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
=NULL
;
541 for (result
= hba
; result
; result
= result
->next
) {
542 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
548 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
550 struct intel_hba
*hba
;
552 /* check if disk attached to Intel HBA */
553 hba
= find_intel_hba(super
->hba
, device
);
556 /* Check if HBA is already attached to super */
557 if (super
->hba
== NULL
) {
558 super
->hba
= alloc_intel_hba(device
);
563 /* Intel metadata allows for all disks attached to the same type HBA.
564 * Do not support HBA types mixing
566 if (device
->type
!= hba
->type
)
569 /* Always forbid spanning between VMD domains (seen as different controllers by mdadm) */
570 if (device
->type
== SYS_DEV_VMD
&& !path_attached_to_hba(device
->path
, hba
->path
))
573 /* Multiple same type HBAs can be used if they share the same OROM */
574 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
576 if (device_orom
!= super
->orom
)
582 hba
->next
= alloc_intel_hba(device
);
586 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
588 struct sys_dev
*list
, *elem
;
591 if ((list
= find_intel_devices()) == NULL
)
595 disk_path
= (char *) devname
;
597 disk_path
= diskfd_to_devpath(fd
);
602 for (elem
= list
; elem
; elem
= elem
->next
)
603 if (path_attached_to_hba(disk_path
, elem
->path
))
606 if (disk_path
!= devname
)
612 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
615 static struct supertype
*match_metadata_desc_imsm(char *arg
)
617 struct supertype
*st
;
619 if (strcmp(arg
, "imsm") != 0 &&
620 strcmp(arg
, "default") != 0
624 st
= xcalloc(1, sizeof(*st
));
625 st
->ss
= &super_imsm
;
626 st
->max_devs
= IMSM_MAX_DEVICES
;
627 st
->minor_version
= 0;
633 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
635 return &mpb
->sig
[MPB_SIG_LEN
];
639 /* retrieve a disk directly from the anchor when the anchor is known to be
640 * up-to-date, currently only at load time
642 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
644 if (index
>= mpb
->num_disks
)
646 return &mpb
->disk
[index
];
649 /* retrieve the disk description based on a index of the disk
652 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
656 for (d
= super
->disks
; d
; d
= d
->next
)
657 if (d
->index
== index
)
662 /* retrieve a disk from the parsed metadata */
663 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
667 dl
= get_imsm_dl_disk(super
, index
);
674 /* generate a checksum directly from the anchor when the anchor is known to be
675 * up-to-date, currently only at load or write_super after coalescing
677 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
679 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
680 __u32
*p
= (__u32
*) mpb
;
684 sum
+= __le32_to_cpu(*p
);
688 return sum
- __le32_to_cpu(mpb
->check_sum
);
691 static size_t sizeof_imsm_map(struct imsm_map
*map
)
693 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
696 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
698 /* A device can have 2 maps if it is in the middle of a migration.
700 * MAP_0 - we return the first map
701 * MAP_1 - we return the second map if it exists, else NULL
702 * MAP_X - we return the second map if it exists, else the first
704 struct imsm_map
*map
= &dev
->vol
.map
[0];
705 struct imsm_map
*map2
= NULL
;
707 if (dev
->vol
.migr_state
)
708 map2
= (void *)map
+ sizeof_imsm_map(map
);
710 switch (second_map
) {
727 /* return the size of the device.
728 * migr_state increases the returned size if map[0] were to be duplicated
730 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
732 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
733 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
735 /* migrating means an additional map */
736 if (dev
->vol
.migr_state
)
737 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
739 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
745 /* retrieve disk serial number list from a metadata update */
746 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
749 struct disk_info
*inf
;
751 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
752 sizeof_imsm_dev(&update
->dev
, 0);
758 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
764 if (index
>= mpb
->num_raid_devs
)
767 /* devices start after all disks */
768 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
770 for (i
= 0; i
<= index
; i
++)
772 return _mpb
+ offset
;
774 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
779 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
781 struct intel_dev
*dv
;
783 if (index
>= super
->anchor
->num_raid_devs
)
785 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
786 if (dv
->index
== index
)
793 * == MAP_0 get first map
794 * == MAP_1 get second map
795 * == MAP_X than get map according to the current migr_state
797 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
801 struct imsm_map
*map
;
803 map
= get_imsm_map(dev
, second_map
);
805 /* top byte identifies disk under rebuild */
806 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
809 #define ord_to_idx(ord) (((ord) << 8) >> 8)
810 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
812 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
814 return ord_to_idx(ord
);
817 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
819 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
822 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
827 for (slot
= 0; slot
< map
->num_members
; slot
++) {
828 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
829 if (ord_to_idx(ord
) == idx
)
836 static int get_imsm_raid_level(struct imsm_map
*map
)
838 if (map
->raid_level
== 1) {
839 if (map
->num_members
== 2)
845 return map
->raid_level
;
848 static int cmp_extent(const void *av
, const void *bv
)
850 const struct extent
*a
= av
;
851 const struct extent
*b
= bv
;
852 if (a
->start
< b
->start
)
854 if (a
->start
> b
->start
)
859 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
864 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
865 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
866 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
868 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
875 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
877 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
879 if (lo
== 0 || hi
== 0)
881 *lo
= __le32_to_cpu((unsigned)n
);
882 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
886 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
888 return (unsigned long long)__le32_to_cpu(lo
) |
889 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
892 static unsigned long long total_blocks(struct imsm_disk
*disk
)
896 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
899 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
903 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
906 static unsigned long long blocks_per_member(struct imsm_map
*map
)
910 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
914 static unsigned long long num_data_stripes(struct imsm_map
*map
)
918 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
921 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
923 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
927 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
929 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
932 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
934 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
937 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
939 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
942 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
944 /* find a list of used extents on the given physical device */
945 struct extent
*rv
, *e
;
947 int memberships
= count_memberships(dl
, super
);
950 /* trim the reserved area for spares, so they can join any array
951 * regardless of whether the OROM has assigned sectors from the
952 * IMSM_RESERVED_SECTORS region
955 reservation
= imsm_min_reserved_sectors(super
);
957 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
959 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
962 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
963 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
964 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
966 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
967 e
->start
= pba_of_lba0(map
);
968 e
->size
= blocks_per_member(map
);
972 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
974 /* determine the start of the metadata
975 * when no raid devices are defined use the default
976 * ...otherwise allow the metadata to truncate the value
977 * as is the case with older versions of imsm
980 struct extent
*last
= &rv
[memberships
- 1];
981 unsigned long long remainder
;
983 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
984 /* round down to 1k block to satisfy precision of the kernel
988 /* make sure remainder is still sane */
989 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
990 remainder
= ROUND_UP(super
->len
, 512) >> 9;
991 if (reservation
> remainder
)
992 reservation
= remainder
;
994 e
->start
= total_blocks(&dl
->disk
) - reservation
;
999 /* try to determine how much space is reserved for metadata from
1000 * the last get_extents() entry, otherwise fallback to the
1003 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1009 /* for spares just return a minimal reservation which will grow
1010 * once the spare is picked up by an array
1012 if (dl
->index
== -1)
1013 return MPB_SECTOR_CNT
;
1015 e
= get_extents(super
, dl
);
1017 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1019 /* scroll to last entry */
1020 for (i
= 0; e
[i
].size
; i
++)
1023 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1030 static int is_spare(struct imsm_disk
*disk
)
1032 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1035 static int is_configured(struct imsm_disk
*disk
)
1037 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1040 static int is_failed(struct imsm_disk
*disk
)
1042 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1045 /* try to determine how much space is reserved for metadata from
1046 * the last get_extents() entry on the smallest active disk,
1047 * otherwise fallback to the default
1049 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1053 unsigned long long min_active
;
1055 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1056 struct dl
*dl
, *dl_min
= NULL
;
1062 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1065 unsigned long long blocks
= total_blocks(&dl
->disk
);
1066 if (blocks
< min_active
|| min_active
== 0) {
1068 min_active
= blocks
;
1074 /* find last lba used by subarrays on the smallest active disk */
1075 e
= get_extents(super
, dl_min
);
1078 for (i
= 0; e
[i
].size
; i
++)
1081 remainder
= min_active
- e
[i
].start
;
1084 /* to give priority to recovery we should not require full
1085 IMSM_RESERVED_SECTORS from the spare */
1086 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1088 /* if real reservation is smaller use that value */
1089 return (remainder
< rv
) ? remainder
: rv
;
1092 /* Return minimum size of a spare that can be used in this array*/
1093 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1095 struct intel_super
*super
= st
->sb
;
1099 unsigned long long rv
= 0;
1103 /* find first active disk in array */
1105 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1109 /* find last lba used by subarrays */
1110 e
= get_extents(super
, dl
);
1113 for (i
= 0; e
[i
].size
; i
++)
1116 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1119 /* add the amount of space needed for metadata */
1120 rv
= rv
+ imsm_min_reserved_sectors(super
);
1125 static int is_gen_migration(struct imsm_dev
*dev
);
1128 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1129 struct imsm_dev
*dev
);
1131 static void print_imsm_dev(struct intel_super
*super
,
1132 struct imsm_dev
*dev
,
1138 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1139 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1143 printf("[%.16s]:\n", dev
->volume
);
1144 printf(" UUID : %s\n", uuid
);
1145 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1147 printf(" <-- %d", get_imsm_raid_level(map2
));
1149 printf(" Members : %d", map
->num_members
);
1151 printf(" <-- %d", map2
->num_members
);
1153 printf(" Slots : [");
1154 for (i
= 0; i
< map
->num_members
; i
++) {
1155 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1156 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1161 for (i
= 0; i
< map2
->num_members
; i
++) {
1162 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1163 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1168 printf(" Failed disk : ");
1169 if (map
->failed_disk_num
== 0xff)
1172 printf("%i", map
->failed_disk_num
);
1174 slot
= get_imsm_disk_slot(map
, disk_idx
);
1176 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1177 printf(" This Slot : %d%s\n", slot
,
1178 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1180 printf(" This Slot : ?\n");
1181 sz
= __le32_to_cpu(dev
->size_high
);
1183 sz
+= __le32_to_cpu(dev
->size_low
);
1184 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1185 human_size(sz
* 512));
1186 sz
= blocks_per_member(map
);
1187 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1188 human_size(sz
* 512));
1189 printf(" Sector Offset : %llu\n",
1191 printf(" Num Stripes : %llu\n",
1192 num_data_stripes(map
));
1193 printf(" Chunk Size : %u KiB",
1194 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1196 printf(" <-- %u KiB",
1197 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1199 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1200 printf(" Migrate State : ");
1201 if (dev
->vol
.migr_state
) {
1202 if (migr_type(dev
) == MIGR_INIT
)
1203 printf("initialize\n");
1204 else if (migr_type(dev
) == MIGR_REBUILD
)
1205 printf("rebuild\n");
1206 else if (migr_type(dev
) == MIGR_VERIFY
)
1208 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1209 printf("general migration\n");
1210 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1211 printf("state change\n");
1212 else if (migr_type(dev
) == MIGR_REPAIR
)
1215 printf("<unknown:%d>\n", migr_type(dev
));
1218 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1219 if (dev
->vol
.migr_state
) {
1220 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1222 printf(" <-- %s", map_state_str
[map
->map_state
]);
1223 printf("\n Checkpoint : %u ",
1224 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1225 if ((is_gen_migration(dev
)) && ((slot
> 1) || (slot
< 0)))
1228 printf("(%llu)", (unsigned long long)
1229 blocks_per_migr_unit(super
, dev
));
1232 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1235 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1237 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1240 if (index
< -1 || !disk
)
1244 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1246 printf(" Disk%02d Serial : %s\n", index
, str
);
1248 printf(" Disk Serial : %s\n", str
);
1249 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1250 is_configured(disk
) ? " active" : "",
1251 is_failed(disk
) ? " failed" : "");
1252 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1253 sz
= total_blocks(disk
) - reserved
;
1254 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1255 human_size(sz
* 512));
1258 void examine_migr_rec_imsm(struct intel_super
*super
)
1260 struct migr_record
*migr_rec
= super
->migr_rec
;
1261 struct imsm_super
*mpb
= super
->anchor
;
1264 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1265 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1266 struct imsm_map
*map
;
1269 if (is_gen_migration(dev
) == 0)
1272 printf("\nMigration Record Information:");
1274 /* first map under migration */
1275 map
= get_imsm_map(dev
, MAP_0
);
1277 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1278 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1279 printf(" Empty\n ");
1280 printf("Examine one of first two disks in array\n");
1283 printf("\n Status : ");
1284 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1287 printf("Contains Data\n");
1288 printf(" Current Unit : %u\n",
1289 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1290 printf(" Family : %u\n",
1291 __le32_to_cpu(migr_rec
->family_num
));
1292 printf(" Ascending : %u\n",
1293 __le32_to_cpu(migr_rec
->ascending_migr
));
1294 printf(" Blocks Per Unit : %u\n",
1295 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1296 printf(" Dest. Depth Per Unit : %u\n",
1297 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1298 printf(" Checkpoint Area pba : %u\n",
1299 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1300 printf(" First member lba : %u\n",
1301 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1302 printf(" Total Number of Units : %u\n",
1303 __le32_to_cpu(migr_rec
->num_migr_units
));
1304 printf(" Size of volume : %u\n",
1305 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1306 printf(" Expansion space for LBA64 : %u\n",
1307 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1308 printf(" Record was read from : %u\n",
1309 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1314 #endif /* MDASSEMBLE */
1315 /*******************************************************************************
1316 * function: imsm_check_attributes
1317 * Description: Function checks if features represented by attributes flags
1318 * are supported by mdadm.
1320 * attributes - Attributes read from metadata
1322 * 0 - passed attributes contains unsupported features flags
1323 * 1 - all features are supported
1324 ******************************************************************************/
1325 static int imsm_check_attributes(__u32 attributes
)
1328 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1330 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1332 not_supported
&= attributes
;
1333 if (not_supported
) {
1334 pr_err("(IMSM): Unsupported attributes : %x\n",
1335 (unsigned)__le32_to_cpu(not_supported
));
1336 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1337 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1338 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1340 if (not_supported
& MPB_ATTRIB_2TB
) {
1341 dprintf("\t\tMPB_ATTRIB_2TB\n");
1342 not_supported
^= MPB_ATTRIB_2TB
;
1344 if (not_supported
& MPB_ATTRIB_RAID0
) {
1345 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1346 not_supported
^= MPB_ATTRIB_RAID0
;
1348 if (not_supported
& MPB_ATTRIB_RAID1
) {
1349 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1350 not_supported
^= MPB_ATTRIB_RAID1
;
1352 if (not_supported
& MPB_ATTRIB_RAID10
) {
1353 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1354 not_supported
^= MPB_ATTRIB_RAID10
;
1356 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1357 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1358 not_supported
^= MPB_ATTRIB_RAID1E
;
1360 if (not_supported
& MPB_ATTRIB_RAID5
) {
1361 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1362 not_supported
^= MPB_ATTRIB_RAID5
;
1364 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1365 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1366 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1368 if (not_supported
& MPB_ATTRIB_BBM
) {
1369 dprintf("\t\tMPB_ATTRIB_BBM\n");
1370 not_supported
^= MPB_ATTRIB_BBM
;
1372 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1373 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1374 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1376 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1377 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1378 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1380 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1381 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1382 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1384 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1385 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1386 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1388 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1389 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1390 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1394 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1403 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1405 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1407 struct intel_super
*super
= st
->sb
;
1408 struct imsm_super
*mpb
= super
->anchor
;
1409 char str
[MAX_SIGNATURE_LENGTH
];
1414 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1417 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1418 printf(" Magic : %s\n", str
);
1419 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1420 printf(" Version : %s\n", get_imsm_version(mpb
));
1421 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1422 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1423 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1424 printf(" Attributes : ");
1425 if (imsm_check_attributes(mpb
->attributes
))
1426 printf("All supported\n");
1428 printf("not supported\n");
1429 getinfo_super_imsm(st
, &info
, NULL
);
1430 fname_from_uuid(st
, &info
, nbuf
, ':');
1431 printf(" UUID : %s\n", nbuf
+ 5);
1432 sum
= __le32_to_cpu(mpb
->check_sum
);
1433 printf(" Checksum : %08x %s\n", sum
,
1434 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1435 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1436 printf(" Disks : %d\n", mpb
->num_disks
);
1437 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1438 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1439 if (super
->bbm_log
) {
1440 struct bbm_log
*log
= super
->bbm_log
;
1443 printf("Bad Block Management Log:\n");
1444 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1445 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1446 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1447 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1448 printf(" First Spare : %llx\n",
1449 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1451 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1453 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1455 super
->current_vol
= i
;
1456 getinfo_super_imsm(st
, &info
, NULL
);
1457 fname_from_uuid(st
, &info
, nbuf
, ':');
1458 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1460 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1461 if (i
== super
->disks
->index
)
1463 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1466 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1467 if (dl
->index
== -1)
1468 print_imsm_disk(&dl
->disk
, -1, reserved
);
1470 examine_migr_rec_imsm(super
);
1473 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1475 /* We just write a generic IMSM ARRAY entry */
1478 struct intel_super
*super
= st
->sb
;
1480 if (!super
->anchor
->num_raid_devs
) {
1481 printf("ARRAY metadata=imsm\n");
1485 getinfo_super_imsm(st
, &info
, NULL
);
1486 fname_from_uuid(st
, &info
, nbuf
, ':');
1487 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1490 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1492 /* We just write a generic IMSM ARRAY entry */
1496 struct intel_super
*super
= st
->sb
;
1499 if (!super
->anchor
->num_raid_devs
)
1502 getinfo_super_imsm(st
, &info
, NULL
);
1503 fname_from_uuid(st
, &info
, nbuf
, ':');
1504 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1505 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1507 super
->current_vol
= i
;
1508 getinfo_super_imsm(st
, &info
, NULL
);
1509 fname_from_uuid(st
, &info
, nbuf1
, ':');
1510 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1511 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1515 static void export_examine_super_imsm(struct supertype
*st
)
1517 struct intel_super
*super
= st
->sb
;
1518 struct imsm_super
*mpb
= super
->anchor
;
1522 getinfo_super_imsm(st
, &info
, NULL
);
1523 fname_from_uuid(st
, &info
, nbuf
, ':');
1524 printf("MD_METADATA=imsm\n");
1525 printf("MD_LEVEL=container\n");
1526 printf("MD_UUID=%s\n", nbuf
+5);
1527 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1530 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
1532 /* The second last 512byte sector of the device contains
1533 * the "struct imsm_super" metadata.
1534 * This contains mpb_size which is the size in bytes of the
1535 * extended metadata. This is located immediately before
1537 * We want to read all that, plus the last sector which
1538 * may contain a migration record, and write it all
1542 unsigned long long dsize
, offset
;
1544 struct imsm_super
*sb
;
1547 if (posix_memalign(&buf
, 4096, 4096) != 0)
1550 if (!get_dev_size(from
, NULL
, &dsize
))
1553 if (lseek64(from
, dsize
-1024, 0) < 0)
1555 if (read(from
, buf
, 512) != 512)
1558 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
1561 sectors
= mpb_sectors(sb
) + 2;
1562 offset
= dsize
- sectors
* 512;
1563 if (lseek64(from
, offset
, 0) < 0 ||
1564 lseek64(to
, offset
, 0) < 0)
1566 while (written
< sectors
* 512) {
1567 int n
= sectors
*512 - written
;
1570 if (read(from
, buf
, n
) != n
)
1572 if (write(to
, buf
, n
) != n
)
1583 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1588 getinfo_super_imsm(st
, &info
, NULL
);
1589 fname_from_uuid(st
, &info
, nbuf
, ':');
1590 printf("\n UUID : %s\n", nbuf
+ 5);
1593 static void brief_detail_super_imsm(struct supertype
*st
)
1597 getinfo_super_imsm(st
, &info
, NULL
);
1598 fname_from_uuid(st
, &info
, nbuf
, ':');
1599 printf(" UUID=%s", nbuf
+ 5);
1602 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1603 static void fd2devname(int fd
, char *name
);
1605 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1607 /* dump an unsorted list of devices attached to AHCI Intel storage
1608 * controller, as well as non-connected ports
1610 int hba_len
= strlen(hba_path
) + 1;
1615 unsigned long port_mask
= (1 << port_count
) - 1;
1617 if (port_count
> (int)sizeof(port_mask
) * 8) {
1619 pr_err("port_count %d out of range\n", port_count
);
1623 /* scroll through /sys/dev/block looking for devices attached to
1626 dir
= opendir("/sys/dev/block");
1630 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1641 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1643 path
= devt_to_devpath(makedev(major
, minor
));
1646 if (!path_attached_to_hba(path
, hba_path
)) {
1652 /* retrieve the scsi device type */
1653 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1655 pr_err("failed to allocate 'device'\n");
1659 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1660 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
1662 pr_err("failed to read device type for %s\n",
1668 type
= strtoul(buf
, NULL
, 10);
1670 /* if it's not a disk print the vendor and model */
1671 if (!(type
== 0 || type
== 7 || type
== 14)) {
1674 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1675 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
1676 strncpy(vendor
, buf
, sizeof(vendor
));
1677 vendor
[sizeof(vendor
) - 1] = '\0';
1678 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1679 while (isspace(*c
) || *c
== '\0')
1683 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1684 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
1685 strncpy(model
, buf
, sizeof(model
));
1686 model
[sizeof(model
) - 1] = '\0';
1687 c
= (char *) &model
[sizeof(model
) - 1];
1688 while (isspace(*c
) || *c
== '\0')
1692 if (vendor
[0] && model
[0])
1693 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1695 switch (type
) { /* numbers from hald/linux/device.c */
1696 case 1: sprintf(buf
, "tape"); break;
1697 case 2: sprintf(buf
, "printer"); break;
1698 case 3: sprintf(buf
, "processor"); break;
1700 case 5: sprintf(buf
, "cdrom"); break;
1701 case 6: sprintf(buf
, "scanner"); break;
1702 case 8: sprintf(buf
, "media_changer"); break;
1703 case 9: sprintf(buf
, "comm"); break;
1704 case 12: sprintf(buf
, "raid"); break;
1705 default: sprintf(buf
, "unknown");
1711 /* chop device path to 'host%d' and calculate the port number */
1712 c
= strchr(&path
[hba_len
], '/');
1715 pr_err("%s - invalid path name\n", path
+ hba_len
);
1720 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
1721 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
1725 *c
= '/'; /* repair the full string */
1726 pr_err("failed to determine port number for %s\n",
1733 /* mark this port as used */
1734 port_mask
&= ~(1 << port
);
1736 /* print out the device information */
1738 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1742 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1744 printf(" Port%d : - disk info unavailable -\n", port
);
1746 fd2devname(fd
, buf
);
1747 printf(" Port%d : %s", port
, buf
);
1748 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1749 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1764 for (i
= 0; i
< port_count
; i
++)
1765 if (port_mask
& (1 << i
))
1766 printf(" Port%d : - no device attached -\n", i
);
1772 static int print_vmd_attached_devs(struct sys_dev
*hba
)
1780 if (hba
->type
!= SYS_DEV_VMD
)
1783 /* scroll through /sys/dev/block looking for devices attached to
1786 dir
= opendir("/sys/bus/pci/drivers/nvme");
1790 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1793 /* is 'ent' a device? check that the 'subsystem' link exists and
1794 * that its target matches 'bus'
1796 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
1798 n
= readlink(path
, link
, sizeof(link
));
1799 if (n
< 0 || n
>= (int)sizeof(link
))
1802 c
= strrchr(link
, '/');
1805 if (strncmp("pci", c
+1, strlen("pci")) != 0)
1808 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
1809 /* if not a intel NVMe - skip it*/
1810 if (devpath_to_vendor(path
) != 0x8086)
1813 rp
= realpath(path
, NULL
);
1817 if (path_attached_to_hba(rp
, hba
->path
)) {
1818 printf(" NVMe under VMD : %s\n", rp
);
1827 static void print_found_intel_controllers(struct sys_dev
*elem
)
1829 for (; elem
; elem
= elem
->next
) {
1830 pr_err("found Intel(R) ");
1831 if (elem
->type
== SYS_DEV_SATA
)
1832 fprintf(stderr
, "SATA ");
1833 else if (elem
->type
== SYS_DEV_SAS
)
1834 fprintf(stderr
, "SAS ");
1835 else if (elem
->type
== SYS_DEV_NVME
)
1836 fprintf(stderr
, "NVMe ");
1838 if (elem
->type
== SYS_DEV_VMD
)
1839 fprintf(stderr
, "VMD domain");
1841 fprintf(stderr
, "RAID controller");
1844 fprintf(stderr
, " at %s", elem
->pci_id
);
1845 fprintf(stderr
, ".\n");
1850 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1857 if ((dir
= opendir(hba_path
)) == NULL
)
1860 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1863 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
1864 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
1866 if (*port_count
== 0)
1868 else if (host
< host_base
)
1871 if (host
+ 1 > *port_count
+ host_base
)
1872 *port_count
= host
+ 1 - host_base
;
1878 static void print_imsm_capability(const struct imsm_orom
*orom
)
1880 printf(" Platform : Intel(R) ");
1881 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
1882 printf("Matrix Storage Manager\n");
1884 printf("Rapid Storage Technology%s\n",
1885 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
1886 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
1887 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
1888 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
1889 printf(" RAID Levels :%s%s%s%s%s\n",
1890 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1891 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1892 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1893 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1894 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1895 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1896 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1897 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1898 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1899 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1900 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1901 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1902 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1903 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1904 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1905 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1906 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1907 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1908 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1909 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1910 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1911 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1912 printf(" 2TB volumes :%s supported\n",
1913 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
1914 printf(" 2TB disks :%s supported\n",
1915 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
1916 printf(" Max Disks : %d\n", orom
->tds
);
1917 printf(" Max Volumes : %d per array, %d per %s\n",
1918 orom
->vpa
, orom
->vphba
,
1919 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
1923 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
1925 printf("MD_FIRMWARE_TYPE=imsm\n");
1926 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
1927 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1928 orom
->hotfix_ver
, orom
->build
);
1929 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
1930 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
1931 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
1932 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
1933 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
1934 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
1935 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1936 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
1937 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
1938 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
1939 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
1940 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
1941 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
1942 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
1943 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
1944 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
1945 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
1946 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
1947 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
1948 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
1949 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
1950 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
1951 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
1952 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
1953 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
1954 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
1955 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
1956 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
1959 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
1961 /* There are two components to imsm platform support, the ahci SATA
1962 * controller and the option-rom. To find the SATA controller we
1963 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1964 * controller with the Intel vendor id is present. This approach
1965 * allows mdadm to leverage the kernel's ahci detection logic, with the
1966 * caveat that if ahci.ko is not loaded mdadm will not be able to
1967 * detect platform raid capabilities. The option-rom resides in a
1968 * platform "Adapter ROM". We scan for its signature to retrieve the
1969 * platform capabilities. If raid support is disabled in the BIOS the
1970 * option-rom capability structure will not be available.
