to use in array */
8316 for (i
= 0; i
< delta_disks
; i
++) {
8317 new_disk
= get_disk_super(super
,
8318 major(u
->new_disks
[i
]),
8319 minor(u
->new_disks
[i
]));
8320 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8321 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8322 new_disk
, new_disk
->index
);
8323 if (new_disk
== NULL
||
8324 (new_disk
->index
>= 0 &&
8325 new_disk
->index
< u
->old_raid_disks
))
8326 goto update_reshape_exit
;
8327 new_disk
->index
= disk_count
++;
8328 /* slot to fill in autolayout
8330 new_disk
->raiddisk
= new_disk
->index
;
8331 new_disk
->disk
.status
|=
8333 new_disk
->disk
.status
&= ~SPARE_DISK
;
8336 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8337 mpb
->num_raid_devs
);
8338 /* manage changes in volume
8340 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8341 void **sp
= *space_list
;
8342 struct imsm_dev
*newdev
;
8343 struct imsm_map
*newmap
, *oldmap
;
8345 for (id
= super
->devlist
; id
; id
= id
->next
) {
8346 if (id
->index
== dev_id
)
8355 /* Copy the dev, but not (all of) the map */
8356 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8357 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8358 newmap
= get_imsm_map(newdev
, MAP_0
);
8359 /* Copy the current map */
8360 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8361 /* update one device only
8363 if (devices_to_reshape
) {
8364 dprintf("imsm: modifying subdev: %i\n",
8366 devices_to_reshape
--;
8367 newdev
->vol
.migr_state
= 1;
8368 newdev
->vol
.curr_migr_unit
= 0;
8369 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8370 newmap
->num_members
= u
->new_raid_disks
;
8371 for (i
= 0; i
< delta_disks
; i
++) {
8372 set_imsm_ord_tbl_ent(newmap
,
8373 u
->old_raid_disks
+ i
,
8374 u
->old_raid_disks
+ i
);
8376 /* New map is correct, now need to save old map
8378 newmap
= get_imsm_map(newdev
, MAP_1
);
8379 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8381 imsm_set_array_size(newdev
, -1);
8384 sp
= (void **)id
->dev
;
8389 /* Clear migration record */
8390 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8393 *space_list
= tofree
;
8396 update_reshape_exit
:
8401 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8402 struct intel_super
*super
,
8405 struct imsm_dev
*dev
= NULL
;
8406 struct intel_dev
*dv
;
8407 struct imsm_dev
*dev_new
;
8408 struct imsm_map
*map
;
8412 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8413 if (dv
->index
== (unsigned int)u
->subarray
) {
8421 map
= get_imsm_map(dev
, MAP_0
);
8423 if (u
->direction
== R10_TO_R0
) {
8424 /* Number of failed disks must be half of initial disk number */
8425 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8426 (map
->num_members
/ 2))
8429 /* iterate through devices to mark removed disks as spare */
8430 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8431 if (dm
->disk
.status
& FAILED_DISK
) {
8432 int idx
= dm
->index
;
8433 /* update indexes on the disk list */
8434 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8435 the index values will end up being correct.... NB */
8436 for (du
= super
->disks
; du
; du
= du
->next
)
8437 if (du
->index
> idx
)
8439 /* mark as spare disk */
8444 map
->num_members
= map
->num_members
/ 2;
8445 map
->map_state
= IMSM_T_STATE_NORMAL
;
8446 map
->num_domains
= 1;
8447 map
->raid_level
= 0;
8448 map
->failed_disk_num
= -1;
8451 if (u
->direction
== R0_TO_R10
) {
8453 /* update slots in current disk list */
8454 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8458 /* create new *missing* disks */
8459 for (i
= 0; i
< map
->num_members
; i
++) {
8460 space
= *space_list
;
8463 *space_list
= *space
;
8465 memcpy(du
, super
->disks
, sizeof(*du
));
8469 du
->index
= (i
* 2) + 1;
8470 sprintf((char *)du
->disk
.serial
,
8471 " MISSING_%d", du
->index
);
8472 sprintf((char *)du
->serial
,
8473 "MISSING_%d", du
->index
);
8474 du
->next
= super
->missing
;
8475 super
->missing
= du
;
8477 /* create new dev and map */
8478 space
= *space_list
;
8481 *space_list
= *space
;
8482 dev_new
= (void *)space
;
8483 memcpy(dev_new
, dev
, sizeof(*dev
));
8484 /* update new map */
8485 map
= get_imsm_map(dev_new
, MAP_0
);
8486 map
->num_members
= map
->num_members
* 2;
8487 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8488 map
->num_domains
= 2;
8489 map
->raid_level
= 1;
8490 /* replace dev<->dev_new */
8493 /* update disk order table */
8494 for (du
= super
->disks
; du
; du
= du
->next
)
8496 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8497 for (du
= super
->missing
; du
; du
= du
->next
)
8498 if (du
->index
>= 0) {
8499 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8500 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8506 static void imsm_process_update(struct supertype
*st
,
8507 struct metadata_update
*update
)
8510 * crack open the metadata_update envelope to find the update record
8511 * update can be one of:
8512 * update_reshape_container_disks - all the arrays in the container
8513 * are being reshaped to have more devices. We need to mark
8514 * the arrays for general migration and convert selected spares
8515 * into active devices.
8516 * update_activate_spare - a spare device has replaced a failed
8517 * device in an array, update the disk_ord_tbl. If this disk is
8518 * present in all member arrays then also clear the SPARE_DISK
8520 * update_create_array
8522 * update_rename_array
8523 * update_add_remove_disk
8525 struct intel_super
*super
= st
->sb
;
8526 struct imsm_super
*mpb
;
8527 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8529 /* update requires a larger buf but the allocation failed */
8530 if (super
->next_len
&& !super
->next_buf
) {
8531 super
->next_len
= 0;
8535 if (super
->next_buf
) {
8536 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8538 super
->len
= super
->next_len
;
8539 super
->buf
= super
->next_buf
;
8541 super
->next_len
= 0;
8542 super
->next_buf
= NULL
;
8545 mpb
= super
->anchor
;
8548 case update_general_migration_checkpoint
: {
8549 struct intel_dev
*id
;
8550 struct imsm_update_general_migration_checkpoint
*u
=
8551 (void *)update
->buf
;
8553 dprintf("called for update_general_migration_checkpoint\n");
8555 /* find device under general migration */
8556 for (id
= super
->devlist
; id
; id
= id
->next
) {
8557 if (is_gen_migration(id
->dev
)) {
8558 id
->dev
->vol
.curr_migr_unit
=
8559 __cpu_to_le32(u
->curr_migr_unit
);
8560 super
->updates_pending
++;
8565 case update_takeover
: {
8566 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8567 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8568 imsm_update_version_info(super
);
8569 super
->updates_pending
++;
8574 case update_reshape_container_disks
: {
8575 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8576 if (apply_reshape_container_disks_update(
8577 u
, super
, &update
->space_list
))
8578 super
->updates_pending
++;
8581 case update_reshape_migration
: {
8582 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8583 if (apply_reshape_migration_update(
8584 u
, super
, &update
->space_list
))
8585 super
->updates_pending
++;
8588 case update_size_change
: {
8589 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8590 if (apply_size_change_update(u
, super
))
8591 super
->updates_pending
++;
8594 case update_activate_spare
: {
8595 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8596 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8597 super
->updates_pending
++;
8600 case update_create_array
: {
8601 /* someone wants to create a new array, we need to be aware of
8602 * a few races/collisions:
8603 * 1/ 'Create' called by two separate instances of mdadm
8604 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8605 * devices that have since been assimilated via
8607 * In the event this update can not be carried out mdadm will
8608 * (FIX ME) notice that its update did not take hold.
