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
;
369 unsigned int sector_size
; /* sector size of used member drives */
373 __u8 serial
[MAX_RAID_SERIAL_LEN
];
376 struct imsm_disk disk
;
379 struct extent
*e
; /* for determining freespace @ create */
380 int raiddisk
; /* slot to fill in autolayout */
382 } *disks
, *current_disk
;
383 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
385 struct dl
*missing
; /* disks removed while we weren't looking */
386 struct bbm_log
*bbm_log
;
387 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
388 const struct imsm_orom
*orom
; /* platform firmware support */
389 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
393 struct imsm_disk disk
;
394 #define IMSM_UNKNOWN_OWNER (-1)
396 struct intel_disk
*next
;
400 unsigned long long start
, size
;
403 /* definitions of reshape process types */
404 enum imsm_reshape_type
{
410 /* definition of messages passed to imsm_process_update */
411 enum imsm_update_type
{
412 update_activate_spare
,
416 update_add_remove_disk
,
417 update_reshape_container_disks
,
418 update_reshape_migration
,
420 update_general_migration_checkpoint
,
424 struct imsm_update_activate_spare
{
425 enum imsm_update_type type
;
429 struct imsm_update_activate_spare
*next
;
435 unsigned long long size
;
442 enum takeover_direction
{
446 struct imsm_update_takeover
{
447 enum imsm_update_type type
;
449 enum takeover_direction direction
;
452 struct imsm_update_reshape
{
453 enum imsm_update_type type
;
457 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
460 struct imsm_update_reshape_migration
{
461 enum imsm_update_type type
;
464 /* fields for array migration changes
471 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
474 struct imsm_update_size_change
{
475 enum imsm_update_type type
;
480 struct imsm_update_general_migration_checkpoint
{
481 enum imsm_update_type type
;
482 __u32 curr_migr_unit
;
486 __u8 serial
[MAX_RAID_SERIAL_LEN
];
489 struct imsm_update_create_array
{
490 enum imsm_update_type type
;
495 struct imsm_update_kill_array
{
496 enum imsm_update_type type
;
500 struct imsm_update_rename_array
{
501 enum imsm_update_type type
;
502 __u8 name
[MAX_RAID_SERIAL_LEN
];
506 struct imsm_update_add_remove_disk
{
507 enum imsm_update_type type
;
510 static const char *_sys_dev_type
[] = {
511 [SYS_DEV_UNKNOWN
] = "Unknown",
512 [SYS_DEV_SAS
] = "SAS",
513 [SYS_DEV_SATA
] = "SATA",
514 [SYS_DEV_NVME
] = "NVMe",
515 [SYS_DEV_VMD
] = "VMD"
518 const char *get_sys_dev_type(enum sys_dev_type type
)
520 if (type
>= SYS_DEV_MAX
)
521 type
= SYS_DEV_UNKNOWN
;
523 return _sys_dev_type
[type
];
526 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
528 struct intel_hba
*result
= xmalloc(sizeof(*result
));
530 result
->type
= device
->type
;
531 result
->path
= xstrdup(device
->path
);
533 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
539 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
541 struct intel_hba
*result
;
543 for (result
= hba
; result
; result
= result
->next
) {
544 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
550 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
552 struct intel_hba
*hba
;
554 /* check if disk attached to Intel HBA */
555 hba
= find_intel_hba(super
->hba
, device
);
558 /* Check if HBA is already attached to super */
559 if (super
->hba
== NULL
) {
560 super
->hba
= alloc_intel_hba(device
);
565 /* Intel metadata allows for all disks attached to the same type HBA.
566 * Do not support HBA types mixing
568 if (device
->type
!= hba
->type
)
571 /* Multiple same type HBAs can be used if they share the same OROM */
572 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
574 if (device_orom
!= super
->orom
)
580 hba
->next
= alloc_intel_hba(device
);
584 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
586 struct sys_dev
*list
, *elem
;
589 if ((list
= find_intel_devices()) == NULL
)
593 disk_path
= (char *) devname
;
595 disk_path
= diskfd_to_devpath(fd
);
600 for (elem
= list
; elem
; elem
= elem
->next
)
601 if (path_attached_to_hba(disk_path
, elem
->path
))
604 if (disk_path
!= devname
)
610 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
613 static struct supertype
*match_metadata_desc_imsm(char *arg
)
615 struct supertype
*st
;
617 if (strcmp(arg
, "imsm") != 0 &&
618 strcmp(arg
, "default") != 0
622 st
= xcalloc(1, sizeof(*st
));
623 st
->ss
= &super_imsm
;
624 st
->max_devs
= IMSM_MAX_DEVICES
;
625 st
->minor_version
= 0;
631 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
633 return &mpb
->sig
[MPB_SIG_LEN
];
637 /* retrieve a disk directly from the anchor when the anchor is known to be
638 * up-to-date, currently only at load time
640 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
642 if (index
>= mpb
->num_disks
)
644 return &mpb
->disk
[index
];
647 /* retrieve the disk description based on a index of the disk
650 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
654 for (d
= super
->disks
; d
; d
= d
->next
)
655 if (d
->index
== index
)
660 /* retrieve a disk from the parsed metadata */
661 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
665 dl
= get_imsm_dl_disk(super
, index
);
672 /* generate a checksum directly from the anchor when the anchor is known to be
673 * up-to-date, currently only at load or write_super after coalescing
675 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
677 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
678 __u32
*p
= (__u32
*) mpb
;
682 sum
+= __le32_to_cpu(*p
);
686 return sum
- __le32_to_cpu(mpb
->check_sum
);
689 static size_t sizeof_imsm_map(struct imsm_map
*map
)
691 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
694 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
696 /* A device can have 2 maps if it is in the middle of a migration.
698 * MAP_0 - we return the first map
699 * MAP_1 - we return the second map if it exists, else NULL
700 * MAP_X - we return the second map if it exists, else the first
702 struct imsm_map
*map
= &dev
->vol
.map
[0];
703 struct imsm_map
*map2
= NULL
;
705 if (dev
->vol
.migr_state
)
706 map2
= (void *)map
+ sizeof_imsm_map(map
);
708 switch (second_map
) {
725 /* return the size of the device.
726 * migr_state increases the returned size if map[0] were to be duplicated
728 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
730 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
731 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
733 /* migrating means an additional map */
734 if (dev
->vol
.migr_state
)
735 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
737 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
743 /* retrieve disk serial number list from a metadata update */
744 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
747 struct disk_info
*inf
;
749 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
750 sizeof_imsm_dev(&update
->dev
, 0);
756 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
762 if (index
>= mpb
->num_raid_devs
)
765 /* devices start after all disks */
766 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
768 for (i
= 0; i
<= index
; i
++)
770 return _mpb
+ offset
;
772 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
777 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
779 struct intel_dev
*dv
;
781 if (index
>= super
->anchor
->num_raid_devs
)
783 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
784 if (dv
->index
== index
)
791 * == MAP_0 get first map
792 * == MAP_1 get second map
793 * == MAP_X than get map according to the current migr_state
795 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
799 struct imsm_map
*map
;
801 map
= get_imsm_map(dev
, second_map
);
803 /* top byte identifies disk under rebuild */
804 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
807 #define ord_to_idx(ord) (((ord) << 8) >> 8)
808 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
810 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
812 return ord_to_idx(ord
);
815 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
817 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
820 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
825 for (slot
= 0; slot
< map
->num_members
; slot
++) {
826 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
827 if (ord_to_idx(ord
) == idx
)
834 static int get_imsm_raid_level(struct imsm_map
*map
)
836 if (map
->raid_level
== 1) {
837 if (map
->num_members
== 2)
843 return map
->raid_level
;
846 static int cmp_extent(const void *av
, const void *bv
)
848 const struct extent
*a
= av
;
849 const struct extent
*b
= bv
;
850 if (a
->start
< b
->start
)
852 if (a
->start
> b
->start
)
857 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
862 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
863 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
864 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
866 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
873 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
875 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
877 if (lo
== 0 || hi
== 0)
879 *lo
= __le32_to_cpu((unsigned)n
);
880 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
884 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
886 return (unsigned long long)__le32_to_cpu(lo
) |
887 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
890 static unsigned long long total_blocks(struct imsm_disk
*disk
)
894 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
897 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
901 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
904 static unsigned long long blocks_per_member(struct imsm_map
*map
)
908 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
912 static unsigned long long num_data_stripes(struct imsm_map
*map
)
916 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
919 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
921 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
925 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
927 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
930 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
932 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
935 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
937 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
940 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
942 /* find a list of used extents on the given physical device */
943 struct extent
*rv
, *e
;
945 int memberships
= count_memberships(dl
, super
);
948 /* trim the reserved area for spares, so they can join any array
949 * regardless of whether the OROM has assigned sectors from the
950 * IMSM_RESERVED_SECTORS region
953 reservation
= imsm_min_reserved_sectors(super
);
955 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
957 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
960 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
961 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
962 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
964 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
965 e
->start
= pba_of_lba0(map
);
966 e
->size
= blocks_per_member(map
);
970 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
972 /* determine the start of the metadata
973 * when no raid devices are defined use the default
974 * ...otherwise allow the metadata to truncate the value
975 * as is the case with older versions of imsm
978 struct extent
*last
= &rv
[memberships
- 1];
979 unsigned long long remainder
;
981 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
982 /* round down to 1k block to satisfy precision of the kernel
986 /* make sure remainder is still sane */
987 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
988 remainder
= ROUND_UP(super
->len
, 512) >> 9;
989 if (reservation
> remainder
)
990 reservation
= remainder
;
992 e
->start
= total_blocks(&dl
->disk
) - reservation
;
997 /* try to determine how much space is reserved for metadata from
998 * the last get_extents() entry, otherwise fallback to the
1001 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1007 /* for spares just return a minimal reservation which will grow
1008 * once the spare is picked up by an array
1010 if (dl
->index
== -1)
1011 return MPB_SECTOR_CNT
;
1013 e
= get_extents(super
, dl
);
1015 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1017 /* scroll to last entry */
1018 for (i
= 0; e
[i
].size
; i
++)
1021 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1028 static int is_spare(struct imsm_disk
*disk
)
1030 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1033 static int is_configured(struct imsm_disk
*disk
)
1035 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1038 static int is_failed(struct imsm_disk
*disk
)
1040 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1043 /* try to determine how much space is reserved for metadata from
1044 * the last get_extents() entry on the smallest active disk,
1045 * otherwise fallback to the default
1047 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1051 unsigned long long min_active
;
1053 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1054 struct dl
*dl
, *dl_min
= NULL
;
1060 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1063 unsigned long long blocks
= total_blocks(&dl
->disk
);
1064 if (blocks
< min_active
|| min_active
== 0) {
1066 min_active
= blocks
;
1072 /* find last lba used by subarrays on the smallest active disk */
1073 e
= get_extents(super
, dl_min
);
1076 for (i
= 0; e
[i
].size
; i
++)
1079 remainder
= min_active
- e
[i
].start
;
1082 /* to give priority to recovery we should not require full
1083 IMSM_RESERVED_SECTORS from the spare */
1084 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1086 /* if real reservation is smaller use that value */
1087 return (remainder
< rv
) ? remainder
: rv
;
1090 /* Return minimum size of a spare that can be used in this array*/
1091 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1093 struct intel_super
*super
= st
->sb
;
1097 unsigned long long rv
= 0;
1101 /* find first active disk in array */
1103 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1107 /* find last lba used by subarrays */
1108 e
= get_extents(super
, dl
);
1111 for (i
= 0; e
[i
].size
; i
++)
1114 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1117 /* add the amount of space needed for metadata */
1118 rv
= rv
+ imsm_min_reserved_sectors(super
);
1123 static int is_gen_migration(struct imsm_dev
*dev
);
1126 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1127 struct imsm_dev
*dev
);
1129 static void print_imsm_dev(struct intel_super
*super
,
1130 struct imsm_dev
*dev
,
1136 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1137 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1141 printf("[%.16s]:\n", dev
->volume
);
1142 printf(" UUID : %s\n", uuid
);
1143 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1145 printf(" <-- %d", get_imsm_raid_level(map2
));
1147 printf(" Members : %d", map
->num_members
);
1149 printf(" <-- %d", map2
->num_members
);
1151 printf(" Slots : [");
1152 for (i
= 0; i
< map
->num_members
; i
++) {
1153 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1154 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1159 for (i
= 0; i
< map2
->num_members
; i
++) {
1160 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1161 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1166 printf(" Failed disk : ");
1167 if (map
->failed_disk_num
== 0xff)
1170 printf("%i", map
->failed_disk_num
);
1172 slot
= get_imsm_disk_slot(map
, disk_idx
);
1174 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1175 printf(" This Slot : %d%s\n", slot
,
1176 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1178 printf(" This Slot : ?\n");
1179 sz
= __le32_to_cpu(dev
->size_high
);
1181 sz
+= __le32_to_cpu(dev
->size_low
);
1182 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1183 human_size(sz
* 512));
1184 sz
= blocks_per_member(map
);
1185 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1186 human_size(sz
* 512));
1187 printf(" Sector Offset : %llu\n",
1189 printf(" Num Stripes : %llu\n",
1190 num_data_stripes(map
));
1191 printf(" Chunk Size : %u KiB",
1192 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1194 printf(" <-- %u KiB",
1195 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1197 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1198 printf(" Migrate State : ");
1199 if (dev
->vol
.migr_state
) {
1200 if (migr_type(dev
) == MIGR_INIT
)
1201 printf("initialize\n");
1202 else if (migr_type(dev
) == MIGR_REBUILD
)
1203 printf("rebuild\n");
1204 else if (migr_type(dev
) == MIGR_VERIFY
)
1206 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1207 printf("general migration\n");
1208 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1209 printf("state change\n");
1210 else if (migr_type(dev
) == MIGR_REPAIR
)
1213 printf("<unknown:%d>\n", migr_type(dev
));
1216 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1217 if (dev
->vol
.migr_state
) {
1218 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1220 printf(" <-- %s", map_state_str
[map
->map_state
]);
1221 printf("\n Checkpoint : %u ",
1222 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1223 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1226 printf("(%llu)", (unsigned long long)
1227 blocks_per_migr_unit(super
, dev
));
1230 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1233 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1235 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1238 if (index
< -1 || !disk
)
1242 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1244 printf(" Disk%02d Serial : %s\n", index
, str
);
1246 printf(" Disk Serial : %s\n", str
);
1247 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1248 is_configured(disk
) ? " active" : "",
1249 is_failed(disk
) ? " failed" : "");
1250 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1251 sz
= total_blocks(disk
) - reserved
;
1252 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1253 human_size(sz
* 512));
1256 void examine_migr_rec_imsm(struct intel_super
*super
)
1258 struct migr_record
*migr_rec
= super
->migr_rec
;
1259 struct imsm_super
*mpb
= super
->anchor
;
1262 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1263 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1264 struct imsm_map
*map
;
1267 if (is_gen_migration(dev
) == 0)
1270 printf("\nMigration Record Information:");
1272 /* first map under migration */
1273 map
= get_imsm_map(dev
, MAP_0
);
1275 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1276 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1277 printf(" Empty\n ");
1278 printf("Examine one of first two disks in array\n");
1281 printf("\n Status : ");
1282 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1285 printf("Contains Data\n");
1286 printf(" Current Unit : %u\n",
1287 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1288 printf(" Family : %u\n",
1289 __le32_to_cpu(migr_rec
->family_num
));
1290 printf(" Ascending : %u\n",
1291 __le32_to_cpu(migr_rec
->ascending_migr
));
1292 printf(" Blocks Per Unit : %u\n",
1293 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1294 printf(" Dest. Depth Per Unit : %u\n",
1295 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1296 printf(" Checkpoint Area pba : %u\n",
1297 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1298 printf(" First member lba : %u\n",
1299 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1300 printf(" Total Number of Units : %u\n",
1301 __le32_to_cpu(migr_rec
->num_migr_units
));
1302 printf(" Size of volume : %u\n",
1303 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1304 printf(" Expansion space for LBA64 : %u\n",
1305 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1306 printf(" Record was read from : %u\n",
1307 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1312 #endif /* MDASSEMBLE */
1313 /*******************************************************************************
1314 * function: imsm_check_attributes
1315 * Description: Function checks if features represented by attributes flags
1316 * are supported by mdadm.
1318 * attributes - Attributes read from metadata
1320 * 0 - passed attributes contains unsupported features flags
1321 * 1 - all features are supported
1322 ******************************************************************************/
1323 static int imsm_check_attributes(__u32 attributes
)
1326 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1328 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1330 not_supported
&= attributes
;
1331 if (not_supported
) {
1332 pr_err("(IMSM): Unsupported attributes : %x\n",
1333 (unsigned)__le32_to_cpu(not_supported
));
1334 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1335 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1336 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1338 if (not_supported
& MPB_ATTRIB_2TB
) {
1339 dprintf("\t\tMPB_ATTRIB_2TB\n");
1340 not_supported
^= MPB_ATTRIB_2TB
;
1342 if (not_supported
& MPB_ATTRIB_RAID0
) {
1343 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1344 not_supported
^= MPB_ATTRIB_RAID0
;
1346 if (not_supported
& MPB_ATTRIB_RAID1
) {
1347 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1348 not_supported
^= MPB_ATTRIB_RAID1
;
1350 if (not_supported
& MPB_ATTRIB_RAID10
) {
1351 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1352 not_supported
^= MPB_ATTRIB_RAID10
;
1354 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1355 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1356 not_supported
^= MPB_ATTRIB_RAID1E
;
1358 if (not_supported
& MPB_ATTRIB_RAID5
) {
1359 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1360 not_supported
^= MPB_ATTRIB_RAID5
;
1362 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1363 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1364 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1366 if (not_supported
& MPB_ATTRIB_BBM
) {
1367 dprintf("\t\tMPB_ATTRIB_BBM\n");
1368 not_supported
^= MPB_ATTRIB_BBM
;
1370 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1371 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1372 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1374 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1375 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1376 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1378 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1379 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1380 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1382 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1383 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1384 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1386 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1387 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1388 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1392 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1401 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1403 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1405 struct intel_super
*super
= st
->sb
;
1406 struct imsm_super
*mpb
= super
->anchor
;
1407 char str
[MAX_SIGNATURE_LENGTH
];
1412 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1415 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1416 printf(" Magic : %s\n", str
);
1417 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1418 printf(" Version : %s\n", get_imsm_version(mpb
));
1419 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1420 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1421 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1422 printf(" Attributes : ");
1423 if (imsm_check_attributes(mpb
->attributes
))
1424 printf("All supported\n");
1426 printf("not supported\n");
1427 getinfo_super_imsm(st
, &info
, NULL
);
1428 fname_from_uuid(st
, &info
, nbuf
, ':');
1429 printf(" UUID : %s\n", nbuf
+ 5);
1430 sum
= __le32_to_cpu(mpb
->check_sum
);
1431 printf(" Checksum : %08x %s\n", sum
,
1432 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1433 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1434 printf(" Disks : %d\n", mpb
->num_disks
);
1435 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1436 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1437 if (super
->bbm_log
) {
1438 struct bbm_log
*log
= super
->bbm_log
;
1441 printf("Bad Block Management Log:\n");
1442 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1443 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1444 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1445 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1446 printf(" First Spare : %llx\n",
1447 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1449 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1451 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1453 super
->current_vol
= i
;
1454 getinfo_super_imsm(st
, &info
, NULL
);
1455 fname_from_uuid(st
, &info
, nbuf
, ':');
1456 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1458 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1459 if (i
== super
->disks
->index
)
1461 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1464 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1465 if (dl
->index
== -1)
1466 print_imsm_disk(&dl
->disk
, -1, reserved
);
1468 examine_migr_rec_imsm(super
);
1471 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1473 /* We just write a generic IMSM ARRAY entry */
1476 struct intel_super
*super
= st
->sb
;
1478 if (!super
->anchor
->num_raid_devs
) {
1479 printf("ARRAY metadata=imsm\n");
1483 getinfo_super_imsm(st
, &info
, NULL
);
1484 fname_from_uuid(st
, &info
, nbuf
, ':');
1485 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1488 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1490 /* We just write a generic IMSM ARRAY entry */
1494 struct intel_super
*super
= st
->sb
;
1497 if (!super
->anchor
->num_raid_devs
)
1500 getinfo_super_imsm(st
, &info
, NULL
);
1501 fname_from_uuid(st
, &info
, nbuf
, ':');
1502 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1503 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1505 super
->current_vol
= i
;
1506 getinfo_super_imsm(st
, &info
, NULL
);
1507 fname_from_uuid(st
, &info
, nbuf1
, ':');
1508 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1509 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1513 static void export_examine_super_imsm(struct supertype
*st
)
1515 struct intel_super
*super
= st
->sb
;
1516 struct imsm_super
*mpb
= super
->anchor
;
1520 getinfo_super_imsm(st
, &info
, NULL
);
1521 fname_from_uuid(st
, &info
, nbuf
, ':');
1522 printf("MD_METADATA=imsm\n");
1523 printf("MD_LEVEL=container\n");
1524 printf("MD_UUID=%s\n", nbuf
+5);
1525 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1528 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
1530 /* The second last 512byte sector of the device contains
1531 * the "struct imsm_super" metadata.