1972 struct sys_dev
*list
, *hba
;
1977 if (enumerate_only
) {
1978 if (check_env("IMSM_NO_PLATFORM"))
1980 list
= find_intel_devices();
1983 for (hba
= list
; hba
; hba
= hba
->next
) {
1984 if (find_imsm_capability(hba
)) {
1994 list
= find_intel_devices();
1997 pr_err("no active Intel(R) RAID controller found.\n");
1999 } else if (verbose
> 0)
2000 print_found_intel_controllers(list
);
2002 for (hba
= list
; hba
; hba
= hba
->next
) {
2003 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2005 if (!find_imsm_capability(hba
)) {
2007 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2008 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2009 get_sys_dev_type(hba
->type
));
2015 if (controller_path
&& result
== 1) {
2016 pr_err("no active Intel(R) RAID controller found under %s\n",
2021 const struct orom_entry
*entry
;
2023 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2024 if (entry
->type
== SYS_DEV_VMD
) {
2025 for (hba
= list
; hba
; hba
= hba
->next
) {
2026 if (hba
->type
== SYS_DEV_VMD
) {
2028 print_imsm_capability(&entry
->orom
);
2029 printf(" I/O Controller : %s (%s)\n",
2030 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2031 if (print_vmd_attached_devs(hba
)) {
2033 pr_err("failed to get devices attached to VMD domain.\n");
2042 print_imsm_capability(&entry
->orom
);
2043 if (entry
->type
== SYS_DEV_NVME
) {
2044 for (hba
= list
; hba
; hba
= hba
->next
) {
2045 if (hba
->type
== SYS_DEV_NVME
)
2046 printf(" NVMe Device : %s\n", hba
->path
);
2052 struct devid_list
*devid
;
2053 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2054 hba
= device_by_id(devid
->devid
);
2058 printf(" I/O Controller : %s (%s)\n",
2059 hba
->path
, get_sys_dev_type(hba
->type
));
2060 if (hba
->type
== SYS_DEV_SATA
) {
2061 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2062 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2064 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2075 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2077 struct sys_dev
*list
, *hba
;
2080 list
= find_intel_devices();
2083 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2088 for (hba
= list
; hba
; hba
= hba
->next
) {
2089 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2091 if (!find_imsm_capability(hba
) && verbose
> 0) {
2093 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2094 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2100 const struct orom_entry
*entry
;
2102 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2103 if (entry
->type
== SYS_DEV_VMD
) {
2104 for (hba
= list
; hba
; hba
= hba
->next
)
2105 print_imsm_capability_export(&entry
->orom
);
2108 print_imsm_capability_export(&entry
->orom
);
2116 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2118 /* the imsm metadata format does not specify any host
2119 * identification information. We return -1 since we can never
2120 * confirm nor deny whether a given array is "meant" for this
2121 * host. We rely on compare_super and the 'family_num' fields to
2122 * exclude member disks that do not belong, and we rely on
2123 * mdadm.conf to specify the arrays that should be assembled.
2124 * Auto-assembly may still pick up "foreign" arrays.
2130 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2132 /* The uuid returned here is used for:
2133 * uuid to put into bitmap file (Create, Grow)
2134 * uuid for backup header when saving critical section (Grow)
2135 * comparing uuids when re-adding a device into an array
2136 * In these cases the uuid required is that of the data-array,
2137 * not the device-set.
2138 * uuid to recognise same set when adding a missing device back
2139 * to an array. This is a uuid for the device-set.
2141 * For each of these we can make do with a truncated
2142 * or hashed uuid rather than the original, as long as
2144 * In each case the uuid required is that of the data-array,
2145 * not the device-set.
2147 /* imsm does not track uuid's so we synthesis one using sha1 on
2148 * - The signature (Which is constant for all imsm array, but no matter)
2149 * - the orig_family_num of the container
2150 * - the index number of the volume
2151 * - the 'serial' number of the volume.
2152 * Hopefully these are all constant.
2154 struct intel_super
*super
= st
->sb
;
2157 struct sha1_ctx ctx
;
2158 struct imsm_dev
*dev
= NULL
;
2161 /* some mdadm versions failed to set ->orig_family_num, in which
2162 * case fall back to ->family_num. orig_family_num will be
2163 * fixed up with the first metadata update.
2165 family_num
= super
->anchor
->orig_family_num
;
2166 if (family_num
== 0)
2167 family_num
= super
->anchor
->family_num
;
2168 sha1_init_ctx(&ctx
);
2169 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2170 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2171 if (super
->current_vol
>= 0)
2172 dev
= get_imsm_dev(super
, super
->current_vol
);
2174 __u32 vol
= super
->current_vol
;
2175 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2176 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2178 sha1_finish_ctx(&ctx
, buf
);
2179 memcpy(uuid
, buf
, 4*4);
2184 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2186 __u8
*v
= get_imsm_version(mpb
);
2187 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2188 char major
[] = { 0, 0, 0 };
2189 char minor
[] = { 0 ,0, 0 };
2190 char patch
[] = { 0, 0, 0 };
2191 char *ver_parse
[] = { major
, minor
, patch
};
2195 while (*v
!= '\0' && v
< end
) {
2196 if (*v
!= '.' && j
< 2)
2197 ver_parse
[i
][j
++] = *v
;
2205 *m
= strtol(minor
, NULL
, 0);
2206 *p
= strtol(patch
, NULL
, 0);
2210 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2212 /* migr_strip_size when repairing or initializing parity */
2213 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2214 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2216 switch (get_imsm_raid_level(map
)) {
2221 return 128*1024 >> 9;
2225 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2227 /* migr_strip_size when rebuilding a degraded disk, no idea why
2228 * this is different than migr_strip_size_resync(), but it's good
2231 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2232 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2234 switch (get_imsm_raid_level(map
)) {
2237 if (map
->num_members
% map
->num_domains
== 0)
2238 return 128*1024 >> 9;
2242 return max((__u32
) 64*1024 >> 9, chunk
);
2244 return 128*1024 >> 9;
2248 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2250 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2251 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2252 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2253 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2255 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2258 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2260 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2261 int level
= get_imsm_raid_level(lo
);
2263 if (level
== 1 || level
== 10) {
2264 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2266 return hi
->num_domains
;
2268 return num_stripes_per_unit_resync(dev
);
2271 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2273 /* named 'imsm_' because raid0, raid1 and raid10
2274 * counter-intuitively have the same number of data disks
2276 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2278 switch (get_imsm_raid_level(map
)) {
2280 return map
->num_members
;
2284 return map
->num_members
/2;
2286 return map
->num_members
- 1;
2288 dprintf("unsupported raid level\n");
2293 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2295 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2296 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2298 switch(get_imsm_raid_level(map
)) {
2301 return chunk
* map
->num_domains
;
2303 return chunk
* map
->num_members
;
2309 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2311 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2312 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2313 __u32 strip
= block
/ chunk
;
2315 switch (get_imsm_raid_level(map
)) {
2318 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2319 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2321 return vol_stripe
* chunk
+ block
% chunk
;
2323 __u32 stripe
= strip
/ (map
->num_members
- 1);
2325 return stripe
* chunk
+ block
% chunk
;
2332 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2333 struct imsm_dev
*dev
)
2335 /* calculate the conversion factor between per member 'blocks'
2336 * (md/{resync,rebuild}_start) and imsm migration units, return
2337 * 0 for the 'not migrating' and 'unsupported migration' cases
2339 if (!dev
->vol
.migr_state
)
2342 switch (migr_type(dev
)) {
2343 case MIGR_GEN_MIGR
: {
2344 struct migr_record
*migr_rec
= super
->migr_rec
;
2345 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2350 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2351 __u32 stripes_per_unit
;
2352 __u32 blocks_per_unit
;
2361 /* yes, this is really the translation of migr_units to
2362 * per-member blocks in the 'resync' case
2364 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2365 migr_chunk
= migr_strip_blocks_resync(dev
);
2366 disks
= imsm_num_data_members(dev
, MAP_0
);
2367 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2368 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2369 segment
= blocks_per_unit
/ stripe
;
2370 block_rel
= blocks_per_unit
- segment
* stripe
;
2371 parity_depth
= parity_segment_depth(dev
);
2372 block_map
= map_migr_block(dev
, block_rel
);
2373 return block_map
+ parity_depth
* segment
;
2375 case MIGR_REBUILD
: {
2376 __u32 stripes_per_unit
;
2379 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2380 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2381 return migr_chunk
* stripes_per_unit
;
2383 case MIGR_STATE_CHANGE
:
2389 static int imsm_level_to_layout(int level
)
2397 return ALGORITHM_LEFT_ASYMMETRIC
;
2404 /*******************************************************************************
2405 * Function: read_imsm_migr_rec
2406 * Description: Function reads imsm migration record from last sector of disk
2408 * fd : disk descriptor
2409 * super : metadata info
2413 ******************************************************************************/
2414 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2417 unsigned long long dsize
;
2419 get_dev_size(fd
, NULL
, &dsize
);
2420 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2421 pr_err("Cannot seek to anchor block: %s\n",
2425 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2426 MIGR_REC_BUF_SIZE
) {
2427 pr_err("Cannot read migr record block: %s\n",
2437 static struct imsm_dev
*imsm_get_device_during_migration(
2438 struct intel_super
*super
)
2441 struct intel_dev
*dv
;
2443 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2444 if (is_gen_migration(dv
->dev
))
2450 /*******************************************************************************
2451 * Function: load_imsm_migr_rec
2452 * Description: Function reads imsm migration record (it is stored at the last
2455 * super : imsm internal array info
2456 * info : general array info
2460 * -2 : no migration in progress
2461 ******************************************************************************/
2462 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2465 struct dl
*dl
= NULL
;
2469 struct imsm_dev
*dev
;
2470 struct imsm_map
*map
= NULL
;
2473 /* find map under migration */
2474 dev
= imsm_get_device_during_migration(super
);
2475 /* nothing to load,no migration in progress?
2479 map
= get_imsm_map(dev
, MAP_0
);
2482 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2483 /* skip spare and failed disks
2485 if (sd
->disk
.raid_disk
< 0)
2487 /* read only from one of the first two slots */
2489 slot
= get_imsm_disk_slot(map
,
2490 sd
->disk
.raid_disk
);
2491 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2494 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2495 fd
= dev_open(nm
, O_RDONLY
);
2501 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2502 /* skip spare and failed disks
2506 /* read only from one of the first two slots */
2508 slot
= get_imsm_disk_slot(map
, dl
->index
);
2509 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2511 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2512 fd
= dev_open(nm
, O_RDONLY
);
2519 retval
= read_imsm_migr_rec(fd
, super
);
2528 /*******************************************************************************
2529 * function: imsm_create_metadata_checkpoint_update
2530 * Description: It creates update for checkpoint change.
2532 * super : imsm internal array info
2533 * u : pointer to prepared update
2536 * If length is equal to 0, input pointer u contains no update
2537 ******************************************************************************/
2538 static int imsm_create_metadata_checkpoint_update(
2539 struct intel_super
*super
,
2540 struct imsm_update_general_migration_checkpoint
**u
)
2543 int update_memory_size
= 0;
2545 dprintf("(enter)\n");
2551 /* size of all update data without anchor */
2552 update_memory_size
=
2553 sizeof(struct imsm_update_general_migration_checkpoint
);
2555 *u
= xcalloc(1, update_memory_size
);
2557 dprintf("error: cannot get memory\n");
2560 (*u
)->type
= update_general_migration_checkpoint
;
2561 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2562 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
2564 return update_memory_size
;
2567 static void imsm_update_metadata_locally(struct supertype
*st
,
2568 void *buf
, int len
);
2570 /*******************************************************************************
2571 * Function: write_imsm_migr_rec
2572 * Description: Function writes imsm migration record
2573 * (at the last sector of disk)
2575 * super : imsm internal array info
2579 ******************************************************************************/
2580 static int write_imsm_migr_rec(struct supertype
*st
)
2582 struct intel_super
*super
= st
->sb
;
2583 unsigned long long dsize
;
2589 struct imsm_update_general_migration_checkpoint
*u
;
2590 struct imsm_dev
*dev
;
2591 struct imsm_map
*map
= NULL
;
2593 /* find map under migration */
2594 dev
= imsm_get_device_during_migration(super
);
2595 /* if no migration, write buffer anyway to clear migr_record
2596 * on disk based on first available device
2599 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2600 super
->current_vol
);
2602 map
= get_imsm_map(dev
, MAP_0
);
2604 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2607 /* skip failed and spare devices */
2610 /* write to 2 first slots only */
2612 slot
= get_imsm_disk_slot(map
, sd
->index
);
2613 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2616 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2617 fd
= dev_open(nm
, O_RDWR
);
2620 get_dev_size(fd
, NULL
, &dsize
);
2621 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2622 pr_err("Cannot seek to anchor block: %s\n",
2626 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2627 MIGR_REC_BUF_SIZE
) {
2628 pr_err("Cannot write migr record block: %s\n",
2635 /* update checkpoint information in metadata */
2636 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2639 dprintf("imsm: Cannot prepare update\n");
2642 /* update metadata locally */
2643 imsm_update_metadata_locally(st
, u
, len
);
2644 /* and possibly remotely */
2645 if (st
->update_tail
) {
2646 append_metadata_update(st
, u
, len
);
2647 /* during reshape we do all work inside metadata handler
2648 * manage_reshape(), so metadata update has to be triggered
2651 flush_metadata_updates(st
);
2652 st
->update_tail
= &st
->updates
;
2662 #endif /* MDASSEMBLE */
2664 /* spare/missing disks activations are not allowe when
2665 * array/container performs reshape operation, because
2666 * all arrays in container works on the same disks set
2668 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2671 struct intel_dev
*i_dev
;
2672 struct imsm_dev
*dev
;
2674 /* check whole container
2676 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2678 if (is_gen_migration(dev
)) {
2679 /* No repair during any migration in container
2687 static unsigned long long imsm_component_size_aligment_check(int level
,
2689 unsigned long long component_size
)
2691 unsigned int component_size_alligment
;
2693 /* check component size aligment
2695 component_size_alligment
= component_size
% (chunk_size
/512);
2697 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
2698 level
, chunk_size
, component_size
,
2699 component_size_alligment
);
2701 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
2702 dprintf("imsm: reported component size alligned from %llu ",
2704 component_size
-= component_size_alligment
;
2705 dprintf_cont("to %llu (%i).\n",
2706 component_size
, component_size_alligment
);
2709 return component_size
;
2712 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2714 struct intel_super
*super
= st
->sb
;
2715 struct migr_record
*migr_rec
= super
->migr_rec
;
2716 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2717 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2718 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2719 struct imsm_map
*map_to_analyse
= map
;
2721 int map_disks
= info
->array
.raid_disks
;
2723 memset(info
, 0, sizeof(*info
));
2725 map_to_analyse
= prev_map
;
2727 dl
= super
->current_disk
;
2729 info
->container_member
= super
->current_vol
;
2730 info
->array
.raid_disks
= map
->num_members
;
2731 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2732 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2733 info
->array
.md_minor
= -1;
2734 info
->array
.ctime
= 0;
2735 info
->array
.utime
= 0;
2736 info
->array
.chunk_size
=
2737 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2738 info
->array
.state
= !dev
->vol
.dirty
;
2739 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2740 info
->custom_array_size
<<= 32;
2741 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2742 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2744 if (is_gen_migration(dev
)) {
2745 info
->reshape_active
= 1;
2746 info
->new_level
= get_imsm_raid_level(map
);
2747 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2748 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2749 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2750 if (info
->delta_disks
) {
2751 /* this needs to be applied to every array
2754 info
->reshape_active
= CONTAINER_RESHAPE
;
2756 /* We shape information that we give to md might have to be
2757 * modify to cope with md's requirement for reshaping arrays.
2758 * For example, when reshaping a RAID0, md requires it to be
2759 * presented as a degraded RAID4.
2760 * Also if a RAID0 is migrating to a RAID5 we need to specify
2761 * the array as already being RAID5, but the 'before' layout
2762 * is a RAID4-like layout.
2764 switch (info
->array
.level
) {
2766 switch(info
->new_level
) {
2768 /* conversion is happening as RAID4 */
2769 info
->array
.level
= 4;
2770 info
->array
.raid_disks
+= 1;
2773 /* conversion is happening as RAID5 */
2774 info
->array
.level
= 5;
2775 info
->array
.layout
= ALGORITHM_PARITY_N
;
2776 info
->delta_disks
-= 1;
2779 /* FIXME error message */
2780 info
->array
.level
= UnSet
;
2786 info
->new_level
= UnSet
;
2787 info
->new_layout
= UnSet
;
2788 info
->new_chunk
= info
->array
.chunk_size
;
2789 info
->delta_disks
= 0;
2793 info
->disk
.major
= dl
->major
;
2794 info
->disk
.minor
= dl
->minor
;
2795 info
->disk
.number
= dl
->index
;
2796 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2800 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2801 info
->component_size
= blocks_per_member(map_to_analyse
);
2803 info
->component_size
= imsm_component_size_aligment_check(
2805 info
->array
.chunk_size
,
2806 info
->component_size
);
2808 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2809 info
->recovery_start
= MaxSector
;
2811 info
->reshape_progress
= 0;
2812 info
->resync_start
= MaxSector
;
2813 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2815 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2816 info
->resync_start
= 0;
2818 if (dev
->vol
.migr_state
) {
2819 switch (migr_type(dev
)) {
2822 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2824 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2826 info
->resync_start
= blocks_per_unit
* units
;
2829 case MIGR_GEN_MIGR
: {
2830 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2832 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2833 unsigned long long array_blocks
;
2836 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2838 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2839 (super
->migr_rec
->rec_status
==
2840 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2843 info
->reshape_progress
= blocks_per_unit
* units
;
2845 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
2846 (unsigned long long)units
,
2847 (unsigned long long)blocks_per_unit
,
2848 info
->reshape_progress
);
2850 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2851 if (used_disks
> 0) {
2852 array_blocks
= blocks_per_member(map
) *
2854 /* round array size down to closest MB
2856 info
->custom_array_size
= (array_blocks
2857 >> SECT_PER_MB_SHIFT
)
2858 << SECT_PER_MB_SHIFT
;
2862 /* we could emulate the checkpointing of
2863 * 'sync_action=check' migrations, but for now
2864 * we just immediately complete them
2867 /* this is handled by container_content_imsm() */
2868 case MIGR_STATE_CHANGE
:
2869 /* FIXME handle other migrations */
2871 /* we are not dirty, so... */
2872 info
->resync_start
= MaxSector
;
2876 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2877 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2879 info
->array
.major_version
= -1;
2880 info
->array
.minor_version
= -2;
2881 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
2882 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2883 uuid_from_super_imsm(st
, info
->uuid
);
2887 for (i
=0; i
<map_disks
; i
++) {
2889 if (i
< info
->array
.raid_disks
) {
2890 struct imsm_disk
*dsk
;
2891 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2892 dsk
= get_imsm_disk(super
, j
);
2893 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2900 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2901 int failed
, int look_in_map
);
2903 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2907 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2909 if (is_gen_migration(dev
)) {
2912 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2914 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2915 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2916 if (map2
->map_state
!= map_state
) {
2917 map2
->map_state
= map_state
;
2918 super
->updates_pending
++;
2924 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2928 for (d
= super
->missing
; d
; d
= d
->next
)
2929 if (d
->index
== index
)
2934 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2936 struct intel_super
*super
= st
->sb
;
2937 struct imsm_disk
*disk
;
2938 int map_disks
= info
->array
.raid_disks
;
2939 int max_enough
= -1;
2941 struct imsm_super
*mpb
;
2943 if (super
->current_vol
>= 0) {
2944 getinfo_super_imsm_volume(st
, info
, map
);
2947 memset(info
, 0, sizeof(*info
));
2949 /* Set raid_disks to zero so that Assemble will always pull in valid
2952 info
->array
.raid_disks
= 0;
2953 info
->array
.level
= LEVEL_CONTAINER
;
2954 info
->array
.layout
= 0;
2955 info
->array
.md_minor
= -1;
2956 info
->array
.ctime
= 0; /* N/A for imsm */
2957 info
->array
.utime
= 0;
2958 info
->array
.chunk_size
= 0;
2960 info
->disk
.major
= 0;
2961 info
->disk
.minor
= 0;
2962 info
->disk
.raid_disk
= -1;
2963 info
->reshape_active
= 0;
2964 info
->array
.major_version
= -1;
2965 info
->array
.minor_version
= -2;
2966 strcpy(info
->text_version
, "imsm");
2967 info
->safe_mode_delay
= 0;
2968 info
->disk
.number
= -1;
2969 info
->disk
.state
= 0;
2971 info
->recovery_start
= MaxSector
;
2972 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2974 /* do we have the all the insync disks that we expect? */
2975 mpb
= super
->anchor
;
2977 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2978 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2979 int failed
, enough
, j
, missing
= 0;
2980 struct imsm_map
*map
;
2983 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2984 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2985 map
= get_imsm_map(dev
, MAP_0
);
2987 /* any newly missing disks?
2988 * (catches single-degraded vs double-degraded)
2990 for (j
= 0; j
< map
->num_members
; j
++) {
2991 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2992 __u32 idx
= ord_to_idx(ord
);
2994 if (!(ord
& IMSM_ORD_REBUILD
) &&
2995 get_imsm_missing(super
, idx
)) {
3001 if (state
== IMSM_T_STATE_FAILED
)
3003 else if (state
== IMSM_T_STATE_DEGRADED
&&
3004 (state
!= map
->map_state
|| missing
))
3006 else /* we're normal, or already degraded */
3008 if (is_gen_migration(dev
) && missing
) {
3009 /* during general migration we need all disks
3010 * that process is running on.
3011 * No new missing disk is allowed.
3015 /* no more checks necessary
3019 /* in the missing/failed disk case check to see
3020 * if at least one array is runnable
3022 max_enough
= max(max_enough
, enough
);
3024 dprintf("enough: %d\n", max_enough
);
3025 info
->container_enough
= max_enough
;
3028 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3030 disk
= &super
->disks
->disk
;
3031 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3032 info
->component_size
= reserved
;
3033 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3034 /* we don't change info->disk.raid_disk here because
3035 * this state will be finalized in mdmon after we have
3036 * found the 'most fresh' version of the metadata
3038 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3039 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3042 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3043 * ->compare_super may have updated the 'num_raid_devs' field for spares
3045 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3046 uuid_from_super_imsm(st
, info
->uuid
);
3048 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3050 /* I don't know how to compute 'map' on imsm, so use safe default */
3053 for (i
= 0; i
< map_disks
; i
++)
3059 /* allocates memory and fills disk in mdinfo structure
3060 * for each disk in array */
3061 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3063 struct mdinfo
*mddev
= NULL
;
3064 struct intel_super
*super
= st
->sb
;
3065 struct imsm_disk
*disk
;
3068 if (!super
|| !super
->disks
)
3071 mddev
= xcalloc(1, sizeof(*mddev
));
3075 tmp
= xcalloc(1, sizeof(*tmp
));
3077 tmp
->next
= mddev
->devs
;
3079 tmp
->disk
.number
= count
++;
3080 tmp
->disk
.major
= dl
->major
;
3081 tmp
->disk
.minor
= dl
->minor
;
3082 tmp
->disk
.state
= is_configured(disk
) ?
3083 (1 << MD_DISK_ACTIVE
) : 0;
3084 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3085 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3086 tmp
->disk
.raid_disk
= -1;
3092 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3093 char *update
, char *devname
, int verbose
,
3094 int uuid_set
, char *homehost
)
3096 /* For 'assemble' and 'force' we need to return non-zero if any
3097 * change was made. For others, the return value is ignored.
3098 * Update options are:
3099 * force-one : This device looks a bit old but needs to be included,
3100 * update age info appropriately.
3101 * assemble: clear any 'faulty' flag to allow this device to
3103 * force-array: Array is degraded but being forced, mark it clean
3104 * if that will be needed to assemble it.
3106 * newdev: not used ????
3107 * grow: Array has gained a new device - this is currently for
3109 * resync: mark as dirty so a resync will happen.
3110 * name: update the name - preserving the homehost
3111 * uuid: Change the uuid of the array to match watch is given
3113 * Following are not relevant for this imsm:
3114 * sparc2.2 : update from old dodgey metadata
3115 * super-minor: change the preferred_minor number
3116 * summaries: update redundant counters.
3117 * homehost: update the recorded homehost
3118 * _reshape_progress: record new reshape_progress position.
3121 struct intel_super
*super
= st
->sb
;
3122 struct imsm_super
*mpb
;
3124 /* we can only update container info */
3125 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3128 mpb
= super
->anchor
;
3130 if (strcmp(update
, "uuid") == 0) {
3131 /* We take this to mean that the family_num should be updated.
3132 * However that is much smaller than the uuid so we cannot really
3133 * allow an explicit uuid to be given. And it is hard to reliably
3135 * So if !uuid_set we know the current uuid is random and just used
3136 * the first 'int' and copy it to the other 3 positions.
3137 * Otherwise we require the 4 'int's to be the same as would be the
3138 * case if we are using a random uuid. So an explicit uuid will be
3139 * accepted as long as all for ints are the same... which shouldn't hurt
3142 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3145 if (info
->uuid
[0] != info
->uuid
[1] ||
3146 info
->uuid
[1] != info
->uuid
[2] ||
3147 info
->uuid
[2] != info
->uuid
[3])
3153 mpb
->orig_family_num
= info
->uuid
[0];
3154 } else if (strcmp(update
, "assemble") == 0)
3159 /* successful update? recompute checksum */
3161 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3166 static size_t disks_to_mpb_size(int disks
)
3170 size
= sizeof(struct imsm_super
);
3171 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3172 size
+= 2 * sizeof(struct imsm_dev
);
3173 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3174 size
+= (4 - 2) * sizeof(struct imsm_map
);
3175 /* 4 possible disk_ord_tbl's */
3176 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3181 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3182 unsigned long long data_offset
)
3184 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3187 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3190 static void free_devlist(struct intel_super
*super
)
3192 struct intel_dev
*dv
;
3194 while (super
->devlist
) {
3195 dv
= super
->devlist
->next
;
3196 free(super
->devlist
->dev
);
3197 free(super
->devlist
);
3198 super
->devlist
= dv
;
3202 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3204 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3207 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3211 * 0 same, or first was empty, and second was copied
3212 * 1 second had wrong number
3214 * 3 wrong other info
3216 struct intel_super
*first
= st
->sb
;
3217 struct intel_super
*sec
= tst
->sb
;
3224 /* in platform dependent environment test if the disks
3225 * use the same Intel hba
3226 * If not on Intel hba at all, allow anything.