8610 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8611 struct intel_dev
*dv
;
8612 struct imsm_dev
*dev
;
8613 struct imsm_map
*map
, *new_map
;
8614 unsigned long long start
, end
;
8615 unsigned long long new_start
, new_end
;
8617 struct disk_info
*inf
;
8620 /* handle racing creates: first come first serve */
8621 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8622 dprintf("subarray %d already defined\n", u
->dev_idx
);
8626 /* check update is next in sequence */
8627 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8628 dprintf("can not create array %d expected index %d\n",
8629 u
->dev_idx
, mpb
->num_raid_devs
);
8633 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8634 new_start
= pba_of_lba0(new_map
);
8635 new_end
= new_start
+ blocks_per_member(new_map
);
8636 inf
= get_disk_info(u
);
8638 /* handle activate_spare versus create race:
8639 * check to make sure that overlapping arrays do not include
8642 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8643 dev
= get_imsm_dev(super
, i
);
8644 map
= get_imsm_map(dev
, MAP_0
);
8645 start
= pba_of_lba0(map
);
8646 end
= start
+ blocks_per_member(map
);
8647 if ((new_start
>= start
&& new_start
<= end
) ||
8648 (start
>= new_start
&& start
<= new_end
))
8653 if (disks_overlap(super
, i
, u
)) {
8654 dprintf("arrays overlap\n");
8659 /* check that prepare update was successful */
8660 if (!update
->space
) {
8661 dprintf("prepare update failed\n");
8665 /* check that all disks are still active before committing
8666 * changes. FIXME: could we instead handle this by creating a
8667 * degraded array? That's probably not what the user expects,
8668 * so better to drop this update on the floor.
8670 for (i
= 0; i
< new_map
->num_members
; i
++) {
8671 dl
= serial_to_dl(inf
[i
].serial
, super
);
8673 dprintf("disk disappeared\n");
8678 super
->updates_pending
++;
8680 /* convert spares to members and fixup ord_tbl */
8681 for (i
= 0; i
< new_map
->num_members
; i
++) {
8682 dl
= serial_to_dl(inf
[i
].serial
, super
);
8683 if (dl
->index
== -1) {
8684 dl
->index
= mpb
->num_disks
;
8686 dl
->disk
.status
|= CONFIGURED_DISK
;
8687 dl
->disk
.status
&= ~SPARE_DISK
;
8689 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8694 update
->space
= NULL
;
8695 imsm_copy_dev(dev
, &u
->dev
);
8696 dv
->index
= u
->dev_idx
;
8697 dv
->next
= super
->devlist
;
8698 super
->devlist
= dv
;
8699 mpb
->num_raid_devs
++;
8701 imsm_update_version_info(super
);
8704 /* mdmon knows how to release update->space, but not
8705 * ((struct intel_dev *) update->space)->dev
8707 if (update
->space
) {
8713 case update_kill_array
: {
8714 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8715 int victim
= u
->dev_idx
;
8716 struct active_array
*a
;
8717 struct intel_dev
**dp
;
8718 struct imsm_dev
*dev
;
8720 /* sanity check that we are not affecting the uuid of
8721 * active arrays, or deleting an active array
8723 * FIXME when immutable ids are available, but note that
8724 * we'll also need to fixup the invalidated/active
8725 * subarray indexes in mdstat
8727 for (a
= st
->arrays
; a
; a
= a
->next
)
8728 if (a
->info
.container_member
>= victim
)
8730 /* by definition if mdmon is running at least one array
8731 * is active in the container, so checking
8732 * mpb->num_raid_devs is just extra paranoia
8734 dev
= get_imsm_dev(super
, victim
);
8735 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8736 dprintf("failed to delete subarray-%d\n", victim
);
8740 for (dp
= &super
->devlist
; *dp
;)
8741 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8744 if ((*dp
)->index
> (unsigned)victim
)
8748 mpb
->num_raid_devs
--;
8749 super
->updates_pending
++;
8752 case update_rename_array
: {
8753 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8754 char name
[MAX_RAID_SERIAL_LEN
+1];
8755 int target
= u
->dev_idx
;
8756 struct active_array
*a
;
8757 struct imsm_dev
*dev
;
8759 /* sanity check that we are not affecting the uuid of
8762 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8763 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8764 for (a
= st
->arrays
; a
; a
= a
->next
)
8765 if (a
->info
.container_member
== target
)
8767 dev
= get_imsm_dev(super
, u
->dev_idx
);
8768 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8769 dprintf("failed to rename subarray-%d\n", target
);
8773 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8774 super
->updates_pending
++;
8777 case update_add_remove_disk
: {
8778 /* we may be able to repair some arrays if disks are
8779 * being added, check the status of add_remove_disk
8780 * if discs has been added.
8782 if (add_remove_disk_update(super
)) {
8783 struct active_array
*a
;
8785 super
->updates_pending
++;
8786 for (a
= st
->arrays
; a
; a
= a
->next
)
8787 a
->check_degraded
= 1;
8792 pr_err("error: unsuported process update type:(type: %d)\n", type
);
8796 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8798 static int imsm_prepare_update(struct supertype
*st
,
8799 struct metadata_update
*update
)
8802 * Allocate space to hold new disk entries, raid-device entries or a new
8803 * mpb if necessary. The manager synchronously waits for updates to
8804 * complete in the monitor, so new mpb buffers allocated here can be
8805 * integrated by the monitor thread without worrying about live pointers
8806 * in the manager thread.
8808 enum imsm_update_type type
;
8809 struct intel_super
*super
= st
->sb
;
8810 struct imsm_super
*mpb
= super
->anchor
;
8814 if (update
->len
< (int)sizeof(type
))
8817 type
= *(enum imsm_update_type
*) update
->buf
;
8820 case update_general_migration_checkpoint
:
8821 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
8823 dprintf("called for update_general_migration_checkpoint\n");
8825 case update_takeover
: {
8826 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8827 if (update
->len
< (int)sizeof(*u
))
8829 if (u
->direction
== R0_TO_R10
) {
8830 void **tail
= (void **)&update
->space_list
;
8831 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8832 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8833 int num_members
= map
->num_members
;
8836 /* allocate memory for added disks */
8837 for (i
= 0; i
< num_members
; i
++) {
8838 size
= sizeof(struct dl
);
8839 space
= xmalloc(size
);
8844 /* allocate memory for new device */
8845 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8846 (num_members
* sizeof(__u32
));
8847 space
= xmalloc(size
);
8851 len
= disks_to_mpb_size(num_members
* 2);
8856 case update_reshape_container_disks
: {
8857 /* Every raid device in the container is about to
8858 * gain some more devices, and we will enter a
8860 * So each 'imsm_map' will be bigger, and the imsm_vol
8861 * will now hold 2 of them.
8862 * Thus we need new 'struct imsm_dev' allocations sized
8863 * as sizeof_imsm_dev but with more devices in both maps.
8865 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8866 struct intel_dev
*dl
;
8867 void **space_tail
= (void**)&update
->space_list
;
8869 if (update
->len
< (int)sizeof(*u
))
8872 dprintf("for update_reshape\n");
8874 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8875 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8877 if (u
->new_raid_disks
> u
->old_raid_disks
)
8878 size
+= sizeof(__u32
)*2*
8879 (u
->new_raid_disks
- u
->old_raid_disks
);
8886 len
= disks_to_mpb_size(u
->new_raid_disks
);
8887 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8890 case update_reshape_migration
: {
8891 /* for migration level 0->5 we need to add disks
8892 * so the same as for container operation we will copy
8893 * device to the bigger location.