1532 * This contains mpb_size which is the size in bytes of the
1533 * extended metadata. This is located immediately before
1535 * We want to read all that, plus the last sector which
1536 * may contain a migration record, and write it all
1540 unsigned long long dsize
, offset
;
1542 struct imsm_super
*sb
;
1545 if (posix_memalign(&buf
, 4096, 4096) != 0)
1548 if (!get_dev_size(from
, NULL
, &dsize
))
1551 if (lseek64(from
, dsize
-1024, 0) < 0)
1553 if (read(from
, buf
, 512) != 512)
1556 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
1559 sectors
= mpb_sectors(sb
) + 2;
1560 offset
= dsize
- sectors
* 512;
1561 if (lseek64(from
, offset
, 0) < 0 ||
1562 lseek64(to
, offset
, 0) < 0)
1564 while (written
< sectors
* 512) {
1565 int n
= sectors
*512 - written
;
1568 if (read(from
, buf
, n
) != n
)
1570 if (write(to
, buf
, n
) != n
)
1581 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1586 getinfo_super_imsm(st
, &info
, NULL
);
1587 fname_from_uuid(st
, &info
, nbuf
, ':');
1588 printf("\n UUID : %s\n", nbuf
+ 5);
1591 static void brief_detail_super_imsm(struct supertype
*st
)
1595 getinfo_super_imsm(st
, &info
, NULL
);
1596 fname_from_uuid(st
, &info
, nbuf
, ':');
1597 printf(" UUID=%s", nbuf
+ 5);
1600 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1601 static void fd2devname(int fd
, char *name
);
1603 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1605 /* dump an unsorted list of devices attached to AHCI Intel storage
1606 * controller, as well as non-connected ports
1608 int hba_len
= strlen(hba_path
) + 1;
1613 unsigned long port_mask
= (1 << port_count
) - 1;
1615 if (port_count
> (int)sizeof(port_mask
) * 8) {
1617 pr_err("port_count %d out of range\n", port_count
);
1621 /* scroll through /sys/dev/block looking for devices attached to
1624 dir
= opendir("/sys/dev/block");
1628 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1639 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1641 path
= devt_to_devpath(makedev(major
, minor
));
1644 if (!path_attached_to_hba(path
, hba_path
)) {
1650 /* retrieve the scsi device type */
1651 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1653 pr_err("failed to allocate 'device'\n");
1657 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1658 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
1660 pr_err("failed to read device type for %s\n",
1666 type
= strtoul(buf
, NULL
, 10);
1668 /* if it's not a disk print the vendor and model */
1669 if (!(type
== 0 || type
== 7 || type
== 14)) {
1672 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1673 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
1674 strncpy(vendor
, buf
, sizeof(vendor
));
1675 vendor
[sizeof(vendor
) - 1] = '\0';
1676 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1677 while (isspace(*c
) || *c
== '\0')
1681 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1682 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
1683 strncpy(model
, buf
, sizeof(model
));
1684 model
[sizeof(model
) - 1] = '\0';
1685 c
= (char *) &model
[sizeof(model
) - 1];
1686 while (isspace(*c
) || *c
== '\0')
1690 if (vendor
[0] && model
[0])
1691 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1693 switch (type
) { /* numbers from hald/linux/device.c */
1694 case 1: sprintf(buf
, "tape"); break;
1695 case 2: sprintf(buf
, "printer"); break;
1696 case 3: sprintf(buf
, "processor"); break;
1698 case 5: sprintf(buf
, "cdrom"); break;
1699 case 6: sprintf(buf
, "scanner"); break;
1700 case 8: sprintf(buf
, "media_changer"); break;
1701 case 9: sprintf(buf
, "comm"); break;
1702 case 12: sprintf(buf
, "raid"); break;
1703 default: sprintf(buf
, "unknown");
1709 /* chop device path to 'host%d' and calculate the port number */
1710 c
= strchr(&path
[hba_len
], '/');
1713 pr_err("%s - invalid path name\n", path
+ hba_len
);
1718 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
1719 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
1723 *c
= '/'; /* repair the full string */
1724 pr_err("failed to determine port number for %s\n",
1731 /* mark this port as used */
1732 port_mask
&= ~(1 << port
);
1734 /* print out the device information */
1736 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1740 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1742 printf(" Port%d : - disk info unavailable -\n", port
);
1744 fd2devname(fd
, buf
);
1745 printf(" Port%d : %s", port
, buf
);
1746 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1747 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1762 for (i
= 0; i
< port_count
; i
++)
1763 if (port_mask
& (1 << i
))
1764 printf(" Port%d : - no device attached -\n", i
);
1770 static int print_vmd_attached_devs(struct sys_dev
*hba
)
1778 if (hba
->type
!= SYS_DEV_VMD
)
1781 /* scroll through /sys/dev/block looking for devices attached to
1784 dir
= opendir("/sys/bus/pci/drivers/nvme");
1788 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1791 /* is 'ent' a device? check that the 'subsystem' link exists and
1792 * that its target matches 'bus'
1794 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
1796 n
= readlink(path
, link
, sizeof(link
));
1797 if (n
< 0 || n
>= (int)sizeof(link
))
1800 c
= strrchr(link
, '/');
1803 if (strncmp("pci", c
+1, strlen("pci")) != 0)
1806 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
1807 /* if not a intel NVMe - skip it*/
1808 if (devpath_to_vendor(path
) != 0x8086)
1811 rp
= realpath(path
, NULL
);
1815 if (path_attached_to_hba(rp
, hba
->path
)) {
1816 printf(" NVMe under VMD : %s\n", rp
);
1825 static void print_found_intel_controllers(struct sys_dev
*elem
)
1827 for (; elem
; elem
= elem
->next
) {
1828 pr_err("found Intel(R) ");
1829 if (elem
->type
== SYS_DEV_SATA
)
1830 fprintf(stderr
, "SATA ");
1831 else if (elem
->type
== SYS_DEV_SAS
)
1832 fprintf(stderr
, "SAS ");
1833 else if (elem
->type
== SYS_DEV_NVME
)
1834 fprintf(stderr
, "NVMe ");
1836 if (elem
->type
== SYS_DEV_VMD
)
1837 fprintf(stderr
, "VMD domain");
1839 fprintf(stderr
, "RAID controller");
1842 fprintf(stderr
, " at %s", elem
->pci_id
);
1843 fprintf(stderr
, ".\n");
1848 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1855 if ((dir
= opendir(hba_path
)) == NULL
)
1858 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1861 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
1862 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
1864 if (*port_count
== 0)
1866 else if (host
< host_base
)
1869 if (host
+ 1 > *port_count
+ host_base
)
1870 *port_count
= host
+ 1 - host_base
;
1876 static void print_imsm_capability(const struct imsm_orom
*orom
)
1878 printf(" Platform : Intel(R) ");
1879 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
1880 printf("Matrix Storage Manager\n");
1882 printf("Rapid Storage Technology%s\n",
1883 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
1884 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
1885 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
1886 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
1887 printf(" RAID Levels :%s%s%s%s%s\n",
1888 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1889 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1890 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1891 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1892 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1893 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1894 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1895 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1896 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1897 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1898 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1899 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1900 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1901 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1902 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1903 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1904 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1905 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1906 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1907 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1908 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1909 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1910 printf(" 2TB volumes :%s supported\n",
1911 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
1912 printf(" 2TB disks :%s supported\n",
1913 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
1914 printf(" Max Disks : %d\n", orom
->tds
);
1915 printf(" Max Volumes : %d per array, %d per %s\n",
1916 orom
->vpa
, orom
->vphba
,
1917 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
1921 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
1923 printf("MD_FIRMWARE_TYPE=imsm\n");
1924 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
1925 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1926 orom
->hotfix_ver
, orom
->build
);
1927 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
1928 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
1929 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
1930 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
1931 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
1932 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
1933 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1934 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
1935 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
1936 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
1937 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
1938 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
1939 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
1940 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
1941 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
1942 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
1943 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
1944 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
1945 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
1946 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
1947 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
1948 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
1949 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
1950 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
1951 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
1952 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
1953 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
1954 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
1957 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
1959 /* There are two components to imsm platform support, the ahci SATA
1960 * controller and the option-rom. To find the SATA controller we
1961 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1962 * controller with the Intel vendor id is present. This approach
1963 * allows mdadm to leverage the kernel's ahci detection logic, with the
1964 * caveat that if ahci.ko is not loaded mdadm will not be able to
1965 * detect platform raid capabilities. The option-rom resides in a
1966 * platform "Adapter ROM". We scan for its signature to retrieve the
1967 * platform capabilities. If raid support is disabled in the BIOS the
1968 * option-rom capability structure will not be available.
1970 struct sys_dev
*list
, *hba
;
1975 if (enumerate_only
) {
1976 if (check_env("IMSM_NO_PLATFORM"))
1978 list
= find_intel_devices();
1981 for (hba
= list
; hba
; hba
= hba
->next
) {
1982 if (find_imsm_capability(hba
)) {
1992 list
= find_intel_devices();
1995 pr_err("no active Intel(R) RAID controller found.\n");
1997 } else if (verbose
> 0)
1998 print_found_intel_controllers(list
);
2000 for (hba
= list
; hba
; hba
= hba
->next
) {
2001 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2003 if (!find_imsm_capability(hba
)) {
2005 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2006 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2007 get_sys_dev_type(hba
->type
));
2013 if (controller_path
&& result
== 1) {
2014 pr_err("no active Intel(R) RAID controller found under %s\n",
2019 const struct orom_entry
*entry
;
2021 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2022 if (entry
->type
== SYS_DEV_VMD
) {
2023 print_imsm_capability(&entry
->orom
);
2024 for (hba
= list
; hba
; hba
= hba
->next
) {
2025 if (hba
->type
== SYS_DEV_VMD
) {
2027 printf(" I/O Controller : %s (%s)\n",
2028 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2029 if (print_vmd_attached_devs(hba
)) {
2031 pr_err("failed to get devices attached to VMD domain.\n");
2040 print_imsm_capability(&entry
->orom
);
2041 if (entry
->type
== SYS_DEV_NVME
) {
2042 for (hba
= list
; hba
; hba
= hba
->next
) {
2043 if (hba
->type
== SYS_DEV_NVME
)
2044 printf(" NVMe Device : %s\n", hba
->path
);
2050 struct devid_list
*devid
;
2051 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2052 hba
= device_by_id(devid
->devid
);
2056 printf(" I/O Controller : %s (%s)\n",
2057 hba
->path
, get_sys_dev_type(hba
->type
));
2058 if (hba
->type
== SYS_DEV_SATA
) {
2059 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2060 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2062 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2073 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2075 struct sys_dev
*list
, *hba
;
2078 list
= find_intel_devices();
2081 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2086 for (hba
= list
; hba
; hba
= hba
->next
) {
2087 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2089 if (!find_imsm_capability(hba
) && verbose
> 0) {
2091 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2092 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2098 const struct orom_entry
*entry
;
2100 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2101 if (entry
->type
== SYS_DEV_VMD
) {
2102 for (hba
= list
; hba
; hba
= hba
->next
)
2103 print_imsm_capability_export(&entry
->orom
);
2106 print_imsm_capability_export(&entry
->orom
);
2114 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2116 /* the imsm metadata format does not specify any host
2117 * identification information. We return -1 since we can never
2118 * confirm nor deny whether a given array is "meant" for this
2119 * host. We rely on compare_super and the 'family_num' fields to
2120 * exclude member disks that do not belong, and we rely on
2121 * mdadm.conf to specify the arrays that should be assembled.
2122 * Auto-assembly may still pick up "foreign" arrays.
2128 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2130 /* The uuid returned here is used for:
2131 * uuid to put into bitmap file (Create, Grow)
2132 * uuid for backup header when saving critical section (Grow)
2133 * comparing uuids when re-adding a device into an array
2134 * In these cases the uuid required is that of the data-array,
2135 * not the device-set.
2136 * uuid to recognise same set when adding a missing device back
2137 * to an array. This is a uuid for the device-set.
2139 * For each of these we can make do with a truncated
2140 * or hashed uuid rather than the original, as long as
2142 * In each case the uuid required is that of the data-array,
2143 * not the device-set.
2145 /* imsm does not track uuid's so we synthesis one using sha1 on
2146 * - The signature (Which is constant for all imsm array, but no matter)
2147 * - the orig_family_num of the container
2148 * - the index number of the volume
2149 * - the 'serial' number of the volume.
2150 * Hopefully these are all constant.
2152 struct intel_super
*super
= st
->sb
;
2155 struct sha1_ctx ctx
;
2156 struct imsm_dev
*dev
= NULL
;
2159 /* some mdadm versions failed to set ->orig_family_num, in which
2160 * case fall back to ->family_num. orig_family_num will be
2161 * fixed up with the first metadata update.
2163 family_num
= super
->anchor
->orig_family_num
;
2164 if (family_num
== 0)
2165 family_num
= super
->anchor
->family_num
;
2166 sha1_init_ctx(&ctx
);
2167 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2168 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2169 if (super
->current_vol
>= 0)
2170 dev
= get_imsm_dev(super
, super
->current_vol
);
2172 __u32 vol
= super
->current_vol
;
2173 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2174 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2176 sha1_finish_ctx(&ctx
, buf
);
2177 memcpy(uuid
, buf
, 4*4);
2182 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2184 __u8
*v
= get_imsm_version(mpb
);
2185 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2186 char major
[] = { 0, 0, 0 };
2187 char minor
[] = { 0 ,0, 0 };
2188 char patch
[] = { 0, 0, 0 };
2189 char *ver_parse
[] = { major
, minor
, patch
};
2193 while (*v
!= '\0' && v
< end
) {
2194 if (*v
!= '.' && j
< 2)
2195 ver_parse
[i
][j
++] = *v
;
2203 *m
= strtol(minor
, NULL
, 0);
2204 *p
= strtol(patch
, NULL
, 0);
2208 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2210 /* migr_strip_size when repairing or initializing parity */
2211 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2212 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2214 switch (get_imsm_raid_level(map
)) {
2219 return 128*1024 >> 9;
2223 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2225 /* migr_strip_size when rebuilding a degraded disk, no idea why
2226 * this is different than migr_strip_size_resync(), but it's good
2229 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2230 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2232 switch (get_imsm_raid_level(map
)) {
2235 if (map
->num_members
% map
->num_domains
== 0)
2236 return 128*1024 >> 9;
2240 return max((__u32
) 64*1024 >> 9, chunk
);
2242 return 128*1024 >> 9;
2246 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2248 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2249 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2250 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2251 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2253 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2256 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2258 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2259 int level
= get_imsm_raid_level(lo
);
2261 if (level
== 1 || level
== 10) {
2262 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2264 return hi
->num_domains
;
2266 return num_stripes_per_unit_resync(dev
);
2269 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2271 /* named 'imsm_' because raid0, raid1 and raid10
2272 * counter-intuitively have the same number of data disks
2274 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2276 switch (get_imsm_raid_level(map
)) {
2278 return map
->num_members
;
2282 return map
->num_members
/2;
2284 return map
->num_members
- 1;
2286 dprintf("unsupported raid level\n");
2291 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2293 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2294 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2296 switch(get_imsm_raid_level(map
)) {
2299 return chunk
* map
->num_domains
;
2301 return chunk
* map
->num_members
;
2307 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2309 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2310 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2311 __u32 strip
= block
/ chunk
;
2313 switch (get_imsm_raid_level(map
)) {
2316 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2317 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2319 return vol_stripe
* chunk
+ block
% chunk
;
2321 __u32 stripe
= strip
/ (map
->num_members
- 1);
2323 return stripe
* chunk
+ block
% chunk
;
2330 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2331 struct imsm_dev
*dev
)
2333 /* calculate the conversion factor between per member 'blocks'
2334 * (md/{resync,rebuild}_start) and imsm migration units, return
2335 * 0 for the 'not migrating' and 'unsupported migration' cases
2337 if (!dev
->vol
.migr_state
)
2340 switch (migr_type(dev
)) {
2341 case MIGR_GEN_MIGR
: {
2342 struct migr_record
*migr_rec
= super
->migr_rec
;
2343 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2348 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2349 __u32 stripes_per_unit
;
2350 __u32 blocks_per_unit
;
2359 /* yes, this is really the translation of migr_units to
2360 * per-member blocks in the 'resync' case
2362 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2363 migr_chunk
= migr_strip_blocks_resync(dev
);
2364 disks
= imsm_num_data_members(dev
, MAP_0
);
2365 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2366 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2367 segment
= blocks_per_unit
/ stripe
;
2368 block_rel
= blocks_per_unit
- segment
* stripe
;
2369 parity_depth
= parity_segment_depth(dev
);
2370 block_map
= map_migr_block(dev
, block_rel
);
2371 return block_map
+ parity_depth
* segment
;
2373 case MIGR_REBUILD
: {
2374 __u32 stripes_per_unit
;
2377 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2378 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2379 return migr_chunk
* stripes_per_unit
;
2381 case MIGR_STATE_CHANGE
:
2387 static int imsm_level_to_layout(int level
)
2395 return ALGORITHM_LEFT_ASYMMETRIC
;
2402 /*******************************************************************************
2403 * Function: read_imsm_migr_rec
2404 * Description: Function reads imsm migration record from last sector of disk
2406 * fd : disk descriptor
2407 * super : metadata info
2411 ******************************************************************************/
2412 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2415 unsigned long long dsize
;
2417 get_dev_size(fd
, NULL
, &dsize
);
2418 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2419 pr_err("Cannot seek to anchor block: %s\n",
2423 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2424 MIGR_REC_BUF_SIZE
) {
2425 pr_err("Cannot read migr record block: %s\n",
2435 static struct imsm_dev
*imsm_get_device_during_migration(
2436 struct intel_super
*super
)
2439 struct intel_dev
*dv
;
2441 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2442 if (is_gen_migration(dv
->dev
))
2448 /*******************************************************************************
2449 * Function: load_imsm_migr_rec
2450 * Description: Function reads imsm migration record (it is stored at the last
2453 * super : imsm internal array info
2454 * info : general array info
2458 * -2 : no migration in progress
2459 ******************************************************************************/
2460 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2467 struct imsm_dev
*dev
;
2468 struct imsm_map
*map
;
2471 /* find map under migration */
2472 dev
= imsm_get_device_during_migration(super
);
2473 /* nothing to load,no migration in progress?
2479 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2480 /* read only from one of the first two slots */
2481 if ((sd
->disk
.raid_disk
< 0) ||
2482 (sd
->disk
.raid_disk
> 1))
2485 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2486 fd
= dev_open(nm
, O_RDONLY
);
2492 map
= get_imsm_map(dev
, MAP_0
);
2493 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2494 /* skip spare and failed disks
2498 /* read only from one of the first two slots */
2500 slot
= get_imsm_disk_slot(map
, dl
->index
);
2501 if (map
== NULL
|| slot
> 1 || slot
< 0)
2503 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2504 fd
= dev_open(nm
, O_RDONLY
);
2511 retval
= read_imsm_migr_rec(fd
, super
);
2520 /*******************************************************************************
2521 * function: imsm_create_metadata_checkpoint_update
2522 * Description: It creates update for checkpoint change.
2524 * super : imsm internal array info
2525 * u : pointer to prepared update
2528 * If length is equal to 0, input pointer u contains no update
2529 ******************************************************************************/
2530 static int imsm_create_metadata_checkpoint_update(
2531 struct intel_super
*super
,
2532 struct imsm_update_general_migration_checkpoint
**u
)
2535 int update_memory_size
= 0;
2537 dprintf("(enter)\n");
2543 /* size of all update data without anchor */
2544 update_memory_size
=
2545 sizeof(struct imsm_update_general_migration_checkpoint
);
2547 *u
= xcalloc(1, update_memory_size
);
2549 dprintf("error: cannot get memory\n");
2552 (*u
)->type
= update_general_migration_checkpoint
;
2553 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2554 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
2556 return update_memory_size
;
2559 static void imsm_update_metadata_locally(struct supertype
*st
,
2560 void *buf
, int len
);
2562 /*******************************************************************************
2563 * Function: write_imsm_migr_rec
2564 * Description: Function writes imsm migration record
2565 * (at the last sector of disk)
2567 * super : imsm internal array info
2571 ******************************************************************************/
2572 static int write_imsm_migr_rec(struct supertype
*st
)
2574 struct intel_super
*super
= st
->sb
;
2575 unsigned long long dsize
;
2581 struct imsm_update_general_migration_checkpoint
*u
;
2582 struct imsm_dev
*dev
;
2583 struct imsm_map
*map
;
2585 /* find map under migration */
2586 dev
= imsm_get_device_during_migration(super
);
2587 /* if no migration, write buffer anyway to clear migr_record
2588 * on disk based on first available device
2591 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2592 super
->current_vol
);
2594 map
= get_imsm_map(dev
, MAP_0
);
2596 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2599 /* skip failed and spare devices */
2602 /* write to 2 first slots only */
2604 slot
= get_imsm_disk_slot(map
, sd
->index
);
2605 if (map
== NULL
|| slot
> 1 || slot
< 0)
2608 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2609 fd
= dev_open(nm
, O_RDWR
);
2612 get_dev_size(fd
, NULL
, &dsize
);
2613 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2614 pr_err("Cannot seek to anchor block: %s\n",
2618 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2619 MIGR_REC_BUF_SIZE
) {
2620 pr_err("Cannot write migr record block: %s\n",
2627 /* update checkpoint information in metadata */
2628 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2631 dprintf("imsm: Cannot prepare update\n");
2634 /* update metadata locally */
2635 imsm_update_metadata_locally(st
, u
, len
);
2636 /* and possibly remotely */
2637 if (st
->update_tail
) {
2638 append_metadata_update(st
, u
, len
);
2639 /* during reshape we do all work inside metadata handler
2640 * manage_reshape(), so metadata update has to be triggered
2643 flush_metadata_updates(st
);
2644 st
->update_tail
= &st
->updates
;
2654 #endif /* MDASSEMBLE */
2656 /* spare/missing disks activations are not allowe when
2657 * array/container performs reshape operation, because
2658 * all arrays in container works on the same disks set
2660 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2663 struct intel_dev
*i_dev
;
2664 struct imsm_dev
*dev
;
2666 /* check whole container
2668 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2670 if (is_gen_migration(dev
)) {
2671 /* No repair during any migration in container
2679 static unsigned long long imsm_component_size_aligment_check(int level
,
2681 unsigned long long component_size
)
2683 unsigned int component_size_alligment
;
2685 /* check component size aligment
2687 component_size_alligment
= component_size
% (chunk_size
/512);
2689 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
2690 level
, chunk_size
, component_size
,
2691 component_size_alligment
);
2693 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
2694 dprintf("imsm: reported component size alligned from %llu ",
2696 component_size
-= component_size_alligment
;
2697 dprintf_cont("to %llu (%i).\n",
2698 component_size
, component_size_alligment
);
2701 return component_size
;
2704 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2706 struct intel_super
*super
= st
->sb
;
2707 struct migr_record
*migr_rec
= super
->migr_rec
;
2708 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2709 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2710 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2711 struct imsm_map
*map_to_analyse
= map
;
2713 int map_disks
= info
->array
.raid_disks
;
2715 memset(info
, 0, sizeof(*info
));
2717 map_to_analyse
= prev_map
;
2719 dl
= super
->current_disk
;
2721 info
->container_member
= super
->current_vol
;
2722 info
->array
.raid_disks
= map
->num_members
;
2723 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2724 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2725 info
->array
.md_minor
= -1;
2726 info
->array
.ctime
= 0;
2727 info
->array
.utime
= 0;
2728 info
->array
.chunk_size
=
2729 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2730 info
->array
.state
= !dev
->vol
.dirty
;
2731 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2732 info
->custom_array_size
<<= 32;
2733 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2734 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2736 if (is_gen_migration(dev
)) {
2737 info
->reshape_active
= 1;
2738 info
->new_level
= get_imsm_raid_level(map
);
2739 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2740 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2741 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2742 if (info
->delta_disks
) {
2743 /* this needs to be applied to every array
2746 info
->reshape_active
= CONTAINER_RESHAPE
;
2748 /* We shape information that we give to md might have to be
2749 * modify to cope with md's requirement for reshaping arrays.
2750 * For example, when reshaping a RAID0, md requires it to be
2751 * presented as a degraded RAID4.
2752 * Also if a RAID0 is migrating to a RAID5 we need to specify
2753 * the array as already being RAID5, but the 'before' layout
2754 * is a RAID4-like layout.
2756 switch (info
->array
.level
) {
2758 switch(info
->new_level
) {
2760 /* conversion is happening as RAID4 */
2761 info
->array
.level
= 4;
2762 info
->array
.raid_disks
+= 1;
2765 /* conversion is happening as RAID5 */
2766 info
->array
.level
= 5;
2767 info
->array
.layout
= ALGORITHM_PARITY_N
;
2768 info
->delta_disks
-= 1;
2771 /* FIXME error message */
2772 info
->array
.level
= UnSet
;
2778 info
->new_level
= UnSet
;
2779 info
->new_layout
= UnSet
;
2780 info
->new_chunk
= info
->array
.chunk_size
;
2781 info
->delta_disks
= 0;
2785 info
->disk
.major
= dl
->major
;
2786 info
->disk
.minor
= dl
->minor
;
2787 info
->disk
.number
= dl
->index
;
2788 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2792 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2793 info
->component_size
= blocks_per_member(map_to_analyse
);
2795 info
->component_size
= imsm_component_size_aligment_check(
2797 info
->array
.chunk_size
,
2798 info
->component_size
);
2800 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2801 info
->recovery_start
= MaxSector
;
2803 info
->reshape_progress
= 0;
2804 info
->resync_start
= MaxSector
;
2805 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2807 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2808 info
->resync_start
= 0;
2810 if (dev
->vol
.migr_state
) {
2811 switch (migr_type(dev
)) {
2814 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2816 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2818 info
->resync_start
= blocks_per_unit
* units
;
2821 case MIGR_GEN_MIGR
: {
2822 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2824 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2825 unsigned long long array_blocks
;
2828 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2830 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2831 (super
->migr_rec
->rec_status
==
2832 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2835 info
->reshape_progress
= blocks_per_unit
* units
;
2837 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
2838 (unsigned long long)units
,
2839 (unsigned long long)blocks_per_unit
,
2840 info
->reshape_progress
);
2842 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2843 if (used_disks
> 0) {
2844 array_blocks
= blocks_per_member(map
) *
2846 /* round array size down to closest MB
2848 info
->custom_array_size
= (array_blocks
2849 >> SECT_PER_MB_SHIFT
)
2850 << SECT_PER_MB_SHIFT
;
2854 /* we could emulate the checkpointing of
2855 * 'sync_action=check' migrations, but for now
2856 * we just immediately complete them
2859 /* this is handled by container_content_imsm() */
2860 case MIGR_STATE_CHANGE
:
2861 /* FIXME handle other migrations */
2863 /* we are not dirty, so... */
2864 info
->resync_start
= MaxSector
;
2868 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2869 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2871 info
->array
.major_version
= -1;
2872 info
->array
.minor_version
= -2;
2873 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
2874 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2875 uuid_from_super_imsm(st
, info
->uuid
);
2879 for (i
=0; i
<map_disks
; i
++) {
2881 if (i
< info
->array
.raid_disks
) {
2882 struct imsm_disk
*dsk
;
2883 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2884 dsk
= get_imsm_disk(super
, j
);
2885 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2892 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2893 int failed
, int look_in_map
);
2895 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2899 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2901 if (is_gen_migration(dev
)) {
2904 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2906 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2907 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2908 if (map2
->map_state
!= map_state
) {
2909 map2
->map_state
= map_state
;
2910 super
->updates_pending
++;
2916 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2920 for (d
= super
->missing
; d
; d
= d
->next
)
2921 if (d
->index
== index
)
2926 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2928 struct intel_super
*super
= st
->sb
;
2929 struct imsm_disk
*disk
;
2930 int map_disks
= info
->array
.raid_disks
;
2931 int max_enough
= -1;
2933 struct imsm_super
*mpb
;
2935 if (super
->current_vol
>= 0) {
2936 getinfo_super_imsm_volume(st
, info
, map
);
2939 memset(info
, 0, sizeof(*info
));
2941 /* Set raid_disks to zero so that Assemble will always pull in valid
2944 info
->array
.raid_disks
= 0;
2945 info
->array
.level
= LEVEL_CONTAINER
;
2946 info
->array
.layout
= 0;
2947 info
->array
.md_minor
= -1;
2948 info
->array
.ctime
= 0; /* N/A for imsm */
2949 info
->array
.utime
= 0;
2950 info
->array
.chunk_size
= 0;
2952 info
->disk
.major
= 0;
2953 info
->disk
.minor
= 0;
2954 info
->disk
.raid_disk
= -1;
2955 info
->reshape_active
= 0;
2956 info
->array
.major_version
= -1;
2957 info
->array
.minor_version
= -2;
2958 strcpy(info
->text_version
, "imsm");
2959 info
->safe_mode_delay
= 0;
2960 info
->disk
.number
= -1;
2961 info
->disk
.state
= 0;
2963 info
->recovery_start
= MaxSector
;
2964 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2966 /* do we have the all the insync disks that we expect? */
2967 mpb
= super
->anchor
;
2969 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2970 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2971 int failed
, enough
, j
, missing
= 0;
2972 struct imsm_map
*map
;
2975 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2976 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2977 map
= get_imsm_map(dev
, MAP_0
);
2979 /* any newly missing disks?