3228 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3229 if (first
->hba
->type
!= sec
->hba
->type
) {
3231 "HBAs of devices do not match %s != %s\n",
3232 get_sys_dev_type(first
->hba
->type
),
3233 get_sys_dev_type(sec
->hba
->type
));
3236 if (first
->orom
!= sec
->orom
) {
3238 "HBAs of devices do not match %s != %s\n",
3239 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3244 /* if an anchor does not have num_raid_devs set then it is a free
3247 if (first
->anchor
->num_raid_devs
> 0 &&
3248 sec
->anchor
->num_raid_devs
> 0) {
3249 /* Determine if these disks might ever have been
3250 * related. Further disambiguation can only take place
3251 * in load_super_imsm_all
3253 __u32 first_family
= first
->anchor
->orig_family_num
;
3254 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3256 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3257 MAX_SIGNATURE_LENGTH
) != 0)
3260 if (first_family
== 0)
3261 first_family
= first
->anchor
->family_num
;
3262 if (sec_family
== 0)
3263 sec_family
= sec
->anchor
->family_num
;
3265 if (first_family
!= sec_family
)
3270 /* if 'first' is a spare promote it to a populated mpb with sec's
3273 if (first
->anchor
->num_raid_devs
== 0 &&
3274 sec
->anchor
->num_raid_devs
> 0) {
3276 struct intel_dev
*dv
;
3277 struct imsm_dev
*dev
;
3279 /* we need to copy raid device info from sec if an allocation
3280 * fails here we don't associate the spare
3282 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3283 dv
= xmalloc(sizeof(*dv
));
3284 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3287 dv
->next
= first
->devlist
;
3288 first
->devlist
= dv
;
3290 if (i
< sec
->anchor
->num_raid_devs
) {
3291 /* allocation failure */
3292 free_devlist(first
);
3293 pr_err("imsm: failed to associate spare\n");
3296 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3297 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3298 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3299 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3300 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3301 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3307 static void fd2devname(int fd
, char *name
)
3311 char dname
[PATH_MAX
];
3316 if (fstat(fd
, &st
) != 0)
3318 sprintf(path
, "/sys/dev/block/%d:%d",
3319 major(st
.st_rdev
), minor(st
.st_rdev
));
3321 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3326 nm
= strrchr(dname
, '/');
3329 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3333 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3335 static int imsm_read_serial(int fd
, char *devname
,
3336 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3338 unsigned char scsi_serial
[255];
3347 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3349 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3351 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3352 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3353 fd2devname(fd
, (char *) serial
);
3359 pr_err("Failed to retrieve serial for %s\n",
3364 rsp_len
= scsi_serial
[3];
3367 pr_err("Failed to retrieve serial for %s\n",
3371 rsp_buf
= (char *) &scsi_serial
[4];
3373 /* trim all whitespace and non-printable characters and convert
3376 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3379 /* ':' is reserved for use in placeholder serial
3380 * numbers for missing disks
3388 len
= dest
- rsp_buf
;
3391 /* truncate leading characters */
3392 if (len
> MAX_RAID_SERIAL_LEN
) {
3393 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3394 len
= MAX_RAID_SERIAL_LEN
;
3397 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3398 memcpy(serial
, dest
, len
);
3403 static int serialcmp(__u8
*s1
, __u8
*s2
)
3405 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3408 static void serialcpy(__u8
*dest
, __u8
*src
)
3410 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3413 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3417 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3418 if (serialcmp(dl
->serial
, serial
) == 0)
3424 static struct imsm_disk
*
3425 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3429 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3430 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3432 if (serialcmp(disk
->serial
, serial
) == 0) {
3443 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3445 struct imsm_disk
*disk
;
3450 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3452 rv
= imsm_read_serial(fd
, devname
, serial
);
3457 dl
= xcalloc(1, sizeof(*dl
));
3460 dl
->major
= major(stb
.st_rdev
);
3461 dl
->minor
= minor(stb
.st_rdev
);
3462 dl
->next
= super
->disks
;
3463 dl
->fd
= keep_fd
? fd
: -1;
3464 assert(super
->disks
== NULL
);
3466 serialcpy(dl
->serial
, serial
);
3469 fd2devname(fd
, name
);
3471 dl
->devname
= xstrdup(devname
);
3473 dl
->devname
= xstrdup(name
);
3475 /* look up this disk's index in the current anchor */
3476 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3479 /* only set index on disks that are a member of a
3480 * populated contianer, i.e. one with raid_devs
3482 if (is_failed(&dl
->disk
))
3484 else if (is_spare(&dl
->disk
))
3492 /* When migrating map0 contains the 'destination' state while map1
3493 * contains the current state. When not migrating map0 contains the
3494 * current state. This routine assumes that map[0].map_state is set to
3495 * the current array state before being called.
3497 * Migration is indicated by one of the following states
3498 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3499 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3500 * map1state=unitialized)
3501 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3503 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3504 * map1state=degraded)
3505 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3508 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3509 __u8 to_state
, int migr_type
)
3511 struct imsm_map
*dest
;
3512 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3514 dev
->vol
.migr_state
= 1;
3515 set_migr_type(dev
, migr_type
);
3516 dev
->vol
.curr_migr_unit
= 0;
3517 dest
= get_imsm_map(dev
, MAP_1
);
3519 /* duplicate and then set the target end state in map[0] */
3520 memcpy(dest
, src
, sizeof_imsm_map(src
));
3521 if ((migr_type
== MIGR_REBUILD
) ||
3522 (migr_type
== MIGR_GEN_MIGR
)) {
3526 for (i
= 0; i
< src
->num_members
; i
++) {
3527 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3528 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3532 if (migr_type
== MIGR_GEN_MIGR
)
3533 /* Clear migration record */
3534 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3536 src
->map_state
= to_state
;
3539 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3542 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3543 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3547 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3548 * completed in the last migration.
3550 * FIXME add support for raid-level-migration
3552 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3553 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3554 /* when final map state is other than expected
3555 * merge maps (not for migration)
3559 for (i
= 0; i
< prev
->num_members
; i
++)
3560 for (j
= 0; j
< map
->num_members
; j
++)
3561 /* during online capacity expansion
3562 * disks position can be changed
3563 * if takeover is used
3565 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3566 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3567 map
->disk_ord_tbl
[j
] |=
3568 prev
->disk_ord_tbl
[i
];
3571 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3572 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3575 dev
->vol
.migr_state
= 0;
3576 set_migr_type(dev
, 0);
3577 dev
->vol
.curr_migr_unit
= 0;
3578 map
->map_state
= map_state
;
3582 static int parse_raid_devices(struct intel_super
*super
)
3585 struct imsm_dev
*dev_new
;
3586 size_t len
, len_migr
;
3588 size_t space_needed
= 0;
3589 struct imsm_super
*mpb
= super
->anchor
;
3591 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3592 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3593 struct intel_dev
*dv
;
3595 len
= sizeof_imsm_dev(dev_iter
, 0);
3596 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3598 space_needed
+= len_migr
- len
;
3600 dv
= xmalloc(sizeof(*dv
));
3601 if (max_len
< len_migr
)
3603 if (max_len
> len_migr
)
3604 space_needed
+= max_len
- len_migr
;
3605 dev_new
= xmalloc(max_len
);
3606 imsm_copy_dev(dev_new
, dev_iter
);
3609 dv
->next
= super
->devlist
;
3610 super
->devlist
= dv
;
3613 /* ensure that super->buf is large enough when all raid devices
3616 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3619 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3620 if (posix_memalign(&buf
, 512, len
) != 0)
3623 memcpy(buf
, super
->buf
, super
->len
);
3624 memset(buf
+ super
->len
, 0, len
- super
->len
);
3633 /* retrieve a pointer to the bbm log which starts after all raid devices */
3634 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3638 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3640 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3646 /*******************************************************************************
3647 * Function: check_mpb_migr_compatibility
3648 * Description: Function checks for unsupported migration features:
3649 * - migration optimization area (pba_of_lba0)
3650 * - descending reshape (ascending_migr)
3652 * super : imsm metadata information
3654 * 0 : migration is compatible
3655 * -1 : migration is not compatible
3656 ******************************************************************************/
3657 int check_mpb_migr_compatibility(struct intel_super
*super
)
3659 struct imsm_map
*map0
, *map1
;
3660 struct migr_record
*migr_rec
= super
->migr_rec
;
3663 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3664 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3667 dev_iter
->vol
.migr_state
== 1 &&
3668 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3669 /* This device is migrating */
3670 map0
= get_imsm_map(dev_iter
, MAP_0
);
3671 map1
= get_imsm_map(dev_iter
, MAP_1
);
3672 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3673 /* migration optimization area was used */
3675 if (migr_rec
->ascending_migr
== 0
3676 && migr_rec
->dest_depth_per_unit
> 0)
3677 /* descending reshape not supported yet */
3684 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3686 /* load_imsm_mpb - read matrix metadata
3687 * allocates super->mpb to be freed by free_imsm
3689 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3691 unsigned long long dsize
;
3692 unsigned long long sectors
;
3694 struct imsm_super
*anchor
;
3697 get_dev_size(fd
, NULL
, &dsize
);
3700 pr_err("%s: device to small for imsm\n",
3705 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3707 pr_err("Cannot seek to anchor block on %s: %s\n",
3708 devname
, strerror(errno
));
3712 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3714 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
3717 if (read(fd
, anchor
, 512) != 512) {
3719 pr_err("Cannot read anchor block on %s: %s\n",
3720 devname
, strerror(errno
));
3725 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3727 pr_err("no IMSM anchor on %s\n", devname
);
3732 __free_imsm(super
, 0);
3733 /* reload capability and hba */
3735 /* capability and hba must be updated with new super allocation */
3736 find_intel_hba_capability(fd
, super
, devname
);
3737 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3738 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3740 pr_err("unable to allocate %zu byte mpb buffer\n",
3745 memcpy(super
->buf
, anchor
, 512);
3747 sectors
= mpb_sectors(anchor
) - 1;
3750 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3751 pr_err("could not allocate migr_rec buffer\n");
3755 super
->clean_migration_record_by_mdmon
= 0;
3758 check_sum
= __gen_imsm_checksum(super
->anchor
);
3759 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3761 pr_err("IMSM checksum %x != %x on %s\n",
3763 __le32_to_cpu(super
->anchor
->check_sum
),
3771 /* read the extended mpb */
3772 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3774 pr_err("Cannot seek to extended mpb on %s: %s\n",
3775 devname
, strerror(errno
));
3779 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3781 pr_err("Cannot read extended mpb on %s: %s\n",
3782 devname
, strerror(errno
));
3786 check_sum
= __gen_imsm_checksum(super
->anchor
);
3787 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3789 pr_err("IMSM checksum %x != %x on %s\n",
3790 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3795 /* FIXME the BBM log is disk specific so we cannot use this global
3796 * buffer for all disks. Ok for now since we only look at the global
3797 * bbm_log_size parameter to gate assembly
3799 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3804 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3806 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3807 static void clear_hi(struct intel_super
*super
)
3809 struct imsm_super
*mpb
= super
->anchor
;
3811 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3813 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3814 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3815 disk
->total_blocks_hi
= 0;
3817 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3818 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3821 for (n
= 0; n
< 2; ++n
) {
3822 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3825 map
->pba_of_lba0_hi
= 0;
3826 map
->blocks_per_member_hi
= 0;
3827 map
->num_data_stripes_hi
= 0;
3833 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3837 err
= load_imsm_mpb(fd
, super
, devname
);
3840 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3843 err
= parse_raid_devices(super
);
3848 static void __free_imsm_disk(struct dl
*d
)
3860 static void free_imsm_disks(struct intel_super
*super
)
3864 while (super
->disks
) {
3866 super
->disks
= d
->next
;
3867 __free_imsm_disk(d
);
3869 while (super
->disk_mgmt_list
) {
3870 d
= super
->disk_mgmt_list
;
3871 super
->disk_mgmt_list
= d
->next
;
3872 __free_imsm_disk(d
);
3874 while (super
->missing
) {
3876 super
->missing
= d
->next
;
3877 __free_imsm_disk(d
);
3882 /* free all the pieces hanging off of a super pointer */
3883 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3885 struct intel_hba
*elem
, *next
;
3891 /* unlink capability description */
3893 if (super
->migr_rec_buf
) {
3894 free(super
->migr_rec_buf
);
3895 super
->migr_rec_buf
= NULL
;
3898 free_imsm_disks(super
);
3899 free_devlist(super
);
3903 free((void *)elem
->path
);
3911 static void free_imsm(struct intel_super
*super
)
3913 __free_imsm(super
, 1);
3917 static void free_super_imsm(struct supertype
*st
)
3919 struct intel_super
*super
= st
->sb
;
3928 static struct intel_super
*alloc_super(void)
3930 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
3932 super
->current_vol
= -1;
3933 super
->create_offset
= ~((unsigned long long) 0);
3938 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3940 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3942 struct sys_dev
*hba_name
;
3945 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3950 hba_name
= find_disk_attached_hba(fd
, NULL
);
3953 pr_err("%s is not attached to Intel(R) RAID controller.\n",
3957 rv
= attach_hba_to_super(super
, hba_name
);
3960 struct intel_hba
*hba
= super
->hba
;
3962 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
3963 " but the container is assigned to Intel(R) %s %s (",
3965 get_sys_dev_type(hba_name
->type
),
3966 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
3967 hba_name
->pci_id
? : "Err!",
3968 get_sys_dev_type(super
->hba
->type
),
3969 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
3972 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3974 fprintf(stderr
, ", ");
3977 fprintf(stderr
, ").\n"
3978 " Mixing devices attached to different %s is not allowed.\n",
3979 hba_name
->type
== SYS_DEV_VMD
? "VMD domains" : "controllers");
3983 super
->orom
= find_imsm_capability(hba_name
);
3990 /* find_missing - helper routine for load_super_imsm_all that identifies
3991 * disks that have disappeared from the system. This routine relies on
3992 * the mpb being uptodate, which it is at load time.
3994 static int find_missing(struct intel_super
*super
)
3997 struct imsm_super
*mpb
= super
->anchor
;
3999 struct imsm_disk
*disk
;
4001 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4002 disk
= __get_imsm_disk(mpb
, i
);
4003 dl
= serial_to_dl(disk
->serial
, super
);
4007 dl
= xmalloc(sizeof(*dl
));
4011 dl
->devname
= xstrdup("missing");
4013 serialcpy(dl
->serial
, disk
->serial
);
4016 dl
->next
= super
->missing
;
4017 super
->missing
= dl
;
4024 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4026 struct intel_disk
*idisk
= disk_list
;
4029 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4031 idisk
= idisk
->next
;
4037 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4038 struct intel_super
*super
,
4039 struct intel_disk
**disk_list
)
4041 struct imsm_disk
*d
= &super
->disks
->disk
;
4042 struct imsm_super
*mpb
= super
->anchor
;
4045 for (i
= 0; i
< tbl_size
; i
++) {
4046 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4047 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4049 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4050 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4051 dprintf("mpb from %d:%d matches %d:%d\n",
4052 super
->disks
->major
,
4053 super
->disks
->minor
,
4054 table
[i
]->disks
->major
,
4055 table
[i
]->disks
->minor
);
4059 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4060 is_configured(d
) == is_configured(tbl_d
)) &&
4061 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4062 /* current version of the mpb is a
4063 * better candidate than the one in
4064 * super_table, but copy over "cross
4065 * generational" status
4067 struct intel_disk
*idisk
;
4069 dprintf("mpb from %d:%d replaces %d:%d\n",
4070 super
->disks
->major
,
4071 super
->disks
->minor
,
4072 table
[i
]->disks
->major
,
4073 table
[i
]->disks
->minor
);
4075 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4076 if (idisk
&& is_failed(&idisk
->disk
))
4077 tbl_d
->status
|= FAILED_DISK
;
4080 struct intel_disk
*idisk
;
4081 struct imsm_disk
*disk
;
4083 /* tbl_mpb is more up to date, but copy
4084 * over cross generational status before
4087 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4088 if (disk
&& is_failed(disk
))
4089 d
->status
|= FAILED_DISK
;
4091 idisk
= disk_list_get(d
->serial
, *disk_list
);
4094 if (disk
&& is_configured(disk
))
4095 idisk
->disk
.status
|= CONFIGURED_DISK
;
4098 dprintf("mpb from %d:%d prefer %d:%d\n",
4099 super
->disks
->major
,
4100 super
->disks
->minor
,
4101 table
[i
]->disks
->major
,
4102 table
[i
]->disks
->minor
);
4110 table
[tbl_size
++] = super
;
4114 /* update/extend the merged list of imsm_disk records */
4115 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4116 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4117 struct intel_disk
*idisk
;
4119 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4121 idisk
->disk
.status
|= disk
->status
;
4122 if (is_configured(&idisk
->disk
) ||
4123 is_failed(&idisk
->disk
))
4124 idisk
->disk
.status
&= ~(SPARE_DISK
);
4126 idisk
= xcalloc(1, sizeof(*idisk
));
4127 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4128 idisk
->disk
= *disk
;
4129 idisk
->next
= *disk_list
;
4133 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4140 static struct intel_super
*
4141 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4144 struct imsm_super
*mpb
= super
->anchor
;
4148 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4149 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4150 struct intel_disk
*idisk
;
4152 idisk
= disk_list_get(disk
->serial
, disk_list
);
4154 if (idisk
->owner
== owner
||
4155 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4158 dprintf("'%.16s' owner %d != %d\n",
4159 disk
->serial
, idisk
->owner
,
4162 dprintf("unknown disk %x [%d]: %.16s\n",
4163 __le32_to_cpu(mpb
->family_num
), i
,
4169 if (ok_count
== mpb
->num_disks
)
4174 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4176 struct intel_super
*s
;
4178 for (s
= super_list
; s
; s
= s
->next
) {
4179 if (family_num
!= s
->anchor
->family_num
)
4181 pr_err("Conflict, offlining family %#x on '%s'\n",
4182 __le32_to_cpu(family_num
), s
->disks
->devname
);
4186 static struct intel_super
*
4187 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4189 struct intel_super
*super_table
[len
];
4190 struct intel_disk
*disk_list
= NULL
;
4191 struct intel_super
*champion
, *spare
;
4192 struct intel_super
*s
, **del
;
4197 memset(super_table
, 0, sizeof(super_table
));
4198 for (s
= *super_list
; s
; s
= s
->next
)
4199 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4201 for (i
= 0; i
< tbl_size
; i
++) {
4202 struct imsm_disk
*d
;
4203 struct intel_disk
*idisk
;
4204 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4207 d
= &s
->disks
->disk
;
4209 /* 'd' must appear in merged disk list for its
4210 * configuration to be valid
4212 idisk
= disk_list_get(d
->serial
, disk_list
);
4213 if (idisk
&& idisk
->owner
== i
)
4214 s
= validate_members(s
, disk_list
, i
);
4219 dprintf("marking family: %#x from %d:%d offline\n",
4221 super_table
[i
]->disks
->major
,
4222 super_table
[i
]->disks
->minor
);
4226 /* This is where the mdadm implementation differs from the Windows
4227 * driver which has no strict concept of a container. We can only
4228 * assemble one family from a container, so when returning a prodigal
4229 * array member to this system the code will not be able to disambiguate
4230 * the container contents that should be assembled ("foreign" versus
4231 * "local"). It requires user intervention to set the orig_family_num
4232 * to a new value to establish a new container. The Windows driver in
4233 * this situation fixes up the volume name in place and manages the
4234 * foreign array as an independent entity.
4239 for (i
= 0; i
< tbl_size
; i
++) {
4240 struct intel_super
*tbl_ent
= super_table
[i
];
4246 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4251 if (s
&& !is_spare
) {
4252 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4254 } else if (!s
&& !is_spare
)
4267 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4268 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4270 /* collect all dl's onto 'champion', and update them to
4271 * champion's version of the status
4273 for (s
= *super_list
; s
; s
= s
->next
) {
4274 struct imsm_super
*mpb
= champion
->anchor
;
4275 struct dl
*dl
= s
->disks
;
4280 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4282 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4283 struct imsm_disk
*disk
;
4285 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4288 /* only set index on disks that are a member of
4289 * a populated contianer, i.e. one with
4292 if (is_failed(&dl
->disk
))
4294 else if (is_spare(&dl
->disk
))
4300 if (i
>= mpb
->num_disks
) {
4301 struct intel_disk
*idisk
;
4303 idisk
= disk_list_get(dl
->serial
, disk_list
);
4304 if (idisk
&& is_spare(&idisk
->disk
) &&
4305 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4313 dl
->next
= champion
->disks
;
4314 champion
->disks
= dl
;
4318 /* delete 'champion' from super_list */
4319 for (del
= super_list
; *del
; ) {
4320 if (*del
== champion
) {
4321 *del
= (*del
)->next
;
4324 del
= &(*del
)->next
;
4326 champion
->next
= NULL
;
4330 struct intel_disk
*idisk
= disk_list
;
4332 disk_list
= disk_list
->next
;
4340 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4341 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4342 int major
, int minor
, int keep_fd
);
4344 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4345 int *max
, int keep_fd
);
4347 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4348 char *devname
, struct md_list
*devlist
,
4351 struct intel_super
*super_list
= NULL
;
4352 struct intel_super
*super
= NULL
;
4357 /* 'fd' is an opened container */
4358 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4360 /* get super block from devlist devices */
4361 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4364 /* all mpbs enter, maybe one leaves */
4365 super
= imsm_thunderdome(&super_list
, i
);
4371 if (find_missing(super
) != 0) {
4377 /* load migration record */
4378 err
= load_imsm_migr_rec(super
, NULL
);
4380 /* migration is in progress,
4381 * but migr_rec cannot be loaded,
4387 /* Check migration compatibility */
4388 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4389 pr_err("Unsupported migration detected");
4391 fprintf(stderr
, " on %s\n", devname
);
4393 fprintf(stderr
, " (IMSM).\n");
4402 while (super_list
) {
4403 struct intel_super
*s
= super_list
;
4405 super_list
= super_list
->next
;
4414 strcpy(st
->container_devnm
, fd2devnm(fd
));
4416 st
->container_devnm
[0] = 0;
4417 if (err
== 0 && st
->ss
== NULL
) {
4418 st
->ss
= &super_imsm
;
4419 st
->minor_version
= 0;
4420 st
->max_devs
= IMSM_MAX_DEVICES
;
4426 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4427 int *max
, int keep_fd
)
4429 struct md_list
*tmpdev
;
4433 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4434 if (tmpdev
->used
!= 1)
4436 if (tmpdev
->container
== 1) {
4438 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4440 pr_err("cannot open device %s: %s\n",
4441 tmpdev
->devname
, strerror(errno
));
4445 err
= get_sra_super_block(fd
, super_list
,
4446 tmpdev
->devname
, &lmax
,
4455 int major
= major(tmpdev
->st_rdev
);
4456 int minor
= minor(tmpdev
->st_rdev
);
4457 err
= get_super_block(super_list
,
4474 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4475 int major
, int minor
, int keep_fd
)
4477 struct intel_super
*s
= NULL
;
4489 sprintf(nm
, "%d:%d", major
, minor
);
4490 dfd
= dev_open(nm
, O_RDWR
);
4496 find_intel_hba_capability(dfd
, s
, devname
);
4497 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4499 /* retry the load if we might have raced against mdmon */
4500 if (err
== 3 && devnm
&& mdmon_running(devnm
))
4501 for (retry
= 0; retry
< 3; retry
++) {
4503 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4509 s
->next
= *super_list
;
4517 if ((dfd
>= 0) && (!keep_fd
))
4524 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4531 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4535 if (sra
->array
.major_version
!= -1 ||
4536 sra
->array
.minor_version
!= -2 ||
4537 strcmp(sra
->text_version
, "imsm") != 0) {
4542 devnm
= fd2devnm(fd
);
4543 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4544 if (get_super_block(super_list
, devnm
, devname
,
4545 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4556 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4558 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4562 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4564 struct intel_super
*super
;
4568 if (test_partition(fd
))
4569 /* IMSM not allowed on partitions */
4572 free_super_imsm(st
);
4574 super
= alloc_super();
4575 /* Load hba and capabilities if they exist.
4576 * But do not preclude loading metadata in case capabilities or hba are
4577 * non-compliant and ignore_hw_compat is set.
4579 rv
= find_intel_hba_capability(fd
, super
, devname
);
4580 /* no orom/efi or non-intel hba of the disk */
4581 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4583 pr_err("No OROM/EFI properties for %s\n", devname
);
4587 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4589 /* retry the load if we might have raced against mdmon */
4591 struct mdstat_ent
*mdstat
= mdstat_by_component(fd2devnm(fd
));
4593 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
4594 for (retry
= 0; retry
< 3; retry
++) {
4596 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4602 free_mdstat(mdstat
);
4607 pr_err("Failed to load all information sections on %s\n", devname
);
4613 if (st
->ss
== NULL
) {
4614 st
->ss
= &super_imsm
;
4615 st
->minor_version
= 0;
4616 st
->max_devs
= IMSM_MAX_DEVICES
;
4619 /* load migration record */
4620 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4621 /* Check for unsupported migration features */
4622 if (check_mpb_migr_compatibility(super
) != 0) {
4623 pr_err("Unsupported migration detected");
4625 fprintf(stderr
, " on %s\n", devname
);
4627 fprintf(stderr
, " (IMSM).\n");
4635 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4637 if (info
->level
== 1)
4639 return info
->chunk_size
>> 9;
4642 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4643 unsigned long long size
)
4645 if (info
->level
== 1)
4648 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4651 static void imsm_update_version_info(struct intel_super
*super
)
4653 /* update the version and attributes */
4654 struct imsm_super
*mpb
= super
->anchor
;
4656 struct imsm_dev
*dev
;
4657 struct imsm_map
*map
;
4660 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4661 dev
= get_imsm_dev(super
, i
);
4662 map
= get_imsm_map(dev
, MAP_0
);
4663 if (__le32_to_cpu(dev
->size_high
) > 0)
4664 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4666 /* FIXME detect when an array spans a port multiplier */
4668 mpb
->attributes
|= MPB_ATTRIB_PM
;
4671 if (mpb
->num_raid_devs
> 1 ||
4672 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4673 version
= MPB_VERSION_ATTRIBS
;
4674 switch (get_imsm_raid_level(map
)) {
4675 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4676 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4677 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4678 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4681 if (map
->num_members
>= 5)
4682 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4683 else if (dev
->status
== DEV_CLONE_N_GO
)
4684 version
= MPB_VERSION_CNG
;
4685 else if (get_imsm_raid_level(map
) == 5)
4686 version
= MPB_VERSION_RAID5
;
4687 else if (map
->num_members
>= 3)
4688 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4689 else if (get_imsm_raid_level(map
) == 1)
4690 version
= MPB_VERSION_RAID1
;
4692 version
= MPB_VERSION_RAID0
;
4694 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4698 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4700 struct imsm_super
*mpb
= super
->anchor
;
4701 char *reason
= NULL
;
4704 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4705 reason
= "must be 16 characters or less";
4707 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4708 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4710 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4711 reason
= "already exists";
4716 if (reason
&& !quiet
)
4717 pr_err("imsm volume name %s\n", reason
);
4722 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4723 unsigned long long size
, char *name
,
4724 char *homehost
, int *uuid
,
4725 long long data_offset
)
4727 /* We are creating a volume inside a pre-existing container.