8894 * in memory prepared device and new disk area are prepared
8895 * for usage in process update
8897 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8898 struct intel_dev
*id
;
8899 void **space_tail
= (void **)&update
->space_list
;
8902 int current_level
= -1;
8904 if (update
->len
< (int)sizeof(*u
))
8907 dprintf("for update_reshape\n");
8909 /* add space for bigger array in update
8911 for (id
= super
->devlist
; id
; id
= id
->next
) {
8912 if (id
->index
== (unsigned)u
->subdev
) {
8913 size
= sizeof_imsm_dev(id
->dev
, 1);
8914 if (u
->new_raid_disks
> u
->old_raid_disks
)
8915 size
+= sizeof(__u32
)*2*
8916 (u
->new_raid_disks
- u
->old_raid_disks
);
8924 if (update
->space_list
== NULL
)
8927 /* add space for disk in update
8929 size
= sizeof(struct dl
);
8935 /* add spare device to update
8937 for (id
= super
->devlist
; id
; id
= id
->next
)
8938 if (id
->index
== (unsigned)u
->subdev
) {
8939 struct imsm_dev
*dev
;
8940 struct imsm_map
*map
;
8942 dev
= get_imsm_dev(super
, u
->subdev
);
8943 map
= get_imsm_map(dev
, MAP_0
);
8944 current_level
= map
->raid_level
;
8947 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
8948 struct mdinfo
*spares
;
8950 spares
= get_spares_for_grow(st
);
8958 makedev(dev
->disk
.major
,
8960 dl
= get_disk_super(super
,
8963 dl
->index
= u
->old_raid_disks
;
8969 len
= disks_to_mpb_size(u
->new_raid_disks
);
8970 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8973 case update_size_change
: {
8974 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
8978 case update_activate_spare
: {
8979 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
8983 case update_create_array
: {
8984 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8985 struct intel_dev
*dv
;
8986 struct imsm_dev
*dev
= &u
->dev
;
8987 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8989 struct disk_info
*inf
;
8993 if (update
->len
< (int)sizeof(*u
))
8996 inf
= get_disk_info(u
);
8997 len
= sizeof_imsm_dev(dev
, 1);
8998 /* allocate a new super->devlist entry */
8999 dv
= xmalloc(sizeof(*dv
));
9000 dv
->dev
= xmalloc(len
);
9003 /* count how many spares will be converted to members */
9004 for (i
= 0; i
< map
->num_members
; i
++) {
9005 dl
= serial_to_dl(inf
[i
].serial
, super
);
9007 /* hmm maybe it failed?, nothing we can do about
9012 if (count_memberships(dl
, super
) == 0)
9015 len
+= activate
* sizeof(struct imsm_disk
);
9018 case update_kill_array
: {
9019 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
9023 case update_rename_array
: {
9024 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
9028 case update_add_remove_disk
:
9029 /* no update->len needed */
9035 /* check if we need a larger metadata buffer */
9036 if (super
->next_buf
)
9037 buf_len
= super
->next_len
;
9039 buf_len
= super
->len
;
9041 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
9042 /* ok we need a larger buf than what is currently allocated
9043 * if this allocation fails process_update will notice that
9044 * ->next_len is set and ->next_buf is NULL
9046 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
9047 if (super
->next_buf
)
9048 free(super
->next_buf
);
9050 super
->next_len
= buf_len
;
9051 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
9052 memset(super
->next_buf
, 0, buf_len
);
9054 super
->next_buf
= NULL
;
9059 /* must be called while manager is quiesced */
9060 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
9062 struct imsm_super
*mpb
= super
->anchor
;
9064 struct imsm_dev
*dev
;
9065 struct imsm_map
*map
;
9066 int i
, j
, num_members
;
9069 dprintf("deleting device[%d] from imsm_super\n", index
);
9071 /* shift all indexes down one */
9072 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
9073 if (iter
->index
> (int)index
)
9075 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
9076 if (iter
->index
> (int)index
)
9079 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9080 dev
= get_imsm_dev(super
, i
);
9081 map
= get_imsm_map(dev
, MAP_0
);
9082 num_members
= map
->num_members
;
9083 for (j
= 0; j
< num_members
; j
++) {
9084 /* update ord entries being careful not to propagate
9085 * ord-flags to the first map
9087 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
9089 if (ord_to_idx(ord
) <= index
)
9092 map
= get_imsm_map(dev
, MAP_0
);
9093 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
9094 map
= get_imsm_map(dev
, MAP_1
);
9096 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
9101 super
->updates_pending
++;
9103 struct dl
*dl
= *dlp
;
9105 *dlp
= (*dlp
)->next
;
9106 __free_imsm_disk(dl
);
9109 #endif /* MDASSEMBLE */
9111 static void close_targets(int *targets
, int new_disks
)
9118 for (i
= 0; i
< new_disks
; i
++) {
9119 if (targets
[i
] >= 0) {
9126 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
9127 struct intel_super
*super
,
9128 struct imsm_dev
*dev
)
9134 struct imsm_map
*map
;
9137 ret_val
= raid_disks
/2;
9138 /* check map if all disks pairs not failed
9141 map
= get_imsm_map(dev
, MAP_0
);
9142 for (i
= 0; i
< ret_val
; i
++) {
9143 int degradation
= 0;
9144 if (get_imsm_disk(super
, i
) == NULL
)
9146 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9148 if (degradation
== 2)
9151 map
= get_imsm_map(dev
, MAP_1
);
9152 /* if there is no second map
9153 * result can be returned
9157 /* check degradation in second map
9159 for (i
= 0; i
< ret_val
; i
++) {
9160 int degradation
= 0;
9161 if (get_imsm_disk(super
, i
) == NULL
)
9163 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9165 if (degradation
== 2)
9179 /*******************************************************************************
9180 * Function: open_backup_targets
9181 * Description: Function opens file descriptors for all devices given in
9184 * info : general array info
9185 * raid_disks : number of disks
9186 * raid_fds : table of device's file descriptors
9187 * super : intel super for raid10 degradation check
9188 * dev : intel device for raid10 degradation check
9192 ******************************************************************************/
9193 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9194 struct intel_super
*super
, struct imsm_dev
*dev
)
9200 for (i
= 0; i
< raid_disks
; i
++)
9203 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9206 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9207 dprintf("disk is faulty!!\n");
9211 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
9214 dn
= map_dev(sd
->disk
.major
,
9216 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9217 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9218 pr_err("cannot open component\n");
9223 /* check if maximum array degradation level is not exceeded
9225 if ((raid_disks
- opened
) >
9226 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
9228 pr_err("Not enough disks can be opened.\n");
9229 close_targets(raid_fds
, raid_disks
);
9235 /*******************************************************************************
9236 * Function: validate_container_imsm
9237 * Description: This routine validates container after assemble,
9238 * eg. if devices in container are under the same controller.