2980 * (catches single-degraded vs double-degraded)
2982 for (j
= 0; j
< map
->num_members
; j
++) {
2983 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2984 __u32 idx
= ord_to_idx(ord
);
2986 if (!(ord
& IMSM_ORD_REBUILD
) &&
2987 get_imsm_missing(super
, idx
)) {
2993 if (state
== IMSM_T_STATE_FAILED
)
2995 else if (state
== IMSM_T_STATE_DEGRADED
&&
2996 (state
!= map
->map_state
|| missing
))
2998 else /* we're normal, or already degraded */
3000 if (is_gen_migration(dev
) && missing
) {
3001 /* during general migration we need all disks
3002 * that process is running on.
3003 * No new missing disk is allowed.
3007 /* no more checks necessary
3011 /* in the missing/failed disk case check to see
3012 * if at least one array is runnable
3014 max_enough
= max(max_enough
, enough
);
3016 dprintf("enough: %d\n", max_enough
);
3017 info
->container_enough
= max_enough
;
3020 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3022 disk
= &super
->disks
->disk
;
3023 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3024 info
->component_size
= reserved
;
3025 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3026 /* we don't change info->disk.raid_disk here because
3027 * this state will be finalized in mdmon after we have
3028 * found the 'most fresh' version of the metadata
3030 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3031 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3034 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3035 * ->compare_super may have updated the 'num_raid_devs' field for spares
3037 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3038 uuid_from_super_imsm(st
, info
->uuid
);
3040 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3042 /* I don't know how to compute 'map' on imsm, so use safe default */
3045 for (i
= 0; i
< map_disks
; i
++)
3051 /* allocates memory and fills disk in mdinfo structure
3052 * for each disk in array */
3053 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3055 struct mdinfo
*mddev
;
3056 struct intel_super
*super
= st
->sb
;
3057 struct imsm_disk
*disk
;
3060 if (!super
|| !super
->disks
)
3063 mddev
= xcalloc(1, sizeof(*mddev
));
3067 tmp
= xcalloc(1, sizeof(*tmp
));
3069 tmp
->next
= mddev
->devs
;
3071 tmp
->disk
.number
= count
++;
3072 tmp
->disk
.major
= dl
->major
;
3073 tmp
->disk
.minor
= dl
->minor
;
3074 tmp
->disk
.state
= is_configured(disk
) ?
3075 (1 << MD_DISK_ACTIVE
) : 0;
3076 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3077 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3078 tmp
->disk
.raid_disk
= -1;
3084 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3085 char *update
, char *devname
, int verbose
,
3086 int uuid_set
, char *homehost
)
3088 /* For 'assemble' and 'force' we need to return non-zero if any
3089 * change was made. For others, the return value is ignored.
3090 * Update options are:
3091 * force-one : This device looks a bit old but needs to be included,
3092 * update age info appropriately.
3093 * assemble: clear any 'faulty' flag to allow this device to
3095 * force-array: Array is degraded but being forced, mark it clean
3096 * if that will be needed to assemble it.
3098 * newdev: not used ????
3099 * grow: Array has gained a new device - this is currently for
3101 * resync: mark as dirty so a resync will happen.
3102 * name: update the name - preserving the homehost
3103 * uuid: Change the uuid of the array to match watch is given
3105 * Following are not relevant for this imsm:
3106 * sparc2.2 : update from old dodgey metadata
3107 * super-minor: change the preferred_minor number
3108 * summaries: update redundant counters.
3109 * homehost: update the recorded homehost
3110 * _reshape_progress: record new reshape_progress position.
3113 struct intel_super
*super
= st
->sb
;
3114 struct imsm_super
*mpb
;
3116 /* we can only update container info */
3117 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3120 mpb
= super
->anchor
;
3122 if (strcmp(update
, "uuid") == 0) {
3123 /* We take this to mean that the family_num should be updated.
3124 * However that is much smaller than the uuid so we cannot really
3125 * allow an explicit uuid to be given. And it is hard to reliably
3127 * So if !uuid_set we know the current uuid is random and just used
3128 * the first 'int' and copy it to the other 3 positions.
3129 * Otherwise we require the 4 'int's to be the same as would be the
3130 * case if we are using a random uuid. So an explicit uuid will be
3131 * accepted as long as all for ints are the same... which shouldn't hurt
3134 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3137 if (info
->uuid
[0] != info
->uuid
[1] ||
3138 info
->uuid
[1] != info
->uuid
[2] ||
3139 info
->uuid
[2] != info
->uuid
[3])
3145 mpb
->orig_family_num
= info
->uuid
[0];
3146 } else if (strcmp(update
, "assemble") == 0)
3151 /* successful update? recompute checksum */
3153 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3158 static size_t disks_to_mpb_size(int disks
)
3162 size
= sizeof(struct imsm_super
);
3163 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3164 size
+= 2 * sizeof(struct imsm_dev
);
3165 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3166 size
+= (4 - 2) * sizeof(struct imsm_map
);
3167 /* 4 possible disk_ord_tbl's */
3168 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3173 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3174 unsigned long long data_offset
)
3176 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3179 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3182 static void free_devlist(struct intel_super
*super
)
3184 struct intel_dev
*dv
;
3186 while (super
->devlist
) {
3187 dv
= super
->devlist
->next
;
3188 free(super
->devlist
->dev
);
3189 free(super
->devlist
);
3190 super
->devlist
= dv
;
3194 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3196 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3199 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3203 * 0 same, or first was empty, and second was copied
3204 * 1 second had wrong number
3206 * 3 wrong other info
3208 struct intel_super
*first
= st
->sb
;
3209 struct intel_super
*sec
= tst
->sb
;
3216 /* in platform dependent environment test if the disks
3217 * use the same Intel hba
3218 * If not on Intel hba at all, allow anything.
3220 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3221 if (first
->hba
->type
!= sec
->hba
->type
) {
3223 "HBAs of devices do not match %s != %s\n",
3224 get_sys_dev_type(first
->hba
->type
),
3225 get_sys_dev_type(sec
->hba
->type
));
3228 if (first
->orom
!= sec
->orom
) {
3230 "HBAs of devices do not match %s != %s\n",
3231 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3236 /* if an anchor does not have num_raid_devs set then it is a free
3239 if (first
->anchor
->num_raid_devs
> 0 &&
3240 sec
->anchor
->num_raid_devs
> 0) {
3241 /* Determine if these disks might ever have been
3242 * related. Further disambiguation can only take place
3243 * in load_super_imsm_all
3245 __u32 first_family
= first
->anchor
->orig_family_num
;
3246 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3248 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3249 MAX_SIGNATURE_LENGTH
) != 0)
3252 if (first_family
== 0)
3253 first_family
= first
->anchor
->family_num
;
3254 if (sec_family
== 0)
3255 sec_family
= sec
->anchor
->family_num
;
3257 if (first_family
!= sec_family
)
3262 /* if 'first' is a spare promote it to a populated mpb with sec's
3265 if (first
->anchor
->num_raid_devs
== 0 &&
3266 sec
->anchor
->num_raid_devs
> 0) {
3268 struct intel_dev
*dv
;
3269 struct imsm_dev
*dev
;
3271 /* we need to copy raid device info from sec if an allocation
3272 * fails here we don't associate the spare
3274 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3275 dv
= xmalloc(sizeof(*dv
));
3276 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3279 dv
->next
= first
->devlist
;
3280 first
->devlist
= dv
;
3282 if (i
< sec
->anchor
->num_raid_devs
) {
3283 /* allocation failure */
3284 free_devlist(first
);
3285 pr_err("imsm: failed to associate spare\n");
3288 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3289 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3290 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3291 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3292 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3293 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3299 static void fd2devname(int fd
, char *name
)
3303 char dname
[PATH_MAX
];
3308 if (fstat(fd
, &st
) != 0)
3310 sprintf(path
, "/sys/dev/block/%d:%d",
3311 major(st
.st_rdev
), minor(st
.st_rdev
));
3313 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3318 nm
= strrchr(dname
, '/');
3321 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3325 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3328 char *name
= fd2kname(fd
);
3333 if (strncmp(name
, "nvme", 4) != 0)
3336 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3338 return load_sys(path
, buf
, buf_len
);
3341 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3343 static int imsm_read_serial(int fd
, char *devname
,
3344 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3353 memset(buf
, 0, sizeof(buf
));
3355 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3358 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3360 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3361 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3362 fd2devname(fd
, (char *) serial
);
3368 pr_err("Failed to retrieve serial for %s\n",
3373 /* trim all whitespace and non-printable characters and convert
3376 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
3379 /* ':' is reserved for use in placeholder serial
3380 * numbers for missing disks
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
|| migr_type
== MIGR_GEN_MIGR
) {
3525 for (i
= 0; i
< src
->num_members
; i
++) {
3526 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3527 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3531 if (migr_type
== MIGR_GEN_MIGR
)
3532 /* Clear migration record */
3533 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3535 src
->map_state
= to_state
;
3538 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3541 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3542 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3546 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3547 * completed in the last migration.
3549 * FIXME add support for raid-level-migration
3551 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
3552 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
3553 /* when final map state is other than expected
3554 * merge maps (not for migration)
3558 for (i
= 0; i
< prev
->num_members
; i
++)
3559 for (j
= 0; j
< map
->num_members
; j
++)
3560 /* during online capacity expansion
3561 * disks position can be changed
3562 * if takeover is used
3564 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3565 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3566 map
->disk_ord_tbl
[j
] |=
3567 prev
->disk_ord_tbl
[i
];
3570 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3571 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3574 dev
->vol
.migr_state
= 0;
3575 set_migr_type(dev
, 0);
3576 dev
->vol
.curr_migr_unit
= 0;
3577 map
->map_state
= map_state
;
3581 static int parse_raid_devices(struct intel_super
*super
)
3584 struct imsm_dev
*dev_new
;
3585 size_t len
, len_migr
;
3587 size_t space_needed
= 0;
3588 struct imsm_super
*mpb
= super
->anchor
;
3590 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3591 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3592 struct intel_dev
*dv
;
3594 len
= sizeof_imsm_dev(dev_iter
, 0);
3595 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3597 space_needed
+= len_migr
- len
;
3599 dv
= xmalloc(sizeof(*dv
));
3600 if (max_len
< len_migr
)
3602 if (max_len
> len_migr
)
3603 space_needed
+= max_len
- len_migr
;
3604 dev_new
= xmalloc(max_len
);
3605 imsm_copy_dev(dev_new
, dev_iter
);
3608 dv
->next
= super
->devlist
;
3609 super
->devlist
= dv
;
3612 /* ensure that super->buf is large enough when all raid devices
3615 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3618 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3619 if (posix_memalign(&buf
, 512, len
) != 0)
3622 memcpy(buf
, super
->buf
, super
->len
);
3623 memset(buf
+ super
->len
, 0, len
- super
->len
);
3632 /* retrieve a pointer to the bbm log which starts after all raid devices */
3633 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3637 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3639 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3645 /*******************************************************************************
3646 * Function: check_mpb_migr_compatibility
3647 * Description: Function checks for unsupported migration features:
3648 * - migration optimization area (pba_of_lba0)
3649 * - descending reshape (ascending_migr)
3651 * super : imsm metadata information
3653 * 0 : migration is compatible
3654 * -1 : migration is not compatible
3655 ******************************************************************************/
3656 int check_mpb_migr_compatibility(struct intel_super
*super
)
3658 struct imsm_map
*map0
, *map1
;
3659 struct migr_record
*migr_rec
= super
->migr_rec
;
3662 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3663 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3666 dev_iter
->vol
.migr_state
== 1 &&
3667 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3668 /* This device is migrating */
3669 map0
= get_imsm_map(dev_iter
, MAP_0
);
3670 map1
= get_imsm_map(dev_iter
, MAP_1
);
3671 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3672 /* migration optimization area was used */
3674 if (migr_rec
->ascending_migr
== 0
3675 && migr_rec
->dest_depth_per_unit
> 0)
3676 /* descending reshape not supported yet */
3683 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3685 /* load_imsm_mpb - read matrix metadata
3686 * allocates super->mpb to be freed by free_imsm
3688 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3690 unsigned long long dsize
;
3691 unsigned long long sectors
;
3693 struct imsm_super
*anchor
;
3696 get_dev_size(fd
, NULL
, &dsize
);
3699 pr_err("%s: device to small for imsm\n",
3704 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3706 pr_err("Cannot seek to anchor block on %s: %s\n",
3707 devname
, strerror(errno
));
3711 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3713 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
3716 if (read(fd
, anchor
, 512) != 512) {
3718 pr_err("Cannot read anchor block on %s: %s\n",
3719 devname
, strerror(errno
));
3724 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3726 pr_err("no IMSM anchor on %s\n", devname
);
3731 __free_imsm(super
, 0);
3732 /* reload capability and hba */
3734 /* capability and hba must be updated with new super allocation */
3735 find_intel_hba_capability(fd
, super
, devname
);
3736 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3737 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3739 pr_err("unable to allocate %zu byte mpb buffer\n",
3744 memcpy(super
->buf
, anchor
, 512);
3746 sectors
= mpb_sectors(anchor
) - 1;
3749 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3750 pr_err("could not allocate migr_rec buffer\n");
3754 super
->clean_migration_record_by_mdmon
= 0;
3757 check_sum
= __gen_imsm_checksum(super
->anchor
);
3758 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3760 pr_err("IMSM checksum %x != %x on %s\n",
3762 __le32_to_cpu(super
->anchor
->check_sum
),
3770 /* read the extended mpb */
3771 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3773 pr_err("Cannot seek to extended mpb on %s: %s\n",
3774 devname
, strerror(errno
));
3778 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3780 pr_err("Cannot read extended mpb on %s: %s\n",
3781 devname
, strerror(errno
));
3785 check_sum
= __gen_imsm_checksum(super
->anchor
);
3786 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3788 pr_err("IMSM checksum %x != %x on %s\n",
3789 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3794 /* FIXME the BBM log is disk specific so we cannot use this global
3795 * buffer for all disks. Ok for now since we only look at the global
3796 * bbm_log_size parameter to gate assembly
3798 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3803 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3805 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3806 static void clear_hi(struct intel_super
*super
)
3808 struct imsm_super
*mpb
= super
->anchor
;
3810 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3812 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3813 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3814 disk
->total_blocks_hi
= 0;
3816 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3817 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3820 for (n
= 0; n
< 2; ++n
) {
3821 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3824 map
->pba_of_lba0_hi
= 0;
3825 map
->blocks_per_member_hi
= 0;
3826 map
->num_data_stripes_hi
= 0;
3832 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3836 err
= load_imsm_mpb(fd
, super
, devname
);
3839 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3842 err
= parse_raid_devices(super
);
3847 static void __free_imsm_disk(struct dl
*d
)
3859 static void free_imsm_disks(struct intel_super
*super
)
3863 while (super
->disks
) {
3865 super
->disks
= d
->next
;
3866 __free_imsm_disk(d
);
3868 while (super
->disk_mgmt_list
) {
3869 d
= super
->disk_mgmt_list
;
3870 super
->disk_mgmt_list
= d
->next
;
3871 __free_imsm_disk(d
);
3873 while (super
->missing
) {
3875 super
->missing
= d
->next
;
3876 __free_imsm_disk(d
);
3881 /* free all the pieces hanging off of a super pointer */
3882 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3884 struct intel_hba
*elem
, *next
;
3890 /* unlink capability description */
3892 if (super
->migr_rec_buf
) {
3893 free(super
->migr_rec_buf
);
3894 super
->migr_rec_buf
= NULL
;
3897 free_imsm_disks(super
);
3898 free_devlist(super
);
3902 free((void *)elem
->path
);
3910 static void free_imsm(struct intel_super
*super
)
3912 __free_imsm(super
, 1);
3916 static void free_super_imsm(struct supertype
*st
)
3918 struct intel_super
*super
= st
->sb
;
3927 static struct intel_super
*alloc_super(void)
3929 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
3931 super
->current_vol
= -1;
3932 super
->create_offset
= ~((unsigned long long) 0);
3937 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3939 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3941 struct sys_dev
*hba_name
;
3944 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
3949 hba_name
= find_disk_attached_hba(fd
, NULL
);
3952 pr_err("%s is not attached to Intel(R) RAID controller.\n",
3956 rv
= attach_hba_to_super(super
, hba_name
);
3959 struct intel_hba
*hba
= super
->hba
;
3961 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
3962 " but the container is assigned to Intel(R) %s %s (",
3964 get_sys_dev_type(hba_name
->type
),
3965 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
3966 hba_name
->pci_id
? : "Err!",
3967 get_sys_dev_type(super
->hba
->type
),
3968 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
3971 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3973 fprintf(stderr
, ", ");
3976 fprintf(stderr
, ").\n"
3977 " Mixing devices attached to different %s is not allowed.\n",
3978 hba_name
->type
== SYS_DEV_VMD
? "VMD domains" : "controllers");
3982 super
->orom
= find_imsm_capability(hba_name
);
3989 /* find_missing - helper routine for load_super_imsm_all that identifies
3990 * disks that have disappeared from the system. This routine relies on
3991 * the mpb being uptodate, which it is at load time.
3993 static int find_missing(struct intel_super
*super
)
3996 struct imsm_super
*mpb
= super
->anchor
;
3998 struct imsm_disk
*disk
;
4000 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4001 disk
= __get_imsm_disk(mpb
, i
);
4002 dl
= serial_to_dl(disk
->serial
, super
);
4006 dl
= xmalloc(sizeof(*dl
));
4010 dl
->devname
= xstrdup("missing");
4012 serialcpy(dl
->serial
, disk
->serial
);
4015 dl
->next
= super
->missing
;
4016 super
->missing
= dl
;
4023 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4025 struct intel_disk
*idisk
= disk_list
;
4028 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4030 idisk
= idisk
->next
;
4036 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4037 struct intel_super
*super
,
4038 struct intel_disk
**disk_list
)
4040 struct imsm_disk
*d
= &super
->disks
->disk
;
4041 struct imsm_super
*mpb
= super
->anchor
;
4044 for (i
= 0; i
< tbl_size
; i
++) {
4045 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4046 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4048 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4049 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4050 dprintf("mpb from %d:%d matches %d:%d\n",
4051 super
->disks
->major
,
4052 super
->disks
->minor
,
4053 table
[i
]->disks
->major
,
4054 table
[i
]->disks
->minor
);
4058 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4059 is_configured(d
) == is_configured(tbl_d
)) &&
4060 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4061 /* current version of the mpb is a
4062 * better candidate than the one in
4063 * super_table, but copy over "cross
4064 * generational" status
4066 struct intel_disk
*idisk
;
4068 dprintf("mpb from %d:%d replaces %d:%d\n",
4069 super
->disks
->major
,
4070 super
->disks
->minor
,
4071 table
[i
]->disks
->major
,
4072 table
[i
]->disks
->minor
);
4074 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4075 if (idisk
&& is_failed(&idisk
->disk
))
4076 tbl_d
->status
|= FAILED_DISK
;
4079 struct intel_disk
*idisk
;
4080 struct imsm_disk
*disk
;
4082 /* tbl_mpb is more up to date, but copy
4083 * over cross generational status before
4086 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4087 if (disk
&& is_failed(disk
))
4088 d
->status
|= FAILED_DISK
;
4090 idisk
= disk_list_get(d
->serial
, *disk_list
);
4093 if (disk
&& is_configured(disk
))
4094 idisk
->disk
.status
|= CONFIGURED_DISK
;
4097 dprintf("mpb from %d:%d prefer %d:%d\n",
4098 super
->disks
->major
,
4099 super
->disks
->minor
,
4100 table
[i
]->disks
->major
,
4101 table
[i
]->disks
->minor
);
4109 table
[tbl_size
++] = super
;
4113 /* update/extend the merged list of imsm_disk records */
4114 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4115 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4116 struct intel_disk
*idisk
;
4118 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4120 idisk
->disk
.status
|= disk
->status
;
4121 if (is_configured(&idisk
->disk
) ||
4122 is_failed(&idisk
->disk
))
4123 idisk
->disk
.status
&= ~(SPARE_DISK
);
4125 idisk
= xcalloc(1, sizeof(*idisk
));
4126 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4127 idisk
->disk
= *disk
;
4128 idisk
->next
= *disk_list
;
4132 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4139 static struct intel_super
*
4140 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4143 struct imsm_super
*mpb
= super
->anchor
;
4147 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4148 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4149 struct intel_disk
*idisk
;
4151 idisk
= disk_list_get(disk
->serial
, disk_list
);
4153 if (idisk
->owner
== owner
||
4154 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4157 dprintf("'%.16s' owner %d != %d\n",
4158 disk
->serial
, idisk
->owner
,
4161 dprintf("unknown disk %x [%d]: %.16s\n",
4162 __le32_to_cpu(mpb
->family_num
), i
,
4168 if (ok_count
== mpb
->num_disks
)
4173 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4175 struct intel_super
*s
;
4177 for (s
= super_list
; s
; s
= s
->next
) {
4178 if (family_num
!= s
->anchor
->family_num
)
4180 pr_err("Conflict, offlining family %#x on '%s'\n",
4181 __le32_to_cpu(family_num
), s
->disks
->devname
);
4185 static struct intel_super
*
4186 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4188 struct intel_super
*super_table
[len
];
4189 struct intel_disk
*disk_list
= NULL
;
4190 struct intel_super
*champion
, *spare
;
4191 struct intel_super
*s
, **del
;
4196 memset(super_table
, 0, sizeof(super_table
));
4197 for (s
= *super_list
; s
; s
= s
->next
)
4198 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4200 for (i
= 0; i
< tbl_size
; i
++) {
4201 struct imsm_disk
*d
;
4202 struct intel_disk
*idisk
;
4203 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4206 d
= &s
->disks
->disk
;
4208 /* 'd' must appear in merged disk list for its
4209 * configuration to be valid
4211 idisk
= disk_list_get(d
->serial
, disk_list
);
4212 if (idisk
&& idisk
->owner
== i
)
4213 s
= validate_members(s
, disk_list
, i
);
4218 dprintf("marking family: %#x from %d:%d offline\n",
4220 super_table
[i
]->disks
->major
,
4221 super_table
[i
]->disks
->minor
);
4225 /* This is where the mdadm implementation differs from the Windows
4226 * driver which has no strict concept of a container. We can only
4227 * assemble one family from a container, so when returning a prodigal
4228 * array member to this system the code will not be able to disambiguate
4229 * the container contents that should be assembled ("foreign" versus
4230 * "local"). It requires user intervention to set the orig_family_num
4231 * to a new value to establish a new container. The Windows driver in
4232 * this situation fixes up the volume name in place and manages the
4233 * foreign array as an independent entity.
4238 for (i
= 0; i
< tbl_size
; i
++) {
4239 struct intel_super
*tbl_ent
= super_table
[i
];
4245 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4250 if (s
&& !is_spare
) {
4251 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4253 } else if (!s
&& !is_spare
)
4266 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4267 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4269 /* collect all dl's onto 'champion', and update them to
4270 * champion's version of the status
4272 for (s
= *super_list
; s
; s
= s
->next
) {
4273 struct imsm_super
*mpb
= champion
->anchor
;
4274 struct dl
*dl
= s
->disks
;
4279 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4281 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4282 struct imsm_disk
*disk
;
4284 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4287 /* only set index on disks that are a member of
4288 * a populated contianer, i.e. one with
4291 if (is_failed(&dl
->disk
))
4293 else if (is_spare(&dl
->disk
))
4299 if (i
>= mpb
->num_disks
) {
4300 struct intel_disk
*idisk
;
4302 idisk
= disk_list_get(dl
->serial
, disk_list
);
4303 if (idisk
&& is_spare(&idisk
->disk
) &&
4304 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4312 dl
->next
= champion
->disks
;
4313 champion
->disks
= dl
;
4317 /* delete 'champion' from super_list */
4318 for (del
= super_list
; *del
; ) {
4319 if (*del
== champion
) {
4320 *del
= (*del
)->next
;
4323 del
= &(*del
)->next
;
4325 champion
->next
= NULL
;
4329 struct intel_disk
*idisk
= disk_list
;
4331 disk_list
= disk_list
->next
;
4339 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4340 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4341 int major
, int minor
, int keep_fd
);
4343 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4344 int *max
, int keep_fd
);
4346 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4347 char *devname
, struct md_list
*devlist
,
4350 struct intel_super
*super_list
= NULL
;
4351 struct intel_super
*super
= NULL
;
4356 /* 'fd' is an opened container */
4357 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4359 /* get super block from devlist devices */
4360 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4363 /* all mpbs enter, maybe one leaves */
4364 super
= imsm_thunderdome(&super_list
, i
);
4370 if (find_missing(super
) != 0) {
4376 /* load migration record */
4377 err
= load_imsm_migr_rec(super
, NULL
);
4379 /* migration is in progress,
4380 * but migr_rec cannot be loaded,
4386 /* Check migration compatibility */
4387 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
4388 pr_err("Unsupported migration detected");
4390 fprintf(stderr
, " on %s\n", devname
);
4392 fprintf(stderr
, " (IMSM).\n");
4401 while (super_list
) {
4402 struct intel_super
*s
= super_list
;
4404 super_list
= super_list
->next
;
4413 strcpy(st
->container_devnm
, fd2devnm(fd
));
4415 st
->container_devnm
[0] = 0;
4416 if (err
== 0 && st
->ss
== NULL
) {
4417 st
->ss
= &super_imsm
;
4418 st
->minor_version
= 0;
4419 st
->max_devs
= IMSM_MAX_DEVICES
;
4425 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4426 int *max
, int keep_fd
)
4428 struct md_list
*tmpdev
;
4432 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4433 if (tmpdev
->used
!= 1)
4435 if (tmpdev
->container
== 1) {
4437 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4439 pr_err("cannot open device %s: %s\n",
4440 tmpdev
->devname
, strerror(errno
));
4444 err
= get_sra_super_block(fd
, super_list
,
4445 tmpdev
->devname
, &lmax
,
4454 int major
= major(tmpdev
->st_rdev
);
4455 int minor
= minor(tmpdev
->st_rdev
);
4456 err
= get_super_block(super_list
,
4473 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4474 int major
, int minor
, int keep_fd
)
4476 struct intel_super
*s
;
4488 sprintf(nm
, "%d:%d", major
, minor
);
4489 dfd
= dev_open(nm
, O_RDWR
);
4495 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
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 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
4576 /* Load hba and capabilities if they exist.