4728 * so st->sb is already set.
4730 struct intel_super
*super
= st
->sb
;
4731 struct imsm_super
*mpb
= super
->anchor
;
4732 struct intel_dev
*dv
;
4733 struct imsm_dev
*dev
;
4734 struct imsm_vol
*vol
;
4735 struct imsm_map
*map
;
4736 int idx
= mpb
->num_raid_devs
;
4738 unsigned long long array_blocks
;
4739 size_t size_old
, size_new
;
4740 unsigned long long num_data_stripes
;
4742 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4743 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
4747 /* ensure the mpb is large enough for the new data */
4748 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4749 size_new
= disks_to_mpb_size(info
->nr_disks
);
4750 if (size_new
> size_old
) {
4752 size_t size_round
= ROUND_UP(size_new
, 512);
4754 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4755 pr_err("could not allocate new mpb\n");
4758 if (posix_memalign(&super
->migr_rec_buf
, 512,
4759 MIGR_REC_BUF_SIZE
) != 0) {
4760 pr_err("could not allocate migr_rec buffer\n");
4766 memcpy(mpb_new
, mpb
, size_old
);
4769 super
->anchor
= mpb_new
;
4770 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4771 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4773 super
->current_vol
= idx
;
4775 /* handle 'failed_disks' by either:
4776 * a) create dummy disk entries in the table if this the first
4777 * volume in the array. We add them here as this is the only
4778 * opportunity to add them. add_to_super_imsm_volume()
4779 * handles the non-failed disks and continues incrementing
4781 * b) validate that 'failed_disks' matches the current number
4782 * of missing disks if the container is populated
4784 if (super
->current_vol
== 0) {
4786 for (i
= 0; i
< info
->failed_disks
; i
++) {
4787 struct imsm_disk
*disk
;
4790 disk
= __get_imsm_disk(mpb
, i
);
4791 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4792 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4793 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4796 find_missing(super
);
4801 for (d
= super
->missing
; d
; d
= d
->next
)
4803 if (info
->failed_disks
> missing
) {
4804 pr_err("unable to add 'missing' disk to container\n");
4809 if (!check_name(super
, name
, 0))
4811 dv
= xmalloc(sizeof(*dv
));
4812 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4813 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4814 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4815 info
->layout
, info
->chunk_size
,
4817 /* round array size down to closest MB */
4818 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4820 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4821 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4822 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4824 vol
->migr_state
= 0;
4825 set_migr_type(dev
, MIGR_INIT
);
4826 vol
->dirty
= !info
->state
;
4827 vol
->curr_migr_unit
= 0;
4828 map
= get_imsm_map(dev
, MAP_0
);
4829 set_pba_of_lba0(map
, super
->create_offset
);
4830 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4831 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4832 map
->failed_disk_num
= ~0;
4833 if (info
->level
> 0)
4834 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
4835 : IMSM_T_STATE_UNINITIALIZED
);
4837 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4838 IMSM_T_STATE_NORMAL
;
4841 if (info
->level
== 1 && info
->raid_disks
> 2) {
4844 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
4848 map
->raid_level
= info
->level
;
4849 if (info
->level
== 10) {
4850 map
->raid_level
= 1;
4851 map
->num_domains
= info
->raid_disks
/ 2;
4852 } else if (info
->level
== 1)
4853 map
->num_domains
= info
->raid_disks
;
4855 map
->num_domains
= 1;
4857 /* info->size is only int so use the 'size' parameter instead */
4858 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4859 num_data_stripes
/= map
->num_domains
;
4860 set_num_data_stripes(map
, num_data_stripes
);
4862 map
->num_members
= info
->raid_disks
;
4863 for (i
= 0; i
< map
->num_members
; i
++) {
4864 /* initialized in add_to_super */
4865 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4867 mpb
->num_raid_devs
++;
4870 dv
->index
= super
->current_vol
;
4871 dv
->next
= super
->devlist
;
4872 super
->devlist
= dv
;
4874 imsm_update_version_info(super
);
4879 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4880 unsigned long long size
, char *name
,
4881 char *homehost
, int *uuid
,
4882 unsigned long long data_offset
)
4884 /* This is primarily called by Create when creating a new array.
4885 * We will then get add_to_super called for each component, and then
4886 * write_init_super called to write it out to each device.
4887 * For IMSM, Create can create on fresh devices or on a pre-existing
4889 * To create on a pre-existing array a different method will be called.
4890 * This one is just for fresh drives.
4892 struct intel_super
*super
;
4893 struct imsm_super
*mpb
;
4897 if (data_offset
!= INVALID_SECTORS
) {
4898 pr_err("data-offset not supported by imsm\n");
4903 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
,
4907 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4911 super
= alloc_super();
4912 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4917 pr_err("could not allocate superblock\n");
4920 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4921 pr_err("could not allocate migr_rec buffer\n");
4926 memset(super
->buf
, 0, mpb_size
);
4928 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4932 /* zeroing superblock */
4936 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4938 version
= (char *) mpb
->sig
;
4939 strcpy(version
, MPB_SIGNATURE
);
4940 version
+= strlen(MPB_SIGNATURE
);
4941 strcpy(version
, MPB_VERSION_RAID0
);
4947 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4948 int fd
, char *devname
)
4950 struct intel_super
*super
= st
->sb
;
4951 struct imsm_super
*mpb
= super
->anchor
;
4952 struct imsm_disk
*_disk
;
4953 struct imsm_dev
*dev
;
4954 struct imsm_map
*map
;
4958 dev
= get_imsm_dev(super
, super
->current_vol
);
4959 map
= get_imsm_map(dev
, MAP_0
);
4961 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4962 pr_err("%s: Cannot add spare devices to IMSM volume\n",
4968 /* we're doing autolayout so grab the pre-marked (in
4969 * validate_geometry) raid_disk
4971 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4972 if (dl
->raiddisk
== dk
->raid_disk
)
4975 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4976 if (dl
->major
== dk
->major
&&
4977 dl
->minor
== dk
->minor
)
4982 pr_err("%s is not a member of the same container\n", devname
);
4986 /* add a pristine spare to the metadata */
4987 if (dl
->index
< 0) {
4988 dl
->index
= super
->anchor
->num_disks
;
4989 super
->anchor
->num_disks
++;
4991 /* Check the device has not already been added */
4992 slot
= get_imsm_disk_slot(map
, dl
->index
);
4994 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4995 pr_err("%s has been included in this array twice\n",
4999 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5000 dl
->disk
.status
= CONFIGURED_DISK
;
5002 /* update size of 'missing' disks to be at least as large as the
5003 * largest acitve member (we only have dummy missing disks when
5004 * creating the first volume)
5006 if (super
->current_vol
== 0) {
5007 for (df
= super
->missing
; df
; df
= df
->next
) {
5008 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5009 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5010 _disk
= __get_imsm_disk(mpb
, df
->index
);
5015 /* refresh unset/failed slots to point to valid 'missing' entries */
5016 for (df
= super
->missing
; df
; df
= df
->next
)
5017 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5018 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5020 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5022 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5023 if (is_gen_migration(dev
)) {
5024 struct imsm_map
*map2
= get_imsm_map(dev
,
5026 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5027 if ((slot2
< map2
->num_members
) &&
5029 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5032 if ((unsigned)df
->index
==
5034 set_imsm_ord_tbl_ent(map2
,
5040 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5044 /* if we are creating the first raid device update the family number */
5045 if (super
->current_vol
== 0) {
5047 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5049 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5050 if (!_dev
|| !_disk
) {
5051 pr_err("BUG mpb setup error\n");
5057 sum
+= __gen_imsm_checksum(mpb
);
5058 mpb
->family_num
= __cpu_to_le32(sum
);
5059 mpb
->orig_family_num
= mpb
->family_num
;
5061 super
->current_disk
= dl
;
5066 * Function marks disk as spare and restores disk serial
5067 * in case it was previously marked as failed by takeover operation
5069 * -1 : critical error
5070 * 0 : disk is marked as spare but serial is not set
5073 int mark_spare(struct dl
*disk
)
5075 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5082 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5083 /* Restore disk serial number, because takeover marks disk
5084 * as failed and adds to serial ':0' before it becomes
5087 serialcpy(disk
->serial
, serial
);
5088 serialcpy(disk
->disk
.serial
, serial
);
5091 disk
->disk
.status
= SPARE_DISK
;
5097 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5098 int fd
, char *devname
,
5099 unsigned long long data_offset
)
5101 struct intel_super
*super
= st
->sb
;
5103 unsigned long long size
;
5108 /* If we are on an RAID enabled platform check that the disk is
5109 * attached to the raid controller.
5110 * We do not need to test disks attachment for container based additions,
5111 * they shall be already tested when container was created/assembled.
5113 rv
= find_intel_hba_capability(fd
, super
, devname
);
5114 /* no orom/efi or non-intel hba of the disk */
5116 dprintf("capability: %p fd: %d ret: %d\n",
5117 super
->orom
, fd
, rv
);
5121 if (super
->current_vol
>= 0)
5122 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5125 dd
= xcalloc(sizeof(*dd
), 1);
5126 dd
->major
= major(stb
.st_rdev
);
5127 dd
->minor
= minor(stb
.st_rdev
);
5128 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5131 dd
->action
= DISK_ADD
;
5132 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5134 pr_err("failed to retrieve scsi serial, aborting\n");
5139 get_dev_size(fd
, NULL
, &size
);
5140 /* clear migr_rec when adding disk to container */
5141 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5142 if (lseek64(fd
, size
- MIGR_REC_POSITION
, SEEK_SET
) >= 0) {
5143 if (write(fd
, super
->migr_rec_buf
,
5144 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5145 perror("Write migr_rec failed");
5149 serialcpy(dd
->disk
.serial
, dd
->serial
);
5150 set_total_blocks(&dd
->disk
, size
);
5151 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5152 struct imsm_super
*mpb
= super
->anchor
;
5153 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5156 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5157 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5159 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5161 if (st
->update_tail
) {
5162 dd
->next
= super
->disk_mgmt_list
;
5163 super
->disk_mgmt_list
= dd
;
5165 dd
->next
= super
->disks
;
5167 super
->updates_pending
++;
5173 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5175 struct intel_super
*super
= st
->sb
;
5178 /* remove from super works only in mdmon - for communication
5179 * manager - monitor. Check if communication memory buffer
5182 if (!st
->update_tail
) {
5183 pr_err("shall be used in mdmon context only\n");
5186 dd
= xcalloc(1, sizeof(*dd
));
5187 dd
->major
= dk
->major
;
5188 dd
->minor
= dk
->minor
;
5191 dd
->action
= DISK_REMOVE
;
5193 dd
->next
= super
->disk_mgmt_list
;
5194 super
->disk_mgmt_list
= dd
;
5199 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5203 struct imsm_super anchor
;
5204 } spare_record
__attribute__ ((aligned(512)));
5206 /* spare records have their own family number and do not have any defined raid
5209 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5211 struct imsm_super
*mpb
= super
->anchor
;
5212 struct imsm_super
*spare
= &spare_record
.anchor
;
5216 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5217 spare
->generation_num
= __cpu_to_le32(1UL);
5218 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5219 spare
->num_disks
= 1;
5220 spare
->num_raid_devs
= 0;
5221 spare
->cache_size
= mpb
->cache_size
;
5222 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5224 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5225 MPB_SIGNATURE MPB_VERSION_RAID0
);
5227 for (d
= super
->disks
; d
; d
= d
->next
) {
5231 spare
->disk
[0] = d
->disk
;
5232 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5233 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5235 sum
= __gen_imsm_checksum(spare
);
5236 spare
->family_num
= __cpu_to_le32(sum
);
5237 spare
->orig_family_num
= 0;
5238 sum
= __gen_imsm_checksum(spare
);
5239 spare
->check_sum
= __cpu_to_le32(sum
);
5241 if (store_imsm_mpb(d
->fd
, spare
)) {
5242 pr_err("failed for device %d:%d %s\n",
5243 d
->major
, d
->minor
, strerror(errno
));
5255 static int write_super_imsm(struct supertype
*st
, int doclose
)
5257 struct intel_super
*super
= st
->sb
;
5258 struct imsm_super
*mpb
= super
->anchor
;
5264 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5266 int clear_migration_record
= 1;
5268 /* 'generation' is incremented everytime the metadata is written */
5269 generation
= __le32_to_cpu(mpb
->generation_num
);
5271 mpb
->generation_num
= __cpu_to_le32(generation
);
5273 /* fix up cases where previous mdadm releases failed to set
5276 if (mpb
->orig_family_num
== 0)
5277 mpb
->orig_family_num
= mpb
->family_num
;
5279 for (d
= super
->disks
; d
; d
= d
->next
) {
5283 mpb
->disk
[d
->index
] = d
->disk
;
5287 for (d
= super
->missing
; d
; d
= d
->next
) {
5288 mpb
->disk
[d
->index
] = d
->disk
;
5291 mpb
->num_disks
= num_disks
;
5292 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5294 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5295 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5296 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5298 imsm_copy_dev(dev
, dev2
);
5299 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5301 if (is_gen_migration(dev2
))
5302 clear_migration_record
= 0;
5304 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5305 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5307 /* recalculate checksum */
5308 sum
= __gen_imsm_checksum(mpb
);
5309 mpb
->check_sum
= __cpu_to_le32(sum
);
5311 if (super
->clean_migration_record_by_mdmon
) {
5312 clear_migration_record
= 1;
5313 super
->clean_migration_record_by_mdmon
= 0;
5315 if (clear_migration_record
)
5316 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5318 /* write the mpb for disks that compose raid devices */
5319 for (d
= super
->disks
; d
; d
= d
->next
) {
5320 if (d
->index
< 0 || is_failed(&d
->disk
))
5323 if (clear_migration_record
) {
5324 unsigned long long dsize
;
5326 get_dev_size(d
->fd
, NULL
, &dsize
);
5327 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5328 if (write(d
->fd
, super
->migr_rec_buf
,
5329 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5330 perror("Write migr_rec failed");
5334 if (store_imsm_mpb(d
->fd
, mpb
))
5336 "failed for device %d:%d (fd: %d)%s\n",
5338 d
->fd
, strerror(errno
));
5347 return write_super_imsm_spares(super
, doclose
);
5352 static int create_array(struct supertype
*st
, int dev_idx
)
5355 struct imsm_update_create_array
*u
;
5356 struct intel_super
*super
= st
->sb
;
5357 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5358 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5359 struct disk_info
*inf
;
5360 struct imsm_disk
*disk
;
5363 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5364 sizeof(*inf
) * map
->num_members
;
5366 u
->type
= update_create_array
;
5367 u
->dev_idx
= dev_idx
;
5368 imsm_copy_dev(&u
->dev
, dev
);
5369 inf
= get_disk_info(u
);
5370 for (i
= 0; i
< map
->num_members
; i
++) {
5371 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5373 disk
= get_imsm_disk(super
, idx
);
5375 disk
= get_imsm_missing(super
, idx
);
5376 serialcpy(inf
[i
].serial
, disk
->serial
);
5378 append_metadata_update(st
, u
, len
);
5383 static int mgmt_disk(struct supertype
*st
)
5385 struct intel_super
*super
= st
->sb
;
5387 struct imsm_update_add_remove_disk
*u
;
5389 if (!super
->disk_mgmt_list
)
5394 u
->type
= update_add_remove_disk
;
5395 append_metadata_update(st
, u
, len
);
5400 static int write_init_super_imsm(struct supertype
*st
)
5402 struct intel_super
*super
= st
->sb
;
5403 int current_vol
= super
->current_vol
;
5405 /* we are done with current_vol reset it to point st at the container */
5406 super
->current_vol
= -1;
5408 if (st
->update_tail
) {
5409 /* queue the recently created array / added disk
5410 * as a metadata update */
5413 /* determine if we are creating a volume or adding a disk */
5414 if (current_vol
< 0) {
5415 /* in the mgmt (add/remove) disk case we are running
5416 * in mdmon context, so don't close fd's
5418 return mgmt_disk(st
);
5420 rv
= create_array(st
, current_vol
);
5425 for (d
= super
->disks
; d
; d
= d
->next
)
5426 Kill(d
->devname
, NULL
, 0, -1, 1);
5427 return write_super_imsm(st
, 1);
5432 static int store_super_imsm(struct supertype
*st
, int fd
)
5434 struct intel_super
*super
= st
->sb
;
5435 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5441 return store_imsm_mpb(fd
, mpb
);
5447 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5449 return __le32_to_cpu(mpb
->bbm_log_size
);
5453 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5454 int layout
, int raiddisks
, int chunk
,
5455 unsigned long long size
,
5456 unsigned long long data_offset
,
5458 unsigned long long *freesize
,
5462 unsigned long long ldsize
;
5463 struct intel_super
*super
=NULL
;
5466 if (level
!= LEVEL_CONTAINER
)
5471 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5474 pr_err("imsm: Cannot open %s: %s\n",
5475 dev
, strerror(errno
));
5478 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5483 /* capabilities retrieve could be possible
5484 * note that there is no fd for the disks in array.
5486 super
= alloc_super();
5487 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
5491 fd2devname(fd
, str
);
5492 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5493 fd
, str
, super
->orom
, rv
, raiddisks
);
5495 /* no orom/efi or non-intel hba of the disk */
5502 if (raiddisks
> super
->orom
->tds
) {
5504 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
5505 raiddisks
, super
->orom
->tds
);
5509 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5510 (ldsize
>> 9) >> 32 > 0) {
5512 pr_err("%s exceeds maximum platform supported size\n", dev
);
5518 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
5524 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5526 const unsigned long long base_start
= e
[*idx
].start
;
5527 unsigned long long end
= base_start
+ e
[*idx
].size
;
5530 if (base_start
== end
)
5534 for (i
= *idx
; i
< num_extents
; i
++) {
5535 /* extend overlapping extents */
5536 if (e
[i
].start
>= base_start
&&
5537 e
[i
].start
<= end
) {
5540 if (e
[i
].start
+ e
[i
].size
> end
)
5541 end
= e
[i
].start
+ e
[i
].size
;
5542 } else if (e
[i
].start
> end
) {
5548 return end
- base_start
;
5551 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5553 /* build a composite disk with all known extents and generate a new
5554 * 'maxsize' given the "all disks in an array must share a common start
5555 * offset" constraint
5557 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
5561 unsigned long long pos
;
5562 unsigned long long start
= 0;
5563 unsigned long long maxsize
;
5564 unsigned long reserve
;
5566 /* coalesce and sort all extents. also, check to see if we need to
5567 * reserve space between member arrays
5570 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5573 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5576 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5581 while (i
< sum_extents
) {
5582 e
[j
].start
= e
[i
].start
;
5583 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5585 if (e
[j
-1].size
== 0)
5594 unsigned long long esize
;
5596 esize
= e
[i
].start
- pos
;
5597 if (esize
>= maxsize
) {
5602 pos
= e
[i
].start
+ e
[i
].size
;
5604 } while (e
[i
-1].size
);
5610 /* FIXME assumes volume at offset 0 is the first volume in a
5613 if (start_extent
> 0)
5614 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5618 if (maxsize
< reserve
)
5621 super
->create_offset
= ~((unsigned long long) 0);
5622 if (start
+ reserve
> super
->create_offset
)
5623 return 0; /* start overflows create_offset */
5624 super
->create_offset
= start
+ reserve
;
5626 return maxsize
- reserve
;
5629 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5631 if (level
< 0 || level
== 6 || level
== 4)
5634 /* if we have an orom prevent invalid raid levels */
5637 case 0: return imsm_orom_has_raid0(orom
);
5640 return imsm_orom_has_raid1e(orom
);
5641 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5642 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5643 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5646 return 1; /* not on an Intel RAID platform so anything goes */
5652 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5653 int dpa
, int verbose
)
5655 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5656 struct mdstat_ent
*memb
= NULL
;
5659 struct md_list
*dv
= NULL
;
5662 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5663 if (memb
->metadata_version
&&
5664 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5665 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5666 !is_subarray(memb
->metadata_version
+9) &&
5668 struct dev_member
*dev
= memb
->members
;
5670 while(dev
&& (fd
< 0)) {
5671 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5672 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5674 fd
= open(path
, O_RDONLY
, 0);
5675 if ((num
<= 0) || (fd
< 0)) {
5676 pr_vrb(": Cannot open %s: %s\n",
5677 dev
->name
, strerror(errno
));
5683 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5684 struct mdstat_ent
*vol
;
5685 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5686 if ((vol
->active
> 0) &&
5687 vol
->metadata_version
&&
5688 is_container_member(vol
, memb
->devnm
)) {
5693 if (*devlist
&& (found
< dpa
)) {
5694 dv
= xcalloc(1, sizeof(*dv
));
5695 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
5696 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
5699 dv
->next
= *devlist
;
5707 free_mdstat(mdstat
);
5712 static struct md_list
*
5713 get_loop_devices(void)
5716 struct md_list
*devlist
= NULL
;
5717 struct md_list
*dv
= NULL
;
5719 for(i
= 0; i
< 12; i
++) {
5720 dv
= xcalloc(1, sizeof(*dv
));
5721 dv
->devname
= xmalloc(40);
5722 sprintf(dv
->devname
, "/dev/loop%d", i
);
5730 static struct md_list
*
5731 get_devices(const char *hba_path
)
5733 struct md_list
*devlist
= NULL
;
5734 struct md_list
*dv
= NULL
;
5740 devlist
= get_loop_devices();
5743 /* scroll through /sys/dev/block looking for devices attached to
5746 dir
= opendir("/sys/dev/block");
5747 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5752 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5754 path
= devt_to_devpath(makedev(major
, minor
));
5757 if (!path_attached_to_hba(path
, hba_path
)) {
5764 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5766 fd2devname(fd
, buf
);
5769 pr_err("cannot open device: %s\n",
5774 dv
= xcalloc(1, sizeof(*dv
));
5775 dv
->devname
= xstrdup(buf
);
5782 devlist
= devlist
->next
;
5792 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5793 int verbose
, int *found
)
5795 struct md_list
*tmpdev
;
5797 struct supertype
*st
= NULL
;
5799 /* first walk the list of devices to find a consistent set
5800 * that match the criterea, if that is possible.
5801 * We flag the ones we like with 'used'.
5804 st
= match_metadata_desc_imsm("imsm");
5806 pr_vrb(": cannot allocate memory for imsm supertype\n");
5810 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5811 char *devname
= tmpdev
->devname
;
5813 struct supertype
*tst
;
5815 if (tmpdev
->used
> 1)
5817 tst
= dup_super(st
);
5819 pr_vrb(": cannot allocate memory for imsm supertype\n");
5822 tmpdev
->container
= 0;
5823 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5825 dprintf("cannot open device %s: %s\n",
5826 devname
, strerror(errno
));
5828 } else if (fstat(dfd
, &stb
)< 0) {
5830 dprintf("fstat failed for %s: %s\n",
5831 devname
, strerror(errno
));
5833 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5834 dprintf("%s is not a block device.\n",
5837 } else if (must_be_container(dfd
)) {
5838 struct supertype
*cst
;
5839 cst
= super_by_fd(dfd
, NULL
);
5841 dprintf("cannot recognize container type %s\n",
5844 } else if (tst
->ss
!= st
->ss
) {
5845 dprintf("non-imsm container - ignore it: %s\n",
5848 } else if (!tst
->ss
->load_container
||
5849 tst
->ss
->load_container(tst
, dfd
, NULL
))
5852 tmpdev
->container
= 1;
5855 cst
->ss
->free_super(cst
);
5857 tmpdev
->st_rdev
= stb
.st_rdev
;
5858 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5859 dprintf("no RAID superblock on %s\n",
5862 } else if (tst
->ss
->compare_super
== NULL
) {
5863 dprintf("Cannot assemble %s metadata on %s\n",
5864 tst
->ss
->name
, devname
);
5870 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5871 /* Ignore unrecognised devices during auto-assembly */
5876 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5878 if (st
->minor_version
== -1)
5879 st
->minor_version
= tst
->minor_version
;
5881 if (memcmp(info
.uuid
, uuid_zero
,
5882 sizeof(int[4])) == 0) {
5883 /* this is a floating spare. It cannot define
5884 * an array unless there are no more arrays of
5885 * this type to be found. It can be included
5886 * in an array of this type though.
5892 if (st
->ss
!= tst
->ss
||
5893 st
->minor_version
!= tst
->minor_version
||
5894 st
->ss
->compare_super(st
, tst
) != 0) {
5895 /* Some mismatch. If exactly one array matches this host,
5896 * we can resolve on that one.
5897 * Or, if we are auto assembling, we just ignore the second
5900 dprintf("superblock on %s doesn't match others - assembly aborted\n",
5906 dprintf("found: devname: %s\n", devname
);
5910 tst
->ss
->free_super(tst
);
5914 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5915 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5916 for (iter
= head
; iter
; iter
= iter
->next
) {
5917 dprintf("content->text_version: %s vol\n",
5918 iter
->text_version
);
5919 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5920 /* do not assemble arrays with unsupported
5922 dprintf("Cannot activate member %s.\n",
5923 iter
->text_version
);
5930 dprintf("No valid super block on device list: err: %d %p\n",
5934 dprintf("no more devices to examine\n");
5937 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5938 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5940 if (count
< tmpdev
->found
)
5943 count
-= tmpdev
->found
;
5946 if (tmpdev
->used
== 1)
5951 st
->ss
->free_super(st
);
5956 count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
5958 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
5960 const struct orom_entry
*entry
;
5961 struct devid_list
*dv
, *devid_list
;
5963 if (!hba
|| !hba
->path
)
5966 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
5967 if (strstr(idev
->path
, hba
->path
))
5971 if (!idev
|| !idev
->dev_id
)
5974 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
5976 if (!entry
|| !entry
->devid_list
)
5979 devid_list
= entry
->devid_list
;
5980 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
5981 struct md_list
*devlist
= NULL
;
5982 struct sys_dev
*device
= device_by_id(dv
->devid
);
5987 hba_path
= device
->path
;
5991 /* VMD has one orom entry for all domain, but spanning is not allowed.