9241 * info : linked list with info about devices used in array
9245 ******************************************************************************/
9246 int validate_container_imsm(struct mdinfo
*info
)
9248 if (check_env("IMSM_NO_PLATFORM"))
9251 struct sys_dev
*idev
;
9252 struct sys_dev
*hba
= NULL
;
9253 struct sys_dev
*intel_devices
= find_intel_devices();
9254 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
9257 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9258 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9267 pr_err("WARNING - Cannot detect HBA for device %s!\n",
9268 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
9272 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
9275 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
9276 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
9278 struct sys_dev
*hba2
= NULL
;
9279 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9280 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9288 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
9289 get_orom_by_device_id(hba2
->dev_id
);
9291 if (hba2
&& hba
->type
!= hba2
->type
) {
9292 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
9293 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
9297 if (orom
!= orom2
) {
9298 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
9299 " This operation is not supported and can lead to data loss.\n");
9304 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
9305 " This operation is not supported and can lead to data loss.\n");
9313 /*******************************************************************************
9314 * Function: init_migr_record_imsm
9315 * Description: Function inits imsm migration record
9317 * super : imsm internal array info
9318 * dev : device under migration
9319 * info : general array info to find the smallest device
9322 ******************************************************************************/
9323 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9324 struct mdinfo
*info
)
9326 struct intel_super
*super
= st
->sb
;
9327 struct migr_record
*migr_rec
= super
->migr_rec
;
9329 unsigned long long dsize
, dev_sectors
;
9330 long long unsigned min_dev_sectors
= -1LLU;
9334 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9335 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9336 unsigned long long num_migr_units
;
9337 unsigned long long array_blocks
;
9339 memset(migr_rec
, 0, sizeof(struct migr_record
));
9340 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9342 /* only ascending reshape supported now */
9343 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9345 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9346 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9347 migr_rec
->dest_depth_per_unit
*=
9348 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9349 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9350 migr_rec
->blocks_per_unit
=
9351 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9352 migr_rec
->dest_depth_per_unit
=
9353 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9354 array_blocks
= info
->component_size
* new_data_disks
;
9356 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9358 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9360 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9362 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9363 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9365 /* Find the smallest dev */
9366 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9367 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9368 fd
= dev_open(nm
, O_RDONLY
);
9371 get_dev_size(fd
, NULL
, &dsize
);
9372 dev_sectors
= dsize
/ 512;
9373 if (dev_sectors
< min_dev_sectors
)
9374 min_dev_sectors
= dev_sectors
;
9377 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9378 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9380 write_imsm_migr_rec(st
);
9385 /*******************************************************************************
9386 * Function: save_backup_imsm
9387 * Description: Function saves critical data stripes to Migration Copy Area
9388 * and updates the current migration unit status.
9389 * Use restore_stripes() to form a destination stripe,
9390 * and to write it to the Copy Area.
9392 * st : supertype information
9393 * dev : imsm device that backup is saved for
9394 * info : general array info
9395 * buf : input buffer
9396 * length : length of data to backup (blocks_per_unit)
9400 ******************************************************************************/
9401 int save_backup_imsm(struct supertype
*st
,
9402 struct imsm_dev
*dev
,
9403 struct mdinfo
*info
,
9408 struct intel_super
*super
= st
->sb
;
9409 unsigned long long *target_offsets
;
9412 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9413 int new_disks
= map_dest
->num_members
;
9414 int dest_layout
= 0;
9416 unsigned long long start
;
9417 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9419 targets
= xmalloc(new_disks
* sizeof(int));
9421 for (i
= 0; i
< new_disks
; i
++)
9424 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
9426 start
= info
->reshape_progress
* 512;
9427 for (i
= 0; i
< new_disks
; i
++) {
9428 target_offsets
[i
] = (unsigned long long)
9429 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9430 /* move back copy area adderss, it will be moved forward
9431 * in restore_stripes() using start input variable
9433 target_offsets
[i
] -= start
/data_disks
;
9436 if (open_backup_targets(info
, new_disks
, targets
,
9440 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9441 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9443 if (restore_stripes(targets
, /* list of dest devices */
9444 target_offsets
, /* migration record offsets */
9447 map_dest
->raid_level
,
9449 -1, /* source backup file descriptor */
9450 0, /* input buf offset
9451 * always 0 buf is already offseted */
9455 pr_err("Error restoring stripes\n");
9463 close_targets(targets
, new_disks
);
9466 free(target_offsets
);
9471 /*******************************************************************************
9472 * Function: save_checkpoint_imsm
9473 * Description: Function called for current unit status update
9474 * in the migration record. It writes it to disk.
9476 * super : imsm internal array info
9477 * info : general array info
9481 * 2: failure, means no valid migration record
9482 * / no general migration in progress /
9483 ******************************************************************************/
9484 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9486 struct intel_super
*super
= st
->sb
;
9487 unsigned long long blocks_per_unit
;
9488 unsigned long long curr_migr_unit
;
9490 if (load_imsm_migr_rec(super
, info
) != 0) {
9491 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
9495 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9496 if (blocks_per_unit
== 0) {
9497 dprintf("imsm: no migration in progress.\n");
9500 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9501 /* check if array is alligned to copy area
9502 * if it is not alligned, add one to current migration unit value
9503 * this can happend on array reshape finish only
9505 if (info
->reshape_progress
% blocks_per_unit
)
9508 super
->migr_rec
->curr_migr_unit
=
9509 __cpu_to_le32(curr_migr_unit
);
9510 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9511 super
->migr_rec
->dest_1st_member_lba
=
9512 __cpu_to_le32(curr_migr_unit
*
9513 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9514 if (write_imsm_migr_rec(st
) < 0) {
9515 dprintf("imsm: Cannot write migration record outside backup area\n");
9522 /*******************************************************************************
9523 * Function: recover_backup_imsm
9524 * Description: Function recovers critical data from the Migration Copy Area
9525 * while assembling an array.
9527 * super : imsm internal array info
9528 * info : general array info
9530 * 0 : success (or there is no data to recover)
9532 ******************************************************************************/
9533 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9535 struct intel_super
*super
= st
->sb
;
9536 struct migr_record
*migr_rec
= super
->migr_rec
;
9537 struct imsm_map
*map_dest
;
9538 struct intel_dev
*id
= NULL
;
9539 unsigned long long read_offset
;
9540 unsigned long long write_offset
;
9542 int *targets
= NULL
;
9543 int new_disks
, i
, err
;
9546 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9547 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9549 int skipped_disks
= 0;
9551 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9555 /* recover data only during assemblation */
9556 if (strncmp(buffer
, "inactive", 8) != 0)
9558 /* no data to recover */
9559 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9561 if (curr_migr_unit
>= num_migr_units
)
9564 /* find device during reshape */
9565 for (id
= super
->devlist
; id
; id
= id
->next
)
9566 if (is_gen_migration(id
->dev
))
9571 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9572 new_disks
= map_dest
->num_members
;
9574 read_offset
= (unsigned long long)
9575 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9577 write_offset
= ((unsigned long long)
9578 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9579 pba_of_lba0(map_dest
)) * 512;
9581 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9582 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9584 targets
= xcalloc(new_disks
, sizeof(int));
9586 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9587 pr_err("Cannot open some devices belonging to array.\n");
9591 for (i
= 0; i
< new_disks
; i
++) {
9592 if (targets
[i
] < 0) {
9596 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9597 pr_err("Cannot seek to block: %s\n",
9602 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9603 pr_err("Cannot read copy area block: %s\n",
9608 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9609 pr_err("Cannot seek to block: %s\n",
9614 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9615 pr_err("Cannot restore block: %s\n",
9622 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9626 pr_err("Cannot restore data from backup. Too many failed disks\n");
9630 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9631 /* ignore error == 2, this can mean end of reshape here
9633 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
9639 for (i
= 0; i
< new_disks
; i
++)
9648 static char disk_by_path
[] = "/dev/disk/by-path/";
9650 static const char *imsm_get_disk_controller_domain(const char *path
)
9652 char disk_path
[PATH_MAX
];
9656 strcpy(disk_path
, disk_by_path
);
9657 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9658 if (stat(disk_path
, &st
) == 0) {
9659 struct sys_dev
* hba
;
9662 path
= devt_to_devpath(st
.st_rdev
);
9665 hba
= find_disk_attached_hba(-1, path
);
9666 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9668 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9672 dprintf("path: %s hba: %s attached: %s\n",
9673 path
, (hba
) ? hba
->path
: "NULL", drv
);
9679 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
9681 static char devnm
[32];
9682 char subdev_name
[20];
9683 struct mdstat_ent
*mdstat
;
9685 sprintf(subdev_name
, "%d", subdev
);
9686 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9690 strcpy(devnm
, mdstat
->devnm
);
9691 free_mdstat(mdstat
);
9695 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9696 struct geo_params
*geo
,
9697 int *old_raid_disks
,
9700 /* currently we only support increasing the number of devices
9701 * for a container. This increases the number of device for each
9702 * member array. They must all be RAID0 or RAID5.