4577 * But do not preclude loading metadata in case capabilities or hba are
4578 * non-compliant and ignore_hw_compat is set.
4580 rv
= find_intel_hba_capability(fd
, super
, devname
);
4581 /* no orom/efi or non-intel hba of the disk */
4582 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
4584 pr_err("No OROM/EFI properties for %s\n", devname
);
4588 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4590 /* retry the load if we might have raced against mdmon */
4592 struct mdstat_ent
*mdstat
= NULL
;
4593 char *name
= fd2kname(fd
);
4596 mdstat
= mdstat_by_component(name
);
4598 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
4599 for (retry
= 0; retry
< 3; retry
++) {
4601 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4607 free_mdstat(mdstat
);
4612 pr_err("Failed to load all information sections on %s\n", devname
);
4618 if (st
->ss
== NULL
) {
4619 st
->ss
= &super_imsm
;
4620 st
->minor_version
= 0;
4621 st
->max_devs
= IMSM_MAX_DEVICES
;
4624 /* load migration record */
4625 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4626 /* Check for unsupported migration features */
4627 if (check_mpb_migr_compatibility(super
) != 0) {
4628 pr_err("Unsupported migration detected");
4630 fprintf(stderr
, " on %s\n", devname
);
4632 fprintf(stderr
, " (IMSM).\n");
4640 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4642 if (info
->level
== 1)
4644 return info
->chunk_size
>> 9;
4647 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4648 unsigned long long size
)
4650 if (info
->level
== 1)
4653 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4656 static void imsm_update_version_info(struct intel_super
*super
)
4658 /* update the version and attributes */
4659 struct imsm_super
*mpb
= super
->anchor
;
4661 struct imsm_dev
*dev
;
4662 struct imsm_map
*map
;
4665 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4666 dev
= get_imsm_dev(super
, i
);
4667 map
= get_imsm_map(dev
, MAP_0
);
4668 if (__le32_to_cpu(dev
->size_high
) > 0)
4669 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4671 /* FIXME detect when an array spans a port multiplier */
4673 mpb
->attributes
|= MPB_ATTRIB_PM
;
4676 if (mpb
->num_raid_devs
> 1 ||
4677 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4678 version
= MPB_VERSION_ATTRIBS
;
4679 switch (get_imsm_raid_level(map
)) {
4680 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4681 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4682 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4683 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4686 if (map
->num_members
>= 5)
4687 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4688 else if (dev
->status
== DEV_CLONE_N_GO
)
4689 version
= MPB_VERSION_CNG
;
4690 else if (get_imsm_raid_level(map
) == 5)
4691 version
= MPB_VERSION_RAID5
;
4692 else if (map
->num_members
>= 3)
4693 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4694 else if (get_imsm_raid_level(map
) == 1)
4695 version
= MPB_VERSION_RAID1
;
4697 version
= MPB_VERSION_RAID0
;
4699 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4703 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4705 struct imsm_super
*mpb
= super
->anchor
;
4706 char *reason
= NULL
;
4709 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4710 reason
= "must be 16 characters or less";
4712 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4713 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4715 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4716 reason
= "already exists";
4721 if (reason
&& !quiet
)
4722 pr_err("imsm volume name %s\n", reason
);
4727 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4728 unsigned long long size
, char *name
,
4729 char *homehost
, int *uuid
,
4730 long long data_offset
)
4732 /* We are creating a volume inside a pre-existing container.
4733 * so st->sb is already set.
4735 struct intel_super
*super
= st
->sb
;
4736 struct imsm_super
*mpb
= super
->anchor
;
4737 struct intel_dev
*dv
;
4738 struct imsm_dev
*dev
;
4739 struct imsm_vol
*vol
;
4740 struct imsm_map
*map
;
4741 int idx
= mpb
->num_raid_devs
;
4743 unsigned long long array_blocks
;
4744 size_t size_old
, size_new
;
4745 unsigned long long num_data_stripes
;
4747 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4748 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
4752 /* ensure the mpb is large enough for the new data */
4753 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4754 size_new
= disks_to_mpb_size(info
->nr_disks
);
4755 if (size_new
> size_old
) {
4757 size_t size_round
= ROUND_UP(size_new
, 512);
4759 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4760 pr_err("could not allocate new mpb\n");
4763 if (posix_memalign(&super
->migr_rec_buf
, 512,
4764 MIGR_REC_BUF_SIZE
) != 0) {
4765 pr_err("could not allocate migr_rec buffer\n");
4771 memcpy(mpb_new
, mpb
, size_old
);
4774 super
->anchor
= mpb_new
;
4775 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4776 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4778 super
->current_vol
= idx
;
4780 /* handle 'failed_disks' by either:
4781 * a) create dummy disk entries in the table if this the first
4782 * volume in the array. We add them here as this is the only
4783 * opportunity to add them. add_to_super_imsm_volume()
4784 * handles the non-failed disks and continues incrementing
4786 * b) validate that 'failed_disks' matches the current number
4787 * of missing disks if the container is populated
4789 if (super
->current_vol
== 0) {
4791 for (i
= 0; i
< info
->failed_disks
; i
++) {
4792 struct imsm_disk
*disk
;
4795 disk
= __get_imsm_disk(mpb
, i
);
4796 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4797 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4798 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4801 find_missing(super
);
4806 for (d
= super
->missing
; d
; d
= d
->next
)
4808 if (info
->failed_disks
> missing
) {
4809 pr_err("unable to add 'missing' disk to container\n");
4814 if (!check_name(super
, name
, 0))
4816 dv
= xmalloc(sizeof(*dv
));
4817 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4818 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4819 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4820 info
->layout
, info
->chunk_size
,
4822 /* round array size down to closest MB */
4823 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4825 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4826 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4827 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4829 vol
->migr_state
= 0;
4830 set_migr_type(dev
, MIGR_INIT
);
4831 vol
->dirty
= !info
->state
;
4832 vol
->curr_migr_unit
= 0;
4833 map
= get_imsm_map(dev
, MAP_0
);
4834 set_pba_of_lba0(map
, super
->create_offset
);
4835 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4836 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4837 map
->failed_disk_num
= ~0;
4838 if (info
->level
> 0)
4839 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
4840 : IMSM_T_STATE_UNINITIALIZED
);
4842 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4843 IMSM_T_STATE_NORMAL
;
4846 if (info
->level
== 1 && info
->raid_disks
> 2) {
4849 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
4853 map
->raid_level
= info
->level
;
4854 if (info
->level
== 10) {
4855 map
->raid_level
= 1;
4856 map
->num_domains
= info
->raid_disks
/ 2;
4857 } else if (info
->level
== 1)
4858 map
->num_domains
= info
->raid_disks
;
4860 map
->num_domains
= 1;
4862 /* info->size is only int so use the 'size' parameter instead */
4863 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4864 num_data_stripes
/= map
->num_domains
;
4865 set_num_data_stripes(map
, num_data_stripes
);
4867 map
->num_members
= info
->raid_disks
;
4868 for (i
= 0; i
< map
->num_members
; i
++) {
4869 /* initialized in add_to_super */
4870 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4872 mpb
->num_raid_devs
++;
4875 dv
->index
= super
->current_vol
;
4876 dv
->next
= super
->devlist
;
4877 super
->devlist
= dv
;
4879 imsm_update_version_info(super
);
4884 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4885 unsigned long long size
, char *name
,
4886 char *homehost
, int *uuid
,
4887 unsigned long long data_offset
)
4889 /* This is primarily called by Create when creating a new array.
4890 * We will then get add_to_super called for each component, and then
4891 * write_init_super called to write it out to each device.
4892 * For IMSM, Create can create on fresh devices or on a pre-existing
4894 * To create on a pre-existing array a different method will be called.
4895 * This one is just for fresh drives.
4897 struct intel_super
*super
;
4898 struct imsm_super
*mpb
;
4902 if (data_offset
!= INVALID_SECTORS
) {
4903 pr_err("data-offset not supported by imsm\n");
4908 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
,
4912 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4916 super
= alloc_super();
4917 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4922 pr_err("could not allocate superblock\n");
4925 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4926 pr_err("could not allocate migr_rec buffer\n");
4931 memset(super
->buf
, 0, mpb_size
);
4933 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4937 /* zeroing superblock */
4941 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4943 version
= (char *) mpb
->sig
;
4944 strcpy(version
, MPB_SIGNATURE
);
4945 version
+= strlen(MPB_SIGNATURE
);
4946 strcpy(version
, MPB_VERSION_RAID0
);
4952 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4953 int fd
, char *devname
)
4955 struct intel_super
*super
= st
->sb
;
4956 struct imsm_super
*mpb
= super
->anchor
;
4957 struct imsm_disk
*_disk
;
4958 struct imsm_dev
*dev
;
4959 struct imsm_map
*map
;
4963 dev
= get_imsm_dev(super
, super
->current_vol
);
4964 map
= get_imsm_map(dev
, MAP_0
);
4966 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4967 pr_err("%s: Cannot add spare devices to IMSM volume\n",
4973 /* we're doing autolayout so grab the pre-marked (in
4974 * validate_geometry) raid_disk
4976 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4977 if (dl
->raiddisk
== dk
->raid_disk
)
4980 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4981 if (dl
->major
== dk
->major
&&
4982 dl
->minor
== dk
->minor
)
4987 pr_err("%s is not a member of the same container\n", devname
);
4991 /* add a pristine spare to the metadata */
4992 if (dl
->index
< 0) {
4993 dl
->index
= super
->anchor
->num_disks
;
4994 super
->anchor
->num_disks
++;
4996 /* Check the device has not already been added */
4997 slot
= get_imsm_disk_slot(map
, dl
->index
);
4999 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5000 pr_err("%s has been included in this array twice\n",
5004 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5005 dl
->disk
.status
= CONFIGURED_DISK
;
5007 /* update size of 'missing' disks to be at least as large as the
5008 * largest acitve member (we only have dummy missing disks when
5009 * creating the first volume)
5011 if (super
->current_vol
== 0) {
5012 for (df
= super
->missing
; df
; df
= df
->next
) {
5013 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5014 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5015 _disk
= __get_imsm_disk(mpb
, df
->index
);
5020 /* refresh unset/failed slots to point to valid 'missing' entries */
5021 for (df
= super
->missing
; df
; df
= df
->next
)
5022 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5023 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5025 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5027 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5028 if (is_gen_migration(dev
)) {
5029 struct imsm_map
*map2
= get_imsm_map(dev
,
5031 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5032 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5033 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5036 if ((unsigned)df
->index
==
5038 set_imsm_ord_tbl_ent(map2
,
5044 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5048 /* if we are creating the first raid device update the family number */
5049 if (super
->current_vol
== 0) {
5051 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5053 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5054 if (!_dev
|| !_disk
) {
5055 pr_err("BUG mpb setup error\n");
5061 sum
+= __gen_imsm_checksum(mpb
);
5062 mpb
->family_num
= __cpu_to_le32(sum
);
5063 mpb
->orig_family_num
= mpb
->family_num
;
5065 super
->current_disk
= dl
;
5070 * Function marks disk as spare and restores disk serial
5071 * in case it was previously marked as failed by takeover operation
5073 * -1 : critical error
5074 * 0 : disk is marked as spare but serial is not set
5077 int mark_spare(struct dl
*disk
)
5079 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5086 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5087 /* Restore disk serial number, because takeover marks disk
5088 * as failed and adds to serial ':0' before it becomes
5091 serialcpy(disk
->serial
, serial
);
5092 serialcpy(disk
->disk
.serial
, serial
);
5095 disk
->disk
.status
= SPARE_DISK
;
5101 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5102 int fd
, char *devname
,
5103 unsigned long long data_offset
)
5105 struct intel_super
*super
= st
->sb
;
5107 unsigned long long size
;
5108 unsigned int member_sector_size
;
5113 /* If we are on an RAID enabled platform check that the disk is
5114 * attached to the raid controller.
5115 * We do not need to test disks attachment for container based additions,
5116 * they shall be already tested when container was created/assembled.
5118 rv
= find_intel_hba_capability(fd
, super
, devname
);
5119 /* no orom/efi or non-intel hba of the disk */
5121 dprintf("capability: %p fd: %d ret: %d\n",
5122 super
->orom
, fd
, rv
);
5126 if (super
->current_vol
>= 0)
5127 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5130 dd
= xcalloc(sizeof(*dd
), 1);
5131 dd
->major
= major(stb
.st_rdev
);
5132 dd
->minor
= minor(stb
.st_rdev
);
5133 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5136 dd
->action
= DISK_ADD
;
5137 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5139 pr_err("failed to retrieve scsi serial, aborting\n");
5145 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5146 (super
->hba
->type
== SYS_DEV_VMD
))) {
5148 char *devpath
= diskfd_to_devpath(fd
);
5149 char controller_path
[PATH_MAX
];
5152 pr_err("failed to get devpath, aborting\n");
5159 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5162 if (devpath_to_vendor(controller_path
) == 0x8086) {
5164 * If Intel's NVMe drive has serial ended with
5165 * "-A","-B","-1" or "-2" it means that this is "x8"
5166 * device (double drive on single PCIe card).
5167 * User should be warned about potential data loss.
5169 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5170 /* Skip empty character at the end */
5171 if (dd
->serial
[i
] == 0)
5174 if (((dd
->serial
[i
] == 'A') ||
5175 (dd
->serial
[i
] == 'B') ||
5176 (dd
->serial
[i
] == '1') ||
5177 (dd
->serial
[i
] == '2')) &&
5178 (dd
->serial
[i
-1] == '-'))
5179 pr_err("\tThe action you are about to take may put your data at risk.\n"
5180 "\tPlease note that x8 devices may consist of two separate x4 devices "
5181 "located on a single PCIe port.\n"
5182 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5188 get_dev_size(fd
, NULL
, &size
);
5189 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5191 if (super
->sector_size
== 0) {
5192 /* this a first device, so sector_size is not set yet */
5193 super
->sector_size
= member_sector_size
;
5194 } else if (member_sector_size
!= super
->sector_size
) {
5195 pr_err("Mixing between different sector size is forbidden, aborting...\n");
5202 /* clear migr_rec when adding disk to container */
5203 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5204 if (lseek64(fd
, size
- MIGR_REC_POSITION
, SEEK_SET
) >= 0) {
5205 if (write(fd
, super
->migr_rec_buf
,
5206 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5207 perror("Write migr_rec failed");
5211 serialcpy(dd
->disk
.serial
, dd
->serial
);
5212 set_total_blocks(&dd
->disk
, size
);
5213 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5214 struct imsm_super
*mpb
= super
->anchor
;
5215 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5218 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5219 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5221 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5223 if (st
->update_tail
) {
5224 dd
->next
= super
->disk_mgmt_list
;
5225 super
->disk_mgmt_list
= dd
;
5227 dd
->next
= super
->disks
;
5229 super
->updates_pending
++;
5235 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5237 struct intel_super
*super
= st
->sb
;
5240 /* remove from super works only in mdmon - for communication
5241 * manager - monitor. Check if communication memory buffer
5244 if (!st
->update_tail
) {
5245 pr_err("shall be used in mdmon context only\n");
5248 dd
= xcalloc(1, sizeof(*dd
));
5249 dd
->major
= dk
->major
;
5250 dd
->minor
= dk
->minor
;
5253 dd
->action
= DISK_REMOVE
;
5255 dd
->next
= super
->disk_mgmt_list
;
5256 super
->disk_mgmt_list
= dd
;
5261 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5265 struct imsm_super anchor
;
5266 } spare_record
__attribute__ ((aligned(512)));
5268 /* spare records have their own family number and do not have any defined raid
5271 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5273 struct imsm_super
*mpb
= super
->anchor
;
5274 struct imsm_super
*spare
= &spare_record
.anchor
;
5278 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5279 spare
->generation_num
= __cpu_to_le32(1UL);
5280 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5281 spare
->num_disks
= 1;
5282 spare
->num_raid_devs
= 0;
5283 spare
->cache_size
= mpb
->cache_size
;
5284 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5286 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5287 MPB_SIGNATURE MPB_VERSION_RAID0
);
5289 for (d
= super
->disks
; d
; d
= d
->next
) {
5293 spare
->disk
[0] = d
->disk
;
5294 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5295 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5297 sum
= __gen_imsm_checksum(spare
);
5298 spare
->family_num
= __cpu_to_le32(sum
);
5299 spare
->orig_family_num
= 0;
5300 sum
= __gen_imsm_checksum(spare
);
5301 spare
->check_sum
= __cpu_to_le32(sum
);
5303 if (store_imsm_mpb(d
->fd
, spare
)) {
5304 pr_err("failed for device %d:%d %s\n",
5305 d
->major
, d
->minor
, strerror(errno
));
5317 static int write_super_imsm(struct supertype
*st
, int doclose
)
5319 struct intel_super
*super
= st
->sb
;
5320 struct imsm_super
*mpb
= super
->anchor
;
5326 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5328 int clear_migration_record
= 1;
5330 /* 'generation' is incremented everytime the metadata is written */
5331 generation
= __le32_to_cpu(mpb
->generation_num
);
5333 mpb
->generation_num
= __cpu_to_le32(generation
);
5335 /* fix up cases where previous mdadm releases failed to set
5338 if (mpb
->orig_family_num
== 0)
5339 mpb
->orig_family_num
= mpb
->family_num
;
5341 for (d
= super
->disks
; d
; d
= d
->next
) {
5345 mpb
->disk
[d
->index
] = d
->disk
;
5349 for (d
= super
->missing
; d
; d
= d
->next
) {
5350 mpb
->disk
[d
->index
] = d
->disk
;
5353 mpb
->num_disks
= num_disks
;
5354 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5356 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5357 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5358 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5360 imsm_copy_dev(dev
, dev2
);
5361 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5363 if (is_gen_migration(dev2
))
5364 clear_migration_record
= 0;
5366 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5367 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5369 /* recalculate checksum */
5370 sum
= __gen_imsm_checksum(mpb
);
5371 mpb
->check_sum
= __cpu_to_le32(sum
);
5373 if (super
->clean_migration_record_by_mdmon
) {
5374 clear_migration_record
= 1;
5375 super
->clean_migration_record_by_mdmon
= 0;
5377 if (clear_migration_record
)
5378 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5380 /* write the mpb for disks that compose raid devices */
5381 for (d
= super
->disks
; d
; d
= d
->next
) {
5382 if (d
->index
< 0 || is_failed(&d
->disk
))
5385 if (clear_migration_record
) {
5386 unsigned long long dsize
;
5388 get_dev_size(d
->fd
, NULL
, &dsize
);
5389 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5390 if (write(d
->fd
, super
->migr_rec_buf
,
5391 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5392 perror("Write migr_rec failed");
5396 if (store_imsm_mpb(d
->fd
, mpb
))
5398 "failed for device %d:%d (fd: %d)%s\n",
5400 d
->fd
, strerror(errno
));
5409 return write_super_imsm_spares(super
, doclose
);
5414 static int create_array(struct supertype
*st
, int dev_idx
)
5417 struct imsm_update_create_array
*u
;
5418 struct intel_super
*super
= st
->sb
;
5419 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5420 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5421 struct disk_info
*inf
;
5422 struct imsm_disk
*disk
;
5425 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5426 sizeof(*inf
) * map
->num_members
;
5428 u
->type
= update_create_array
;
5429 u
->dev_idx
= dev_idx
;
5430 imsm_copy_dev(&u
->dev
, dev
);
5431 inf
= get_disk_info(u
);
5432 for (i
= 0; i
< map
->num_members
; i
++) {
5433 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5435 disk
= get_imsm_disk(super
, idx
);
5437 disk
= get_imsm_missing(super
, idx
);
5438 serialcpy(inf
[i
].serial
, disk
->serial
);
5440 append_metadata_update(st
, u
, len
);
5445 static int mgmt_disk(struct supertype
*st
)
5447 struct intel_super
*super
= st
->sb
;
5449 struct imsm_update_add_remove_disk
*u
;
5451 if (!super
->disk_mgmt_list
)
5456 u
->type
= update_add_remove_disk
;
5457 append_metadata_update(st
, u
, len
);
5462 static int write_init_super_imsm(struct supertype
*st
)
5464 struct intel_super
*super
= st
->sb
;
5465 int current_vol
= super
->current_vol
;
5467 /* we are done with current_vol reset it to point st at the container */
5468 super
->current_vol
= -1;
5470 if (st
->update_tail
) {
5471 /* queue the recently created array / added disk
5472 * as a metadata update */
5475 /* determine if we are creating a volume or adding a disk */
5476 if (current_vol
< 0) {
5477 /* in the mgmt (add/remove) disk case we are running
5478 * in mdmon context, so don't close fd's
5480 return mgmt_disk(st
);
5482 rv
= create_array(st
, current_vol
);
5487 for (d
= super
->disks
; d
; d
= d
->next
)
5488 Kill(d
->devname
, NULL
, 0, -1, 1);
5489 return write_super_imsm(st
, 1);
5494 static int store_super_imsm(struct supertype
*st
, int fd
)
5496 struct intel_super
*super
= st
->sb
;
5497 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5503 return store_imsm_mpb(fd
, mpb
);
5509 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5511 return __le32_to_cpu(mpb
->bbm_log_size
);
5515 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5516 int layout
, int raiddisks
, int chunk
,
5517 unsigned long long size
,
5518 unsigned long long data_offset
,
5520 unsigned long long *freesize
,
5524 unsigned long long ldsize
;
5525 struct intel_super
*super
;
5528 if (level
!= LEVEL_CONTAINER
)
5533 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5536 pr_err("imsm: Cannot open %s: %s\n",
5537 dev
, strerror(errno
));
5540 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5545 /* capabilities retrieve could be possible
5546 * note that there is no fd for the disks in array.