5992 * VMD arrays should be counted per domain (controller), so skip
5993 * domains that are not the given one.
5995 if ((hba
->type
== SYS_DEV_VMD
) &&
5996 (strncmp(device
->path
, hba
->path
, strlen(device
->path
)) != 0))
5999 devlist
= get_devices(hba_path
);
6000 /* if no intel devices return zero volumes */
6001 if (devlist
== NULL
)
6004 count
+= active_arrays_by_format("imsm", hba_path
, &devlist
, dpa
, verbose
);
6005 dprintf("path: %s active arrays: %d\n", hba_path
, count
);
6006 if (devlist
== NULL
)
6010 count
+= count_volumes_list(devlist
,
6014 dprintf("found %d count: %d\n", found
, count
);
6017 dprintf("path: %s total number of volumes: %d\n", hba_path
, count
);
6020 struct md_list
*dv
= devlist
;
6021 devlist
= devlist
->next
;
6029 static int imsm_default_chunk(const struct imsm_orom
*orom
)
6031 /* up to 512 if the plaform supports it, otherwise the platform max.
6032 * 128 if no platform detected
6034 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
6036 return min(512, (1 << fs
));
6040 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
6041 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
6043 /* check/set platform and metadata limits/defaults */
6044 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
6045 pr_vrb(": platform supports a maximum of %d disks per array\n",
6050 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6051 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
6052 pr_vrb(": platform does not support raid%d with %d disk%s\n",
6053 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
6057 if (*chunk
== 0 || *chunk
== UnSet
)
6058 *chunk
= imsm_default_chunk(super
->orom
);
6060 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
6061 pr_vrb(": platform does not support a chunk size of: %d\n", *chunk
);
6065 if (layout
!= imsm_level_to_layout(level
)) {
6067 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
6068 else if (level
== 10)
6069 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
6071 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
6076 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
6077 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
6078 pr_vrb(": platform does not support a volume size over 2TB\n");
6085 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
6086 * FIX ME add ahci details
6088 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
6089 int layout
, int raiddisks
, int *chunk
,
6090 unsigned long long size
,
6091 unsigned long long data_offset
,
6093 unsigned long long *freesize
,
6097 struct intel_super
*super
= st
->sb
;
6098 struct imsm_super
*mpb
;
6100 unsigned long long pos
= 0;
6101 unsigned long long maxsize
;
6105 /* We must have the container info already read in. */
6109 mpb
= super
->anchor
;
6111 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
6112 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6116 /* General test: make sure there is space for
6117 * 'raiddisks' device extents of size 'size' at a given
6120 unsigned long long minsize
= size
;
6121 unsigned long long start_offset
= MaxSector
;
6124 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
6125 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6130 e
= get_extents(super
, dl
);
6133 unsigned long long esize
;
6134 esize
= e
[i
].start
- pos
;
6135 if (esize
>= minsize
)
6137 if (found
&& start_offset
== MaxSector
) {
6140 } else if (found
&& pos
!= start_offset
) {
6144 pos
= e
[i
].start
+ e
[i
].size
;
6146 } while (e
[i
-1].size
);
6151 if (dcnt
< raiddisks
) {
6153 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6160 /* This device must be a member of the set */
6161 if (stat(dev
, &stb
) < 0)
6163 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6165 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6166 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6167 dl
->minor
== (int)minor(stb
.st_rdev
))
6172 pr_err("%s is not in the same imsm set\n", dev
);
6174 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6175 /* If a volume is present then the current creation attempt
6176 * cannot incorporate new spares because the orom may not
6177 * understand this configuration (all member disks must be
6178 * members of each array in the container).
6180 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6181 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6183 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6184 mpb
->num_disks
!= raiddisks
) {
6185 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6189 /* retrieve the largest free space block */
6190 e
= get_extents(super
, dl
);
6195 unsigned long long esize
;
6197 esize
= e
[i
].start
- pos
;
6198 if (esize
>= maxsize
)
6200 pos
= e
[i
].start
+ e
[i
].size
;
6202 } while (e
[i
-1].size
);
6207 pr_err("unable to determine free space for: %s\n",
6211 if (maxsize
< size
) {
6213 pr_err("%s not enough space (%llu < %llu)\n",
6214 dev
, maxsize
, size
);
6218 /* count total number of extents for merge */
6220 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6222 i
+= dl
->extent_cnt
;
6224 maxsize
= merge_extents(super
, i
);
6226 if (!check_env("IMSM_NO_PLATFORM") &&
6227 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6228 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6232 if (maxsize
< size
|| maxsize
== 0) {
6235 pr_err("no free space left on device. Aborting...\n");
6237 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
6243 *freesize
= maxsize
;
6246 int count
= count_volumes(super
->hba
,
6247 super
->orom
->dpa
, verbose
);
6248 if (super
->orom
->vphba
<= count
) {
6249 pr_vrb(": platform does not support more than %d raid volumes.\n",
6250 super
->orom
->vphba
);
6257 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6258 unsigned long long size
, int chunk
,
6259 unsigned long long *freesize
)
6261 struct intel_super
*super
= st
->sb
;
6262 struct imsm_super
*mpb
= super
->anchor
;
6267 unsigned long long maxsize
;
6268 unsigned long long minsize
;
6272 /* find the largest common start free region of the possible disks */
6276 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6282 /* don't activate new spares if we are orom constrained
6283 * and there is already a volume active in the container
6285 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6288 e
= get_extents(super
, dl
);
6291 for (i
= 1; e
[i
-1].size
; i
++)
6299 maxsize
= merge_extents(super
, extent_cnt
);
6303 minsize
= chunk
* 2;
6305 if (cnt
< raiddisks
||
6306 (super
->orom
&& used
&& used
!= raiddisks
) ||
6307 maxsize
< minsize
||
6309 pr_err("not enough devices with space to create array.\n");
6310 return 0; /* No enough free spaces large enough */
6321 if (!check_env("IMSM_NO_PLATFORM") &&
6322 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6323 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6327 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6329 dl
->raiddisk
= cnt
++;
6333 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6338 static int reserve_space(struct supertype
*st
, int raiddisks
,
6339 unsigned long long size
, int chunk
,
6340 unsigned long long *freesize
)
6342 struct intel_super
*super
= st
->sb
;
6347 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6350 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6352 dl
->raiddisk
= cnt
++;
6359 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6360 int raiddisks
, int *chunk
, unsigned long long size
,
6361 unsigned long long data_offset
,
6362 char *dev
, unsigned long long *freesize
,
6370 * if given unused devices create a container
6371 * if given given devices in a container create a member volume
6373 if (level
== LEVEL_CONTAINER
) {
6374 /* Must be a fresh device to add to a container */
6375 return validate_geometry_imsm_container(st
, level
, layout
,
6385 struct intel_super
*super
= st
->sb
;
6386 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6387 raiddisks
, chunk
, size
,
6390 /* we are being asked to automatically layout a
6391 * new volume based on the current contents of
6392 * the container. If the the parameters can be
6393 * satisfied reserve_space will record the disks,
6394 * start offset, and size of the volume to be
6395 * created. add_to_super and getinfo_super
6396 * detect when autolayout is in progress.
6398 /* assuming that freesize is always given when array is
6400 if (super
->orom
&& freesize
) {
6402 count
= count_volumes(super
->hba
,
6403 super
->orom
->dpa
, verbose
);
6404 if (super
->orom
->vphba
<= count
) {
6405 pr_vrb(": platform does not support more than %d raid volumes.\n",
6406 super
->orom
->vphba
);
6411 return reserve_space(st
, raiddisks
, size
,
6417 /* creating in a given container */
6418 return validate_geometry_imsm_volume(st
, level
, layout
,
6419 raiddisks
, chunk
, size
,
6421 dev
, freesize
, verbose
);
6424 /* This device needs to be a device in an 'imsm' container */
6425 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6428 pr_err("Cannot create this array on device %s\n",
6433 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6435 pr_err("Cannot open %s: %s\n",
6436 dev
, strerror(errno
));
6439 /* Well, it is in use by someone, maybe an 'imsm' container. */
6440 cfd
= open_container(fd
);
6444 pr_err("Cannot use %s: It is busy\n",
6448 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
6449 if (sra
&& sra
->array
.major_version
== -1 &&
6450 strcmp(sra
->text_version
, "imsm") == 0)
6454 /* This is a member of a imsm container. Load the container
6455 * and try to create a volume
6457 struct intel_super
*super
;
6459 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6461 strcpy(st
->container_devnm
, fd2devnm(cfd
));
6463 return validate_geometry_imsm_volume(st
, level
, layout
,
6465 size
, data_offset
, dev
,
6472 pr_err("failed container membership check\n");
6478 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6480 struct intel_super
*super
= st
->sb
;
6482 if (level
&& *level
== UnSet
)
6483 *level
= LEVEL_CONTAINER
;
6485 if (level
&& layout
&& *layout
== UnSet
)
6486 *layout
= imsm_level_to_layout(*level
);
6488 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6489 *chunk
= imsm_default_chunk(super
->orom
);
6492 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6494 static int kill_subarray_imsm(struct supertype
*st
)
6496 /* remove the subarray currently referenced by ->current_vol */
6498 struct intel_dev
**dp
;
6499 struct intel_super
*super
= st
->sb
;
6500 __u8 current_vol
= super
->current_vol
;
6501 struct imsm_super
*mpb
= super
->anchor
;
6503 if (super
->current_vol
< 0)
6505 super
->current_vol
= -1; /* invalidate subarray cursor */
6507 /* block deletions that would change the uuid of active subarrays
6509 * FIXME when immutable ids are available, but note that we'll
6510 * also need to fixup the invalidated/active subarray indexes in
6513 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6516 if (i
< current_vol
)
6518 sprintf(subarray
, "%u", i
);
6519 if (is_subarray_active(subarray
, st
->devnm
)) {
6520 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6527 if (st
->update_tail
) {
6528 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
6530 u
->type
= update_kill_array
;
6531 u
->dev_idx
= current_vol
;
6532 append_metadata_update(st
, u
, sizeof(*u
));
6537 for (dp
= &super
->devlist
; *dp
;)
6538 if ((*dp
)->index
== current_vol
) {
6541 handle_missing(super
, (*dp
)->dev
);
6542 if ((*dp
)->index
> current_vol
)
6547 /* no more raid devices, all active components are now spares,
6548 * but of course failed are still failed
6550 if (--mpb
->num_raid_devs
== 0) {
6553 for (d
= super
->disks
; d
; d
= d
->next
)
6558 super
->updates_pending
++;
6563 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6564 char *update
, struct mddev_ident
*ident
)
6566 /* update the subarray currently referenced by ->current_vol */
6567 struct intel_super
*super
= st
->sb
;
6568 struct imsm_super
*mpb
= super
->anchor
;
6570 if (strcmp(update
, "name") == 0) {
6571 char *name
= ident
->name
;
6575 if (is_subarray_active(subarray
, st
->devnm
)) {
6576 pr_err("Unable to update name of active subarray\n");
6580 if (!check_name(super
, name
, 0))
6583 vol
= strtoul(subarray
, &ep
, 10);
6584 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6587 if (st
->update_tail
) {
6588 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
6590 u
->type
= update_rename_array
;
6592 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6593 append_metadata_update(st
, u
, sizeof(*u
));
6595 struct imsm_dev
*dev
;
6598 dev
= get_imsm_dev(super
, vol
);
6599 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6600 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6601 dev
= get_imsm_dev(super
, i
);
6602 handle_missing(super
, dev
);
6604 super
->updates_pending
++;
6611 #endif /* MDASSEMBLE */
6613 static int is_gen_migration(struct imsm_dev
*dev
)
6618 if (!dev
->vol
.migr_state
)
6621 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6627 static int is_rebuilding(struct imsm_dev
*dev
)
6629 struct imsm_map
*migr_map
;
6631 if (!dev
->vol
.migr_state
)
6634 if (migr_type(dev
) != MIGR_REBUILD
)
6637 migr_map
= get_imsm_map(dev
, MAP_1
);
6639 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6646 static int is_initializing(struct imsm_dev
*dev
)
6648 struct imsm_map
*migr_map
;
6650 if (!dev
->vol
.migr_state
)
6653 if (migr_type(dev
) != MIGR_INIT
)
6656 migr_map
= get_imsm_map(dev
, MAP_1
);
6658 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6665 static void update_recovery_start(struct intel_super
*super
,
6666 struct imsm_dev
*dev
,
6667 struct mdinfo
*array
)
6669 struct mdinfo
*rebuild
= NULL
;
6673 if (!is_rebuilding(dev
))
6676 /* Find the rebuild target, but punt on the dual rebuild case */
6677 for (d
= array
->devs
; d
; d
= d
->next
)
6678 if (d
->recovery_start
== 0) {
6685 /* (?) none of the disks are marked with
6686 * IMSM_ORD_REBUILD, so assume they are missing and the
6687 * disk_ord_tbl was not correctly updated
6689 dprintf("failed to locate out-of-sync disk\n");
6693 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6694 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6698 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6701 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6703 /* Given a container loaded by load_super_imsm_all,
6704 * extract information about all the arrays into
6706 * If 'subarray' is given, just extract info about that array.
6708 * For each imsm_dev create an mdinfo, fill it in,
6709 * then look for matching devices in super->disks
6710 * and create appropriate device mdinfo.
6712 struct intel_super
*super
= st
->sb
;
6713 struct imsm_super
*mpb
= super
->anchor
;
6714 struct mdinfo
*rest
= NULL
;
6718 int spare_disks
= 0;
6720 /* do not assemble arrays when not all attributes are supported */
6721 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6723 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
6726 /* check for bad blocks */
6727 if (imsm_bbm_log_size(super
->anchor
)) {
6728 pr_err("BBM log found in IMSM metadata.Arrays activation is blocked.\n");
6732 /* count spare devices, not used in maps
6734 for (d
= super
->disks
; d
; d
= d
->next
)
6738 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6739 struct imsm_dev
*dev
;
6740 struct imsm_map
*map
;
6741 struct imsm_map
*map2
;
6742 struct mdinfo
*this;
6750 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6753 dev
= get_imsm_dev(super
, i
);
6754 map
= get_imsm_map(dev
, MAP_0
);
6755 map2
= get_imsm_map(dev
, MAP_1
);
6757 /* do not publish arrays that are in the middle of an
6758 * unsupported migration
6760 if (dev
->vol
.migr_state
&&
6761 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6762 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
6766 /* do not publish arrays that are not support by controller's
6770 this = xmalloc(sizeof(*this));
6772 super
->current_vol
= i
;
6773 getinfo_super_imsm_volume(st
, this, NULL
);
6776 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6777 /* mdadm does not support all metadata features- set the bit in all arrays state */
6778 if (!validate_geometry_imsm_orom(super
,
6779 get_imsm_raid_level(map
), /* RAID level */
6780 imsm_level_to_layout(get_imsm_raid_level(map
)),
6781 map
->num_members
, /* raid disks */
6782 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6784 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
6786 this->array
.state
|=
6787 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6788 (1<<MD_SB_BLOCK_VOLUME
);
6792 /* if array has bad blocks, set suitable bit in all arrays state */
6794 this->array
.state
|=
6795 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6796 (1<<MD_SB_BLOCK_VOLUME
);
6798 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6799 unsigned long long recovery_start
;
6800 struct mdinfo
*info_d
;
6807 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6808 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6809 for (d
= super
->disks
; d
; d
= d
->next
)
6810 if (d
->index
== idx
)
6813 recovery_start
= MaxSector
;
6816 if (d
&& is_failed(&d
->disk
))
6818 if (ord
& IMSM_ORD_REBUILD
)
6822 * if we skip some disks the array will be assmebled degraded;
6823 * reset resync start to avoid a dirty-degraded
6824 * situation when performing the intial sync
6826 * FIXME handle dirty degraded
6828 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6829 this->resync_start
= MaxSector
;
6833 info_d
= xcalloc(1, sizeof(*info_d
));
6834 info_d
->next
= this->devs
;
6835 this->devs
= info_d
;
6837 info_d
->disk
.number
= d
->index
;
6838 info_d
->disk
.major
= d
->major
;
6839 info_d
->disk
.minor
= d
->minor
;
6840 info_d
->disk
.raid_disk
= slot
;
6841 info_d
->recovery_start
= recovery_start
;
6843 if (slot
< map2
->num_members
)
6844 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6846 this->array
.spare_disks
++;
6848 if (slot
< map
->num_members
)
6849 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6851 this->array
.spare_disks
++;
6853 if (info_d
->recovery_start
== MaxSector
)
6854 this->array
.working_disks
++;
6856 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6857 info_d
->data_offset
= pba_of_lba0(map
);
6858 info_d
->component_size
= blocks_per_member(map
);
6860 /* now that the disk list is up-to-date fixup recovery_start */
6861 update_recovery_start(super
, dev
, this);
6862 this->array
.spare_disks
+= spare_disks
;
6865 /* check for reshape */
6866 if (this->reshape_active
== 1)
6867 recover_backup_imsm(st
, this);
6875 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6876 int failed
, int look_in_map
)
6878 struct imsm_map
*map
;
6880 map
= get_imsm_map(dev
, look_in_map
);
6883 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6884 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6886 switch (get_imsm_raid_level(map
)) {
6888 return IMSM_T_STATE_FAILED
;
6891 if (failed
< map
->num_members
)
6892 return IMSM_T_STATE_DEGRADED
;
6894 return IMSM_T_STATE_FAILED
;
6899 * check to see if any mirrors have failed, otherwise we
6900 * are degraded. Even numbered slots are mirrored on
6904 /* gcc -Os complains that this is unused */
6905 int insync
= insync
;
6907 for (i
= 0; i
< map
->num_members
; i
++) {
6908 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6909 int idx
= ord_to_idx(ord
);
6910 struct imsm_disk
*disk
;
6912 /* reset the potential in-sync count on even-numbered
6913 * slots. num_copies is always 2 for imsm raid10
6918 disk
= get_imsm_disk(super
, idx
);
6919 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6922 /* no in-sync disks left in this mirror the
6926 return IMSM_T_STATE_FAILED
;
6929 return IMSM_T_STATE_DEGRADED
;
6933 return IMSM_T_STATE_DEGRADED
;
6935 return IMSM_T_STATE_FAILED
;
6941 return map
->map_state
;
6944 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6949 struct imsm_disk
*disk
;
6950 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6951 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6952 struct imsm_map
*map_for_loop
;
6957 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6958 * disks that are being rebuilt. New failures are recorded to
6959 * map[0]. So we look through all the disks we started with and
6960 * see if any failures are still present, or if any new ones
6964 if (prev
&& (map
->num_members
< prev
->num_members
))
6965 map_for_loop
= prev
;
6967 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6969 /* when MAP_X is passed both maps failures are counted
6972 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6973 (i
< prev
->num_members
)) {
6974 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6975 idx_1
= ord_to_idx(ord
);
6977 disk
= get_imsm_disk(super
, idx_1
);
6978 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6981 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6982 (i
< map
->num_members
)) {
6983 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6984 idx
= ord_to_idx(ord
);
6987 disk
= get_imsm_disk(super
, idx
);
6988 if (!disk
|| is_failed(disk
) ||
6989 ord
& IMSM_ORD_REBUILD
)
6999 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
7002 struct intel_super
*super
= c
->sb
;
7003 struct imsm_super
*mpb
= super
->anchor
;
7005 if (atoi(inst
) >= mpb
->num_raid_devs
) {
7006 pr_err("subarry index %d, out of range\n", atoi(inst
));
7010 dprintf("imsm: open_new %s\n", inst
);
7011 a
->info
.container_member
= atoi(inst
);
7015 static int is_resyncing(struct imsm_dev
*dev
)
7017 struct imsm_map
*migr_map
;
7019 if (!dev
->vol
.migr_state
)
7022 if (migr_type(dev
) == MIGR_INIT
||
7023 migr_type(dev
) == MIGR_REPAIR
)
7026 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7029 migr_map
= get_imsm_map(dev
, MAP_1
);
7031 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
7032 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
7038 /* return true if we recorded new information */
7039 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7043 struct imsm_map
*map
;
7044 char buf
[MAX_RAID_SERIAL_LEN
+3];
7045 unsigned int len
, shift
= 0;
7047 /* new failures are always set in map[0] */
7048 map
= get_imsm_map(dev
, MAP_0
);
7050 slot
= get_imsm_disk_slot(map
, idx
);
7054 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
7055 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
7058 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
7059 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
7061 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
7062 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
7063 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
7065 disk
->status
|= FAILED_DISK
;
7066 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
7067 /* mark failures in second map if second map exists and this disk
7069 * This is valid for migration, initialization and rebuild
7071 if (dev
->vol
.migr_state
) {
7072 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
7073 int slot2
= get_imsm_disk_slot(map2
, idx
);
7075 if ((slot2
< map2
->num_members
) &&
7077 set_imsm_ord_tbl_ent(map2
, slot2
,
7078 idx
| IMSM_ORD_REBUILD
);
7080 if (map
->failed_disk_num
== 0xff)
7081 map
->failed_disk_num
= slot
;
7085 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7087 mark_failure(dev
, disk
, idx
);
7089 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
7092 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
7093 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
7096 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7100 if (!super
->missing
)
7103 /* When orom adds replacement for missing disk it does
7104 * not remove entry of missing disk, but just updates map with
7105 * new added disk. So it is not enough just to test if there is
7106 * any missing disk, we have to look if there are any failed disks
7107 * in map to stop migration */
7109 dprintf("imsm: mark missing\n");
7110 /* end process for initialization and rebuild only
7112 if (is_gen_migration(dev
) == 0) {
7116 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7117 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7120 end_migration(dev
, super
, map_state
);
7122 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7123 mark_missing(dev
, &dl
->disk
, dl
->index
);
7124 super
->updates_pending
++;
7127 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7130 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7131 unsigned long long array_blocks
;
7132 struct imsm_map
*map
;
7134 if (used_disks
== 0) {
7135 /* when problems occures
7136 * return current array_blocks value
7138 array_blocks
= __le32_to_cpu(dev
->size_high
);
7139 array_blocks
= array_blocks
<< 32;
7140 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7142 return array_blocks
;
7145 /* set array size in metadata
7147 if (new_size
<= 0) {
7148 /* OLCE size change is caused by added disks
7150 map
= get_imsm_map(dev
, MAP_0
);
7151 array_blocks
= blocks_per_member(map
) * used_disks
;
7153 /* Online Volume Size Change
7154 * Using available free space
7156 array_blocks
= new_size
;
7159 /* round array size down to closest MB
7161 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7162 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7163 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7165 return array_blocks
;
7168 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7170 static void imsm_progress_container_reshape(struct intel_super
*super
)
7172 /* if no device has a migr_state, but some device has a
7173 * different number of members than the previous device, start
7174 * changing the number of devices in this device to match
7177 struct imsm_super
*mpb
= super
->anchor
;
7178 int prev_disks
= -1;
7182 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7183 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7184 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7185 struct imsm_map
*map2
;
7186 int prev_num_members
;
7188 if (dev
->vol
.migr_state
)
7191 if (prev_disks
== -1)
7192 prev_disks
= map
->num_members
;
7193 if (prev_disks
== map
->num_members
)
7196 /* OK, this array needs to enter reshape mode.
7197 * i.e it needs a migr_state
7200 copy_map_size
= sizeof_imsm_map(map
);
7201 prev_num_members
= map
->num_members
;
7202 map
->num_members
= prev_disks
;
7203 dev
->vol
.migr_state
= 1;
7204 dev
->vol
.curr_migr_unit
= 0;
7205 set_migr_type(dev
, MIGR_GEN_MIGR
);
7206 for (i
= prev_num_members
;
7207 i
< map
->num_members
; i
++)
7208 set_imsm_ord_tbl_ent(map
, i
, i
);
7209 map2
= get_imsm_map(dev
, MAP_1
);
7210 /* Copy the current map */
7211 memcpy(map2
, map
, copy_map_size
);
7212 map2
->num_members
= prev_num_members
;
7214 imsm_set_array_size(dev
, -1);
7215 super
->clean_migration_record_by_mdmon
= 1;
7216 super
->updates_pending
++;
7220 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7221 * states are handled in imsm_set_disk() with one exception, when a
7222 * resync is stopped due to a new failure this routine will set the
7223 * 'degraded' state for the array.
7225 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7227 int inst
= a
->info
.container_member
;
7228 struct intel_super
*super
= a
->container
->sb
;
7229 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7230 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7231 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7232 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7233 __u32 blocks_per_unit
;
7235 if (dev
->vol
.migr_state
&&
7236 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7237 /* array state change is blocked due to reshape action
7239 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7240 * - finish the reshape (if last_checkpoint is big and action != reshape)
7241 * - update curr_migr_unit
7243 if (a
->curr_action
== reshape
) {
7244 /* still reshaping, maybe update curr_migr_unit */
7245 goto mark_checkpoint
;
7247 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7248 /* for some reason we aborted the reshape.