9705 struct mdinfo
*info
, *member
;
9706 int devices_that_can_grow
= 0;
9708 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
9710 if (geo
->size
> 0 ||
9711 geo
->level
!= UnSet
||
9712 geo
->layout
!= UnSet
||
9713 geo
->chunksize
!= 0 ||
9714 geo
->raid_disks
== UnSet
) {
9715 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
9719 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9720 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
9724 info
= container_content_imsm(st
, NULL
);
9725 for (member
= info
; member
; member
= member
->next
) {
9728 dprintf("imsm: checking device_num: %i\n",
9729 member
->container_member
);
9731 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9732 /* we work on container for Online Capacity Expansion
9733 * only so raid_disks has to grow
9735 dprintf("imsm: for container operation raid disks increase is required\n");
9739 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
9740 /* we cannot use this container with other raid level
9742 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
9746 /* check for platform support
9747 * for this raid level configuration
9749 struct intel_super
*super
= st
->sb
;
9750 if (!is_raid_level_supported(super
->orom
,
9751 member
->array
.level
,
9753 dprintf("platform does not support raid%d with %d disk%s\n",
9756 geo
->raid_disks
> 1 ? "s" : "");
9759 /* check if component size is aligned to chunk size
9761 if (info
->component_size
%
9762 (info
->array
.chunk_size
/512)) {
9763 dprintf("Component size is not aligned to chunk size\n");
9768 if (*old_raid_disks
&&
9769 info
->array
.raid_disks
!= *old_raid_disks
)
9771 *old_raid_disks
= info
->array
.raid_disks
;
9773 /* All raid5 and raid0 volumes in container
9774 * have to be ready for Online Capacity Expansion
9775 * so they need to be assembled. We have already
9776 * checked that no recovery etc is happening.
9778 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
9779 st
->container_devnm
);
9780 if (result
== NULL
) {
9781 dprintf("imsm: cannot find array\n");
9784 devices_that_can_grow
++;
9787 if (!member
&& devices_that_can_grow
)
9791 dprintf("Container operation allowed\n");
9793 dprintf("Error: %i\n", ret_val
);
9798 /* Function: get_spares_for_grow
9799 * Description: Allocates memory and creates list of spare devices
9800 * avaliable in container. Checks if spare drive size is acceptable.
9801 * Parameters: Pointer to the supertype structure
9802 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9805 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9807 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9808 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9811 /******************************************************************************
9812 * function: imsm_create_metadata_update_for_reshape
9813 * Function creates update for whole IMSM container.
9815 ******************************************************************************/
9816 static int imsm_create_metadata_update_for_reshape(
9817 struct supertype
*st
,
9818 struct geo_params
*geo
,
9820 struct imsm_update_reshape
**updatep
)
9822 struct intel_super
*super
= st
->sb
;
9823 struct imsm_super
*mpb
= super
->anchor
;
9824 int update_memory_size
;
9825 struct imsm_update_reshape
*u
;
9826 struct mdinfo
*spares
;
9831 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
9833 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9835 /* size of all update data without anchor */
9836 update_memory_size
= sizeof(struct imsm_update_reshape
);
9838 /* now add space for spare disks that we need to add. */
9839 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9841 u
= xcalloc(1, update_memory_size
);
9842 u
->type
= update_reshape_container_disks
;
9843 u
->old_raid_disks
= old_raid_disks
;
9844 u
->new_raid_disks
= geo
->raid_disks
;
9846 /* now get spare disks list
9848 spares
= get_spares_for_grow(st
);
9851 || delta_disks
> spares
->array
.spare_disks
) {
9852 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
9857 /* we have got spares
9858 * update disk list in imsm_disk list table in anchor
9860 dprintf("imsm: %i spares are available.\n\n",
9861 spares
->array
.spare_disks
);
9864 for (i
= 0; i
< delta_disks
; i
++) {
9869 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9871 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9872 dl
->index
= mpb
->num_disks
;
9882 dprintf("imsm: reshape update preparation :");
9883 if (i
== delta_disks
) {
9884 dprintf_cont(" OK\n");
9886 return update_memory_size
;
9889 dprintf_cont(" Error\n");
9894 /******************************************************************************
9895 * function: imsm_create_metadata_update_for_size_change()
9896 * Creates update for IMSM array for array size change.
9898 ******************************************************************************/
9899 static int imsm_create_metadata_update_for_size_change(
9900 struct supertype
*st
,
9901 struct geo_params
*geo
,
9902 struct imsm_update_size_change
**updatep
)
9904 struct intel_super
*super
= st
->sb
;
9905 int update_memory_size
;
9906 struct imsm_update_size_change
*u
;
9908 dprintf("(enter) New size = %llu\n", geo
->size
);
9910 /* size of all update data without anchor */
9911 update_memory_size
= sizeof(struct imsm_update_size_change
);
9913 u
= xcalloc(1, update_memory_size
);
9914 u
->type
= update_size_change
;
9915 u
->subdev
= super
->current_vol
;
9916 u
->new_size
= geo
->size
;
9918 dprintf("imsm: reshape update preparation : OK\n");
9921 return update_memory_size
;
9924 /******************************************************************************
9925 * function: imsm_create_metadata_update_for_migration()
9926 * Creates update for IMSM array.