5548 super
= alloc_super();
5549 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
5555 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
5559 fd2devname(fd
, str
);
5560 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5561 fd
, str
, super
->orom
, rv
, raiddisks
);
5563 /* no orom/efi or non-intel hba of the disk */
5570 if (raiddisks
> super
->orom
->tds
) {
5572 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
5573 raiddisks
, super
->orom
->tds
);
5577 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5578 (ldsize
>> 9) >> 32 > 0) {
5580 pr_err("%s exceeds maximum platform supported size\n", dev
);
5586 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
5592 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5594 const unsigned long long base_start
= e
[*idx
].start
;
5595 unsigned long long end
= base_start
+ e
[*idx
].size
;
5598 if (base_start
== end
)
5602 for (i
= *idx
; i
< num_extents
; i
++) {
5603 /* extend overlapping extents */
5604 if (e
[i
].start
>= base_start
&&
5605 e
[i
].start
<= end
) {
5608 if (e
[i
].start
+ e
[i
].size
> end
)
5609 end
= e
[i
].start
+ e
[i
].size
;
5610 } else if (e
[i
].start
> end
) {
5616 return end
- base_start
;
5619 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5621 /* build a composite disk with all known extents and generate a new
5622 * 'maxsize' given the "all disks in an array must share a common start
5623 * offset" constraint
5625 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
5629 unsigned long long pos
;
5630 unsigned long long start
= 0;
5631 unsigned long long maxsize
;
5632 unsigned long reserve
;
5634 /* coalesce and sort all extents. also, check to see if we need to
5635 * reserve space between member arrays
5638 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5641 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5644 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5649 while (i
< sum_extents
) {
5650 e
[j
].start
= e
[i
].start
;
5651 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5653 if (e
[j
-1].size
== 0)
5662 unsigned long long esize
;
5664 esize
= e
[i
].start
- pos
;
5665 if (esize
>= maxsize
) {
5670 pos
= e
[i
].start
+ e
[i
].size
;
5672 } while (e
[i
-1].size
);
5678 /* FIXME assumes volume at offset 0 is the first volume in a
5681 if (start_extent
> 0)
5682 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5686 if (maxsize
< reserve
)
5689 super
->create_offset
= ~((unsigned long long) 0);
5690 if (start
+ reserve
> super
->create_offset
)
5691 return 0; /* start overflows create_offset */
5692 super
->create_offset
= start
+ reserve
;
5694 return maxsize
- reserve
;
5697 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5699 if (level
< 0 || level
== 6 || level
== 4)
5702 /* if we have an orom prevent invalid raid levels */
5705 case 0: return imsm_orom_has_raid0(orom
);
5708 return imsm_orom_has_raid1e(orom
);
5709 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5710 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5711 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5714 return 1; /* not on an Intel RAID platform so anything goes */
5720 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5721 int dpa
, int verbose
)
5723 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5724 struct mdstat_ent
*memb
;
5730 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5731 if (memb
->metadata_version
&&
5732 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5733 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5734 !is_subarray(memb
->metadata_version
+9) &&
5736 struct dev_member
*dev
= memb
->members
;
5738 while(dev
&& (fd
< 0)) {
5739 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5740 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5742 fd
= open(path
, O_RDONLY
, 0);
5743 if (num
<= 0 || fd
< 0) {
5744 pr_vrb("Cannot open %s: %s\n",
5745 dev
->name
, strerror(errno
));
5751 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
5752 struct mdstat_ent
*vol
;
5753 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5754 if (vol
->active
> 0 &&
5755 vol
->metadata_version
&&
5756 is_container_member(vol
, memb
->devnm
)) {
5761 if (*devlist
&& (found
< dpa
)) {
5762 dv
= xcalloc(1, sizeof(*dv
));
5763 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
5764 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
5767 dv
->next
= *devlist
;
5775 free_mdstat(mdstat
);
5780 static struct md_list
*
5781 get_loop_devices(void)
5784 struct md_list
*devlist
= NULL
;
5787 for(i
= 0; i
< 12; i
++) {
5788 dv
= xcalloc(1, sizeof(*dv
));
5789 dv
->devname
= xmalloc(40);
5790 sprintf(dv
->devname
, "/dev/loop%d", i
);
5798 static struct md_list
*
5799 get_devices(const char *hba_path
)
5801 struct md_list
*devlist
= NULL
;
5808 devlist
= get_loop_devices();
5811 /* scroll through /sys/dev/block looking for devices attached to
5814 dir
= opendir("/sys/dev/block");
5815 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5820 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5822 path
= devt_to_devpath(makedev(major
, minor
));
5825 if (!path_attached_to_hba(path
, hba_path
)) {
5832 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5834 fd2devname(fd
, buf
);
5837 pr_err("cannot open device: %s\n",
5842 dv
= xcalloc(1, sizeof(*dv
));
5843 dv
->devname
= xstrdup(buf
);
5850 devlist
= devlist
->next
;
5860 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5861 int verbose
, int *found
)
5863 struct md_list
*tmpdev
;
5865 struct supertype
*st
;
5867 /* first walk the list of devices to find a consistent set
5868 * that match the criterea, if that is possible.
5869 * We flag the ones we like with 'used'.
5872 st
= match_metadata_desc_imsm("imsm");
5874 pr_vrb("cannot allocate memory for imsm supertype\n");
5878 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5879 char *devname
= tmpdev
->devname
;
5881 struct supertype
*tst
;
5883 if (tmpdev
->used
> 1)
5885 tst
= dup_super(st
);
5887 pr_vrb("cannot allocate memory for imsm supertype\n");
5890 tmpdev
->container
= 0;
5891 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5893 dprintf("cannot open device %s: %s\n",
5894 devname
, strerror(errno
));
5896 } else if (fstat(dfd
, &stb
)< 0) {
5898 dprintf("fstat failed for %s: %s\n",
5899 devname
, strerror(errno
));
5901 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5902 dprintf("%s is not a block device.\n",
5905 } else if (must_be_container(dfd
)) {
5906 struct supertype
*cst
;
5907 cst
= super_by_fd(dfd
, NULL
);
5909 dprintf("cannot recognize container type %s\n",
5912 } else if (tst
->ss
!= st
->ss
) {
5913 dprintf("non-imsm container - ignore it: %s\n",
5916 } else if (!tst
->ss
->load_container
||
5917 tst
->ss
->load_container(tst
, dfd
, NULL
))
5920 tmpdev
->container
= 1;
5923 cst
->ss
->free_super(cst
);
5925 tmpdev
->st_rdev
= stb
.st_rdev
;
5926 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5927 dprintf("no RAID superblock on %s\n",
5930 } else if (tst
->ss
->compare_super
== NULL
) {
5931 dprintf("Cannot assemble %s metadata on %s\n",
5932 tst
->ss
->name
, devname
);
5938 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5939 /* Ignore unrecognised devices during auto-assembly */
5944 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5946 if (st
->minor_version
== -1)
5947 st
->minor_version
= tst
->minor_version
;
5949 if (memcmp(info
.uuid
, uuid_zero
,
5950 sizeof(int[4])) == 0) {
5951 /* this is a floating spare. It cannot define
5952 * an array unless there are no more arrays of
5953 * this type to be found. It can be included
5954 * in an array of this type though.
5960 if (st
->ss
!= tst
->ss
||
5961 st
->minor_version
!= tst
->minor_version
||
5962 st
->ss
->compare_super(st
, tst
) != 0) {
5963 /* Some mismatch. If exactly one array matches this host,
5964 * we can resolve on that one.
5965 * Or, if we are auto assembling, we just ignore the second
5968 dprintf("superblock on %s doesn't match others - assembly aborted\n",
5974 dprintf("found: devname: %s\n", devname
);
5978 tst
->ss
->free_super(tst
);
5982 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5983 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5984 for (iter
= head
; iter
; iter
= iter
->next
) {
5985 dprintf("content->text_version: %s vol\n",
5986 iter
->text_version
);
5987 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5988 /* do not assemble arrays with unsupported
5990 dprintf("Cannot activate member %s.\n",
5991 iter
->text_version
);
5998 dprintf("No valid super block on device list: err: %d %p\n",
6002 dprintf("no more devices to examine\n");
6005 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6006 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6008 if (count
< tmpdev
->found
)
6011 count
-= tmpdev
->found
;
6014 if (tmpdev
->used
== 1)
6019 st
->ss
->free_super(st
);
6024 count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
6026 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6028 const struct orom_entry
*entry
;
6029 struct devid_list
*dv
, *devid_list
;
6031 if (!hba
|| !hba
->path
)
6034 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
6035 if (strstr(idev
->path
, hba
->path
))
6039 if (!idev
|| !idev
->dev_id
)
6042 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
6044 if (!entry
|| !entry
->devid_list
)
6047 devid_list
= entry
->devid_list
;
6048 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
6049 struct md_list
*devlist
;
6050 struct sys_dev
*device
= device_by_id(dv
->devid
);
6055 hba_path
= device
->path
;
6059 devlist
= get_devices(hba_path
);
6060 /* if no intel devices return zero volumes */
6061 if (devlist
== NULL
)
6064 count
+= active_arrays_by_format("imsm", hba_path
, &devlist
, dpa
, verbose
);
6065 dprintf("path: %s active arrays: %d\n", hba_path
, count
);
6066 if (devlist
== NULL
)
6070 count
+= count_volumes_list(devlist
,
6074 dprintf("found %d count: %d\n", found
, count
);
6077 dprintf("path: %s total number of volumes: %d\n", hba_path
, count
);
6080 struct md_list
*dv
= devlist
;
6081 devlist
= devlist
->next
;
6089 static int imsm_default_chunk(const struct imsm_orom
*orom
)
6091 /* up to 512 if the plaform supports it, otherwise the platform max.
6092 * 128 if no platform detected
6094 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
6096 return min(512, (1 << fs
));
6100 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
6101 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
6103 /* check/set platform and metadata limits/defaults */
6104 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
6105 pr_vrb("platform supports a maximum of %d disks per array\n",
6110 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6111 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
6112 pr_vrb("platform does not support raid%d with %d disk%s\n",
6113 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
6117 if (*chunk
== 0 || *chunk
== UnSet
)
6118 *chunk
= imsm_default_chunk(super
->orom
);
6120 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
6121 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
6125 if (layout
!= imsm_level_to_layout(level
)) {
6127 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
6128 else if (level
== 10)
6129 pr_vrb("imsm raid 10 only supports the n2 layout\n");
6131 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
6136 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
6137 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
6138 pr_vrb("platform does not support a volume size over 2TB\n");
6145 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
6146 * FIX ME add ahci details
6148 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
6149 int layout
, int raiddisks
, int *chunk
,
6150 unsigned long long size
,
6151 unsigned long long data_offset
,
6153 unsigned long long *freesize
,
6157 struct intel_super
*super
= st
->sb
;
6158 struct imsm_super
*mpb
;
6160 unsigned long long pos
= 0;
6161 unsigned long long maxsize
;
6165 /* We must have the container info already read in. */
6169 mpb
= super
->anchor
;
6171 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
6172 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6176 /* General test: make sure there is space for
6177 * 'raiddisks' device extents of size 'size' at a given
6180 unsigned long long minsize
= size
;
6181 unsigned long long start_offset
= MaxSector
;
6184 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
6185 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6190 e
= get_extents(super
, dl
);
6193 unsigned long long esize
;
6194 esize
= e
[i
].start
- pos
;
6195 if (esize
>= minsize
)
6197 if (found
&& start_offset
== MaxSector
) {
6200 } else if (found
&& pos
!= start_offset
) {
6204 pos
= e
[i
].start
+ e
[i
].size
;
6206 } while (e
[i
-1].size
);
6211 if (dcnt
< raiddisks
) {
6213 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6220 /* This device must be a member of the set */
6221 if (stat(dev
, &stb
) < 0)
6223 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6225 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6226 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6227 dl
->minor
== (int)minor(stb
.st_rdev
))
6232 pr_err("%s is not in the same imsm set\n", dev
);
6234 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6235 /* If a volume is present then the current creation attempt
6236 * cannot incorporate new spares because the orom may not
6237 * understand this configuration (all member disks must be
6238 * members of each array in the container).
6240 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6241 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6243 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6244 mpb
->num_disks
!= raiddisks
) {
6245 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6249 /* retrieve the largest free space block */
6250 e
= get_extents(super
, dl
);
6255 unsigned long long esize
;
6257 esize
= e
[i
].start
- pos
;
6258 if (esize
>= maxsize
)
6260 pos
= e
[i
].start
+ e
[i
].size
;
6262 } while (e
[i
-1].size
);
6267 pr_err("unable to determine free space for: %s\n",
6271 if (maxsize
< size
) {
6273 pr_err("%s not enough space (%llu < %llu)\n",
6274 dev
, maxsize
, size
);
6278 /* count total number of extents for merge */
6280 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6282 i
+= dl
->extent_cnt
;
6284 maxsize
= merge_extents(super
, i
);
6286 if (!check_env("IMSM_NO_PLATFORM") &&
6287 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6288 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6292 if (maxsize
< size
|| maxsize
== 0) {
6295 pr_err("no free space left on device. Aborting...\n");
6297 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
6303 *freesize
= maxsize
;
6306 int count
= count_volumes(super
->hba
,
6307 super
->orom
->dpa
, verbose
);
6308 if (super
->orom
->vphba
<= count
) {
6309 pr_vrb("platform does not support more than %d raid volumes.\n",
6310 super
->orom
->vphba
);
6317 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6318 unsigned long long size
, int chunk
,
6319 unsigned long long *freesize
)
6321 struct intel_super
*super
= st
->sb
;
6322 struct imsm_super
*mpb
= super
->anchor
;
6327 unsigned long long maxsize
;
6328 unsigned long long minsize
;
6332 /* find the largest common start free region of the possible disks */
6336 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6342 /* don't activate new spares if we are orom constrained
6343 * and there is already a volume active in the container
6345 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6348 e
= get_extents(super
, dl
);
6351 for (i
= 1; e
[i
-1].size
; i
++)
6359 maxsize
= merge_extents(super
, extent_cnt
);
6363 minsize
= chunk
* 2;
6365 if (cnt
< raiddisks
||
6366 (super
->orom
&& used
&& used
!= raiddisks
) ||
6367 maxsize
< minsize
||
6369 pr_err("not enough devices with space to create array.\n");
6370 return 0; /* No enough free spaces large enough */
6381 if (!check_env("IMSM_NO_PLATFORM") &&
6382 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6383 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6387 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6389 dl
->raiddisk
= cnt
++;
6393 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6398 static int reserve_space(struct supertype
*st
, int raiddisks
,
6399 unsigned long long size
, int chunk
,
6400 unsigned long long *freesize
)
6402 struct intel_super
*super
= st
->sb
;
6407 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6410 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6412 dl
->raiddisk
= cnt
++;
6419 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6420 int raiddisks
, int *chunk
, unsigned long long size
,
6421 unsigned long long data_offset
,
6422 char *dev
, unsigned long long *freesize
,
6430 * if given unused devices create a container
6431 * if given given devices in a container create a member volume
6433 if (level
== LEVEL_CONTAINER
) {
6434 /* Must be a fresh device to add to a container */
6435 return validate_geometry_imsm_container(st
, level
, layout
,
6445 struct intel_super
*super
= st
->sb
;
6446 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6447 raiddisks
, chunk
, size
,
6450 /* we are being asked to automatically layout a
6451 * new volume based on the current contents of
6452 * the container. If the the parameters can be
6453 * satisfied reserve_space will record the disks,
6454 * start offset, and size of the volume to be
6455 * created. add_to_super and getinfo_super
6456 * detect when autolayout is in progress.
6458 /* assuming that freesize is always given when array is
6460 if (super
->orom
&& freesize
) {
6462 count
= count_volumes(super
->hba
,
6463 super
->orom
->dpa
, verbose
);
6464 if (super
->orom
->vphba
<= count
) {
6465 pr_vrb("platform does not support more than %d raid volumes.\n",
6466 super
->orom
->vphba
);
6471 return reserve_space(st
, raiddisks
, size
,
6477 /* creating in a given container */
6478 return validate_geometry_imsm_volume(st
, level
, layout
,
6479 raiddisks
, chunk
, size
,
6481 dev
, freesize
, verbose
);
6484 /* This device needs to be a device in an 'imsm' container */
6485 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6488 pr_err("Cannot create this array on device %s\n",
6493 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6495 pr_err("Cannot open %s: %s\n",
6496 dev
, strerror(errno
));
6499 /* Well, it is in use by someone, maybe an 'imsm' container. */
6500 cfd
= open_container(fd
);
6504 pr_err("Cannot use %s: It is busy\n",
6508 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
6509 if (sra
&& sra
->array
.major_version
== -1 &&
6510 strcmp(sra
->text_version
, "imsm") == 0)
6514 /* This is a member of a imsm container. Load the container
6515 * and try to create a volume
6517 struct intel_super
*super
;
6519 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6521 strcpy(st
->container_devnm
, fd2devnm(cfd
));
6523 return validate_geometry_imsm_volume(st
, level
, layout
,
6525 size
, data_offset
, dev
,
6532 pr_err("failed container membership check\n");
6538 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6540 struct intel_super
*super
= st
->sb
;
6542 if (level
&& *level
== UnSet
)
6543 *level
= LEVEL_CONTAINER
;
6545 if (level
&& layout
&& *layout
== UnSet
)
6546 *layout
= imsm_level_to_layout(*level
);
6548 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6549 *chunk
= imsm_default_chunk(super
->orom
);
6552 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6554 static int kill_subarray_imsm(struct supertype
*st
)
6556 /* remove the subarray currently referenced by ->current_vol */
6558 struct intel_dev
**dp
;
6559 struct intel_super
*super
= st
->sb
;
6560 __u8 current_vol
= super
->current_vol
;
6561 struct imsm_super
*mpb
= super
->anchor
;
6563 if (super
->current_vol
< 0)
6565 super
->current_vol
= -1; /* invalidate subarray cursor */
6567 /* block deletions that would change the uuid of active subarrays
6569 * FIXME when immutable ids are available, but note that we'll
6570 * also need to fixup the invalidated/active subarray indexes in
6573 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6576 if (i
< current_vol
)
6578 sprintf(subarray
, "%u", i
);
6579 if (is_subarray_active(subarray
, st
->devnm
)) {
6580 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6587 if (st
->update_tail
) {
6588 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
6590 u
->type
= update_kill_array
;
6591 u
->dev_idx
= current_vol
;
6592 append_metadata_update(st
, u
, sizeof(*u
));
6597 for (dp
= &super
->devlist
; *dp
;)
6598 if ((*dp
)->index
== current_vol
) {
6601 handle_missing(super
, (*dp
)->dev
);
6602 if ((*dp
)->index
> current_vol
)
6607 /* no more raid devices, all active components are now spares,
6608 * but of course failed are still failed
6610 if (--mpb
->num_raid_devs
== 0) {
6613 for (d
= super
->disks
; d
; d
= d
->next
)
6618 super
->updates_pending
++;
6623 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6624 char *update
, struct mddev_ident
*ident
)
6626 /* update the subarray currently referenced by ->current_vol */
6627 struct intel_super
*super
= st
->sb
;
6628 struct imsm_super
*mpb
= super
->anchor
;
6630 if (strcmp(update
, "name") == 0) {
6631 char *name
= ident
->name
;
6635 if (is_subarray_active(subarray
, st
->devnm
)) {
6636 pr_err("Unable to update name of active subarray\n");
6640 if (!check_name(super
, name
, 0))
6643 vol
= strtoul(subarray
, &ep
, 10);
6644 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6647 if (st
->update_tail
) {
6648 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
6650 u
->type
= update_rename_array
;
6652 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6653 append_metadata_update(st
, u
, sizeof(*u
));
6655 struct imsm_dev
*dev
;
6658 dev
= get_imsm_dev(super
, vol
);
6659 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6660 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6661 dev
= get_imsm_dev(super
, i
);
6662 handle_missing(super
, dev
);
6664 super
->updates_pending
++;
6671 #endif /* MDASSEMBLE */
6673 static int is_gen_migration(struct imsm_dev
*dev
)
6678 if (!dev
->vol
.migr_state
)
6681 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6687 static int is_rebuilding(struct imsm_dev
*dev
)
6689 struct imsm_map
*migr_map
;
6691 if (!dev
->vol
.migr_state
)
6694 if (migr_type(dev
) != MIGR_REBUILD
)
6697 migr_map
= get_imsm_map(dev
, MAP_1
);
6699 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6706 static int is_initializing(struct imsm_dev
*dev
)
6708 struct imsm_map
*migr_map
;
6710 if (!dev
->vol
.migr_state
)
6713 if (migr_type(dev
) != MIGR_INIT
)
6716 migr_map
= get_imsm_map(dev
, MAP_1
);
6718 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6725 static void update_recovery_start(struct intel_super
*super
,
6726 struct imsm_dev
*dev
,
6727 struct mdinfo
*array
)
6729 struct mdinfo
*rebuild
= NULL
;
6733 if (!is_rebuilding(dev
))
6736 /* Find the rebuild target, but punt on the dual rebuild case */
6737 for (d
= array
->devs
; d
; d
= d
->next
)
6738 if (d
->recovery_start
== 0) {
6745 /* (?) none of the disks are marked with
6746 * IMSM_ORD_REBUILD, so assume they are missing and the
6747 * disk_ord_tbl was not correctly updated
6749 dprintf("failed to locate out-of-sync disk\n");
6753 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6754 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6758 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6761 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6763 /* Given a container loaded by load_super_imsm_all,
6764 * extract information about all the arrays into
6766 * If 'subarray' is given, just extract info about that array.
6768 * For each imsm_dev create an mdinfo, fill it in,
6769 * then look for matching devices in super->disks
6770 * and create appropriate device mdinfo.
6772 struct intel_super
*super
= st
->sb
;
6773 struct imsm_super
*mpb
= super
->anchor
;
6774 struct mdinfo
*rest
= NULL
;
6778 int spare_disks
= 0;
6780 /* do not assemble arrays when not all attributes are supported */
6781 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6783 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
6786 /* check for bad blocks */
6787 if (imsm_bbm_log_size(super
->anchor
)) {
6788 pr_err("BBM log found in IMSM metadata.Arrays activation is blocked.\n");
6792 /* count spare devices, not used in maps
6794 for (d
= super
->disks
; d
; d
= d
->next
)
6798 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6799 struct imsm_dev
*dev
;
6800 struct imsm_map
*map
;
6801 struct imsm_map
*map2
;
6802 struct mdinfo
*this;
6810 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6813 dev
= get_imsm_dev(super
, i
);
6814 map
= get_imsm_map(dev
, MAP_0
);
6815 map2
= get_imsm_map(dev
, MAP_1
);
6817 /* do not publish arrays that are in the middle of an
6818 * unsupported migration
6820 if (dev
->vol
.migr_state
&&
6821 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6822 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
6826 /* do not publish arrays that are not support by controller's
6830 this = xmalloc(sizeof(*this));
6832 super
->current_vol
= i
;
6833 getinfo_super_imsm_volume(st
, this, NULL
);
6836 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6837 /* mdadm does not support all metadata features- set the bit in all arrays state */
6838 if (!validate_geometry_imsm_orom(super
,
6839 get_imsm_raid_level(map
), /* RAID level */
6840 imsm_level_to_layout(get_imsm_raid_level(map
)),
6841 map
->num_members
, /* raid disks */
6842 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6844 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
6846 this->array
.state
|=
6847 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6848 (1<<MD_SB_BLOCK_VOLUME
);
6852 /* if array has bad blocks, set suitable bit in all arrays state */
6854 this->array
.state
|=
6855 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6856 (1<<MD_SB_BLOCK_VOLUME
);
6858 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6859 unsigned long long recovery_start
;
6860 struct mdinfo
*info_d
;
6867 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6868 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6869 for (d
= super
->disks
; d
; d
= d
->next
)
6870 if (d
->index
== idx
)
6873 recovery_start
= MaxSector
;
6876 if (d
&& is_failed(&d
->disk
))
6878 if (ord
& IMSM_ORD_REBUILD
)
6882 * if we skip some disks the array will be assmebled degraded;
6883 * reset resync start to avoid a dirty-degraded
6884 * situation when performing the intial sync
6886 * FIXME handle dirty degraded
6888 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6889 this->resync_start
= MaxSector
;
6893 info_d
= xcalloc(1, sizeof(*info_d
));
6894 info_d
->next
= this->devs
;
6895 this->devs
= info_d
;
6897 info_d
->disk
.number
= d
->index
;
6898 info_d
->disk
.major
= d
->major
;
6899 info_d
->disk
.minor
= d
->minor
;
6900 info_d
->disk
.raid_disk
= slot
;
6901 info_d
->recovery_start
= recovery_start
;
6903 if (slot
< map2
->num_members
)
6904 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6906 this->array
.spare_disks
++;
6908 if (slot
< map
->num_members
)
6909 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6911 this->array
.spare_disks
++;
6913 if (info_d
->recovery_start
== MaxSector
)
6914 this->array
.working_disks
++;
6916 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6917 info_d
->data_offset
= pba_of_lba0(map
);
6918 info_d
->component_size
= blocks_per_member(map
);
6920 /* now that the disk list is up-to-date fixup recovery_start */
6921 update_recovery_start(super
, dev
, this);
6922 this->array
.spare_disks
+= spare_disks
;
6925 /* check for reshape */
6926 if (this->reshape_active
== 1)
6927 recover_backup_imsm(st
, this);
6935 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6936 int failed
, int look_in_map
)
6938 struct imsm_map
*map
;
6940 map
= get_imsm_map(dev
, look_in_map
);
6943 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6944 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6946 switch (get_imsm_raid_level(map
)) {
6948 return IMSM_T_STATE_FAILED
;
6951 if (failed
< map
->num_members
)
6952 return IMSM_T_STATE_DEGRADED
;
6954 return IMSM_T_STATE_FAILED
;
6959 * check to see if any mirrors have failed, otherwise we
6960 * are degraded. Even numbered slots are mirrored on
6964 /* gcc -Os complains that this is unused */
6965 int insync
= insync
;
6967 for (i
= 0; i
< map
->num_members
; i
++) {
6968 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6969 int idx
= ord_to_idx(ord
);
6970 struct imsm_disk
*disk
;
6972 /* reset the potential in-sync count on even-numbered
6973 * slots. num_copies is always 2 for imsm raid10
6978 disk
= get_imsm_disk(super
, idx
);
6979 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6982 /* no in-sync disks left in this mirror the
6986 return IMSM_T_STATE_FAILED
;
6989 return IMSM_T_STATE_DEGRADED
;
6993 return IMSM_T_STATE_DEGRADED
;
6995 return IMSM_T_STATE_FAILED
;
7001 return map
->map_state
;
7004 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
7009 struct imsm_disk
*disk
;
7010 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7011 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
7012 struct imsm_map
*map_for_loop
;
7017 /* at the beginning of migration we set IMSM_ORD_REBUILD on
7018 * disks that are being rebuilt. New failures are recorded to
7019 * map[0]. So we look through all the disks we started with and
7020 * see if any failures are still present, or if any new ones
7024 if (prev
&& (map
->num_members
< prev
->num_members
))
7025 map_for_loop
= prev
;
7027 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
7029 /* when MAP_X is passed both maps failures are counted
7032 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
7033 i
< prev
->num_members
) {
7034 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
7035 idx_1
= ord_to_idx(ord
);
7037 disk
= get_imsm_disk(super
, idx_1
);
7038 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7041 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
7042 i
< map
->num_members
) {
7043 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
7044 idx
= ord_to_idx(ord
);
7047 disk
= get_imsm_disk(super
, idx
);
7048 if (!disk
|| is_failed(disk
) ||
7049 ord
& IMSM_ORD_REBUILD
)
7059 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
7062 struct intel_super
*super
= c
->sb
;
7063 struct imsm_super
*mpb
= super
->anchor
;
7065 if (atoi(inst
) >= mpb
->num_raid_devs
) {
7066 pr_err("subarry index %d, out of range\n", atoi(inst
));
7070 dprintf("imsm: open_new %s\n", inst
);
7071 a
->info
.container_member
= atoi(inst
);
7075 static int is_resyncing(struct imsm_dev
*dev
)
7077 struct imsm_map
*migr_map
;
7079 if (!dev
->vol
.migr_state
)
7082 if (migr_type(dev
) == MIGR_INIT
||
7083 migr_type(dev
) == MIGR_REPAIR
)
7086 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7089 migr_map
= get_imsm_map(dev
, MAP_1
);
7091 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
7092 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
7098 /* return true if we recorded new information */
7099 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7103 struct imsm_map
*map
;
7104 char buf
[MAX_RAID_SERIAL_LEN
+3];
7105 unsigned int len
, shift
= 0;
7107 /* new failures are always set in map[0] */
7108 map
= get_imsm_map(dev
, MAP_0
);
7110 slot
= get_imsm_disk_slot(map
, idx
);
7114 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
7115 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
7118 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
7119 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
7121 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
7122 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
7123 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
7125 disk
->status
|= FAILED_DISK
;
7126 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
7127 /* mark failures in second map if second map exists and this disk
7129 * This is valid for migration, initialization and rebuild
7131 if (dev
->vol
.migr_state
) {
7132 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
7133 int slot2
= get_imsm_disk_slot(map2
, idx
);
7135 if (slot2
< map2
->num_members
&& slot2
>= 0)
7136 set_imsm_ord_tbl_ent(map2
, slot2
,
7137 idx
| IMSM_ORD_REBUILD
);
7139 if (map
->failed_disk_num
== 0xff)
7140 map
->failed_disk_num
= slot
;
7144 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7146 mark_failure(dev
, disk
, idx
);
7148 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
7151 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
7152 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
7155 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7159 if (!super
->missing
)
7162 /* When orom adds replacement for missing disk it does
7163 * not remove entry of missing disk, but just updates map with
7164 * new added disk. So it is not enough just to test if there is
7165 * any missing disk, we have to look if there are any failed disks
7166 * in map to stop migration */
7168 dprintf("imsm: mark missing\n");
7169 /* end process for initialization and rebuild only
7171 if (is_gen_migration(dev
) == 0) {
7175 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7176 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7179 end_migration(dev
, super
, map_state
);
7181 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7182 mark_missing(dev
, &dl
->disk
, dl
->index
);
7183 super
->updates_pending
++;
7186 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7189 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7190 unsigned long long array_blocks
;
7191 struct imsm_map
*map
;
7193 if (used_disks
== 0) {
7194 /* when problems occures
7195 * return current array_blocks value
7197 array_blocks
= __le32_to_cpu(dev
->size_high
);
7198 array_blocks
= array_blocks
<< 32;
7199 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7201 return array_blocks
;
7204 /* set array size in metadata
7206 if (new_size
<= 0) {
7207 /* OLCE size change is caused by added disks
7209 map
= get_imsm_map(dev
, MAP_0
);
7210 array_blocks
= blocks_per_member(map
) * used_disks
;
7212 /* Online Volume Size Change
7213 * Using available free space
7215 array_blocks
= new_size
;
7218 /* round array size down to closest MB
7220 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7221 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7222 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7224 return array_blocks
;
7227 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7229 static void imsm_progress_container_reshape(struct intel_super
*super
)
7231 /* if no device has a migr_state, but some device has a
7232 * different number of members than the previous device, start
7233 * changing the number of devices in this device to match
7236 struct imsm_super
*mpb
= super
->anchor
;
7237 int prev_disks
= -1;
7241 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7242 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7243 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7244 struct imsm_map
*map2
;
7245 int prev_num_members
;
7247 if (dev
->vol
.migr_state
)
7250 if (prev_disks
== -1)
7251 prev_disks
= map
->num_members
;
7252 if (prev_disks
== map
->num_members
)
7255 /* OK, this array needs to enter reshape mode.