7250 * disable automatic metadata rollback
7251 * user action is required to recover process
7254 struct imsm_map
*map2
=
7255 get_imsm_map(dev
, MAP_1
);
7256 dev
->vol
.migr_state
= 0;
7257 set_migr_type(dev
, 0);
7258 dev
->vol
.curr_migr_unit
= 0;
7260 sizeof_imsm_map(map2
));
7261 super
->updates_pending
++;
7264 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7265 unsigned long long array_blocks
;
7269 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7270 if (used_disks
> 0) {
7272 blocks_per_member(map
) *
7274 /* round array size down to closest MB
7276 array_blocks
= (array_blocks
7277 >> SECT_PER_MB_SHIFT
)
7278 << SECT_PER_MB_SHIFT
;
7279 a
->info
.custom_array_size
= array_blocks
;
7280 /* encourage manager to update array
7284 a
->check_reshape
= 1;
7286 /* finalize online capacity expansion/reshape */
7287 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7289 mdi
->disk
.raid_disk
,
7292 imsm_progress_container_reshape(super
);
7297 /* before we activate this array handle any missing disks */
7298 if (consistent
== 2)
7299 handle_missing(super
, dev
);
7301 if (consistent
== 2 &&
7302 (!is_resync_complete(&a
->info
) ||
7303 map_state
!= IMSM_T_STATE_NORMAL
||
7304 dev
->vol
.migr_state
))
7307 if (is_resync_complete(&a
->info
)) {
7308 /* complete intialization / resync,
7309 * recovery and interrupted recovery is completed in
7312 if (is_resyncing(dev
)) {
7313 dprintf("imsm: mark resync done\n");
7314 end_migration(dev
, super
, map_state
);
7315 super
->updates_pending
++;
7316 a
->last_checkpoint
= 0;
7318 } else if ((!is_resyncing(dev
) && !failed
) &&
7319 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7320 /* mark the start of the init process if nothing is failed */
7321 dprintf("imsm: mark resync start\n");
7322 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7323 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7325 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7326 super
->updates_pending
++;
7330 /* skip checkpointing for general migration,
7331 * it is controlled in mdadm
7333 if (is_gen_migration(dev
))
7334 goto skip_mark_checkpoint
;
7336 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7337 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7338 if (blocks_per_unit
) {
7342 units
= a
->last_checkpoint
/ blocks_per_unit
;
7345 /* check that we did not overflow 32-bits, and that
7346 * curr_migr_unit needs updating
7348 if (units32
== units
&&
7350 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7351 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7352 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7353 super
->updates_pending
++;
7357 skip_mark_checkpoint
:
7358 /* mark dirty / clean */
7359 if (dev
->vol
.dirty
!= !consistent
) {
7360 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7365 super
->updates_pending
++;
7371 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7373 int inst
= a
->info
.container_member
;
7374 struct intel_super
*super
= a
->container
->sb
;
7375 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7376 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7377 struct imsm_disk
*disk
;
7379 int recovery_not_finished
= 0;
7384 if (n
> map
->num_members
)
7385 pr_err("imsm: set_disk %d out of range 0..%d\n",
7386 n
, map
->num_members
- 1);
7391 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7393 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7394 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7396 /* check for new failures */
7397 if (state
& DS_FAULTY
) {
7398 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7399 super
->updates_pending
++;
7402 /* check if in_sync */
7403 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7404 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7406 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7407 super
->updates_pending
++;
7410 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7411 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7413 /* check if recovery complete, newly degraded, or failed */
7414 dprintf("imsm: Detected transition to state ");
7415 switch (map_state
) {
7416 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7417 dprintf("normal: ");
7418 if (is_rebuilding(dev
)) {
7419 dprintf_cont("while rebuilding");
7420 /* check if recovery is really finished */
7421 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7422 if (mdi
->recovery_start
!= MaxSector
) {
7423 recovery_not_finished
= 1;
7426 if (recovery_not_finished
) {
7428 dprintf("Rebuild has not finished yet, state not changed");
7429 if (a
->last_checkpoint
< mdi
->recovery_start
) {
7430 a
->last_checkpoint
= mdi
->recovery_start
;
7431 super
->updates_pending
++;
7435 end_migration(dev
, super
, map_state
);
7436 map
= get_imsm_map(dev
, MAP_0
);
7437 map
->failed_disk_num
= ~0;
7438 super
->updates_pending
++;
7439 a
->last_checkpoint
= 0;
7442 if (is_gen_migration(dev
)) {
7443 dprintf_cont("while general migration");
7444 if (a
->last_checkpoint
>= a
->info
.component_size
)
7445 end_migration(dev
, super
, map_state
);
7447 map
->map_state
= map_state
;
7448 map
= get_imsm_map(dev
, MAP_0
);
7449 map
->failed_disk_num
= ~0;
7450 super
->updates_pending
++;
7454 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7455 dprintf_cont("degraded: ");
7456 if ((map
->map_state
!= map_state
) &&
7457 !dev
->vol
.migr_state
) {
7458 dprintf_cont("mark degraded");
7459 map
->map_state
= map_state
;
7460 super
->updates_pending
++;
7461 a
->last_checkpoint
= 0;
7464 if (is_rebuilding(dev
)) {
7465 dprintf_cont("while rebuilding.");
7466 if (map
->map_state
!= map_state
) {
7467 dprintf_cont(" Map state change");
7468 end_migration(dev
, super
, map_state
);
7469 super
->updates_pending
++;
7473 if (is_gen_migration(dev
)) {
7474 dprintf_cont("while general migration");
7475 if (a
->last_checkpoint
>= a
->info
.component_size
)
7476 end_migration(dev
, super
, map_state
);
7478 map
->map_state
= map_state
;
7479 manage_second_map(super
, dev
);
7481 super
->updates_pending
++;
7484 if (is_initializing(dev
)) {
7485 dprintf_cont("while initialization.");
7486 map
->map_state
= map_state
;
7487 super
->updates_pending
++;
7491 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7492 dprintf_cont("failed: ");
7493 if (is_gen_migration(dev
)) {
7494 dprintf_cont("while general migration");
7495 map
->map_state
= map_state
;
7496 super
->updates_pending
++;
7499 if (map
->map_state
!= map_state
) {
7500 dprintf_cont("mark failed");
7501 end_migration(dev
, super
, map_state
);
7502 super
->updates_pending
++;
7503 a
->last_checkpoint
= 0;
7508 dprintf_cont("state %i\n", map_state
);
7513 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7516 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7517 unsigned long long dsize
;
7518 unsigned long long sectors
;
7520 get_dev_size(fd
, NULL
, &dsize
);
7522 if (mpb_size
> 512) {
7523 /* -1 to account for anchor */
7524 sectors
= mpb_sectors(mpb
) - 1;
7526 /* write the extended mpb to the sectors preceeding the anchor */
7527 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7530 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7535 /* first block is stored on second to last sector of the disk */
7536 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7539 if (write(fd
, buf
, 512) != 512)
7545 static void imsm_sync_metadata(struct supertype
*container
)
7547 struct intel_super
*super
= container
->sb
;
7549 dprintf("sync metadata: %d\n", super
->updates_pending
);
7550 if (!super
->updates_pending
)
7553 write_super_imsm(container
, 0);
7555 super
->updates_pending
= 0;
7558 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7560 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7561 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7564 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7568 if (dl
&& is_failed(&dl
->disk
))
7572 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
7577 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7578 struct active_array
*a
, int activate_new
,
7579 struct mdinfo
*additional_test_list
)
7581 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7582 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7583 struct imsm_super
*mpb
= super
->anchor
;
7584 struct imsm_map
*map
;
7585 unsigned long long pos
;
7590 __u32 array_start
= 0;
7591 __u32 array_end
= 0;
7593 struct mdinfo
*test_list
;
7595 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7596 /* If in this array, skip */
7597 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7598 if (d
->state_fd
>= 0 &&
7599 d
->disk
.major
== dl
->major
&&
7600 d
->disk
.minor
== dl
->minor
) {
7601 dprintf("%x:%x already in array\n",
7602 dl
->major
, dl
->minor
);
7607 test_list
= additional_test_list
;
7609 if (test_list
->disk
.major
== dl
->major
&&
7610 test_list
->disk
.minor
== dl
->minor
) {
7611 dprintf("%x:%x already in additional test list\n",
7612 dl
->major
, dl
->minor
);
7615 test_list
= test_list
->next
;
7620 /* skip in use or failed drives */
7621 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7623 dprintf("%x:%x status (failed: %d index: %d)\n",
7624 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7628 /* skip pure spares when we are looking for partially
7629 * assimilated drives
7631 if (dl
->index
== -1 && !activate_new
)
7634 /* Does this unused device have the requisite free space?
7635 * It needs to be able to cover all member volumes
7637 ex
= get_extents(super
, dl
);
7639 dprintf("cannot get extents\n");
7642 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7643 dev
= get_imsm_dev(super
, i
);
7644 map
= get_imsm_map(dev
, MAP_0
);
7646 /* check if this disk is already a member of
7649 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7655 array_start
= pba_of_lba0(map
);
7656 array_end
= array_start
+
7657 blocks_per_member(map
) - 1;
7660 /* check that we can start at pba_of_lba0 with
7661 * blocks_per_member of space
7663 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7667 pos
= ex
[j
].start
+ ex
[j
].size
;
7669 } while (ex
[j
-1].size
);
7676 if (i
< mpb
->num_raid_devs
) {
7677 dprintf("%x:%x does not have %u to %u available\n",
7678 dl
->major
, dl
->minor
, array_start
, array_end
);
7688 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7690 struct imsm_dev
*dev2
;
7691 struct imsm_map
*map
;
7697 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7699 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7700 if (state
== IMSM_T_STATE_FAILED
) {
7701 map
= get_imsm_map(dev2
, MAP_0
);
7704 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7706 * Check if failed disks are deleted from intel
7707 * disk list or are marked to be deleted
7709 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7710 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7712 * Do not rebuild the array if failed disks
7713 * from failed sub-array are not removed from
7717 is_failed(&idisk
->disk
) &&
7718 (idisk
->action
!= DISK_REMOVE
))
7726 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7727 struct metadata_update
**updates
)
7730 * Find a device with unused free space and use it to replace a
7731 * failed/vacant region in an array. We replace failed regions one a
7732 * array at a time. The result is that a new spare disk will be added
7733 * to the first failed array and after the monitor has finished
7734 * propagating failures the remainder will be consumed.
7736 * FIXME add a capability for mdmon to request spares from another
7740 struct intel_super
*super
= a
->container
->sb
;
7741 int inst
= a
->info
.container_member
;
7742 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7743 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7744 int failed
= a
->info
.array
.raid_disks
;
7745 struct mdinfo
*rv
= NULL
;
7748 struct metadata_update
*mu
;
7750 struct imsm_update_activate_spare
*u
;
7755 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7756 if ((d
->curr_state
& DS_FAULTY
) &&
7758 /* wait for Removal to happen */
7760 if (d
->state_fd
>= 0)
7764 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7765 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7767 if (imsm_reshape_blocks_arrays_changes(super
))
7770 /* Cannot activate another spare if rebuild is in progress already
7772 if (is_rebuilding(dev
)) {
7773 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
7777 if (a
->info
.array
.level
== 4)
7778 /* No repair for takeovered array
7779 * imsm doesn't support raid4
7783 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7784 IMSM_T_STATE_DEGRADED
)
7788 * If there are any failed disks check state of the other volume.
7789 * Block rebuild if the another one is failed until failed disks
7790 * are removed from container.
7793 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
7794 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7795 /* check if states of the other volumes allow for rebuild */
7796 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7798 allowed
= imsm_rebuild_allowed(a
->container
,
7806 /* For each slot, if it is not working, find a spare */
7807 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7808 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7809 if (d
->disk
.raid_disk
== i
)
7811 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7812 if (d
&& (d
->state_fd
>= 0))
7816 * OK, this device needs recovery. Try to re-add the
7817 * previous occupant of this slot, if this fails see if
7818 * we can continue the assimilation of a spare that was
7819 * partially assimilated, finally try to activate a new
7822 dl
= imsm_readd(super
, i
, a
);
7824 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7826 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7830 /* found a usable disk with enough space */
7831 di
= xcalloc(1, sizeof(*di
));
7833 /* dl->index will be -1 in the case we are activating a
7834 * pristine spare. imsm_process_update() will create a
7835 * new index in this case. Once a disk is found to be
7836 * failed in all member arrays it is kicked from the
7839 di
->disk
.number
= dl
->index
;
7841 /* (ab)use di->devs to store a pointer to the device
7844 di
->devs
= (struct mdinfo
*) dl
;
7846 di
->disk
.raid_disk
= i
;
7847 di
->disk
.major
= dl
->major
;
7848 di
->disk
.minor
= dl
->minor
;
7850 di
->recovery_start
= 0;
7851 di
->data_offset
= pba_of_lba0(map
);
7852 di
->component_size
= a
->info
.component_size
;
7853 di
->container_member
= inst
;
7854 super
->random
= random32();
7858 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7859 i
, di
->data_offset
);
7863 /* No spares found */
7865 /* Now 'rv' has a list of devices to return.
7866 * Create a metadata_update record to update the
7867 * disk_ord_tbl for the array
7869 mu
= xmalloc(sizeof(*mu
));
7870 mu
->buf
= xcalloc(num_spares
,
7871 sizeof(struct imsm_update_activate_spare
));
7873 mu
->space_list
= NULL
;
7874 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7875 mu
->next
= *updates
;
7876 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7878 for (di
= rv
; di
; di
= di
->next
) {
7879 u
->type
= update_activate_spare
;
7880 u
->dl
= (struct dl
*) di
->devs
;
7882 u
->slot
= di
->disk
.raid_disk
;
7893 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7895 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7896 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7897 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7898 struct disk_info
*inf
= get_disk_info(u
);
7899 struct imsm_disk
*disk
;
7903 for (i
= 0; i
< map
->num_members
; i
++) {
7904 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7905 for (j
= 0; j
< new_map
->num_members
; j
++)
7906 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7913 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7915 struct dl
*dl
= NULL
;
7916 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7917 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7922 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7924 struct dl
*prev
= NULL
;
7928 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7929 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7932 prev
->next
= dl
->next
;
7934 super
->disks
= dl
->next
;
7936 __free_imsm_disk(dl
);
7937 dprintf("removed %x:%x\n", major
, minor
);
7945 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7947 static int add_remove_disk_update(struct intel_super
*super
)
7949 int check_degraded
= 0;
7950 struct dl
*disk
= NULL
;
7951 /* add/remove some spares to/from the metadata/contrainer */
7952 while (super
->disk_mgmt_list
) {
7953 struct dl
*disk_cfg
;
7955 disk_cfg
= super
->disk_mgmt_list
;
7956 super
->disk_mgmt_list
= disk_cfg
->next
;
7957 disk_cfg
->next
= NULL
;
7959 if (disk_cfg
->action
== DISK_ADD
) {
7960 disk_cfg
->next
= super
->disks
;
7961 super
->disks
= disk_cfg
;
7963 dprintf("added %x:%x\n",
7964 disk_cfg
->major
, disk_cfg
->minor
);
7965 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7966 dprintf("Disk remove action processed: %x.%x\n",
7967 disk_cfg
->major
, disk_cfg
->minor
);
7968 disk
= get_disk_super(super
,
7972 /* store action status */
7973 disk
->action
= DISK_REMOVE
;
7974 /* remove spare disks only */
7975 if (disk
->index
== -1) {
7976 remove_disk_super(super
,
7981 /* release allocate disk structure */
7982 __free_imsm_disk(disk_cfg
);
7985 return check_degraded
;
7988 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7989 struct intel_super
*super
,
7992 struct intel_dev
*id
;
7993 void **tofree
= NULL
;
7996 dprintf("(enter)\n");
7997 if ((u
->subdev
< 0) ||
7999 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8002 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
8003 dprintf("imsm: Error: Memory is not allocated\n");
8007 for (id
= super
->devlist
; id
; id
= id
->next
) {
8008 if (id
->index
== (unsigned)u
->subdev
) {
8009 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8010 struct imsm_map
*map
;
8011 struct imsm_dev
*new_dev
=
8012 (struct imsm_dev
*)*space_list
;
8013 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8015 struct dl
*new_disk
;
8017 if (new_dev
== NULL
)
8019 *space_list
= **space_list
;
8020 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
8021 map
= get_imsm_map(new_dev
, MAP_0
);
8023 dprintf("imsm: Error: migration in progress");
8027 to_state
= map
->map_state
;
8028 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
8030 /* this should not happen */
8031 if (u
->new_disks
[0] < 0) {
8032 map
->failed_disk_num
=
8033 map
->num_members
- 1;
8034 to_state
= IMSM_T_STATE_DEGRADED
;
8036 to_state
= IMSM_T_STATE_NORMAL
;
8038 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
8039 if (u
->new_level
> -1)
8040 map
->raid_level
= u
->new_level
;
8041 migr_map
= get_imsm_map(new_dev
, MAP_1
);
8042 if ((u
->new_level
== 5) &&
8043 (migr_map
->raid_level
== 0)) {
8044 int ord
= map
->num_members
- 1;
8045 migr_map
->num_members
--;
8046 if (u
->new_disks
[0] < 0)
8047 ord
|= IMSM_ORD_REBUILD
;
8048 set_imsm_ord_tbl_ent(map
,
8049 map
->num_members
- 1,
8053 tofree
= (void **)dev
;
8055 /* update chunk size
8057 if (u
->new_chunksize
> 0)
8058 map
->blocks_per_strip
=
8059 __cpu_to_le16(u
->new_chunksize
* 2);
8063 if ((u
->new_level
!= 5) ||
8064 (migr_map
->raid_level
!= 0) ||
8065 (migr_map
->raid_level
== map
->raid_level
))
8068 if (u
->new_disks
[0] >= 0) {
8071 new_disk
= get_disk_super(super
,
8072 major(u
->new_disks
[0]),
8073 minor(u
->new_disks
[0]));
8074 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8075 major(u
->new_disks
[0]),
8076 minor(u
->new_disks
[0]),
8077 new_disk
, new_disk
->index
);
8078 if (new_disk
== NULL
)
8079 goto error_disk_add
;
8081 new_disk
->index
= map
->num_members
- 1;
8082 /* slot to fill in autolayout
8084 new_disk
->raiddisk
= new_disk
->index
;
8085 new_disk
->disk
.status
|= CONFIGURED_DISK
;
8086 new_disk
->disk
.status
&= ~SPARE_DISK
;
8088 goto error_disk_add
;
8091 *tofree
= *space_list
;
8092 /* calculate new size
8094 imsm_set_array_size(new_dev
, -1);
8101 *space_list
= tofree
;
8105 dprintf("Error: imsm: Cannot find disk.\n");
8109 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8110 struct intel_super
*super
)
8112 struct intel_dev
*id
;
8115 dprintf("(enter)\n");
8116 if ((u
->subdev
< 0) ||
8118 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8122 for (id
= super
->devlist
; id
; id
= id
->next
) {
8123 if (id
->index
== (unsigned)u
->subdev
) {
8124 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8125 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8126 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8127 unsigned long long blocks_per_member
;
8129 /* calculate new size
8131 blocks_per_member
= u
->new_size
/ used_disks
;
8132 dprintf("(size: %llu, blocks per member: %llu)\n",
8133 u
->new_size
, blocks_per_member
);
8134 set_blocks_per_member(map
, blocks_per_member
);
8135 imsm_set_array_size(dev
, u
->new_size
);
8145 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8146 struct intel_super
*super
,
8147 struct active_array
*active_array
)
8149 struct imsm_super
*mpb
= super
->anchor
;
8150 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8151 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8152 struct imsm_map
*migr_map
;
8153 struct active_array
*a
;
8154 struct imsm_disk
*disk
;
8161 int second_map_created
= 0;
8163 for (; u
; u
= u
->next
) {
8164 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8169 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8174 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
8179 /* count failures (excluding rebuilds and the victim)
8180 * to determine map[0] state
8183 for (i
= 0; i
< map
->num_members
; i
++) {
8186 disk
= get_imsm_disk(super
,
8187 get_imsm_disk_idx(dev
, i
, MAP_X
));
8188 if (!disk
|| is_failed(disk
))
8192 /* adding a pristine spare, assign a new index */
8193 if (dl
->index
< 0) {
8194 dl
->index
= super
->anchor
->num_disks
;
8195 super
->anchor
->num_disks
++;
8198 disk
->status
|= CONFIGURED_DISK
;
8199 disk
->status
&= ~SPARE_DISK
;
8202 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8203 if (!second_map_created
) {
8204 second_map_created
= 1;
8205 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8206 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8208 map
->map_state
= to_state
;
8209 migr_map
= get_imsm_map(dev
, MAP_1
);
8210 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8211 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8212 dl
->index
| IMSM_ORD_REBUILD
);
8214 /* update the family_num to mark a new container
8215 * generation, being careful to record the existing
8216 * family_num in orig_family_num to clean up after
8217 * earlier mdadm versions that neglected to set it.
8219 if (mpb
->orig_family_num
== 0)
8220 mpb
->orig_family_num
= mpb
->family_num
;
8221 mpb
->family_num
+= super
->random
;
8223 /* count arrays using the victim in the metadata */
8225 for (a
= active_array
; a
; a
= a
->next
) {
8226 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8227 map
= get_imsm_map(dev
, MAP_0
);
8229 if (get_imsm_disk_slot(map
, victim
) >= 0)
8233 /* delete the victim if it is no longer being
8239 /* We know that 'manager' isn't touching anything,
8240 * so it is safe to delete
8242 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8243 if ((*dlp
)->index
== victim
)
8246 /* victim may be on the missing list */
8248 for (dlp
= &super
->missing
; *dlp
;
8249 dlp
= &(*dlp
)->next
)
8250 if ((*dlp
)->index
== victim
)
8252 imsm_delete(super
, dlp
, victim
);
8259 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8260 struct intel_super
*super
,
8263 struct dl
*new_disk
;
8264 struct intel_dev
*id
;
8266 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8267 int disk_count
= u
->old_raid_disks
;
8268 void **tofree
= NULL
;
8269 int devices_to_reshape
= 1;
8270 struct imsm_super
*mpb
= super
->anchor
;
8272 unsigned int dev_id
;
8274 dprintf("(enter)\n");
8276 /* enable spares to use in array */
8277 for (i
= 0; i
< delta_disks
; i
++) {
8278 new_disk
= get_disk_super(super
,
8279 major(u
->new_disks
[i
]),
8280 minor(u
->new_disks
[i
]));
8281 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8282 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8283 new_disk
, new_disk
->index
);
8284 if ((new_disk
== NULL
) ||
8285 ((new_disk
->index
>= 0) &&
8286 (new_disk
->index
< u
->old_raid_disks
)))
8287 goto update_reshape_exit
;
8288 new_disk
->index
= disk_count
++;
8289 /* slot to fill in autolayout
8291 new_disk
->raiddisk
= new_disk
->index
;
8292 new_disk
->disk
.status
|=
8294 new_disk
->disk
.status
&= ~SPARE_DISK
;
8297 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8298 mpb
->num_raid_devs
);
8299 /* manage changes in volume
8301 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8302 void **sp
= *space_list
;
8303 struct imsm_dev
*newdev
;
8304 struct imsm_map
*newmap
, *oldmap
;
8306 for (id
= super
->devlist
; id
; id
= id
->next
) {
8307 if (id
->index
== dev_id
)
8316 /* Copy the dev, but not (all of) the map */
8317 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8318 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8319 newmap
= get_imsm_map(newdev
, MAP_0
);
8320 /* Copy the current map */
8321 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8322 /* update one device only
8324 if (devices_to_reshape
) {
8325 dprintf("imsm: modifying subdev: %i\n",
8327 devices_to_reshape
--;
8328 newdev
->vol
.migr_state
= 1;
8329 newdev
->vol
.curr_migr_unit
= 0;
8330 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8331 newmap
->num_members
= u
->new_raid_disks
;
8332 for (i
= 0; i
< delta_disks
; i
++) {
8333 set_imsm_ord_tbl_ent(newmap
,
8334 u
->old_raid_disks
+ i
,
8335 u
->old_raid_disks
+ i
);
8337 /* New map is correct, now need to save old map
8339 newmap
= get_imsm_map(newdev
, MAP_1
);
8340 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8342 imsm_set_array_size(newdev
, -1);
8345 sp
= (void **)id
->dev
;
8350 /* Clear migration record */
8351 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8354 *space_list
= tofree
;
8357 update_reshape_exit
:
8362 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8363 struct intel_super
*super
,
8366 struct imsm_dev
*dev
= NULL
;
8367 struct intel_dev
*dv
;
8368 struct imsm_dev
*dev_new
;
8369 struct imsm_map
*map
;
8373 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8374 if (dv
->index
== (unsigned int)u
->subarray
) {
8382 map
= get_imsm_map(dev
, MAP_0
);
8384 if (u
->direction
== R10_TO_R0
) {
8385 /* Number of failed disks must be half of initial disk number */
8386 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8387 (map
->num_members
/ 2))
8390 /* iterate through devices to mark removed disks as spare */
8391 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8392 if (dm
->disk
.status
& FAILED_DISK
) {
8393 int idx
= dm
->index
;
8394 /* update indexes on the disk list */
8395 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8396 the index values will end up being correct.... NB */
8397 for (du
= super
->disks
; du
; du
= du
->next
)
8398 if (du
->index
> idx
)
8400 /* mark as spare disk */
8405 map
->num_members
= map
->num_members
/ 2;
8406 map
->map_state
= IMSM_T_STATE_NORMAL
;
8407 map
->num_domains
= 1;
8408 map
->raid_level
= 0;
8409 map
->failed_disk_num
= -1;
8412 if (u
->direction
== R0_TO_R10
) {
8414 /* update slots in current disk list */
8415 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8419 /* create new *missing* disks */
8420 for (i
= 0; i
< map
->num_members
; i
++) {
8421 space
= *space_list
;
8424 *space_list
= *space
;
8426 memcpy(du
, super
->disks
, sizeof(*du
));
8430 du
->index
= (i
* 2) + 1;
8431 sprintf((char *)du
->disk
.serial
,
8432 " MISSING_%d", du
->index
);
8433 sprintf((char *)du
->serial
,
8434 "MISSING_%d", du
->index
);
8435 du
->next
= super
->missing
;
8436 super
->missing
= du
;
8438 /* create new dev and map */
8439 space
= *space_list
;
8442 *space_list
= *space
;
8443 dev_new
= (void *)space
;
8444 memcpy(dev_new
, dev
, sizeof(*dev
));
8445 /* update new map */
8446 map
= get_imsm_map(dev_new
, MAP_0
);
8447 map
->num_members
= map
->num_members
* 2;
8448 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8449 map
->num_domains
= 2;
8450 map
->raid_level
= 1;
8451 /* replace dev<->dev_new */
8454 /* update disk order table */
8455 for (du
= super
->disks
; du
; du
= du
->next
)
8457 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8458 for (du
= super
->missing
; du
; du
= du
->next
)
8459 if (du
->index
>= 0) {
8460 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8461 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8467 static void imsm_process_update(struct supertype
*st
,
8468 struct metadata_update
*update
)
8471 * crack open the metadata_update envelope to find the update record
8472 * update can be one of:
8473 * update_reshape_container_disks - all the arrays in the container
8474 * are being reshaped to have more devices. We need to mark
8475 * the arrays for general migration and convert selected spares
8476 * into active devices.