9928 ******************************************************************************/
9929 static int imsm_create_metadata_update_for_migration(
9930 struct supertype
*st
,
9931 struct geo_params
*geo
,
9932 struct imsm_update_reshape_migration
**updatep
)
9934 struct intel_super
*super
= st
->sb
;
9935 int update_memory_size
;
9936 struct imsm_update_reshape_migration
*u
;
9937 struct imsm_dev
*dev
;
9938 int previous_level
= -1;
9940 dprintf("(enter) New Level = %i\n", geo
->level
);
9942 /* size of all update data without anchor */
9943 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9945 u
= xcalloc(1, update_memory_size
);
9946 u
->type
= update_reshape_migration
;
9947 u
->subdev
= super
->current_vol
;
9948 u
->new_level
= geo
->level
;
9949 u
->new_layout
= geo
->layout
;
9950 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9951 u
->new_disks
[0] = -1;
9952 u
->new_chunksize
= -1;
9954 dev
= get_imsm_dev(super
, u
->subdev
);
9956 struct imsm_map
*map
;
9958 map
= get_imsm_map(dev
, MAP_0
);
9960 int current_chunk_size
=
9961 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9963 if (geo
->chunksize
!= current_chunk_size
) {
9964 u
->new_chunksize
= geo
->chunksize
/ 1024;
9965 dprintf("imsm: chunk size change from %i to %i\n",
9966 current_chunk_size
, u
->new_chunksize
);
9968 previous_level
= map
->raid_level
;
9971 if (geo
->level
== 5 && previous_level
== 0) {
9972 struct mdinfo
*spares
= NULL
;
9974 u
->new_raid_disks
++;
9975 spares
= get_spares_for_grow(st
);
9976 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
9979 update_memory_size
= 0;
9980 dprintf("error: cannot get spare device for requested migration");
9985 dprintf("imsm: reshape update preparation : OK\n");
9988 return update_memory_size
;
9991 static void imsm_update_metadata_locally(struct supertype
*st
,
9994 struct metadata_update mu
;
9999 mu
.space_list
= NULL
;
10001 if (imsm_prepare_update(st
, &mu
))
10002 imsm_process_update(st
, &mu
);
10004 while (mu
.space_list
) {
10005 void **space
= mu
.space_list
;
10006 mu
.space_list
= *space
;
10011 /***************************************************************************
10012 * Function: imsm_analyze_change
10013 * Description: Function analyze change for single volume
10014 * and validate if transition is supported
10015 * Parameters: Geometry parameters, supertype structure,
10016 * metadata change direction (apply/rollback)
10017 * Returns: Operation type code on success, -1 if fail
10018 ****************************************************************************/
10019 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
10020 struct geo_params
*geo
,
10023 struct mdinfo info
;
10025 int check_devs
= 0;
10027 /* number of added/removed disks in operation result */
10028 int devNumChange
= 0;
10029 /* imsm compatible layout value for array geometry verification */
10030 int imsm_layout
= -1;
10032 struct imsm_dev
*dev
;
10033 struct intel_super
*super
;
10034 unsigned long long current_size
;
10035 unsigned long long free_size
;
10036 unsigned long long max_size
;
10039 getinfo_super_imsm_volume(st
, &info
, NULL
);
10040 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
10041 geo
->level
!= UnSet
) {
10042 switch (info
.array
.level
) {
10044 if (geo
->level
== 5) {
10045 change
= CH_MIGRATION
;
10046 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
10047 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
10049 goto analyse_change_exit
;
10051 imsm_layout
= geo
->layout
;
10053 devNumChange
= 1; /* parity disk added */
10054 } else if (geo
->level
== 10) {
10055 change
= CH_TAKEOVER
;
10057 devNumChange
= 2; /* two mirrors added */
10058 imsm_layout
= 0x102; /* imsm supported layout */
10063 if (geo
->level
== 0) {
10064 change
= CH_TAKEOVER
;
10066 devNumChange
= -(geo
->raid_disks
/2);
10067 imsm_layout
= 0; /* imsm raid0 layout */
10071 if (change
== -1) {
10072 pr_err("Error. Level Migration from %d to %d not supported!\n",
10073 info
.array
.level
, geo
->level
);
10074 goto analyse_change_exit
;
10077 geo
->level
= info
.array
.level
;
10079 if (geo
->layout
!= info
.array
.layout
&&
10080 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
10081 change
= CH_MIGRATION
;
10082 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
10083 geo
->layout
== 5) {
10084 /* reshape 5 -> 4 */
10085 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
10086 geo
->layout
== 0) {
10087 /* reshape 4 -> 5 */
10091 pr_err("Error. Layout Migration from %d to %d not supported!\n",
10092 info
.array
.layout
, geo
->layout
);
10094 goto analyse_change_exit
;
10097 geo
->layout
= info
.array
.layout
;
10098 if (imsm_layout
== -1)
10099 imsm_layout
= info
.array
.layout
;
10102 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
10103 geo
->chunksize
!= info
.array
.chunk_size
) {
10104 if (info
.array
.level
== 10) {
10105 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
10107 goto analyse_change_exit
;
10109 change
= CH_MIGRATION
;
10111 geo
->chunksize
= info
.array
.chunk_size
;
10114 chunk
= geo
->chunksize
/ 1024;
10117 dev
= get_imsm_dev(super
, super
->current_vol
);
10118 data_disks
= imsm_num_data_members(dev
, MAP_0
);
10119 /* compute current size per disk member
10121 current_size
= info
.custom_array_size
/ data_disks
;
10123 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
10124 /* align component size
10126 geo
->size
= imsm_component_size_aligment_check(
10127 get_imsm_raid_level(dev
->vol
.map
),
10130 if (geo
->size
== 0) {
10131 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
10133 goto analyse_change_exit
;
10137 if (current_size
!= geo
->size
&& geo
->size
> 0) {
10138 if (change
!= -1) {
10139 pr_err("Error. Size change should be the only one at a time.\n");
10141 goto analyse_change_exit
;
10143 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
10144 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
10145 super
->current_vol
, st
->devnm
);
10146 goto analyse_change_exit
;
10148 /* check the maximum available size
10150 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
10151 0, chunk
, &free_size
);
10153 /* Cannot find maximum available space
10157 max_size
= free_size
+ current_size
;
10158 /* align component size
10160 max_size
= imsm_component_size_aligment_check(
10161 get_imsm_raid_level(dev
->vol
.map
),
10165 if (geo
->size
== MAX_SIZE
) {
10166 /* requested size change to the maximum available size
10168 if (max_size
== 0) {
10169 pr_err("Error. Cannot find maximum available space.\n");
10171 goto analyse_change_exit
;
10173 geo
->size
= max_size
;
10176 if (direction
== ROLLBACK_METADATA_CHANGES
) {
10177 /* accept size for rollback only
10180 /* round size due to metadata compatibility
10182 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
10183 << SECT_PER_MB_SHIFT
;
10184 dprintf("Prepare update for size change to %llu\n",
10186 if (current_size
>= geo
->size
) {
10187 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
10188 current_size
, geo
->size
);
10189 goto analyse_change_exit
;
10191 if (max_size
&& geo
->size
> max_size
) {
10192 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
10193 max_size
, geo
->size
);
10194 goto analyse_change_exit
;
10197 geo
->size
*= data_disks
;
10198 geo
->raid_disks
= dev
->vol
.map
->num_members
;
10199 change
= CH_ARRAY_SIZE
;
10201 if (!validate_geometry_imsm(st
,
10204 geo
->raid_disks
+ devNumChange
,
10206 geo
->size
, INVALID_SECTORS
,
10211 struct intel_super
*super
= st
->sb
;
10212 struct imsm_super
*mpb
= super
->anchor
;
10214 if (mpb
->num_raid_devs
> 1) {
10215 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
10221 analyse_change_exit
:
10222 if (direction
== ROLLBACK_METADATA_CHANGES
&&
10223 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
10224 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
10230 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10232 struct intel_super
*super
= st
->sb
;
10233 struct imsm_update_takeover
*u
;
10235 u
= xmalloc(sizeof(struct imsm_update_takeover
));
10237 u
->type
= update_takeover
;
10238 u
->subarray
= super
->current_vol
;
10240 /* 10->0 transition */
10241 if (geo
->level
== 0)
10242 u
->direction
= R10_TO_R0
;
10244 /* 0->10 transition */
10245 if (geo
->level
== 10)
10246 u
->direction
= R0_TO_R10
;
10248 /* update metadata locally */
10249 imsm_update_metadata_locally(st
, u
,
10250 sizeof(struct imsm_update_takeover
));
10251 /* and possibly remotely */
10252 if (st
->update_tail
)
10253 append_metadata_update(st
, u
,
10254 sizeof(struct imsm_update_takeover
));
10261 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
10263 int layout