7256 * i.e it needs a migr_state
7259 copy_map_size
= sizeof_imsm_map(map
);
7260 prev_num_members
= map
->num_members
;
7261 map
->num_members
= prev_disks
;
7262 dev
->vol
.migr_state
= 1;
7263 dev
->vol
.curr_migr_unit
= 0;
7264 set_migr_type(dev
, MIGR_GEN_MIGR
);
7265 for (i
= prev_num_members
;
7266 i
< map
->num_members
; i
++)
7267 set_imsm_ord_tbl_ent(map
, i
, i
);
7268 map2
= get_imsm_map(dev
, MAP_1
);
7269 /* Copy the current map */
7270 memcpy(map2
, map
, copy_map_size
);
7271 map2
->num_members
= prev_num_members
;
7273 imsm_set_array_size(dev
, -1);
7274 super
->clean_migration_record_by_mdmon
= 1;
7275 super
->updates_pending
++;
7279 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7280 * states are handled in imsm_set_disk() with one exception, when a
7281 * resync is stopped due to a new failure this routine will set the
7282 * 'degraded' state for the array.
7284 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7286 int inst
= a
->info
.container_member
;
7287 struct intel_super
*super
= a
->container
->sb
;
7288 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7289 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7290 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7291 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7292 __u32 blocks_per_unit
;
7294 if (dev
->vol
.migr_state
&&
7295 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7296 /* array state change is blocked due to reshape action
7298 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7299 * - finish the reshape (if last_checkpoint is big and action != reshape)
7300 * - update curr_migr_unit
7302 if (a
->curr_action
== reshape
) {
7303 /* still reshaping, maybe update curr_migr_unit */
7304 goto mark_checkpoint
;
7306 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7307 /* for some reason we aborted the reshape.
7309 * disable automatic metadata rollback
7310 * user action is required to recover process
7313 struct imsm_map
*map2
=
7314 get_imsm_map(dev
, MAP_1
);
7315 dev
->vol
.migr_state
= 0;
7316 set_migr_type(dev
, 0);
7317 dev
->vol
.curr_migr_unit
= 0;
7319 sizeof_imsm_map(map2
));
7320 super
->updates_pending
++;
7323 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7324 unsigned long long array_blocks
;
7328 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7329 if (used_disks
> 0) {
7331 blocks_per_member(map
) *
7333 /* round array size down to closest MB
7335 array_blocks
= (array_blocks
7336 >> SECT_PER_MB_SHIFT
)
7337 << SECT_PER_MB_SHIFT
;
7338 a
->info
.custom_array_size
= array_blocks
;
7339 /* encourage manager to update array
7343 a
->check_reshape
= 1;
7345 /* finalize online capacity expansion/reshape */
7346 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7348 mdi
->disk
.raid_disk
,
7351 imsm_progress_container_reshape(super
);
7356 /* before we activate this array handle any missing disks */
7357 if (consistent
== 2)
7358 handle_missing(super
, dev
);
7360 if (consistent
== 2 &&
7361 (!is_resync_complete(&a
->info
) ||
7362 map_state
!= IMSM_T_STATE_NORMAL
||
7363 dev
->vol
.migr_state
))
7366 if (is_resync_complete(&a
->info
)) {
7367 /* complete intialization / resync,
7368 * recovery and interrupted recovery is completed in
7371 if (is_resyncing(dev
)) {
7372 dprintf("imsm: mark resync done\n");
7373 end_migration(dev
, super
, map_state
);
7374 super
->updates_pending
++;
7375 a
->last_checkpoint
= 0;
7377 } else if ((!is_resyncing(dev
) && !failed
) &&
7378 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7379 /* mark the start of the init process if nothing is failed */
7380 dprintf("imsm: mark resync start\n");
7381 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7382 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7384 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7385 super
->updates_pending
++;
7389 /* skip checkpointing for general migration,
7390 * it is controlled in mdadm
7392 if (is_gen_migration(dev
))
7393 goto skip_mark_checkpoint
;
7395 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7396 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7397 if (blocks_per_unit
) {
7401 units
= a
->last_checkpoint
/ blocks_per_unit
;
7404 /* check that we did not overflow 32-bits, and that
7405 * curr_migr_unit needs updating
7407 if (units32
== units
&&
7409 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7410 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7411 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7412 super
->updates_pending
++;
7416 skip_mark_checkpoint
:
7417 /* mark dirty / clean */
7418 if (dev
->vol
.dirty
!= !consistent
) {
7419 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7424 super
->updates_pending
++;
7430 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7432 int inst
= a
->info
.container_member
;
7433 struct intel_super
*super
= a
->container
->sb
;
7434 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7435 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7436 struct imsm_disk
*disk
;
7438 int recovery_not_finished
= 0;
7443 if (n
> map
->num_members
)
7444 pr_err("imsm: set_disk %d out of range 0..%d\n",
7445 n
, map
->num_members
- 1);
7450 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7452 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7453 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7455 /* check for new failures */
7456 if (state
& DS_FAULTY
) {
7457 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7458 super
->updates_pending
++;
7461 /* check if in_sync */
7462 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7463 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7465 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7466 super
->updates_pending
++;
7469 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7470 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7472 /* check if recovery complete, newly degraded, or failed */
7473 dprintf("imsm: Detected transition to state ");
7474 switch (map_state
) {
7475 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7476 dprintf("normal: ");
7477 if (is_rebuilding(dev
)) {
7478 dprintf_cont("while rebuilding");
7479 /* check if recovery is really finished */
7480 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7481 if (mdi
->recovery_start
!= MaxSector
) {
7482 recovery_not_finished
= 1;
7485 if (recovery_not_finished
) {
7487 dprintf("Rebuild has not finished yet, state not changed");
7488 if (a
->last_checkpoint
< mdi
->recovery_start
) {
7489 a
->last_checkpoint
= mdi
->recovery_start
;
7490 super
->updates_pending
++;
7494 end_migration(dev
, super
, map_state
);
7495 map
= get_imsm_map(dev
, MAP_0
);
7496 map
->failed_disk_num
= ~0;
7497 super
->updates_pending
++;
7498 a
->last_checkpoint
= 0;
7501 if (is_gen_migration(dev
)) {
7502 dprintf_cont("while general migration");
7503 if (a
->last_checkpoint
>= a
->info
.component_size
)
7504 end_migration(dev
, super
, map_state
);
7506 map
->map_state
= map_state
;
7507 map
= get_imsm_map(dev
, MAP_0
);
7508 map
->failed_disk_num
= ~0;
7509 super
->updates_pending
++;
7513 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7514 dprintf_cont("degraded: ");
7515 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
7516 dprintf_cont("mark degraded");
7517 map
->map_state
= map_state
;
7518 super
->updates_pending
++;
7519 a
->last_checkpoint
= 0;
7522 if (is_rebuilding(dev
)) {
7523 dprintf_cont("while rebuilding.");
7524 if (map
->map_state
!= map_state
) {
7525 dprintf_cont(" Map state change");
7526 end_migration(dev
, super
, map_state
);
7527 super
->updates_pending
++;
7531 if (is_gen_migration(dev
)) {
7532 dprintf_cont("while general migration");
7533 if (a
->last_checkpoint
>= a
->info
.component_size
)
7534 end_migration(dev
, super
, map_state
);
7536 map
->map_state
= map_state
;
7537 manage_second_map(super
, dev
);
7539 super
->updates_pending
++;
7542 if (is_initializing(dev
)) {
7543 dprintf_cont("while initialization.");
7544 map
->map_state
= map_state
;
7545 super
->updates_pending
++;
7549 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7550 dprintf_cont("failed: ");
7551 if (is_gen_migration(dev
)) {
7552 dprintf_cont("while general migration");
7553 map
->map_state
= map_state
;
7554 super
->updates_pending
++;
7557 if (map
->map_state
!= map_state
) {
7558 dprintf_cont("mark failed");
7559 end_migration(dev
, super
, map_state
);
7560 super
->updates_pending
++;
7561 a
->last_checkpoint
= 0;
7566 dprintf_cont("state %i\n", map_state
);
7571 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7574 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7575 unsigned long long dsize
;
7576 unsigned long long sectors
;
7578 get_dev_size(fd
, NULL
, &dsize
);
7580 if (mpb_size
> 512) {
7581 /* -1 to account for anchor */
7582 sectors
= mpb_sectors(mpb
) - 1;
7584 /* write the extended mpb to the sectors preceeding the anchor */
7585 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7588 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7593 /* first block is stored on second to last sector of the disk */
7594 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7597 if (write(fd
, buf
, 512) != 512)
7603 static void imsm_sync_metadata(struct supertype
*container
)
7605 struct intel_super
*super
= container
->sb
;
7607 dprintf("sync metadata: %d\n", super
->updates_pending
);
7608 if (!super
->updates_pending
)
7611 write_super_imsm(container
, 0);
7613 super
->updates_pending
= 0;
7616 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7618 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7619 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7622 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7626 if (dl
&& is_failed(&dl
->disk
))
7630 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
7635 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7636 struct active_array
*a
, int activate_new
,
7637 struct mdinfo
*additional_test_list
)
7639 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7640 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7641 struct imsm_super
*mpb
= super
->anchor
;
7642 struct imsm_map
*map
;
7643 unsigned long long pos
;
7648 __u32 array_start
= 0;
7649 __u32 array_end
= 0;
7651 struct mdinfo
*test_list
;
7653 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7654 /* If in this array, skip */
7655 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7656 if (d
->state_fd
>= 0 &&
7657 d
->disk
.major
== dl
->major
&&
7658 d
->disk
.minor
== dl
->minor
) {
7659 dprintf("%x:%x already in array\n",
7660 dl
->major
, dl
->minor
);
7665 test_list
= additional_test_list
;
7667 if (test_list
->disk
.major
== dl
->major
&&
7668 test_list
->disk
.minor
== dl
->minor
) {
7669 dprintf("%x:%x already in additional test list\n",
7670 dl
->major
, dl
->minor
);
7673 test_list
= test_list
->next
;
7678 /* skip in use or failed drives */
7679 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7681 dprintf("%x:%x status (failed: %d index: %d)\n",
7682 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7686 /* skip pure spares when we are looking for partially
7687 * assimilated drives
7689 if (dl
->index
== -1 && !activate_new
)
7692 /* Does this unused device have the requisite free space?
7693 * It needs to be able to cover all member volumes
7695 ex
= get_extents(super
, dl
);
7697 dprintf("cannot get extents\n");
7700 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7701 dev
= get_imsm_dev(super
, i
);
7702 map
= get_imsm_map(dev
, MAP_0
);
7704 /* check if this disk is already a member of
7707 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7713 array_start
= pba_of_lba0(map
);
7714 array_end
= array_start
+
7715 blocks_per_member(map
) - 1;
7718 /* check that we can start at pba_of_lba0 with
7719 * blocks_per_member of space
7721 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7725 pos
= ex
[j
].start
+ ex
[j
].size
;
7727 } while (ex
[j
-1].size
);
7734 if (i
< mpb
->num_raid_devs
) {
7735 dprintf("%x:%x does not have %u to %u available\n",
7736 dl
->major
, dl
->minor
, array_start
, array_end
);
7746 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7748 struct imsm_dev
*dev2
;
7749 struct imsm_map
*map
;
7755 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7757 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7758 if (state
== IMSM_T_STATE_FAILED
) {
7759 map
= get_imsm_map(dev2
, MAP_0
);
7762 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7764 * Check if failed disks are deleted from intel
7765 * disk list or are marked to be deleted
7767 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7768 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7770 * Do not rebuild the array if failed disks
7771 * from failed sub-array are not removed from
7775 is_failed(&idisk
->disk
) &&
7776 (idisk
->action
!= DISK_REMOVE
))
7784 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7785 struct metadata_update
**updates
)
7788 * Find a device with unused free space and use it to replace a
7789 * failed/vacant region in an array. We replace failed regions one a
7790 * array at a time. The result is that a new spare disk will be added
7791 * to the first failed array and after the monitor has finished
7792 * propagating failures the remainder will be consumed.
7794 * FIXME add a capability for mdmon to request spares from another
7798 struct intel_super
*super
= a
->container
->sb
;
7799 int inst
= a
->info
.container_member
;
7800 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7801 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7802 int failed
= a
->info
.array
.raid_disks
;
7803 struct mdinfo
*rv
= NULL
;
7806 struct metadata_update
*mu
;
7808 struct imsm_update_activate_spare
*u
;
7813 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7814 if ((d
->curr_state
& DS_FAULTY
) &&
7816 /* wait for Removal to happen */
7818 if (d
->state_fd
>= 0)
7822 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7823 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7825 if (imsm_reshape_blocks_arrays_changes(super
))
7828 /* Cannot activate another spare if rebuild is in progress already
7830 if (is_rebuilding(dev
)) {
7831 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
7835 if (a
->info
.array
.level
== 4)
7836 /* No repair for takeovered array
7837 * imsm doesn't support raid4
7841 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7842 IMSM_T_STATE_DEGRADED
)
7845 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
7846 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
7851 * If there are any failed disks check state of the other volume.
7852 * Block rebuild if the another one is failed until failed disks
7853 * are removed from container.
7856 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
7857 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7858 /* check if states of the other volumes allow for rebuild */
7859 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7861 allowed
= imsm_rebuild_allowed(a
->container
,
7869 /* For each slot, if it is not working, find a spare */
7870 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7871 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7872 if (d
->disk
.raid_disk
== i
)
7874 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7875 if (d
&& (d
->state_fd
>= 0))
7879 * OK, this device needs recovery. Try to re-add the
7880 * previous occupant of this slot, if this fails see if
7881 * we can continue the assimilation of a spare that was
7882 * partially assimilated, finally try to activate a new
7885 dl
= imsm_readd(super
, i
, a
);
7887 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7889 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7893 /* found a usable disk with enough space */
7894 di
= xcalloc(1, sizeof(*di
));
7896 /* dl->index will be -1 in the case we are activating a
7897 * pristine spare. imsm_process_update() will create a
7898 * new index in this case. Once a disk is found to be
7899 * failed in all member arrays it is kicked from the
7902 di
->disk
.number
= dl
->index
;
7904 /* (ab)use di->devs to store a pointer to the device
7907 di
->devs
= (struct mdinfo
*) dl
;
7909 di
->disk
.raid_disk
= i
;
7910 di
->disk
.major
= dl
->major
;
7911 di
->disk
.minor
= dl
->minor
;
7913 di
->recovery_start
= 0;
7914 di
->data_offset
= pba_of_lba0(map
);
7915 di
->component_size
= a
->info
.component_size
;
7916 di
->container_member
= inst
;
7917 super
->random
= random32();
7921 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7922 i
, di
->data_offset
);
7926 /* No spares found */
7928 /* Now 'rv' has a list of devices to return.
7929 * Create a metadata_update record to update the
7930 * disk_ord_tbl for the array
7932 mu
= xmalloc(sizeof(*mu
));
7933 mu
->buf
= xcalloc(num_spares
,
7934 sizeof(struct imsm_update_activate_spare
));
7936 mu
->space_list
= NULL
;
7937 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7938 mu
->next
= *updates
;
7939 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7941 for (di
= rv
; di
; di
= di
->next
) {
7942 u
->type
= update_activate_spare
;
7943 u
->dl
= (struct dl
*) di
->devs
;
7945 u
->slot
= di
->disk
.raid_disk
;
7956 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7958 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7959 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7960 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7961 struct disk_info
*inf
= get_disk_info(u
);
7962 struct imsm_disk
*disk
;
7966 for (i
= 0; i
< map
->num_members
; i
++) {
7967 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7968 for (j
= 0; j
< new_map
->num_members
; j
++)
7969 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7976 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7980 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7981 if (dl
->major
== major
&& dl
->minor
== minor
)
7986 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7992 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7993 if (dl
->major
== major
&& dl
->minor
== minor
) {
7996 prev
->next
= dl
->next
;
7998 super
->disks
= dl
->next
;
8000 __free_imsm_disk(dl
);
8001 dprintf("removed %x:%x\n", major
, minor
);
8009 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
8011 static int add_remove_disk_update(struct intel_super
*super
)
8013 int check_degraded
= 0;
8016 /* add/remove some spares to/from the metadata/contrainer */
8017 while (super
->disk_mgmt_list
) {
8018 struct dl
*disk_cfg
;
8020 disk_cfg
= super
->disk_mgmt_list
;
8021 super
->disk_mgmt_list
= disk_cfg
->next
;
8022 disk_cfg
->next
= NULL
;
8024 if (disk_cfg
->action
== DISK_ADD
) {
8025 disk_cfg
->next
= super
->disks
;
8026 super
->disks
= disk_cfg
;
8028 dprintf("added %x:%x\n",
8029 disk_cfg
->major
, disk_cfg
->minor
);
8030 } else if (disk_cfg
->action
== DISK_REMOVE
) {
8031 dprintf("Disk remove action processed: %x.%x\n",
8032 disk_cfg
->major
, disk_cfg
->minor
);
8033 disk
= get_disk_super(super
,
8037 /* store action status */
8038 disk
->action
= DISK_REMOVE
;
8039 /* remove spare disks only */
8040 if (disk
->index
== -1) {
8041 remove_disk_super(super
,
8046 /* release allocate disk structure */
8047 __free_imsm_disk(disk_cfg
);
8050 return check_degraded
;
8053 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
8054 struct intel_super
*super
,
8057 struct intel_dev
*id
;
8058 void **tofree
= NULL
;
8061 dprintf("(enter)\n");
8062 if (u
->subdev
< 0 || u
->subdev
> 1) {
8063 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8066 if (space_list
== NULL
|| *space_list
== NULL
) {
8067 dprintf("imsm: Error: Memory is not allocated\n");
8071 for (id
= super
->devlist
; id
; id
= id
->next
) {
8072 if (id
->index
== (unsigned)u
->subdev
) {
8073 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8074 struct imsm_map
*map
;
8075 struct imsm_dev
*new_dev
=
8076 (struct imsm_dev
*)*space_list
;
8077 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8079 struct dl
*new_disk
;
8081 if (new_dev
== NULL
)
8083 *space_list
= **space_list
;
8084 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
8085 map
= get_imsm_map(new_dev
, MAP_0
);
8087 dprintf("imsm: Error: migration in progress");
8091 to_state
= map
->map_state
;
8092 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
8094 /* this should not happen */
8095 if (u
->new_disks
[0] < 0) {
8096 map
->failed_disk_num
=
8097 map
->num_members
- 1;
8098 to_state
= IMSM_T_STATE_DEGRADED
;
8100 to_state
= IMSM_T_STATE_NORMAL
;
8102 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
8103 if (u
->new_level
> -1)
8104 map
->raid_level
= u
->new_level
;
8105 migr_map
= get_imsm_map(new_dev
, MAP_1
);
8106 if ((u
->new_level
== 5) &&
8107 (migr_map
->raid_level
== 0)) {
8108 int ord
= map
->num_members
- 1;
8109 migr_map
->num_members
--;
8110 if (u
->new_disks
[0] < 0)
8111 ord
|= IMSM_ORD_REBUILD
;
8112 set_imsm_ord_tbl_ent(map
,
8113 map
->num_members
- 1,
8117 tofree
= (void **)dev
;
8119 /* update chunk size
8121 if (u
->new_chunksize
> 0)
8122 map
->blocks_per_strip
=
8123 __cpu_to_le16(u
->new_chunksize
* 2);
8127 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
8128 migr_map
->raid_level
== map
->raid_level
)
8131 if (u
->new_disks
[0] >= 0) {
8134 new_disk
= get_disk_super(super
,
8135 major(u
->new_disks
[0]),
8136 minor(u
->new_disks
[0]));
8137 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8138 major(u
->new_disks
[0]),
8139 minor(u
->new_disks
[0]),
8140 new_disk
, new_disk
->index
);
8141 if (new_disk
== NULL
)
8142 goto error_disk_add
;
8144 new_disk
->index
= map
->num_members
- 1;
8145 /* slot to fill in autolayout
8147 new_disk
->raiddisk
= new_disk
->index
;
8148 new_disk
->disk
.status
|= CONFIGURED_DISK
;
8149 new_disk
->disk
.status
&= ~SPARE_DISK
;
8151 goto error_disk_add
;
8154 *tofree
= *space_list
;
8155 /* calculate new size
8157 imsm_set_array_size(new_dev
, -1);
8164 *space_list
= tofree
;
8168 dprintf("Error: imsm: Cannot find disk.\n");
8172 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8173 struct intel_super
*super
)
8175 struct intel_dev
*id
;
8178 dprintf("(enter)\n");
8179 if (u
->subdev
< 0 || u
->subdev
> 1) {
8180 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8184 for (id
= super
->devlist
; id
; id
= id
->next
) {
8185 if (id
->index
== (unsigned)u
->subdev
) {
8186 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8187 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8188 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8189 unsigned long long blocks_per_member
;
8191 /* calculate new size
8193 blocks_per_member
= u
->new_size
/ used_disks
;
8194 dprintf("(size: %llu, blocks per member: %llu)\n",
8195 u
->new_size
, blocks_per_member
);
8196 set_blocks_per_member(map
, blocks_per_member
);
8197 imsm_set_array_size(dev
, u
->new_size
);
8207 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8208 struct intel_super
*super
,
8209 struct active_array
*active_array
)
8211 struct imsm_super
*mpb
= super
->anchor
;
8212 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8213 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8214 struct imsm_map
*migr_map
;
8215 struct active_array
*a
;
8216 struct imsm_disk
*disk
;
8223 int second_map_created
= 0;
8225 for (; u
; u
= u
->next
) {
8226 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8231 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8236 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
8241 /* count failures (excluding rebuilds and the victim)
8242 * to determine map[0] state
8245 for (i
= 0; i
< map
->num_members
; i
++) {
8248 disk
= get_imsm_disk(super
,
8249 get_imsm_disk_idx(dev
, i
, MAP_X
));
8250 if (!disk
|| is_failed(disk
))
8254 /* adding a pristine spare, assign a new index */
8255 if (dl
->index
< 0) {
8256 dl
->index
= super
->anchor
->num_disks
;
8257 super
->anchor
->num_disks
++;
8260 disk
->status
|= CONFIGURED_DISK
;
8261 disk
->status
&= ~SPARE_DISK
;
8264 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8265 if (!second_map_created
) {
8266 second_map_created
= 1;
8267 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8268 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8270 map
->map_state
= to_state
;
8271 migr_map
= get_imsm_map(dev
, MAP_1
);
8272 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8273 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8274 dl
->index
| IMSM_ORD_REBUILD
);
8276 /* update the family_num to mark a new container
8277 * generation, being careful to record the existing
8278 * family_num in orig_family_num to clean up after
8279 * earlier mdadm versions that neglected to set it.