8477 * update_activate_spare - a spare device has replaced a failed
8478 * device in an array, update the disk_ord_tbl. If this disk is
8479 * present in all member arrays then also clear the SPARE_DISK
8481 * update_create_array
8483 * update_rename_array
8484 * update_add_remove_disk
8486 struct intel_super
*super
= st
->sb
;
8487 struct imsm_super
*mpb
;
8488 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8490 /* update requires a larger buf but the allocation failed */
8491 if (super
->next_len
&& !super
->next_buf
) {
8492 super
->next_len
= 0;
8496 if (super
->next_buf
) {
8497 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8499 super
->len
= super
->next_len
;
8500 super
->buf
= super
->next_buf
;
8502 super
->next_len
= 0;
8503 super
->next_buf
= NULL
;
8506 mpb
= super
->anchor
;
8509 case update_general_migration_checkpoint
: {
8510 struct intel_dev
*id
;
8511 struct imsm_update_general_migration_checkpoint
*u
=
8512 (void *)update
->buf
;
8514 dprintf("called for update_general_migration_checkpoint\n");
8516 /* find device under general migration */
8517 for (id
= super
->devlist
; id
; id
= id
->next
) {
8518 if (is_gen_migration(id
->dev
)) {
8519 id
->dev
->vol
.curr_migr_unit
=
8520 __cpu_to_le32(u
->curr_migr_unit
);
8521 super
->updates_pending
++;
8526 case update_takeover
: {
8527 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8528 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8529 imsm_update_version_info(super
);
8530 super
->updates_pending
++;
8535 case update_reshape_container_disks
: {
8536 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8537 if (apply_reshape_container_disks_update(
8538 u
, super
, &update
->space_list
))
8539 super
->updates_pending
++;
8542 case update_reshape_migration
: {
8543 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8544 if (apply_reshape_migration_update(
8545 u
, super
, &update
->space_list
))
8546 super
->updates_pending
++;
8549 case update_size_change
: {
8550 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8551 if (apply_size_change_update(u
, super
))
8552 super
->updates_pending
++;
8555 case update_activate_spare
: {
8556 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8557 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8558 super
->updates_pending
++;
8561 case update_create_array
: {
8562 /* someone wants to create a new array, we need to be aware of
8563 * a few races/collisions:
8564 * 1/ 'Create' called by two separate instances of mdadm
8565 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8566 * devices that have since been assimilated via
8568 * In the event this update can not be carried out mdadm will
8569 * (FIX ME) notice that its update did not take hold.
8571 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8572 struct intel_dev
*dv
;
8573 struct imsm_dev
*dev
;
8574 struct imsm_map
*map
, *new_map
;
8575 unsigned long long start
, end
;
8576 unsigned long long new_start
, new_end
;
8578 struct disk_info
*inf
;
8581 /* handle racing creates: first come first serve */
8582 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8583 dprintf("subarray %d already defined\n", u
->dev_idx
);
8587 /* check update is next in sequence */
8588 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8589 dprintf("can not create array %d expected index %d\n",
8590 u
->dev_idx
, mpb
->num_raid_devs
);
8594 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8595 new_start
= pba_of_lba0(new_map
);
8596 new_end
= new_start
+ blocks_per_member(new_map
);
8597 inf
= get_disk_info(u
);
8599 /* handle activate_spare versus create race:
8600 * check to make sure that overlapping arrays do not include
8603 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8604 dev
= get_imsm_dev(super
, i
);
8605 map
= get_imsm_map(dev
, MAP_0
);
8606 start
= pba_of_lba0(map
);
8607 end
= start
+ blocks_per_member(map
);
8608 if ((new_start
>= start
&& new_start
<= end
) ||
8609 (start
>= new_start
&& start
<= new_end
))
8614 if (disks_overlap(super
, i
, u
)) {
8615 dprintf("arrays overlap\n");
8620 /* check that prepare update was successful */
8621 if (!update
->space
) {
8622 dprintf("prepare update failed\n");
8626 /* check that all disks are still active before committing
8627 * changes. FIXME: could we instead handle this by creating a
8628 * degraded array? That's probably not what the user expects,
8629 * so better to drop this update on the floor.
8631 for (i
= 0; i
< new_map
->num_members
; i
++) {
8632 dl
= serial_to_dl(inf
[i
].serial
, super
);
8634 dprintf("disk disappeared\n");
8639 super
->updates_pending
++;
8641 /* convert spares to members and fixup ord_tbl */
8642 for (i
= 0; i
< new_map
->num_members
; i
++) {
8643 dl
= serial_to_dl(inf
[i
].serial
, super
);
8644 if (dl
->index
== -1) {
8645 dl
->index
= mpb
->num_disks
;
8647 dl
->disk
.status
|= CONFIGURED_DISK
;
8648 dl
->disk
.status
&= ~SPARE_DISK
;
8650 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8655 update
->space
= NULL
;
8656 imsm_copy_dev(dev
, &u
->dev
);
8657 dv
->index
= u
->dev_idx
;
8658 dv
->next
= super
->devlist
;
8659 super
->devlist
= dv
;
8660 mpb
->num_raid_devs
++;
8662 imsm_update_version_info(super
);
8665 /* mdmon knows how to release update->space, but not
8666 * ((struct intel_dev *) update->space)->dev
8668 if (update
->space
) {
8674 case update_kill_array
: {
8675 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8676 int victim
= u
->dev_idx
;
8677 struct active_array
*a
;
8678 struct intel_dev
**dp
;
8679 struct imsm_dev
*dev
;
8681 /* sanity check that we are not affecting the uuid of
8682 * active arrays, or deleting an active array
8684 * FIXME when immutable ids are available, but note that
8685 * we'll also need to fixup the invalidated/active
8686 * subarray indexes in mdstat
8688 for (a
= st
->arrays
; a
; a
= a
->next
)
8689 if (a
->info
.container_member
>= victim
)
8691 /* by definition if mdmon is running at least one array
8692 * is active in the container, so checking
8693 * mpb->num_raid_devs is just extra paranoia
8695 dev
= get_imsm_dev(super
, victim
);
8696 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8697 dprintf("failed to delete subarray-%d\n", victim
);
8701 for (dp
= &super
->devlist
; *dp
;)
8702 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8705 if ((*dp
)->index
> (unsigned)victim
)
8709 mpb
->num_raid_devs
--;
8710 super
->updates_pending
++;
8713 case update_rename_array
: {
8714 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8715 char name
[MAX_RAID_SERIAL_LEN
+1];
8716 int target
= u
->dev_idx
;
8717 struct active_array
*a
;
8718 struct imsm_dev
*dev
;
8720 /* sanity check that we are not affecting the uuid of
8723 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8724 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8725 for (a
= st
->arrays
; a
; a
= a
->next
)
8726 if (a
->info
.container_member
== target
)
8728 dev
= get_imsm_dev(super
, u
->dev_idx
);
8729 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8730 dprintf("failed to rename subarray-%d\n", target
);
8734 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8735 super
->updates_pending
++;
8738 case update_add_remove_disk
: {
8739 /* we may be able to repair some arrays if disks are
8740 * being added, check the status of add_remove_disk
8741 * if discs has been added.
8743 if (add_remove_disk_update(super
)) {
8744 struct active_array
*a
;
8746 super
->updates_pending
++;
8747 for (a
= st
->arrays
; a
; a
= a
->next
)
8748 a
->check_degraded
= 1;
8753 pr_err("error: unsuported process update type:(type: %d)\n", type
);
8757 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8759 static int imsm_prepare_update(struct supertype
*st
,
8760 struct metadata_update
*update
)
8763 * Allocate space to hold new disk entries, raid-device entries or a new
8764 * mpb if necessary. The manager synchronously waits for updates to
8765 * complete in the monitor, so new mpb buffers allocated here can be
8766 * integrated by the monitor thread without worrying about live pointers
8767 * in the manager thread.
8769 enum imsm_update_type type
;
8770 struct intel_super
*super
= st
->sb
;
8771 struct imsm_super
*mpb
= super
->anchor
;
8775 if (update
->len
< (int)sizeof(type
))
8778 type
= *(enum imsm_update_type
*) update
->buf
;
8781 case update_general_migration_checkpoint
:
8782 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
8784 dprintf("called for update_general_migration_checkpoint\n");
8786 case update_takeover
: {
8787 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8788 if (update
->len
< (int)sizeof(*u
))
8790 if (u
->direction
== R0_TO_R10
) {
8791 void **tail
= (void **)&update
->space_list
;
8792 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8793 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8794 int num_members
= map
->num_members
;
8797 /* allocate memory for added disks */
8798 for (i
= 0; i
< num_members
; i
++) {
8799 size
= sizeof(struct dl
);
8800 space
= xmalloc(size
);
8805 /* allocate memory for new device */
8806 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8807 (num_members
* sizeof(__u32
));
8808 space
= xmalloc(size
);
8812 len
= disks_to_mpb_size(num_members
* 2);
8817 case update_reshape_container_disks
: {
8818 /* Every raid device in the container is about to
8819 * gain some more devices, and we will enter a
8821 * So each 'imsm_map' will be bigger, and the imsm_vol
8822 * will now hold 2 of them.
8823 * Thus we need new 'struct imsm_dev' allocations sized
8824 * as sizeof_imsm_dev but with more devices in both maps.
8826 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8827 struct intel_dev
*dl
;
8828 void **space_tail
= (void**)&update
->space_list
;
8830 if (update
->len
< (int)sizeof(*u
))
8833 dprintf("for update_reshape\n");
8835 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8836 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8838 if (u
->new_raid_disks
> u
->old_raid_disks
)
8839 size
+= sizeof(__u32
)*2*
8840 (u
->new_raid_disks
- u
->old_raid_disks
);
8847 len
= disks_to_mpb_size(u
->new_raid_disks
);
8848 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8851 case update_reshape_migration
: {
8852 /* for migration level 0->5 we need to add disks
8853 * so the same as for container operation we will copy
8854 * device to the bigger location.
8855 * in memory prepared device and new disk area are prepared
8856 * for usage in process update
8858 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8859 struct intel_dev
*id
;
8860 void **space_tail
= (void **)&update
->space_list
;
8863 int current_level
= -1;
8865 if (update
->len
< (int)sizeof(*u
))
8868 dprintf("for update_reshape\n");
8870 /* add space for bigger array in update
8872 for (id
= super
->devlist
; id
; id
= id
->next
) {
8873 if (id
->index
== (unsigned)u
->subdev
) {
8874 size
= sizeof_imsm_dev(id
->dev
, 1);
8875 if (u
->new_raid_disks
> u
->old_raid_disks
)
8876 size
+= sizeof(__u32
)*2*
8877 (u
->new_raid_disks
- u
->old_raid_disks
);
8885 if (update
->space_list
== NULL
)
8888 /* add space for disk in update
8890 size
= sizeof(struct dl
);
8896 /* add spare device to update
8898 for (id
= super
->devlist
; id
; id
= id
->next
)
8899 if (id
->index
== (unsigned)u
->subdev
) {
8900 struct imsm_dev
*dev
;
8901 struct imsm_map
*map
;
8903 dev
= get_imsm_dev(super
, u
->subdev
);
8904 map
= get_imsm_map(dev
, MAP_0
);
8905 current_level
= map
->raid_level
;
8908 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8909 struct mdinfo
*spares
;
8911 spares
= get_spares_for_grow(st
);
8919 makedev(dev
->disk
.major
,
8921 dl
= get_disk_super(super
,
8924 dl
->index
= u
->old_raid_disks
;
8930 len
= disks_to_mpb_size(u
->new_raid_disks
);
8931 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8934 case update_size_change
: {
8935 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
8939 case update_activate_spare
: {
8940 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
8944 case update_create_array
: {
8945 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8946 struct intel_dev
*dv
;
8947 struct imsm_dev
*dev
= &u
->dev
;
8948 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8950 struct disk_info
*inf
;
8954 if (update
->len
< (int)sizeof(*u
))
8957 inf
= get_disk_info(u
);
8958 len
= sizeof_imsm_dev(dev
, 1);
8959 /* allocate a new super->devlist entry */
8960 dv
= xmalloc(sizeof(*dv
));
8961 dv
->dev
= xmalloc(len
);
8964 /* count how many spares will be converted to members */
8965 for (i
= 0; i
< map
->num_members
; i
++) {
8966 dl
= serial_to_dl(inf
[i
].serial
, super
);
8968 /* hmm maybe it failed?, nothing we can do about
8973 if (count_memberships(dl
, super
) == 0)
8976 len
+= activate
* sizeof(struct imsm_disk
);
8979 case update_kill_array
: {
8980 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
8984 case update_rename_array
: {
8985 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
8989 case update_add_remove_disk
:
8990 /* no update->len needed */
8996 /* check if we need a larger metadata buffer */
8997 if (super
->next_buf
)
8998 buf_len
= super
->next_len
;
9000 buf_len
= super
->len
;
9002 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
9003 /* ok we need a larger buf than what is currently allocated
9004 * if this allocation fails process_update will notice that
9005 * ->next_len is set and ->next_buf is NULL
9007 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
9008 if (super
->next_buf
)
9009 free(super
->next_buf
);
9011 super
->next_len
= buf_len
;
9012 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
9013 memset(super
->next_buf
, 0, buf_len
);
9015 super
->next_buf
= NULL
;
9020 /* must be called while manager is quiesced */
9021 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
9023 struct imsm_super
*mpb
= super
->anchor
;
9025 struct imsm_dev
*dev
;
9026 struct imsm_map
*map
;
9027 int i
, j
, num_members
;
9030 dprintf("deleting device[%d] from imsm_super\n", index
);
9032 /* shift all indexes down one */
9033 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
9034 if (iter
->index
> (int)index
)
9036 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
9037 if (iter
->index
> (int)index
)
9040 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9041 dev
= get_imsm_dev(super
, i
);
9042 map
= get_imsm_map(dev
, MAP_0
);
9043 num_members
= map
->num_members
;
9044 for (j
= 0; j
< num_members
; j
++) {
9045 /* update ord entries being careful not to propagate
9046 * ord-flags to the first map
9048 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
9050 if (ord_to_idx(ord
) <= index
)
9053 map
= get_imsm_map(dev
, MAP_0
);
9054 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
9055 map
= get_imsm_map(dev
, MAP_1
);
9057 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
9062 super
->updates_pending
++;
9064 struct dl
*dl
= *dlp
;
9066 *dlp
= (*dlp
)->next
;
9067 __free_imsm_disk(dl
);
9070 #endif /* MDASSEMBLE */
9072 static void close_targets(int *targets
, int new_disks
)
9079 for (i
= 0; i
< new_disks
; i
++) {
9080 if (targets
[i
] >= 0) {
9087 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
9088 struct intel_super
*super
,
9089 struct imsm_dev
*dev
)
9095 struct imsm_map
*map
;
9098 ret_val
= raid_disks
/2;
9099 /* check map if all disks pairs not failed
9102 map
= get_imsm_map(dev
, MAP_0
);
9103 for (i
= 0; i
< ret_val
; i
++) {
9104 int degradation
= 0;
9105 if (get_imsm_disk(super
, i
) == NULL
)
9107 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9109 if (degradation
== 2)
9112 map
= get_imsm_map(dev
, MAP_1
);
9113 /* if there is no second map
9114 * result can be returned
9118 /* check degradation in second map
9120 for (i
= 0; i
< ret_val
; i
++) {
9121 int degradation
= 0;
9122 if (get_imsm_disk(super
, i
) == NULL
)
9124 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9126 if (degradation
== 2)
9140 /*******************************************************************************
9141 * Function: open_backup_targets
9142 * Description: Function opens file descriptors for all devices given in
9145 * info : general array info
9146 * raid_disks : number of disks
9147 * raid_fds : table of device's file descriptors
9148 * super : intel super for raid10 degradation check
9149 * dev : intel device for raid10 degradation check
9153 ******************************************************************************/
9154 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9155 struct intel_super
*super
, struct imsm_dev
*dev
)
9161 for (i
= 0; i
< raid_disks
; i
++)
9164 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9167 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9168 dprintf("disk is faulty!!\n");
9172 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
9173 (sd
->disk
.raid_disk
< 0))
9176 dn
= map_dev(sd
->disk
.major
,
9178 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9179 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9180 pr_err("cannot open component\n");
9185 /* check if maximum array degradation level is not exceeded
9187 if ((raid_disks
- opened
) >
9188 imsm_get_allowed_degradation(info
->new_level
,
9191 pr_err("Not enough disks can be opened.\n");
9192 close_targets(raid_fds
, raid_disks
);
9198 /*******************************************************************************
9199 * Function: validate_container_imsm
9200 * Description: This routine validates container after assemble,
9201 * eg. if devices in container are under the same controller.
9204 * info : linked list with info about devices used in array
9208 ******************************************************************************/
9209 int validate_container_imsm(struct mdinfo
*info
)
9211 if (check_env("IMSM_NO_PLATFORM"))
9214 struct sys_dev
*idev
;
9215 struct sys_dev
*hba
= NULL
;
9216 struct sys_dev
*intel_devices
= find_intel_devices();
9217 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
9220 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9221 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9230 pr_err("WARNING - Cannot detect HBA for device %s!\n",
9231 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
9235 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
9238 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
9239 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
9241 struct sys_dev
*hba2
= NULL
;
9242 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9243 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9251 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
9252 get_orom_by_device_id(hba2
->dev_id
);
9254 if (hba2
&& hba
->type
!= hba2
->type
) {
9255 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
9256 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
9260 if ((orom
!= orom2
) || ((hba
->type
== SYS_DEV_VMD
) && (hba
!= hba2
))) {
9261 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
9262 " This operation is not supported and can lead to data loss.\n");
9267 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
9268 " This operation is not supported and can lead to data loss.\n");
9276 /*******************************************************************************
9277 * Function: init_migr_record_imsm
9278 * Description: Function inits imsm migration record
9280 * super : imsm internal array info
9281 * dev : device under migration
9282 * info : general array info to find the smallest device
9285 ******************************************************************************/
9286 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9287 struct mdinfo
*info
)
9289 struct intel_super
*super
= st
->sb
;
9290 struct migr_record
*migr_rec
= super
->migr_rec
;
9292 unsigned long long dsize
, dev_sectors
;
9293 long long unsigned min_dev_sectors
= -1LLU;
9297 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9298 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9299 unsigned long long num_migr_units
;
9300 unsigned long long array_blocks
;
9302 memset(migr_rec
, 0, sizeof(struct migr_record
));
9303 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9305 /* only ascending reshape supported now */
9306 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9308 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9309 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9310 migr_rec
->dest_depth_per_unit
*=
9311 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9312 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9313 migr_rec
->blocks_per_unit
=
9314 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9315 migr_rec
->dest_depth_per_unit
=
9316 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9317 array_blocks
= info
->component_size
* new_data_disks
;
9319 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9321 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9323 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9325 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9326 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9328 /* Find the smallest dev */
9329 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9330 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9331 fd
= dev_open(nm
, O_RDONLY
);
9334 get_dev_size(fd
, NULL
, &dsize
);
9335 dev_sectors
= dsize
/ 512;
9336 if (dev_sectors
< min_dev_sectors
)
9337 min_dev_sectors
= dev_sectors
;
9340 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9341 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9343 write_imsm_migr_rec(st
);
9348 /*******************************************************************************
9349 * Function: save_backup_imsm
9350 * Description: Function saves critical data stripes to Migration Copy Area
9351 * and updates the current migration unit status.
9352 * Use restore_stripes() to form a destination stripe,
9353 * and to write it to the Copy Area.
9355 * st : supertype information
9356 * dev : imsm device that backup is saved for
9357 * info : general array info
9358 * buf : input buffer
9359 * length : length of data to backup (blocks_per_unit)
9363 ******************************************************************************/
9364 int save_backup_imsm(struct supertype
*st
,
9365 struct imsm_dev
*dev
,
9366 struct mdinfo
*info
,
9371 struct intel_super
*super
= st
->sb
;
9372 unsigned long long *target_offsets
= NULL
;
9373 int *targets
= NULL
;
9375 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9376 int new_disks
= map_dest
->num_members
;
9377 int dest_layout
= 0;
9379 unsigned long long start
;
9380 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9382 targets
= xmalloc(new_disks
* sizeof(int));
9384 for (i
= 0; i
< new_disks
; i
++)
9387 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
9389 start
= info
->reshape_progress
* 512;
9390 for (i
= 0; i
< new_disks
; i
++) {
9391 target_offsets
[i
] = (unsigned long long)
9392 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9393 /* move back copy area adderss, it will be moved forward
9394 * in restore_stripes() using start input variable
9396 target_offsets
[i
] -= start
/data_disks
;
9399 if (open_backup_targets(info
, new_disks
, targets
,
9403 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9404 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9406 if (restore_stripes(targets
, /* list of dest devices */
9407 target_offsets
, /* migration record offsets */
9410 map_dest
->raid_level
,
9412 -1, /* source backup file descriptor */
9413 0, /* input buf offset
9414 * always 0 buf is already offseted */
9418 pr_err("Error restoring stripes\n");
9426 close_targets(targets
, new_disks
);
9429 free(target_offsets
);
9434 /*******************************************************************************
9435 * Function: save_checkpoint_imsm
9436 * Description: Function called for current unit status update
9437 * in the migration record. It writes it to disk.
9439 * super : imsm internal array info
9440 * info : general array info
9444 * 2: failure, means no valid migration record
9445 * / no general migration in progress /
9446 ******************************************************************************/
9447 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9449 struct intel_super
*super
= st
->sb
;
9450 unsigned long long blocks_per_unit
;
9451 unsigned long long curr_migr_unit
;
9453 if (load_imsm_migr_rec(super
, info
) != 0) {
9454 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
9458 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9459 if (blocks_per_unit
== 0) {
9460 dprintf("imsm: no migration in progress.\n");
9463 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9464 /* check if array is alligned to copy area
9465 * if it is not alligned, add one to current migration unit value
9466 * this can happend on array reshape finish only
9468 if (info
->reshape_progress
% blocks_per_unit
)
9471 super
->migr_rec
->curr_migr_unit
=
9472 __cpu_to_le32(curr_migr_unit
);
9473 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9474 super
->migr_rec
->dest_1st_member_lba
=
9475 __cpu_to_le32(curr_migr_unit
*
9476 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9477 if (write_imsm_migr_rec(st
) < 0) {
9478 dprintf("imsm: Cannot write migration record outside backup area\n");
9485 /*******************************************************************************
9486 * Function: recover_backup_imsm
9487 * Description: Function recovers critical data from the Migration Copy Area
9488 * while assembling an array.
9490 * super : imsm internal array info
9491 * info : general array info
9493 * 0 : success (or there is no data to recover)
9495 ******************************************************************************/
9496 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9498 struct intel_super
*super
= st
->sb
;
9499 struct migr_record
*migr_rec
= super
->migr_rec
;
9500 struct imsm_map
*map_dest
= NULL
;
9501 struct intel_dev
*id
= NULL
;
9502 unsigned long long read_offset
;
9503 unsigned long long write_offset
;
9505 int *targets
= NULL
;
9506 int new_disks
, i
, err
;
9509 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9510 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9512 int skipped_disks
= 0;
9514 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9518 /* recover data only during assemblation */
9519 if (strncmp(buffer
, "inactive", 8) != 0)
9521 /* no data to recover */
9522 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9524 if (curr_migr_unit
>= num_migr_units
)
9527 /* find device during reshape */
9528 for (id
= super
->devlist
; id
; id
= id
->next
)
9529 if (is_gen_migration(id
->dev
))
9534 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9535 new_disks
= map_dest
->num_members
;
9537 read_offset
= (unsigned long long)
9538 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9540 write_offset
= ((unsigned long long)
9541 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9542 pba_of_lba0(map_dest
)) * 512;
9544 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9545 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9547 targets
= xcalloc(new_disks
, sizeof(int));
9549 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9550 pr_err("Cannot open some devices belonging to array.\n");
9554 for (i
= 0; i
< new_disks
; i
++) {
9555 if (targets
[i
] < 0) {
9559 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9560 pr_err("Cannot seek to block: %s\n",
9565 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9566 pr_err("Cannot read copy area block: %s\n",
9571 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9572 pr_err("Cannot seek to block: %s\n",
9577 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9578 pr_err("Cannot restore block: %s\n",
9585 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9589 pr_err("Cannot restore data from backup. Too many failed disks\n");
9593 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9594 /* ignore error == 2, this can mean end of reshape here
9596 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
9602 for (i
= 0; i
< new_disks
; i
++)
9611 static char disk_by_path
[] = "/dev/disk/by-path/";
9613 static const char *imsm_get_disk_controller_domain(const char *path
)
9615 char disk_path
[PATH_MAX
];
9619 strcpy(disk_path
, disk_by_path
);
9620 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9621 if (stat(disk_path
, &st
) == 0) {
9622 struct sys_dev
* hba
;
9625 path
= devt_to_devpath(st
.st_rdev
);
9628 hba
= find_disk_attached_hba(-1, path
);
9629 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9631 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9635 dprintf("path: %s hba: %s attached: %s\n",
9636 path
, (hba
) ? hba
->path
: "NULL", drv
);
9642 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
9644 static char devnm
[32];
9645 char subdev_name
[20];
9646 struct mdstat_ent
*mdstat
;
9648 sprintf(subdev_name
, "%d", subdev
);
9649 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9653 strcpy(devnm
, mdstat
->devnm
);
9654 free_mdstat(mdstat
);
9658 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9659 struct geo_params
*geo
,
9660 int *old_raid_disks
,
9663 /* currently we only support increasing the number of devices
9664 * for a container. This increases the number of device for each
9665 * member array. They must all be RAID0 or RAID5.
9668 struct mdinfo
*info
, *member
;
9669 int devices_that_can_grow
= 0;
9671 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
9673 if (geo
->size
> 0 ||
9674 geo
->level
!= UnSet
||
9675 geo
->layout
!= UnSet
||
9676 geo
->chunksize
!= 0 ||
9677 geo
->raid_disks
== UnSet
) {
9678 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
9682 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9683 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
9687 info
= container_content_imsm(st
, NULL
);
9688 for (member
= info
; member
; member
= member
->next
) {
9691 dprintf("imsm: checking device_num: %i\n",
9692 member
->container_member
);
9694 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9695 /* we work on container for Online Capacity Expansion
9696 * only so raid_disks has to grow
9698 dprintf("imsm: for container operation raid disks increase is required\n");
9702 if ((info
->array
.level
!= 0) &&
9703 (info
->array
.level
!= 5)) {
9704 /* we cannot use this container with other raid level
9706 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
9710 /* check for platform support
9711 * for this raid level configuration
9713 struct intel_super
*super
= st
->sb
;
9714 if (!is_raid_level_supported(super
->orom
,
9715 member
->array
.level
,
9717 dprintf("platform does not support raid%d with %d disk%s\n",
9720 geo
->raid_disks
> 1 ? "s" : "");
9723 /* check if component size is aligned to chunk size
9725 if (info
->component_size
%
9726 (info
->array
.chunk_size
/512)) {
9727 dprintf("Component size is not aligned to chunk size\n");
9732 if (*old_raid_disks
&&
9733 info
->array
.raid_disks
!= *old_raid_disks
)
9735 *old_raid_disks
= info
->array
.raid_disks
;
9737 /* All raid5 and raid0 volumes in container
9738 * have to be ready for Online Capacity Expansion
9739 * so they need to be assembled. We have already
9740 * checked that no recovery etc is happening.
9742 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
9743 st
->container_devnm
);
9744 if (result
== NULL
) {
9745 dprintf("imsm: cannot find array\n");
9748 devices_that_can_grow
++;
9751 if (!member
&& devices_that_can_grow
)
9755 dprintf("Container operation allowed\n");
9757 dprintf("Error: %i\n", ret_val
);
9762 /* Function: get_spares_for_grow
9763 * Description: Allocates memory and creates list of spare devices
9764 * avaliable in container. Checks if spare drive size is acceptable.