, int chunksize
, int raid_disks
,
10264 int delta_disks
, char *backup
, char *dev
,
10265 int direction
, int verbose
)
10268 struct geo_params geo
;
10270 dprintf("(enter)\n");
10272 memset(&geo
, 0, sizeof(struct geo_params
));
10274 geo
.dev_name
= dev
;
10275 strcpy(geo
.devnm
, st
->devnm
);
10278 geo
.layout
= layout
;
10279 geo
.chunksize
= chunksize
;
10280 geo
.raid_disks
= raid_disks
;
10281 if (delta_disks
!= UnSet
)
10282 geo
.raid_disks
+= delta_disks
;
10284 dprintf("for level : %i\n", geo
.level
);
10285 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
10287 if (experimental() == 0)
10290 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
10291 /* On container level we can only increase number of devices. */
10292 dprintf("imsm: info: Container operation\n");
10293 int old_raid_disks
= 0;
10295 if (imsm_reshape_is_allowed_on_container(
10296 st
, &geo
, &old_raid_disks
, direction
)) {
10297 struct imsm_update_reshape
*u
= NULL
;
10300 len
= imsm_create_metadata_update_for_reshape(
10301 st
, &geo
, old_raid_disks
, &u
);
10304 dprintf("imsm: Cannot prepare update\n");
10305 goto exit_imsm_reshape_super
;
10309 /* update metadata locally */
10310 imsm_update_metadata_locally(st
, u
, len
);
10311 /* and possibly remotely */
10312 if (st
->update_tail
)
10313 append_metadata_update(st
, u
, len
);
10318 pr_err("(imsm) Operation is not allowed on this container\n");
10321 /* On volume level we support following operations
10322 * - takeover: raid10 -> raid0; raid0 -> raid10
10323 * - chunk size migration
10324 * - migration: raid5 -> raid0; raid0 -> raid5
10326 struct intel_super
*super
= st
->sb
;
10327 struct intel_dev
*dev
= super
->devlist
;
10329 dprintf("imsm: info: Volume operation\n");
10330 /* find requested device */
10333 imsm_find_array_devnm_by_subdev(
10334 dev
->index
, st
->container_devnm
);
10335 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
10340 pr_err("Cannot find %s (%s) subarray\n",
10341 geo
.dev_name
, geo
.devnm
);
10342 goto exit_imsm_reshape_super
;
10344 super
->current_vol
= dev
->index
;
10345 change
= imsm_analyze_change(st
, &geo
, direction
);
10348 ret_val
= imsm_takeover(st
, &geo
);
10350 case CH_MIGRATION
: {
10351 struct imsm_update_reshape_migration
*u
= NULL
;
10353 imsm_create_metadata_update_for_migration(
10356 dprintf("imsm: Cannot prepare update\n");
10360 /* update metadata locally */
10361 imsm_update_metadata_locally(st
, u
, len
);
10362 /* and possibly remotely */
10363 if (st
->update_tail
)
10364 append_metadata_update(st
, u
, len
);
10369 case CH_ARRAY_SIZE
: {
10370 struct imsm_update_size_change
*u
= NULL
;
10372 imsm_create_metadata_update_for_size_change(
10375 dprintf("imsm: Cannot prepare update\n");
10379 /* update metadata locally */
10380 imsm_update_metadata_locally(st
, u
, len
);
10381 /* and possibly remotely */
10382 if (st
->update_tail
)
10383 append_metadata_update(st
, u
, len
);
10393 exit_imsm_reshape_super
:
10394 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10398 #define COMPLETED_OK 0
10399 #define COMPLETED_NONE 1
10400 #define COMPLETED_DELAYED 2
10402 static int read_completed(int fd
, unsigned long long *val
)
10407 ret
= sysfs_fd_get_str(fd
, buf
, 50);
10411 ret
= COMPLETED_OK
;
10412 if (strncmp(buf
, "none", 4) == 0) {
10413 ret
= COMPLETED_NONE
;
10414 } else if (strncmp(buf
, "delayed", 7) == 0) {
10415 ret
= COMPLETED_DELAYED
;
10418 *val
= strtoull(buf
, &ep
, 0);
10419 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
10425 /*******************************************************************************
10426 * Function: wait_for_reshape_imsm
10427 * Description: Function writes new sync_max value and waits until
10428 * reshape process reach new position
10430 * sra : general array info
10431 * ndata : number of disks in new array's layout
10434 * 1 : there is no reshape in progress,
10436 ******************************************************************************/
10437 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10439 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
10441 unsigned long long completed
;
10442 /* to_complete : new sync_max position */
10443 unsigned long long to_complete
= sra
->reshape_progress
;
10444 unsigned long long position_to_set
= to_complete
/ ndata
;
10447 dprintf("cannot open reshape_position\n");
10452 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10454 dprintf("cannot read reshape_position (no reshape in progres)\n");
10463 if (completed
> position_to_set
) {
10464 dprintf("wrong next position to set %llu (%llu)\n",
10465 to_complete
, position_to_set
);
10469 dprintf("Position set: %llu\n", position_to_set
);
10470 if (sysfs_set_num(sra
, NULL
, "sync_max",
10471 position_to_set
) != 0) {
10472 dprintf("cannot set reshape position to %llu\n",
10481 int timeout
= 3000;
10483 sysfs_wait(fd
, &timeout
);
10484 if (sysfs_get_str(sra
, NULL
, "sync_action",
10486 strncmp(action
, "reshape", 7) != 0) {
10487 if (strncmp(action
, "idle", 4) == 0)
10493 rc
= read_completed(fd
, &completed
);
10495 dprintf("cannot read reshape_position (in loop)\n");
10498 } else if (rc
== COMPLETED_NONE
)
10500 } while (completed
< position_to_set
);
10506 /*******************************************************************************
10507 * Function: check_degradation_change
10508 * Description: Check that array hasn't become failed.
10510 * info : for sysfs access
10511 * sources : source disks descriptors
10512 * degraded: previous degradation level
10514 * degradation level
10515 ******************************************************************************/
10516 int check_degradation_change(struct mdinfo
*info
,
10520 unsigned long long new_degraded
;
10523 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10524 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
10525 /* check each device to ensure it is still working */
10528 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10529 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10531 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10533 if (sysfs_get_str(info
,
10534 sd
, "state", sbuf
, 20) < 0 ||
10535 strstr(sbuf
, "faulty") ||
10536 strstr(sbuf
, "in_sync") == NULL
) {
10537 /* this device is dead */
10538 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10539 if (sd
->disk
.raid_disk
>= 0 &&
10540 sources
[sd
->disk
.raid_disk
] >= 0) {
10542 sd
->disk
.raid_disk
]);
10543 sources
[sd
->disk
.raid_disk
] =
10552 return new_degraded
;
10555 /*******************************************************************************
10556 * Function: imsm_manage_reshape
10557 * Description: Function finds array under reshape and it manages reshape
10558 * process. It creates stripes backups (if required) and sets
10561 * afd : Backup handle (nattive) - not used
10562 * sra : general array info
10563 * reshape : reshape parameters - not used
10564 * st : supertype structure
10565 * blocks : size of critical section [blocks]
10566 * fds : table of source device descriptor
10567 * offsets : start of array (offest per devices)
10569 * destfd : table of destination device descriptor
10570 * destoffsets : table of destination offsets (per device)
10572 * 1 : success, reshape is done
10574 ******************************************************************************/
10575 static int imsm_manage_reshape(
10576 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10577 struct supertype
*st
, unsigned long backup_blocks
,
10578 int *fds
, unsigned long long *offsets
,
10579 int dests
, int *destfd
, unsigned long long *destoffsets
)
10582 struct intel_super
*super
= st
->sb
;
10583 struct intel_dev
*dv
;
10584 struct imsm_dev
*dev
= NULL
;
10585 struct imsm_map
*map_src
;
10586 int migr_vol_qan
= 0;
10587 int ndata
, odata
; /* [bytes] */
10588 int chunk
; /* [bytes] */
10589 struct migr_record
*migr_rec
;
10591 unsigned int buf_size
; /* [bytes] */
10592 unsigned long long max_position
; /* array size [bytes] */
10593 unsigned long long next_step
; /* [blocks]/[bytes] */
10594 unsigned long long old_data_stripe_length
;
10595 unsigned long long start_src
; /* [bytes] */
10596 unsigned long long start
; /* [bytes] */
10597 unsigned long long start_buf_shift
; /* [bytes] */
10599 int source_layout
= 0;
10604 if (!fds
|| !offsets
)
10607 /* Find volume during the reshape */
10608 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10609 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10610 && dv
->dev
->vol
.migr_state
== 1) {
10615 /* Only one volume can migrate at the same time */
10616 if (migr_vol_qan
!= 1) {
10617 pr_err("%s", migr_vol_qan
?