8281 if (mpb
->orig_family_num
== 0)
8282 mpb
->orig_family_num
= mpb
->family_num
;
8283 mpb
->family_num
+= super
->random
;
8285 /* count arrays using the victim in the metadata */
8287 for (a
= active_array
; a
; a
= a
->next
) {
8288 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8289 map
= get_imsm_map(dev
, MAP_0
);
8291 if (get_imsm_disk_slot(map
, victim
) >= 0)
8295 /* delete the victim if it is no longer being
8301 /* We know that 'manager' isn't touching anything,
8302 * so it is safe to delete
8304 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8305 if ((*dlp
)->index
== victim
)
8308 /* victim may be on the missing list */
8310 for (dlp
= &super
->missing
; *dlp
;
8311 dlp
= &(*dlp
)->next
)
8312 if ((*dlp
)->index
== victim
)
8314 imsm_delete(super
, dlp
, victim
);
8321 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8322 struct intel_super
*super
,
8325 struct dl
*new_disk
;
8326 struct intel_dev
*id
;
8328 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8329 int disk_count
= u
->old_raid_disks
;
8330 void **tofree
= NULL
;
8331 int devices_to_reshape
= 1;
8332 struct imsm_super
*mpb
= super
->anchor
;
8334 unsigned int dev_id
;
8336 dprintf("(enter)\n");
8338 /* enable spares to use in array */
8339 for (i
= 0; i
< delta_disks
; i
++) {
8340 new_disk
= get_disk_super(super
,
8341 major(u
->new_disks
[i
]),
8342 minor(u
->new_disks
[i
]));
8343 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8344 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8345 new_disk
, new_disk
->index
);
8346 if (new_disk
== NULL
||
8347 (new_disk
->index
>= 0 &&
8348 new_disk
->index
< u
->old_raid_disks
))
8349 goto update_reshape_exit
;
8350 new_disk
->index
= disk_count
++;
8351 /* slot to fill in autolayout
8353 new_disk
->raiddisk
= new_disk
->index
;
8354 new_disk
->disk
.status
|=
8356 new_disk
->disk
.status
&= ~SPARE_DISK
;
8359 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8360 mpb
->num_raid_devs
);
8361 /* manage changes in volume
8363 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8364 void **sp
= *space_list
;
8365 struct imsm_dev
*newdev
;
8366 struct imsm_map
*newmap
, *oldmap
;
8368 for (id
= super
->devlist
; id
; id
= id
->next
) {
8369 if (id
->index
== dev_id
)
8378 /* Copy the dev, but not (all of) the map */
8379 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8380 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8381 newmap
= get_imsm_map(newdev
, MAP_0
);
8382 /* Copy the current map */
8383 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8384 /* update one device only
8386 if (devices_to_reshape
) {
8387 dprintf("imsm: modifying subdev: %i\n",
8389 devices_to_reshape
--;
8390 newdev
->vol
.migr_state
= 1;
8391 newdev
->vol
.curr_migr_unit
= 0;
8392 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8393 newmap
->num_members
= u
->new_raid_disks
;
8394 for (i
= 0; i
< delta_disks
; i
++) {
8395 set_imsm_ord_tbl_ent(newmap
,
8396 u
->old_raid_disks
+ i
,
8397 u
->old_raid_disks
+ i
);
8399 /* New map is correct, now need to save old map
8401 newmap
= get_imsm_map(newdev
, MAP_1
);
8402 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8404 imsm_set_array_size(newdev
, -1);
8407 sp
= (void **)id
->dev
;
8412 /* Clear migration record */
8413 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8416 *space_list
= tofree
;
8419 update_reshape_exit
:
8424 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8425 struct intel_super
*super
,
8428 struct imsm_dev
*dev
= NULL
;
8429 struct intel_dev
*dv
;
8430 struct imsm_dev
*dev_new
;
8431 struct imsm_map
*map
;
8435 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8436 if (dv
->index
== (unsigned int)u
->subarray
) {
8444 map
= get_imsm_map(dev
, MAP_0
);
8446 if (u
->direction
== R10_TO_R0
) {
8447 /* Number of failed disks must be half of initial disk number */
8448 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8449 (map
->num_members
/ 2))
8452 /* iterate through devices to mark removed disks as spare */
8453 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8454 if (dm
->disk
.status
& FAILED_DISK
) {
8455 int idx
= dm
->index
;
8456 /* update indexes on the disk list */
8457 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8458 the index values will end up being correct.... NB */
8459 for (du
= super
->disks
; du
; du
= du
->next
)
8460 if (du
->index
> idx
)
8462 /* mark as spare disk */
8467 map
->num_members
= map
->num_members
/ 2;
8468 map
->map_state
= IMSM_T_STATE_NORMAL
;
8469 map
->num_domains
= 1;
8470 map
->raid_level
= 0;
8471 map
->failed_disk_num
= -1;
8474 if (u
->direction
== R0_TO_R10
) {
8476 /* update slots in current disk list */
8477 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8481 /* create new *missing* disks */
8482 for (i
= 0; i
< map
->num_members
; i
++) {
8483 space
= *space_list
;
8486 *space_list
= *space
;
8488 memcpy(du
, super
->disks
, sizeof(*du
));
8492 du
->index
= (i
* 2) + 1;
8493 sprintf((char *)du
->disk
.serial
,
8494 " MISSING_%d", du
->index
);
8495 sprintf((char *)du
->serial
,
8496 "MISSING_%d", du
->index
);
8497 du
->next
= super
->missing
;
8498 super
->missing
= du
;
8500 /* create new dev and map */
8501 space
= *space_list
;
8504 *space_list
= *space
;
8505 dev_new
= (void *)space
;
8506 memcpy(dev_new
, dev
, sizeof(*dev
));
8507 /* update new map */
8508 map
= get_imsm_map(dev_new
, MAP_0
);
8509 map
->num_members
= map
->num_members
* 2;
8510 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8511 map
->num_domains
= 2;
8512 map
->raid_level
= 1;
8513 /* replace dev<->dev_new */
8516 /* update disk order table */
8517 for (du
= super
->disks
; du
; du
= du
->next
)
8519 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8520 for (du
= super
->missing
; du
; du
= du
->next
)
8521 if (du
->index
>= 0) {
8522 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8523 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8529 static void imsm_process_update(struct supertype
*st
,
8530 struct metadata_update
*update
)
8533 * crack open the metadata_update envelope to find the update record
8534 * update can be one of:
8535 * update_reshape_container_disks - all the arrays in the container
8536 * are being reshaped to have more devices. We need to mark
8537 * the arrays for general migration and convert selected spares
8538 * into active devices.
8539 * update_activate_spare - a spare device has replaced a failed
8540 * device in an array, update the disk_ord_tbl. If this disk is
8541 * present in all member arrays then also clear the SPARE_DISK
8543 * update_create_array
8545 * update_rename_array
8546 * update_add_remove_disk
8548 struct intel_super
*super
= st
->sb
;
8549 struct imsm_super
*mpb
;
8550 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8552 /* update requires a larger buf but the allocation failed */
8553 if (super
->next_len
&& !super
->next_buf
) {
8554 super
->next_len
= 0;
8558 if (super
->next_buf
) {
8559 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8561 super
->len
= super
->next_len
;
8562 super
->buf
= super
->next_buf
;
8564 super
->next_len
= 0;
8565 super
->next_buf
= NULL
;
8568 mpb
= super
->anchor
;
8571 case update_general_migration_checkpoint
: {
8572 struct intel_dev
*id
;
8573 struct imsm_update_general_migration_checkpoint
*u
=
8574 (void *)update
->buf
;
8576 dprintf("called for update_general_migration_checkpoint\n");
8578 /* find device under general migration */
8579 for (id
= super
->devlist
; id
; id
= id
->next
) {
8580 if (is_gen_migration(id
->dev
)) {
8581 id
->dev
->vol
.curr_migr_unit
=
8582 __cpu_to_le32(u
->curr_migr_unit
);
8583 super
->updates_pending
++;
8588 case update_takeover
: {
8589 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8590 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8591 imsm_update_version_info(super
);
8592 super
->updates_pending
++;
8597 case update_reshape_container_disks
: {
8598 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8599 if (apply_reshape_container_disks_update(
8600 u
, super
, &update
->space_list
))
8601 super
->updates_pending
++;
8604 case update_reshape_migration
: {
8605 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8606 if (apply_reshape_migration_update(
8607 u
, super
, &update
->space_list
))
8608 super
->updates_pending
++;
8611 case update_size_change
: {
8612 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8613 if (apply_size_change_update(u
, super
))
8614 super
->updates_pending
++;
8617 case update_activate_spare
: {
8618 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8619 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8620 super
->updates_pending
++;
8623 case update_create_array
: {
8624 /* someone wants to create a new array, we need to be aware of
8625 * a few races/collisions:
8626 * 1/ 'Create' called by two separate instances of mdadm
8627 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8628 * devices that have since been assimilated via
8630 * In the event this update can not be carried out mdadm will
8631 * (FIX ME) notice that its update did not take hold.
8633 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8634 struct intel_dev
*dv
;
8635 struct imsm_dev
*dev
;
8636 struct imsm_map
*map
, *new_map
;
8637 unsigned long long start
, end
;
8638 unsigned long long new_start
, new_end
;
8640 struct disk_info
*inf
;
8643 /* handle racing creates: first come first serve */
8644 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8645 dprintf("subarray %d already defined\n", u
->dev_idx
);
8649 /* check update is next in sequence */
8650 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8651 dprintf("can not create array %d expected index %d\n",
8652 u
->dev_idx
, mpb
->num_raid_devs
);
8656 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8657 new_start
= pba_of_lba0(new_map
);
8658 new_end
= new_start
+ blocks_per_member(new_map
);
8659 inf
= get_disk_info(u
);
8661 /* handle activate_spare versus create race:
8662 * check to make sure that overlapping arrays do not include
8665 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8666 dev
= get_imsm_dev(super
, i
);
8667 map
= get_imsm_map(dev
, MAP_0
);
8668 start
= pba_of_lba0(map
);
8669 end
= start
+ blocks_per_member(map
);
8670 if ((new_start
>= start
&& new_start
<= end
) ||
8671 (start
>= new_start
&& start
<= new_end
))
8676 if (disks_overlap(super
, i
, u
)) {
8677 dprintf("arrays overlap\n");
8682 /* check that prepare update was successful */
8683 if (!update
->space
) {
8684 dprintf("prepare update failed\n");
8688 /* check that all disks are still active before committing
8689 * changes. FIXME: could we instead handle this by creating a
8690 * degraded array? That's probably not what the user expects,
8691 * so better to drop this update on the floor.
8693 for (i
= 0; i
< new_map
->num_members
; i
++) {
8694 dl
= serial_to_dl(inf
[i
].serial
, super
);
8696 dprintf("disk disappeared\n");
8701 super
->updates_pending
++;
8703 /* convert spares to members and fixup ord_tbl */
8704 for (i
= 0; i
< new_map
->num_members
; i
++) {
8705 dl
= serial_to_dl(inf
[i
].serial
, super
);
8706 if (dl
->index
== -1) {
8707 dl
->index
= mpb
->num_disks
;
8709 dl
->disk
.status
|= CONFIGURED_DISK
;
8710 dl
->disk
.status
&= ~SPARE_DISK
;
8712 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8717 update
->space
= NULL
;
8718 imsm_copy_dev(dev
, &u
->dev
);
8719 dv
->index
= u
->dev_idx
;
8720 dv
->next
= super
->devlist
;
8721 super
->devlist
= dv
;
8722 mpb
->num_raid_devs
++;
8724 imsm_update_version_info(super
);
8727 /* mdmon knows how to release update->space, but not
8728 * ((struct intel_dev *) update->space)->dev
8730 if (update
->space
) {
8736 case update_kill_array
: {
8737 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8738 int victim
= u
->dev_idx
;
8739 struct active_array
*a
;
8740 struct intel_dev
**dp
;
8741 struct imsm_dev
*dev
;
8743 /* sanity check that we are not affecting the uuid of
8744 * active arrays, or deleting an active array
8746 * FIXME when immutable ids are available, but note that
8747 * we'll also need to fixup the invalidated/active
8748 * subarray indexes in mdstat
8750 for (a
= st
->arrays
; a
; a
= a
->next
)
8751 if (a
->info
.container_member
>= victim
)
8753 /* by definition if mdmon is running at least one array
8754 * is active in the container, so checking
8755 * mpb->num_raid_devs is just extra paranoia
8757 dev
= get_imsm_dev(super
, victim
);
8758 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8759 dprintf("failed to delete subarray-%d\n", victim
);
8763 for (dp
= &super
->devlist
; *dp
;)
8764 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8767 if ((*dp
)->index
> (unsigned)victim
)
8771 mpb
->num_raid_devs
--;
8772 super
->updates_pending
++;
8775 case update_rename_array
: {
8776 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8777 char name
[MAX_RAID_SERIAL_LEN
+1];
8778 int target
= u
->dev_idx
;
8779 struct active_array
*a
;
8780 struct imsm_dev
*dev
;
8782 /* sanity check that we are not affecting the uuid of
8785 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8786 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8787 for (a
= st
->arrays
; a
; a
= a
->next
)
8788 if (a
->info
.container_member
== target
)
8790 dev
= get_imsm_dev(super
, u
->dev_idx
);
8791 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8792 dprintf("failed to rename subarray-%d\n", target
);
8796 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8797 super
->updates_pending
++;
8800 case update_add_remove_disk
: {
8801 /* we may be able to repair some arrays if disks are
8802 * being added, check the status of add_remove_disk
8803 * if discs has been added.
8805 if (add_remove_disk_update(super
)) {
8806 struct active_array
*a
;
8808 super
->updates_pending
++;
8809 for (a
= st
->arrays
; a
; a
= a
->next
)
8810 a
->check_degraded
= 1;
8815 pr_err("error: unsuported process update type:(type: %d)\n", type
);
8819 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8821 static int imsm_prepare_update(struct supertype
*st
,
8822 struct metadata_update
*update
)
8825 * Allocate space to hold new disk entries, raid-device entries or a new
8826 * mpb if necessary. The manager synchronously waits for updates to
8827 * complete in the monitor, so new mpb buffers allocated here can be
8828 * integrated by the monitor thread without worrying about live pointers
8829 * in the manager thread.
8831 enum imsm_update_type type
;
8832 struct intel_super
*super
= st
->sb
;
8833 struct imsm_super
*mpb
= super
->anchor
;
8837 if (update
->len
< (int)sizeof(type
))
8840 type
= *(enum imsm_update_type
*) update
->buf
;
8843 case update_general_migration_checkpoint
:
8844 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
8846 dprintf("called for update_general_migration_checkpoint\n");
8848 case update_takeover
: {
8849 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8850 if (update
->len
< (int)sizeof(*u
))
8852 if (u
->direction
== R0_TO_R10
) {
8853 void **tail
= (void **)&update
->space_list
;
8854 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8855 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8856 int num_members
= map
->num_members
;
8859 /* allocate memory for added disks */
8860 for (i
= 0; i
< num_members
; i
++) {
8861 size
= sizeof(struct dl
);
8862 space
= xmalloc(size
);
8867 /* allocate memory for new device */
8868 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8869 (num_members
* sizeof(__u32
));
8870 space
= xmalloc(size
);
8874 len
= disks_to_mpb_size(num_members
* 2);
8879 case update_reshape_container_disks
: {
8880 /* Every raid device in the container is about to
8881 * gain some more devices, and we will enter a
8883 * So each 'imsm_map' will be bigger, and the imsm_vol
8884 * will now hold 2 of them.
8885 * Thus we need new 'struct imsm_dev' allocations sized
8886 * as sizeof_imsm_dev but with more devices in both maps.
8888 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8889 struct intel_dev
*dl
;
8890 void **space_tail
= (void**)&update
->space_list
;
8892 if (update
->len
< (int)sizeof(*u
))
8895 dprintf("for update_reshape\n");
8897 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8898 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8900 if (u
->new_raid_disks
> u
->old_raid_disks
)
8901 size
+= sizeof(__u32
)*2*
8902 (u
->new_raid_disks
- u
->old_raid_disks
);
8909 len
= disks_to_mpb_size(u
->new_raid_disks
);
8910 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8913 case update_reshape_migration
: {
8914 /* for migration level 0->5 we need to add disks
8915 * so the same as for container operation we will copy
8916 * device to the bigger location.
8917 * in memory prepared device and new disk area are prepared
8918 * for usage in process update
8920 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8921 struct intel_dev
*id
;
8922 void **space_tail
= (void **)&update
->space_list
;
8925 int current_level
= -1;
8927 if (update
->len
< (int)sizeof(*u
))
8930 dprintf("for update_reshape\n");
8932 /* add space for bigger array in update
8934 for (id
= super
->devlist
; id
; id
= id
->next
) {
8935 if (id
->index
== (unsigned)u
->subdev
) {
8936 size
= sizeof_imsm_dev(id
->dev
, 1);
8937 if (u
->new_raid_disks
> u
->old_raid_disks
)
8938 size
+= sizeof(__u32
)*2*
8939 (u
->new_raid_disks
- u
->old_raid_disks
);
8947 if (update
->space_list
== NULL
)
8950 /* add space for disk in update
8952 size
= sizeof(struct dl
);
8958 /* add spare device to update
8960 for (id
= super
->devlist
; id
; id
= id
->next
)
8961 if (id
->index
== (unsigned)u
->subdev
) {
8962 struct imsm_dev
*dev
;
8963 struct imsm_map
*map
;
8965 dev
= get_imsm_dev(super
, u
->subdev
);
8966 map
= get_imsm_map(dev
, MAP_0
);
8967 current_level
= map
->raid_level
;
8970 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
8971 struct mdinfo
*spares
;
8973 spares
= get_spares_for_grow(st
);
8981 makedev(dev
->disk
.major
,
8983 dl
= get_disk_super(super
,
8986 dl
->index
= u
->old_raid_disks
;
8992 len
= disks_to_mpb_size(u
->new_raid_disks
);
8993 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8996 case update_size_change
: {
8997 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
9001 case update_activate_spare
: {
9002 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
9006 case update_create_array
: {
9007 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9008 struct intel_dev
*dv
;
9009 struct imsm_dev
*dev
= &u
->dev
;
9010 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9012 struct disk_info
*inf
;
9016 if (update
->len
< (int)sizeof(*u
))
9019 inf
= get_disk_info(u
);
9020 len
= sizeof_imsm_dev(dev
, 1);
9021 /* allocate a new super->devlist entry */
9022 dv
= xmalloc(sizeof(*dv
));
9023 dv
->dev
= xmalloc(len
);
9026 /* count how many spares will be converted to members */
9027 for (i
= 0; i
< map
->num_members
; i
++) {
9028 dl
= serial_to_dl(inf
[i
].serial
, super
);
9030 /* hmm maybe it failed?, nothing we can do about
9035 if (count_memberships(dl
, super
) == 0)
9038 len
+= activate
* sizeof(struct imsm_disk
);
9041 case update_kill_array
: {
9042 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
9046 case update_rename_array
: {
9047 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
9051 case update_add_remove_disk
:
9052 /* no update->len needed */
9058 /* check if we need a larger metadata buffer */
9059 if (super
->next_buf
)
9060 buf_len
= super
->next_len
;
9062 buf_len
= super
->len
;
9064 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
9065 /* ok we need a larger buf than what is currently allocated
9066 * if this allocation fails process_update will notice that
9067 * ->next_len is set and ->next_buf is NULL
9069 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
9070 if (super
->next_buf
)
9071 free(super
->next_buf
);
9073 super
->next_len
= buf_len
;
9074 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
9075 memset(super
->next_buf
, 0, buf_len
);
9077 super
->next_buf
= NULL
;
9082 /* must be called while manager is quiesced */
9083 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
9085 struct imsm_super
*mpb
= super
->anchor
;
9087 struct imsm_dev
*dev
;
9088 struct imsm_map
*map
;
9089 int i
, j
, num_members
;
9092 dprintf("deleting device[%d] from imsm_super\n", index
);
9094 /* shift all indexes down one */
9095 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
9096 if (iter
->index
> (int)index
)
9098 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
9099 if (iter
->index
> (int)index
)
9102 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9103 dev
= get_imsm_dev(super
, i
);
9104 map
= get_imsm_map(dev
, MAP_0
);
9105 num_members
= map
->num_members
;
9106 for (j
= 0; j
< num_members
; j
++) {
9107 /* update ord entries being careful not to propagate
9108 * ord-flags to the first map
9110 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
9112 if (ord_to_idx(ord
) <= index
)
9115 map
= get_imsm_map(dev
, MAP_0
);
9116 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
9117 map
= get_imsm_map(dev
, MAP_1
);
9119 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
9124 super
->updates_pending
++;
9126 struct dl
*dl
= *dlp
;
9128 *dlp
= (*dlp
)->next
;
9129 __free_imsm_disk(dl
);
9132 #endif /* MDASSEMBLE */
9134 static void close_targets(int *targets
, int new_disks
)
9141 for (i
= 0; i
< new_disks
; i
++) {
9142 if (targets
[i
] >= 0) {
9149 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
9150 struct intel_super
*super
,
9151 struct imsm_dev
*dev
)
9157 struct imsm_map
*map
;
9160 ret_val
= raid_disks
/2;
9161 /* check map if all disks pairs not failed
9164 map
= get_imsm_map(dev
, MAP_0
);
9165 for (i
= 0; i
< ret_val
; i
++) {
9166 int degradation
= 0;
9167 if (get_imsm_disk(super
, i
) == NULL
)
9169 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9171 if (degradation
== 2)
9174 map
= get_imsm_map(dev
, MAP_1
);
9175 /* if there is no second map
9176 * result can be returned
9180 /* check degradation in second map
9182 for (i
= 0; i
< ret_val
; i
++) {
9183 int degradation
= 0;
9184 if (get_imsm_disk(super
, i
) == NULL
)
9186 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9188 if (degradation
== 2)
9202 /*******************************************************************************
9203 * Function: open_backup_targets
9204 * Description: Function opens file descriptors for all devices given in
9207 * info : general array info
9208 * raid_disks : number of disks
9209 * raid_fds : table of device's file descriptors
9210 * super : intel super for raid10 degradation check
9211 * dev : intel device for raid10 degradation check
9215 ******************************************************************************/
9216 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9217 struct intel_super
*super
, struct imsm_dev
*dev
)
9223 for (i
= 0; i
< raid_disks
; i
++)
9226 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9229 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9230 dprintf("disk is faulty!!\n");
9234 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
9237 dn
= map_dev(sd
->disk
.major
,
9239 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9240 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9241 pr_err("cannot open component\n");
9246 /* check if maximum array degradation level is not exceeded
9248 if ((raid_disks
- opened
) >
9249 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
9251 pr_err("Not enough disks can be opened.\n");
9252 close_targets(raid_fds
, raid_disks
);
9258 /*******************************************************************************
9259 * Function: validate_container_imsm
9260 * Description: This routine validates container after assemble,
9261 * eg. if devices in container are under the same controller.