9765 * Parameters: Pointer to the supertype structure
9766 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9769 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9771 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9772 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9775 /******************************************************************************
9776 * function: imsm_create_metadata_update_for_reshape
9777 * Function creates update for whole IMSM container.
9779 ******************************************************************************/
9780 static int imsm_create_metadata_update_for_reshape(
9781 struct supertype
*st
,
9782 struct geo_params
*geo
,
9784 struct imsm_update_reshape
**updatep
)
9786 struct intel_super
*super
= st
->sb
;
9787 struct imsm_super
*mpb
= super
->anchor
;
9788 int update_memory_size
= 0;
9789 struct imsm_update_reshape
*u
= NULL
;
9790 struct mdinfo
*spares
= NULL
;
9792 int delta_disks
= 0;
9795 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
9797 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9799 /* size of all update data without anchor */
9800 update_memory_size
= sizeof(struct imsm_update_reshape
);
9802 /* now add space for spare disks that we need to add. */
9803 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9805 u
= xcalloc(1, update_memory_size
);
9806 u
->type
= update_reshape_container_disks
;
9807 u
->old_raid_disks
= old_raid_disks
;
9808 u
->new_raid_disks
= geo
->raid_disks
;
9810 /* now get spare disks list
9812 spares
= get_spares_for_grow(st
);
9815 || delta_disks
> spares
->array
.spare_disks
) {
9816 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
9821 /* we have got spares
9822 * update disk list in imsm_disk list table in anchor
9824 dprintf("imsm: %i spares are available.\n\n",
9825 spares
->array
.spare_disks
);
9828 for (i
= 0; i
< delta_disks
; i
++) {
9833 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9835 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9836 dl
->index
= mpb
->num_disks
;
9846 dprintf("imsm: reshape update preparation :");
9847 if (i
== delta_disks
) {
9848 dprintf_cont(" OK\n");
9850 return update_memory_size
;
9853 dprintf_cont(" Error\n");
9858 /******************************************************************************
9859 * function: imsm_create_metadata_update_for_size_change()
9860 * Creates update for IMSM array for array size change.
9862 ******************************************************************************/
9863 static int imsm_create_metadata_update_for_size_change(
9864 struct supertype
*st
,
9865 struct geo_params
*geo
,
9866 struct imsm_update_size_change
**updatep
)
9868 struct intel_super
*super
= st
->sb
;
9869 int update_memory_size
= 0;
9870 struct imsm_update_size_change
*u
= NULL
;
9872 dprintf("(enter) New size = %llu\n", geo
->size
);
9874 /* size of all update data without anchor */
9875 update_memory_size
= sizeof(struct imsm_update_size_change
);
9877 u
= xcalloc(1, update_memory_size
);
9878 u
->type
= update_size_change
;
9879 u
->subdev
= super
->current_vol
;
9880 u
->new_size
= geo
->size
;
9882 dprintf("imsm: reshape update preparation : OK\n");
9885 return update_memory_size
;
9888 /******************************************************************************
9889 * function: imsm_create_metadata_update_for_migration()
9890 * Creates update for IMSM array.
9892 ******************************************************************************/
9893 static int imsm_create_metadata_update_for_migration(
9894 struct supertype
*st
,
9895 struct geo_params
*geo
,
9896 struct imsm_update_reshape_migration
**updatep
)
9898 struct intel_super
*super
= st
->sb
;
9899 int update_memory_size
= 0;
9900 struct imsm_update_reshape_migration
*u
= NULL
;
9901 struct imsm_dev
*dev
;
9902 int previous_level
= -1;
9904 dprintf("(enter) New Level = %i\n", geo
->level
);
9906 /* size of all update data without anchor */
9907 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9909 u
= xcalloc(1, update_memory_size
);
9910 u
->type
= update_reshape_migration
;
9911 u
->subdev
= super
->current_vol
;
9912 u
->new_level
= geo
->level
;
9913 u
->new_layout
= geo
->layout
;
9914 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9915 u
->new_disks
[0] = -1;
9916 u
->new_chunksize
= -1;
9918 dev
= get_imsm_dev(super
, u
->subdev
);
9920 struct imsm_map
*map
;
9922 map
= get_imsm_map(dev
, MAP_0
);
9924 int current_chunk_size
=
9925 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9927 if (geo
->chunksize
!= current_chunk_size
) {
9928 u
->new_chunksize
= geo
->chunksize
/ 1024;
9929 dprintf("imsm: chunk size change from %i to %i\n",
9930 current_chunk_size
, u
->new_chunksize
);
9932 previous_level
= map
->raid_level
;
9935 if ((geo
->level
== 5) && (previous_level
== 0)) {
9936 struct mdinfo
*spares
= NULL
;
9938 u
->new_raid_disks
++;
9939 spares
= get_spares_for_grow(st
);
9940 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9943 update_memory_size
= 0;
9944 dprintf("error: cannot get spare device for requested migration");
9949 dprintf("imsm: reshape update preparation : OK\n");
9952 return update_memory_size
;
9955 static void imsm_update_metadata_locally(struct supertype
*st
,
9958 struct metadata_update mu
;
9963 mu
.space_list
= NULL
;
9965 if (imsm_prepare_update(st
, &mu
))
9966 imsm_process_update(st
, &mu
);
9968 while (mu
.space_list
) {
9969 void **space
= mu
.space_list
;
9970 mu
.space_list
= *space
;
9975 /***************************************************************************
9976 * Function: imsm_analyze_change
9977 * Description: Function analyze change for single volume
9978 * and validate if transition is supported
9979 * Parameters: Geometry parameters, supertype structure,
9980 * metadata change direction (apply/rollback)
9981 * Returns: Operation type code on success, -1 if fail
9982 ****************************************************************************/
9983 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9984 struct geo_params
*geo
,
9991 /* number of added/removed disks in operation result */
9992 int devNumChange
= 0;
9993 /* imsm compatible layout value for array geometry verification */
9994 int imsm_layout
= -1;
9996 struct imsm_dev
*dev
;
9997 struct intel_super
*super
;
9998 unsigned long long current_size
;
9999 unsigned long long free_size
;
10000 unsigned long long max_size
;
10003 getinfo_super_imsm_volume(st
, &info
, NULL
);
10004 if ((geo
->level
!= info
.array
.level
) &&
10005 (geo
->level
>= 0) &&
10006 (geo
->level
!= UnSet
)) {
10007 switch (info
.array
.level
) {
10009 if (geo
->level
== 5) {
10010 change
= CH_MIGRATION
;
10011 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
10012 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
10014 goto analyse_change_exit
;
10016 imsm_layout
= geo
->layout
;
10018 devNumChange
= 1; /* parity disk added */
10019 } else if (geo
->level
== 10) {
10020 change
= CH_TAKEOVER
;
10022 devNumChange
= 2; /* two mirrors added */
10023 imsm_layout
= 0x102; /* imsm supported layout */
10028 if (geo
->level
== 0) {
10029 change
= CH_TAKEOVER
;
10031 devNumChange
= -(geo
->raid_disks
/2);
10032 imsm_layout
= 0; /* imsm raid0 layout */
10036 if (change
== -1) {
10037 pr_err("Error. Level Migration from %d to %d not supported!\n",
10038 info
.array
.level
, geo
->level
);
10039 goto analyse_change_exit
;
10042 geo
->level
= info
.array
.level
;
10044 if ((geo
->layout
!= info
.array
.layout
)
10045 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
10046 change
= CH_MIGRATION
;
10047 if ((info
.array
.layout
== 0)
10048 && (info
.array
.level
== 5)
10049 && (geo
->layout
== 5)) {
10050 /* reshape 5 -> 4 */
10051 } else if ((info
.array
.layout
== 5)
10052 && (info
.array
.level
== 5)
10053 && (geo
->layout
== 0)) {
10054 /* reshape 4 -> 5 */
10058 pr_err("Error. Layout Migration from %d to %d not supported!\n",
10059 info
.array
.layout
, geo
->layout
);
10061 goto analyse_change_exit
;
10064 geo
->layout
= info
.array
.layout
;
10065 if (imsm_layout
== -1)
10066 imsm_layout
= info
.array
.layout
;
10069 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
10070 && (geo
->chunksize
!= info
.array
.chunk_size
))
10071 change
= CH_MIGRATION
;
10073 geo
->chunksize
= info
.array
.chunk_size
;
10075 chunk
= geo
->chunksize
/ 1024;
10078 dev
= get_imsm_dev(super
, super
->current_vol
);
10079 data_disks
= imsm_num_data_members(dev
, MAP_0
);
10080 /* compute current size per disk member
10082 current_size
= info
.custom_array_size
/ data_disks
;
10084 if ((geo
->size
> 0) && (geo
->size
!= MAX_SIZE
)) {
10085 /* align component size
10087 geo
->size
= imsm_component_size_aligment_check(
10088 get_imsm_raid_level(dev
->vol
.map
),
10091 if (geo
->size
== 0) {
10092 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
10094 goto analyse_change_exit
;
10098 if ((current_size
!= geo
->size
) && (geo
->size
> 0)) {
10099 if (change
!= -1) {
10100 pr_err("Error. Size change should be the only one at a time.\n");
10102 goto analyse_change_exit
;
10104 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
10105 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
10106 super
->current_vol
, st
->devnm
);
10107 goto analyse_change_exit
;
10109 /* check the maximum available size
10111 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
10112 0, chunk
, &free_size
);
10114 /* Cannot find maximum available space
10118 max_size
= free_size
+ current_size
;
10119 /* align component size
10121 max_size
= imsm_component_size_aligment_check(
10122 get_imsm_raid_level(dev
->vol
.map
),
10126 if (geo
->size
== MAX_SIZE
) {
10127 /* requested size change to the maximum available size
10129 if (max_size
== 0) {
10130 pr_err("Error. Cannot find maximum available space.\n");
10132 goto analyse_change_exit
;
10134 geo
->size
= max_size
;
10137 if ((direction
== ROLLBACK_METADATA_CHANGES
)) {
10138 /* accept size for rollback only
10141 /* round size due to metadata compatibility
10143 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
10144 << SECT_PER_MB_SHIFT
;
10145 dprintf("Prepare update for size change to %llu\n",
10147 if (current_size
>= geo
->size
) {
10148 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
10149 current_size
, geo
->size
);
10150 goto analyse_change_exit
;
10152 if (max_size
&& geo
->size
> max_size
) {
10153 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
10154 max_size
, geo
->size
);
10155 goto analyse_change_exit
;
10158 geo
->size
*= data_disks
;
10159 geo
->raid_disks
= dev
->vol
.map
->num_members
;
10160 change
= CH_ARRAY_SIZE
;
10162 if (!validate_geometry_imsm(st
,
10165 geo
->raid_disks
+ devNumChange
,
10167 geo
->size
, INVALID_SECTORS
,
10172 struct intel_super
*super
= st
->sb
;
10173 struct imsm_super
*mpb
= super
->anchor
;
10175 if (mpb
->num_raid_devs
> 1) {
10176 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
10182 analyse_change_exit
:
10183 if ((direction
== ROLLBACK_METADATA_CHANGES
) &&
10184 ((change
== CH_MIGRATION
) || (change
== CH_TAKEOVER
))) {
10185 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
10191 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10193 struct intel_super
*super
= st
->sb
;
10194 struct imsm_update_takeover
*u
;
10196 u
= xmalloc(sizeof(struct imsm_update_takeover
));
10198 u
->type
= update_takeover
;
10199 u
->subarray
= super
->current_vol
;
10201 /* 10->0 transition */
10202 if (geo
->level
== 0)
10203 u
->direction
= R10_TO_R0
;
10205 /* 0->10 transition */
10206 if (geo
->level
== 10)
10207 u
->direction
= R0_TO_R10
;
10209 /* update metadata locally */
10210 imsm_update_metadata_locally(st
, u
,
10211 sizeof(struct imsm_update_takeover
));
10212 /* and possibly remotely */
10213 if (st
->update_tail
)
10214 append_metadata_update(st
, u
,
10215 sizeof(struct imsm_update_takeover
));
10222 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
10224 int layout
, int chunksize
, int raid_disks
,
10225 int delta_disks
, char *backup
, char *dev
,
10226 int direction
, int verbose
)
10229 struct geo_params geo
;
10231 dprintf("(enter)\n");
10233 memset(&geo
, 0, sizeof(struct geo_params
));
10235 geo
.dev_name
= dev
;
10236 strcpy(geo
.devnm
, st
->devnm
);
10239 geo
.layout
= layout
;
10240 geo
.chunksize
= chunksize
;
10241 geo
.raid_disks
= raid_disks
;
10242 if (delta_disks
!= UnSet
)
10243 geo
.raid_disks
+= delta_disks
;
10245 dprintf("for level : %i\n", geo
.level
);
10246 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
10248 if (experimental() == 0)
10251 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
10252 /* On container level we can only increase number of devices. */
10253 dprintf("imsm: info: Container operation\n");
10254 int old_raid_disks
= 0;
10256 if (imsm_reshape_is_allowed_on_container(
10257 st
, &geo
, &old_raid_disks
, direction
)) {
10258 struct imsm_update_reshape
*u
= NULL
;
10261 len
= imsm_create_metadata_update_for_reshape(
10262 st
, &geo
, old_raid_disks
, &u
);
10265 dprintf("imsm: Cannot prepare update\n");
10266 goto exit_imsm_reshape_super
;
10270 /* update metadata locally */
10271 imsm_update_metadata_locally(st
, u
, len
);
10272 /* and possibly remotely */
10273 if (st
->update_tail
)
10274 append_metadata_update(st
, u
, len
);
10279 pr_err("(imsm) Operation is not allowed on this container\n");
10282 /* On volume level we support following operations
10283 * - takeover: raid10 -> raid0; raid0 -> raid10
10284 * - chunk size migration
10285 * - migration: raid5 -> raid0; raid0 -> raid5
10287 struct intel_super
*super
= st
->sb
;
10288 struct intel_dev
*dev
= super
->devlist
;
10290 dprintf("imsm: info: Volume operation\n");
10291 /* find requested device */
10294 imsm_find_array_devnm_by_subdev(
10295 dev
->index
, st
->container_devnm
);
10296 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
10301 pr_err("Cannot find %s (%s) subarray\n",
10302 geo
.dev_name
, geo
.devnm
);
10303 goto exit_imsm_reshape_super
;
10305 super
->current_vol
= dev
->index
;
10306 change
= imsm_analyze_change(st
, &geo
, direction
);
10309 ret_val
= imsm_takeover(st
, &geo
);
10311 case CH_MIGRATION
: {
10312 struct imsm_update_reshape_migration
*u
= NULL
;
10314 imsm_create_metadata_update_for_migration(
10317 dprintf("imsm: Cannot prepare update\n");
10321 /* update metadata locally */
10322 imsm_update_metadata_locally(st
, u
, len
);
10323 /* and possibly remotely */
10324 if (st
->update_tail
)
10325 append_metadata_update(st
, u
, len
);
10330 case CH_ARRAY_SIZE
: {
10331 struct imsm_update_size_change
*u
= NULL
;
10333 imsm_create_metadata_update_for_size_change(
10336 dprintf("imsm: Cannot prepare update\n");
10340 /* update metadata locally */
10341 imsm_update_metadata_locally(st
, u
, len
);
10342 /* and possibly remotely */
10343 if (st
->update_tail
)
10344 append_metadata_update(st
, u
, len
);
10354 exit_imsm_reshape_super
:
10355 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10359 /*******************************************************************************
10360 * Function: wait_for_reshape_imsm
10361 * Description: Function writes new sync_max value and waits until
10362 * reshape process reach new position
10364 * sra : general array info
10365 * ndata : number of disks in new array's layout
10368 * 1 : there is no reshape in progress,
10370 ******************************************************************************/
10371 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10373 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
10374 unsigned long long completed
;
10375 /* to_complete : new sync_max position */
10376 unsigned long long to_complete
= sra
->reshape_progress
;
10377 unsigned long long position_to_set
= to_complete
/ ndata
;
10380 dprintf("cannot open reshape_position\n");
10384 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10385 dprintf("cannot read reshape_position (no reshape in progres)\n");
10390 if (completed
> position_to_set
) {
10391 dprintf("wrong next position to set %llu (%llu)\n",
10392 to_complete
, position_to_set
);
10396 dprintf("Position set: %llu\n", position_to_set
);
10397 if (sysfs_set_num(sra
, NULL
, "sync_max",
10398 position_to_set
) != 0) {
10399 dprintf("cannot set reshape position to %llu\n",
10407 int timeout
= 3000;
10408 sysfs_wait(fd
, &timeout
);
10409 if (sysfs_get_str(sra
, NULL
, "sync_action",
10411 strncmp(action
, "reshape", 7) != 0) {
10415 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10416 dprintf("cannot read reshape_position (in loop)\n");
10420 } while (completed
< position_to_set
);
10426 /*******************************************************************************
10427 * Function: check_degradation_change
10428 * Description: Check that array hasn't become failed.
10430 * info : for sysfs access
10431 * sources : source disks descriptors
10432 * degraded: previous degradation level
10434 * degradation level
10435 ******************************************************************************/
10436 int check_degradation_change(struct mdinfo
*info
,
10440 unsigned long long new_degraded
;
10443 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10444 if ((rv
== -1) || (new_degraded
!= (unsigned long long)degraded
)) {
10445 /* check each device to ensure it is still working */
10448 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10449 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10451 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10453 if (sysfs_get_str(info
,
10454 sd
, "state", sbuf
, 20) < 0 ||
10455 strstr(sbuf
, "faulty") ||
10456 strstr(sbuf
, "in_sync") == NULL
) {
10457 /* this device is dead */
10458 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10459 if (sd
->disk
.raid_disk
>= 0 &&
10460 sources
[sd
->disk
.raid_disk
] >= 0) {
10462 sd
->disk
.raid_disk
]);
10463 sources
[sd
->disk
.raid_disk
] =
10472 return new_degraded
;
10475 /*******************************************************************************
10476 * Function: imsm_manage_reshape
10477 * Description: Function finds array under reshape and it manages reshape
10478 * process. It creates stripes backups (if required) and sets
10481 * afd : Backup handle (nattive) - not used
10482 * sra : general array info
10483 * reshape : reshape parameters - not used
10484 * st : supertype structure
10485 * blocks : size of critical section [blocks]
10486 * fds : table of source device descriptor
10487 * offsets : start of array (offest per devices)
10489 * destfd : table of destination device descriptor
10490 * destoffsets : table of destination offsets (per device)
10492 * 1 : success, reshape is done
10494 ******************************************************************************/
10495 static int imsm_manage_reshape(
10496 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10497 struct supertype
*st
, unsigned long backup_blocks
,
10498 int *fds
, unsigned long long *offsets
,
10499 int dests
, int *destfd
, unsigned long long *destoffsets
)
10502 struct intel_super
*super
= st
->sb
;
10503 struct intel_dev
*dv
= NULL
;
10504 struct imsm_dev
*dev
= NULL
;
10505 struct imsm_map
*map_src
;
10506 int migr_vol_qan
= 0;
10507 int ndata
, odata
; /* [bytes] */
10508 int chunk
; /* [bytes] */
10509 struct migr_record
*migr_rec
;
10511 unsigned int buf_size
; /* [bytes] */
10512 unsigned long long max_position
; /* array size [bytes] */
10513 unsigned long long next_step
; /* [blocks]/[bytes] */
10514 unsigned long long old_data_stripe_length
;
10515 unsigned long long start_src
; /* [bytes] */
10516 unsigned long long start
; /* [bytes] */
10517 unsigned long long start_buf_shift
; /* [bytes] */
10519 int source_layout
= 0;
10521 if (!fds
|| !offsets
|| !sra
)
10524 /* Find volume during the reshape */
10525 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10526 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10527 && dv
->dev
->vol
.migr_state
== 1) {
10532 /* Only one volume can migrate at the same time */
10533 if (migr_vol_qan
!= 1) {
10534 pr_err(": %s", migr_vol_qan
?
10535 "Number of migrating volumes greater than 1\n" :
10536 "There is no volume during migrationg\n");
10540 map_src
= get_imsm_map(dev
, MAP_1
);
10541 if (map_src
== NULL
)
10544 ndata
= imsm_num_data_members(dev
, MAP_0
);
10545 odata
= imsm_num_data_members(dev
, MAP_1
);
10547 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10548 old_data_stripe_length
= odata
* chunk
;
10550 migr_rec
= super
->migr_rec
;
10552 /* initialize migration record for start condition */
10553 if (sra
->reshape_progress
== 0)
10554 init_migr_record_imsm(st
, dev
, sra
);
10556 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10557 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
10560 /* Save checkpoint to update migration record for current
10561 * reshape position (in md). It can be farther than current
10562 * reshape position in metadata.
10564 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10565 /* ignore error == 2, this can mean end of reshape here
10567 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
10572 /* size for data */
10573 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10574 /* extend buffer size for parity disk */
10575 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10576 /* add space for stripe aligment */
10577 buf_size
+= old_data_stripe_length
;
10578 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10579 dprintf("imsm: Cannot allocate checpoint buffer\n");
10583 max_position
= sra
->component_size
* ndata
;
10584 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10586 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10587 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10588 /* current reshape position [blocks] */
10589 unsigned long long current_position
=
10590 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10591 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10592 unsigned long long border
;
10594 /* Check that array hasn't become failed.
10596 degraded
= check_degradation_change(sra
, fds
, degraded
);
10597 if (degraded
> 1) {
10598 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
10602 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10604 if ((current_position
+ next_step
) > max_position
)
10605 next_step
= max_position
- current_position
;
10607 start
= current_position
* 512;
10609 /* align reading start to old geometry */
10610 start_buf_shift
= start
% old_data_stripe_length
;
10611 start_src
= start
- start_buf_shift
;
10613 border
= (start_src
/ odata
) - (start
/ ndata
);
10615 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10616 /* save critical stripes to buf
10617 * start - start address of current unit
10618 * to backup [bytes]
10619 * start_src - start address of current unit
10620 * to backup alligned to source array
10623 unsigned long long next_step_filler
= 0;
10624 unsigned long long copy_length
= next_step
* 512;
10626 /* allign copy area length to stripe in old geometry */
10627 next_step_filler
= ((copy_length
+ start_buf_shift
)
10628 % old_data_stripe_length
);
10629 if (next_step_filler
)
10630 next_step_filler
= (old_data_stripe_length
10631 - next_step_filler
);
10632 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
10633 start
, start_src
, copy_length
,
10634 start_buf_shift
, next_step_filler
);
10636 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10637 chunk
, map_src
->raid_level
,
10638 source_layout
, 0, NULL
, start_src
,
10640 next_step_filler
+ start_buf_shift
,
10642 dprintf("imsm: Cannot save stripes to buffer\n");
10645 /* Convert data to destination format and store it
10646 * in backup general migration area
10648 if (save_backup_imsm(st
, dev
, sra
,
10649 buf
+ start_buf_shift
, copy_length
)) {
10650 dprintf("imsm: Cannot save stripes to target devices\n");
10653 if (save_checkpoint_imsm(st
, sra
,
10654 UNIT_SRC_IN_CP_AREA
)) {
10655 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
10659 /* set next step to use whole border area */
10660 border
/= next_step
;
10662 next_step
*= border
;
10664 /* When data backed up, checkpoint stored,
10665 * kick the kernel to reshape unit of data
10667 next_step
= next_step
+ sra
->reshape_progress
;
10668 /* limit next step to array max position */
10669 if (next_step
> max_position
)
10670 next_step
= max_position
;
10671 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10672 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10673 sra
->reshape_progress
= next_step
;
10675 /* wait until reshape finish */
10676 if (wait_for_reshape_imsm(sra
, ndata
)) {
10677 dprintf("wait_for_reshape_imsm returned error!\n");
10683 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10684 /* ignore error == 2, this can mean end of reshape here
10686 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
10692 /* clear migr_rec on disks after successful migration */
10695 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
10696 for (d
= super
->disks
; d
; d
= d
->next
) {
10697 if (d
->index
< 0 || is_failed(&d
->disk
))
10699 unsigned long long dsize
;
10701 get_dev_size(d
->fd
, NULL
, &dsize
);
10702 if (lseek64(d
->fd
, dsize
- MIGR_REC_POSITION
,
10704 if (write(d
->fd
, super
->migr_rec_buf
,
10705 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
10706 perror("Write migr_rec failed");
10710 /* return '1' if done */
10714 /* See Grow.c: abort_reshape() for further explanation */
10715 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
10716 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
10717 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
10722 #endif /* MDASSEMBLE */
10724 struct superswitch super_imsm
= {
10726 .examine_super
= examine_super_imsm
,
10727 .brief_examine_super
= brief_examine_super_imsm
,
10728 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10729 .export_examine_super
= export_examine_super_imsm
,
10730 .detail_super
= detail_super_imsm
,
10731 .brief_detail_super
= brief_detail_super_imsm
,
10732 .write_init_super
= write_init_super_imsm
,
10733 .validate_geometry
= validate_geometry_imsm
,
10734 .add_to_super
= add_to_super_imsm
,
10735 .remove_from_super
= remove_from_super_imsm
,
10736 .detail_platform
= detail_platform_imsm
,
10737 .export_detail_platform
= export_detail_platform_imsm
,
10738 .kill_subarray
= kill_subarray_imsm
,
10739 .update_subarray
= update_subarray_imsm
,
10740 .load_container
= load_container_imsm
,
10741 .default_geometry
= default_geometry_imsm
,
10742 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10743 .reshape_super
= imsm_reshape_super
,
10744 .manage_reshape
= imsm_manage_reshape
,
10745 .recover_backup
= recover_backup_imsm
,
10746 .copy_metadata
= copy_metadata_imsm
,
10748 .match_home
= match_home_imsm
,
10749 .uuid_from_super
= uuid_from_super_imsm
,
10750 .getinfo_super
= getinfo_super_imsm
,
10751 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10752 .update_super
= update_super_imsm
,
10754 .avail_size
= avail_size_imsm
,
10755 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10757 .compare_super
= compare_super_imsm
,
10759 .load_super
= load_super_imsm
,
10760 .init_super
= init_super_imsm
,
10761 .store_super
= store_super_imsm
,
10762 .free_super
= free_super_imsm
,
10763 .match_metadata_desc
= match_metadata_desc_imsm
,
10764 .container_content
= container_content_imsm
,
10765 .validate_container
= validate_container_imsm
,
10772 .open_new
= imsm_open_new
,
10773 .set_array_state
= imsm_set_array_state
,
10774 .set_disk
= imsm_set_disk
,
10775 .sync_metadata
= imsm_sync_metadata
,
10776 .activate_spare
= imsm_activate_spare
,
10777 .process_update
= imsm_process_update
,
10778 .prepare_update
= imsm_prepare_update
,
10779 #endif /* MDASSEMBLE */