10618 "Number of migrating volumes greater than 1\n" :
10619 "There is no volume during migrationg\n");
10623 map_src
= get_imsm_map(dev
, MAP_1
);
10624 if (map_src
== NULL
)
10627 ndata
= imsm_num_data_members(dev
, MAP_0
);
10628 odata
= imsm_num_data_members(dev
, MAP_1
);
10630 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10631 old_data_stripe_length
= odata
* chunk
;
10633 migr_rec
= super
->migr_rec
;
10635 /* initialize migration record for start condition */
10636 if (sra
->reshape_progress
== 0)
10637 init_migr_record_imsm(st
, dev
, sra
);
10639 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10640 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
10643 /* Save checkpoint to update migration record for current
10644 * reshape position (in md). It can be farther than current
10645 * reshape position in metadata.
10647 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10648 /* ignore error == 2, this can mean end of reshape here
10650 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
10655 /* size for data */
10656 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10657 /* extend buffer size for parity disk */
10658 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10659 /* add space for stripe aligment */
10660 buf_size
+= old_data_stripe_length
;
10661 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10662 dprintf("imsm: Cannot allocate checpoint buffer\n");
10666 max_position
= sra
->component_size
* ndata
;
10667 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10669 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10670 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10671 /* current reshape position [blocks] */
10672 unsigned long long current_position
=
10673 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10674 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10675 unsigned long long border
;
10677 /* Check that array hasn't become failed.
10679 degraded
= check_degradation_change(sra
, fds
, degraded
);
10680 if (degraded
> 1) {
10681 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
10685 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10687 if ((current_position
+ next_step
) > max_position
)
10688 next_step
= max_position
- current_position
;
10690 start
= current_position
* 512;
10692 /* align reading start to old geometry */
10693 start_buf_shift
= start
% old_data_stripe_length
;
10694 start_src
= start
- start_buf_shift
;
10696 border
= (start_src
/ odata
) - (start
/ ndata
);
10698 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10699 /* save critical stripes to buf
10700 * start - start address of current unit
10701 * to backup [bytes]
10702 * start_src - start address of current unit
10703 * to backup alligned to source array
10706 unsigned long long next_step_filler
;
10707 unsigned long long copy_length
= next_step
* 512;
10709 /* allign copy area length to stripe in old geometry */
10710 next_step_filler
= ((copy_length
+ start_buf_shift
)
10711 % old_data_stripe_length
);
10712 if (next_step_filler
)
10713 next_step_filler
= (old_data_stripe_length
10714 - next_step_filler
);
10715 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
10716 start
, start_src
, copy_length
,
10717 start_buf_shift
, next_step_filler
);
10719 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10720 chunk
, map_src
->raid_level
,
10721 source_layout
, 0, NULL
, start_src
,
10723 next_step_filler
+ start_buf_shift
,
10725 dprintf("imsm: Cannot save stripes to buffer\n");
10728 /* Convert data to destination format and store it
10729 * in backup general migration area
10731 if (save_backup_imsm(st
, dev
, sra
,
10732 buf
+ start_buf_shift
, copy_length
)) {
10733 dprintf("imsm: Cannot save stripes to target devices\n");
10736 if (save_checkpoint_imsm(st
, sra
,
10737 UNIT_SRC_IN_CP_AREA
)) {
10738 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
10742 /* set next step to use whole border area */
10743 border
/= next_step
;
10745 next_step
*= border
;
10747 /* When data backed up, checkpoint stored,
10748 * kick the kernel to reshape unit of data
10750 next_step
= next_step
+ sra
->reshape_progress
;
10751 /* limit next step to array max position */
10752 if (next_step
> max_position
)
10753 next_step
= max_position
;
10754 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10755 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10756 sra
->reshape_progress
= next_step
;
10758 /* wait until reshape finish */
10759 if (wait_for_reshape_imsm(sra
, ndata
)) {
10760 dprintf("wait_for_reshape_imsm returned error!\n");
10766 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10767 /* ignore error == 2, this can mean end of reshape here
10769 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
10775 /* clear migr_rec on disks after successful migration */
10778 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
10779 for (d
= super
->disks
; d
; d
= d
->next
) {
10780 if (d
->index
< 0 || is_failed(&d
->disk
))
10782 unsigned long long dsize
;
10784 get_dev_size(d
->fd
, NULL
, &dsize
);
10785 if (lseek64(d
->fd
, dsize
- MIGR_REC_POSITION
,
10787 if (write(d
->fd
, super
->migr_rec_buf
,
10788 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
10789 perror("Write migr_rec failed");
10793 /* return '1' if done */
10797 /* See Grow.c: abort_reshape() for further explanation */
10798 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
10799 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
10800 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
10805 #endif /* MDASSEMBLE */
10807 struct superswitch super_imsm
= {
10809 .examine_super
= examine_super_imsm
,
10810 .brief_examine_super
= brief_examine_super_imsm
,
10811 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10812 .export_examine_super
= export_examine_super_imsm
,
10813 .detail_super
= detail_super_imsm
,
10814 .brief_detail_super
= brief_detail_super_imsm
,
10815 .write_init_super
= write_init_super_imsm
,
10816 .validate_geometry
= validate_geometry_imsm
,
10817 .add_to_super
= add_to_super_imsm
,
10818 .remove_from_super
= remove_from_super_imsm
,
10819 .detail_platform
= detail_platform_imsm
,
10820 .export_detail_platform
= export_detail_platform_imsm
,
10821 .kill_subarray
= kill_subarray_imsm
,
10822 .update_subarray
= update_subarray_imsm
,
10823 .load_container
= load_container_imsm
,
10824 .default_geometry
= default_geometry_imsm
,
10825 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10826 .reshape_super
= imsm_reshape_super
,
10827 .manage_reshape
= imsm_manage_reshape
,
10828 .recover_backup
= recover_backup_imsm
,
10829 .copy_metadata
= copy_metadata_imsm
,
10831 .match_home
= match_home_imsm
,
10832 .uuid_from_super
= uuid_from_super_imsm
,
10833 .getinfo_super
= getinfo_super_imsm
,
10834 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10835 .update_super
= update_super_imsm
,
10837 .avail_size
= avail_size_imsm
,
10838 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10840 .compare_super
= compare_super_imsm
,
10842 .load_super
= load_super_imsm
,
10843 .init_super
= init_super_imsm
,
10844 .store_super
= store_super_imsm
,
10845 .free_super
= free_super_imsm
,
10846 .match_metadata_desc
= match_metadata_desc_imsm
,
10847 .container_content
= container_content_imsm
,
10848 .validate_container
= validate_container_imsm
,
10855 .open_new
= imsm_open_new
,
10856 .set_array_state
= imsm_set_array_state
,
10857 .set_disk
= imsm_set_disk
,
10858 .sync_metadata
= imsm_sync_metadata
,
10859 .activate_spare
= imsm_activate_spare
,
10860 .process_update
= imsm_process_update
,
10861 .prepare_update
= imsm_prepare_update
,
10862 #endif /* MDASSEMBLE */