9264 * info : linked list with info about devices used in array
9268 ******************************************************************************/
9269 int validate_container_imsm(struct mdinfo
*info
)
9271 if (check_env("IMSM_NO_PLATFORM"))
9274 struct sys_dev
*idev
;
9275 struct sys_dev
*hba
= NULL
;
9276 struct sys_dev
*intel_devices
= find_intel_devices();
9277 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
9280 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9281 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9290 pr_err("WARNING - Cannot detect HBA for device %s!\n",
9291 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
9295 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
9298 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
9299 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
9301 struct sys_dev
*hba2
= NULL
;
9302 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9303 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9311 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
9312 get_orom_by_device_id(hba2
->dev_id
);
9314 if (hba2
&& hba
->type
!= hba2
->type
) {
9315 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
9316 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
9320 if (orom
!= orom2
) {
9321 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
9322 " This operation is not supported and can lead to data loss.\n");
9327 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
9328 " This operation is not supported and can lead to data loss.\n");
9336 /*******************************************************************************
9337 * Function: init_migr_record_imsm
9338 * Description: Function inits imsm migration record
9340 * super : imsm internal array info
9341 * dev : device under migration
9342 * info : general array info to find the smallest device
9345 ******************************************************************************/
9346 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9347 struct mdinfo
*info
)
9349 struct intel_super
*super
= st
->sb
;
9350 struct migr_record
*migr_rec
= super
->migr_rec
;
9352 unsigned long long dsize
, dev_sectors
;
9353 long long unsigned min_dev_sectors
= -1LLU;
9357 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9358 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9359 unsigned long long num_migr_units
;
9360 unsigned long long array_blocks
;
9362 memset(migr_rec
, 0, sizeof(struct migr_record
));
9363 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9365 /* only ascending reshape supported now */
9366 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9368 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9369 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9370 migr_rec
->dest_depth_per_unit
*=
9371 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9372 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9373 migr_rec
->blocks_per_unit
=
9374 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9375 migr_rec
->dest_depth_per_unit
=
9376 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9377 array_blocks
= info
->component_size
* new_data_disks
;
9379 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9381 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9383 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9385 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9386 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9388 /* Find the smallest dev */
9389 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9390 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9391 fd
= dev_open(nm
, O_RDONLY
);
9394 get_dev_size(fd
, NULL
, &dsize
);
9395 dev_sectors
= dsize
/ 512;
9396 if (dev_sectors
< min_dev_sectors
)
9397 min_dev_sectors
= dev_sectors
;
9400 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9401 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9403 write_imsm_migr_rec(st
);
9408 /*******************************************************************************
9409 * Function: save_backup_imsm
9410 * Description: Function saves critical data stripes to Migration Copy Area
9411 * and updates the current migration unit status.
9412 * Use restore_stripes() to form a destination stripe,
9413 * and to write it to the Copy Area.
9415 * st : supertype information
9416 * dev : imsm device that backup is saved for
9417 * info : general array info
9418 * buf : input buffer
9419 * length : length of data to backup (blocks_per_unit)
9423 ******************************************************************************/
9424 int save_backup_imsm(struct supertype
*st
,
9425 struct imsm_dev
*dev
,
9426 struct mdinfo
*info
,
9431 struct intel_super
*super
= st
->sb
;
9432 unsigned long long *target_offsets
;
9435 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9436 int new_disks
= map_dest
->num_members
;
9437 int dest_layout
= 0;
9439 unsigned long long start
;
9440 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9442 targets
= xmalloc(new_disks
* sizeof(int));
9444 for (i
= 0; i
< new_disks
; i
++)
9447 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
9449 start
= info
->reshape_progress
* 512;
9450 for (i
= 0; i
< new_disks
; i
++) {
9451 target_offsets
[i
] = (unsigned long long)
9452 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9453 /* move back copy area adderss, it will be moved forward
9454 * in restore_stripes() using start input variable
9456 target_offsets
[i
] -= start
/data_disks
;
9459 if (open_backup_targets(info
, new_disks
, targets
,
9463 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9464 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9466 if (restore_stripes(targets
, /* list of dest devices */
9467 target_offsets
, /* migration record offsets */
9470 map_dest
->raid_level
,
9472 -1, /* source backup file descriptor */
9473 0, /* input buf offset
9474 * always 0 buf is already offseted */
9478 pr_err("Error restoring stripes\n");
9486 close_targets(targets
, new_disks
);
9489 free(target_offsets
);
9494 /*******************************************************************************
9495 * Function: save_checkpoint_imsm
9496 * Description: Function called for current unit status update
9497 * in the migration record. It writes it to disk.
9499 * super : imsm internal array info
9500 * info : general array info
9504 * 2: failure, means no valid migration record
9505 * / no general migration in progress /
9506 ******************************************************************************/
9507 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9509 struct intel_super
*super
= st
->sb
;
9510 unsigned long long blocks_per_unit
;
9511 unsigned long long curr_migr_unit
;
9513 if (load_imsm_migr_rec(super
, info
) != 0) {
9514 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
9518 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9519 if (blocks_per_unit
== 0) {
9520 dprintf("imsm: no migration in progress.\n");
9523 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9524 /* check if array is alligned to copy area
9525 * if it is not alligned, add one to current migration unit value
9526 * this can happend on array reshape finish only
9528 if (info
->reshape_progress
% blocks_per_unit
)
9531 super
->migr_rec
->curr_migr_unit
=
9532 __cpu_to_le32(curr_migr_unit
);
9533 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9534 super
->migr_rec
->dest_1st_member_lba
=
9535 __cpu_to_le32(curr_migr_unit
*
9536 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9537 if (write_imsm_migr_rec(st
) < 0) {
9538 dprintf("imsm: Cannot write migration record outside backup area\n");
9545 /*******************************************************************************
9546 * Function: recover_backup_imsm
9547 * Description: Function recovers critical data from the Migration Copy Area
9548 * while assembling an array.
9550 * super : imsm internal array info
9551 * info : general array info
9553 * 0 : success (or there is no data to recover)
9555 ******************************************************************************/
9556 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9558 struct intel_super
*super
= st
->sb
;
9559 struct migr_record
*migr_rec
= super
->migr_rec
;
9560 struct imsm_map
*map_dest
;
9561 struct intel_dev
*id
= NULL
;
9562 unsigned long long read_offset
;
9563 unsigned long long write_offset
;
9565 int *targets
= NULL
;
9566 int new_disks
, i
, err
;
9569 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9570 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9572 int skipped_disks
= 0;
9574 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9578 /* recover data only during assemblation */
9579 if (strncmp(buffer
, "inactive", 8) != 0)
9581 /* no data to recover */
9582 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9584 if (curr_migr_unit
>= num_migr_units
)
9587 /* find device during reshape */
9588 for (id
= super
->devlist
; id
; id
= id
->next
)
9589 if (is_gen_migration(id
->dev
))
9594 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9595 new_disks
= map_dest
->num_members
;
9597 read_offset
= (unsigned long long)
9598 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9600 write_offset
= ((unsigned long long)
9601 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9602 pba_of_lba0(map_dest
)) * 512;
9604 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9605 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9607 targets
= xcalloc(new_disks
, sizeof(int));
9609 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9610 pr_err("Cannot open some devices belonging to array.\n");
9614 for (i
= 0; i
< new_disks
; i
++) {
9615 if (targets
[i
] < 0) {
9619 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9620 pr_err("Cannot seek to block: %s\n",
9625 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9626 pr_err("Cannot read copy area block: %s\n",
9631 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9632 pr_err("Cannot seek to block: %s\n",
9637 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9638 pr_err("Cannot restore block: %s\n",
9645 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9649 pr_err("Cannot restore data from backup. Too many failed disks\n");
9653 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9654 /* ignore error == 2, this can mean end of reshape here
9656 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
9662 for (i
= 0; i
< new_disks
; i
++)
9671 static char disk_by_path
[] = "/dev/disk/by-path/";
9673 static const char *imsm_get_disk_controller_domain(const char *path
)
9675 char disk_path
[PATH_MAX
];
9679 strcpy(disk_path
, disk_by_path
);
9680 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9681 if (stat(disk_path
, &st
) == 0) {
9682 struct sys_dev
* hba
;
9685 path
= devt_to_devpath(st
.st_rdev
);
9688 hba
= find_disk_attached_hba(-1, path
);
9689 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9691 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9695 dprintf("path: %s hba: %s attached: %s\n",
9696 path
, (hba
) ? hba
->path
: "NULL", drv
);
9702 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
9704 static char devnm
[32];
9705 char subdev_name
[20];
9706 struct mdstat_ent
*mdstat
;
9708 sprintf(subdev_name
, "%d", subdev
);
9709 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9713 strcpy(devnm
, mdstat
->devnm
);
9714 free_mdstat(mdstat
);
9718 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9719 struct geo_params
*geo
,
9720 int *old_raid_disks
,
9723 /* currently we only support increasing the number of devices
9724 * for a container. This increases the number of device for each
9725 * member array. They must all be RAID0 or RAID5.
9728 struct mdinfo
*info
, *member
;
9729 int devices_that_can_grow
= 0;
9731 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
9733 if (geo
->size
> 0 ||
9734 geo
->level
!= UnSet
||
9735 geo
->layout
!= UnSet
||
9736 geo
->chunksize
!= 0 ||
9737 geo
->raid_disks
== UnSet
) {
9738 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
9742 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9743 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
9747 info
= container_content_imsm(st
, NULL
);
9748 for (member
= info
; member
; member
= member
->next
) {
9751 dprintf("imsm: checking device_num: %i\n",
9752 member
->container_member
);
9754 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9755 /* we work on container for Online Capacity Expansion
9756 * only so raid_disks has to grow
9758 dprintf("imsm: for container operation raid disks increase is required\n");
9762 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
9763 /* we cannot use this container with other raid level
9765 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
9769 /* check for platform support
9770 * for this raid level configuration
9772 struct intel_super
*super
= st
->sb
;
9773 if (!is_raid_level_supported(super
->orom
,
9774 member
->array
.level
,
9776 dprintf("platform does not support raid%d with %d disk%s\n",
9779 geo
->raid_disks
> 1 ? "s" : "");
9782 /* check if component size is aligned to chunk size
9784 if (info
->component_size
%
9785 (info
->array
.chunk_size
/512)) {
9786 dprintf("Component size is not aligned to chunk size\n");
9791 if (*old_raid_disks
&&
9792 info
->array
.raid_disks
!= *old_raid_disks
)
9794 *old_raid_disks
= info
->array
.raid_disks
;
9796 /* All raid5 and raid0 volumes in container
9797 * have to be ready for Online Capacity Expansion
9798 * so they need to be assembled. We have already
9799 * checked that no recovery etc is happening.
9801 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
9802 st
->container_devnm
);
9803 if (result
== NULL
) {
9804 dprintf("imsm: cannot find array\n");
9807 devices_that_can_grow
++;
9810 if (!member
&& devices_that_can_grow
)
9814 dprintf("Container operation allowed\n");
9816 dprintf("Error: %i\n", ret_val
);
9821 /* Function: get_spares_for_grow
9822 * Description: Allocates memory and creates list of spare devices
9823 * avaliable in container. Checks if spare drive size is acceptable.
9824 * Parameters: Pointer to the supertype structure
9825 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9828 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9830 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9831 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9834 /******************************************************************************
9835 * function: imsm_create_metadata_update_for_reshape
9836 * Function creates update for whole IMSM container.
9838 ******************************************************************************/
9839 static int imsm_create_metadata_update_for_reshape(
9840 struct supertype
*st
,
9841 struct geo_params
*geo
,
9843 struct imsm_update_reshape
**updatep
)
9845 struct intel_super
*super
= st
->sb
;
9846 struct imsm_super
*mpb
= super
->anchor
;
9847 int update_memory_size
;
9848 struct imsm_update_reshape
*u
;
9849 struct mdinfo
*spares
;
9854 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
9856 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9858 /* size of all update data without anchor */
9859 update_memory_size
= sizeof(struct imsm_update_reshape
);
9861 /* now add space for spare disks that we need to add. */
9862 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9864 u
= xcalloc(1, update_memory_size
);
9865 u
->type
= update_reshape_container_disks
;
9866 u
->old_raid_disks
= old_raid_disks
;
9867 u
->new_raid_disks
= geo
->raid_disks
;
9869 /* now get spare disks list
9871 spares
= get_spares_for_grow(st
);
9874 || delta_disks
> spares
->array
.spare_disks
) {
9875 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
9880 /* we have got spares
9881 * update disk list in imsm_disk list table in anchor
9883 dprintf("imsm: %i spares are available.\n\n",
9884 spares
->array
.spare_disks
);
9887 for (i
= 0; i
< delta_disks
; i
++) {
9892 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9894 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9895 dl
->index
= mpb
->num_disks
;
9905 dprintf("imsm: reshape update preparation :");
9906 if (i
== delta_disks
) {
9907 dprintf_cont(" OK\n");
9909 return update_memory_size
;
9912 dprintf_cont(" Error\n");
9917 /******************************************************************************
9918 * function: imsm_create_metadata_update_for_size_change()
9919 * Creates update for IMSM array for array size change.
9921 ******************************************************************************/
9922 static int imsm_create_metadata_update_for_size_change(
9923 struct supertype
*st
,
9924 struct geo_params
*geo
,
9925 struct imsm_update_size_change
**updatep
)
9927 struct intel_super
*super
= st
->sb
;
9928 int update_memory_size
;
9929 struct imsm_update_size_change
*u
;
9931 dprintf("(enter) New size = %llu\n", geo
->size
);
9933 /* size of all update data without anchor */
9934 update_memory_size
= sizeof(struct imsm_update_size_change
);
9936 u
= xcalloc(1, update_memory_size
);
9937 u
->type
= update_size_change
;
9938 u
->subdev
= super
->current_vol
;
9939 u
->new_size
= geo
->size
;
9941 dprintf("imsm: reshape update preparation : OK\n");
9944 return update_memory_size
;
9947 /******************************************************************************
9948 * function: imsm_create_metadata_update_for_migration()
9949 * Creates update for IMSM array.
9951 ******************************************************************************/
9952 static int imsm_create_metadata_update_for_migration(
9953 struct supertype
*st
,
9954 struct geo_params
*geo
,
9955 struct imsm_update_reshape_migration
**updatep
)
9957 struct intel_super
*super
= st
->sb
;
9958 int update_memory_size
;
9959 struct imsm_update_reshape_migration
*u
;
9960 struct imsm_dev
*dev
;
9961 int previous_level
= -1;
9963 dprintf("(enter) New Level = %i\n", geo
->level
);
9965 /* size of all update data without anchor */
9966 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9968 u
= xcalloc(1, update_memory_size
);
9969 u
->type
= update_reshape_migration
;
9970 u
->subdev
= super
->current_vol
;
9971 u
->new_level
= geo
->level
;
9972 u
->new_layout
= geo
->layout
;
9973 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9974 u
->new_disks
[0] = -1;
9975 u
->new_chunksize
= -1;
9977 dev
= get_imsm_dev(super
, u
->subdev
);
9979 struct imsm_map
*map
;
9981 map
= get_imsm_map(dev
, MAP_0
);
9983 int current_chunk_size
=
9984 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9986 if (geo
->chunksize
!= current_chunk_size
) {
9987 u
->new_chunksize
= geo
->chunksize
/ 1024;
9988 dprintf("imsm: chunk size change from %i to %i\n",
9989 current_chunk_size
, u
->new_chunksize
);
9991 previous_level
= map
->raid_level
;
9994 if (geo
->level
== 5 && previous_level
== 0) {
9995 struct mdinfo
*spares
= NULL
;
9997 u
->new_raid_disks
++;
9998 spares
= get_spares_for_grow(st
);
9999 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
10001 sysfs_free(spares
);
10002 update_memory_size
= 0;
10003 dprintf("error: cannot get spare device for requested migration");
10006 sysfs_free(spares
);
10008 dprintf("imsm: reshape update preparation : OK\n");
10011 return update_memory_size
;
10014 static void imsm_update_metadata_locally(struct supertype
*st
,
10015 void *buf
, int len
)
10017 struct metadata_update mu
;
10022 mu
.space_list
= NULL
;
10024 if (imsm_prepare_update(st
, &mu
))
10025 imsm_process_update(st
, &mu
);
10027 while (mu
.space_list
) {
10028 void **space
= mu
.space_list
;
10029 mu
.space_list
= *space
;
10034 /***************************************************************************
10035 * Function: imsm_analyze_change
10036 * Description: Function analyze change for single volume
10037 * and validate if transition is supported
10038 * Parameters: Geometry parameters, supertype structure,
10039 * metadata change direction (apply/rollback)
10040 * Returns: Operation type code on success, -1 if fail
10041 ****************************************************************************/
10042 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
10043 struct geo_params
*geo
,
10046 struct mdinfo info
;
10048 int check_devs
= 0;
10050 /* number of added/removed disks in operation result */
10051 int devNumChange
= 0;
10052 /* imsm compatible layout value for array geometry verification */
10053 int imsm_layout
= -1;
10055 struct imsm_dev
*dev
;
10056 struct intel_super
*super
;
10057 unsigned long long current_size
;
10058 unsigned long long free_size
;
10059 unsigned long long max_size
;
10062 getinfo_super_imsm_volume(st
, &info
, NULL
);
10063 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
10064 geo
->level
!= UnSet
) {
10065 switch (info
.array
.level
) {
10067 if (geo
->level
== 5) {
10068 change
= CH_MIGRATION
;
10069 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
10070 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
10072 goto analyse_change_exit
;
10074 imsm_layout
= geo
->layout
;
10076 devNumChange
= 1; /* parity disk added */
10077 } else if (geo
->level
== 10) {
10078 change
= CH_TAKEOVER
;
10080 devNumChange
= 2; /* two mirrors added */
10081 imsm_layout
= 0x102; /* imsm supported layout */
10086 if (geo
->level
== 0) {
10087 change
= CH_TAKEOVER
;
10089 devNumChange
= -(geo
->raid_disks
/2);
10090 imsm_layout
= 0; /* imsm raid0 layout */
10094 if (change
== -1) {
10095 pr_err("Error. Level Migration from %d to %d not supported!\n",
10096 info
.array
.level
, geo
->level
);
10097 goto analyse_change_exit
;
10100 geo
->level
= info
.array
.level
;
10102 if (geo
->layout
!= info
.array
.layout
&&
10103 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
10104 change
= CH_MIGRATION
;
10105 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
10106 geo
->layout
== 5) {
10107 /* reshape 5 -> 4 */
10108 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
10109 geo
->layout
== 0) {
10110 /* reshape 4 -> 5 */
10114 pr_err("Error. Layout Migration from %d to %d not supported!\n",
10115 info
.array
.layout
, geo
->layout
);
10117 goto analyse_change_exit
;
10120 geo
->layout
= info
.array
.layout
;
10121 if (imsm_layout
== -1)
10122 imsm_layout
= info
.array
.layout
;
10125 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
10126 geo
->chunksize
!= info
.array
.chunk_size
) {
10127 if (info
.array
.level
== 10) {
10128 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
10130 goto analyse_change_exit
;
10132 change
= CH_MIGRATION
;
10134 geo
->chunksize
= info
.array
.chunk_size
;
10137 chunk
= geo
->chunksize
/ 1024;
10140 dev
= get_imsm_dev(super
, super
->current_vol
);
10141 data_disks
= imsm_num_data_members(dev
, MAP_0
);
10142 /* compute current size per disk member
10144 current_size
= info
.custom_array_size
/ data_disks
;
10146 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
10147 /* align component size
10149 geo
->size
= imsm_component_size_aligment_check(
10150 get_imsm_raid_level(dev
->vol
.map
),
10153 if (geo
->size
== 0) {
10154 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
10156 goto analyse_change_exit
;
10160 if (current_size
!= geo
->size
&& geo
->size
> 0) {
10161 if (change
!= -1) {
10162 pr_err("Error. Size change should be the only one at a time.\n");
10164 goto analyse_change_exit
;
10166 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
10167 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
10168 super
->current_vol
, st
->devnm
);
10169 goto analyse_change_exit
;
10171 /* check the maximum available size
10173 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
10174 0, chunk
, &free_size
);
10176 /* Cannot find maximum available space
10180 max_size
= free_size
+ current_size
;
10181 /* align component size
10183 max_size
= imsm_component_size_aligment_check(
10184 get_imsm_raid_level(dev
->vol
.map
),
10188 if (geo
->size
== MAX_SIZE
) {
10189 /* requested size change to the maximum available size
10191 if (max_size
== 0) {
10192 pr_err("Error. Cannot find maximum available space.\n");
10194 goto analyse_change_exit
;
10196 geo
->size
= max_size
;
10199 if (direction
== ROLLBACK_METADATA_CHANGES
) {
10200 /* accept size for rollback only
10203 /* round size due to metadata compatibility
10205 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
10206 << SECT_PER_MB_SHIFT
;
10207 dprintf("Prepare update for size change to %llu\n",
10209 if (current_size
>= geo
->size
) {
10210 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
10211 current_size
, geo
->size
);
10212 goto analyse_change_exit
;
10214 if (max_size
&& geo
->size
> max_size
) {
10215 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
10216 max_size
, geo
->size
);
10217 goto analyse_change_exit
;
10220 geo
->size
*= data_disks
;
10221 geo
->raid_disks
= dev
->vol
.map
->num_members
;
10222 change
= CH_ARRAY_SIZE
;
10224 if (!validate_geometry_imsm(st
,
10227 geo
->raid_disks
+ devNumChange
,
10229 geo
->size
, INVALID_SECTORS
,
10234 struct intel_super
*super
= st
->sb
;
10235 struct imsm_super
*mpb
= super
->anchor
;
10237 if (mpb
->num_raid_devs
> 1) {
10238 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
10244 analyse_change_exit
:
10245 if (direction
== ROLLBACK_METADATA_CHANGES
&&
10246 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
10247 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
10253 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10255 struct intel_super
*super
= st
->sb
;
10256 struct imsm_update_takeover
*u
;
10258 u
= xmalloc(sizeof(struct imsm_update_takeover
));
10260 u
->type
= update_takeover
;
10261 u
->subarray
= super
->current_vol
;
10263 /* 10->0 transition */
10264 if (geo
->level
== 0)
10265 u
->direction
= R10_TO_R0
;
10267 /* 0->10 transition */
10268 if (geo
->level
== 10)
10269 u
->direction
= R0_TO_R10
;
10271 /* update metadata locally */
10272 imsm_update_metadata_locally(st
, u
,
10273 sizeof(struct imsm_update_takeover
));
10274 /* and possibly remotely */
10275 if (st
->update_tail
)
10276 append_metadata_update(st
, u
,
10277 sizeof(struct imsm_update_takeover
));
10284 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
10286 int layout
, int chunksize
, int raid_disks
,
10287 int delta_disks
, char *backup
, char *dev
,
10288 int direction
, int verbose
)
10291 struct geo_params geo
;
10293 dprintf("(enter)\n");
10295 memset(&geo
, 0, sizeof(struct geo_params
));
10297 geo
.dev_name
= dev
;
10298 strcpy(geo
.devnm
, st
->devnm
);
10301 geo
.layout
= layout
;
10302 geo
.chunksize
= chunksize
;
10303 geo
.raid_disks
= raid_disks
;
10304 if (delta_disks
!= UnSet
)
10305 geo
.raid_disks
+= delta_disks
;
10307 dprintf("for level : %i\n", geo
.level
);
10308 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
10310 if (experimental() == 0)
10313 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
10314 /* On container level we can only increase number of devices. */
10315 dprintf("imsm: info: Container operation\n");
10316 int old_raid_disks
= 0;
10318 if (imsm_reshape_is_allowed_on_container(
10319 st
, &geo
, &old_raid_disks
, direction
)) {
10320 struct imsm_update_reshape
*u
= NULL
;
10323 len
= imsm_create_metadata_update_for_reshape(
10324 st
, &geo
, old_raid_disks
, &u
);
10327 dprintf("imsm: Cannot prepare update\n");
10328 goto exit_imsm_reshape_super
;
10332 /* update metadata locally */
10333 imsm_update_metadata_locally(st
, u
, len
);
10334 /* and possibly remotely */
10335 if (st
->update_tail
)
10336 append_metadata_update(st
, u
, len
);
10341 pr_err("(imsm) Operation is not allowed on this container\n");
10344 /* On volume level we support following operations
10345 * - takeover: raid10 -> raid0; raid0 -> raid10
10346 * - chunk size migration
10347 * - migration: raid5 -> raid0; raid0 -> raid5
10349 struct intel_super
*super
= st
->sb
;
10350 struct intel_dev
*dev
= super
->devlist
;
10352 dprintf("imsm: info: Volume operation\n");
10353 /* find requested device */
10356 imsm_find_array_devnm_by_subdev(
10357 dev
->index
, st
->container_devnm
);
10358 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
10363 pr_err("Cannot find %s (%s) subarray\n",
10364 geo
.dev_name
, geo
.devnm
);
10365 goto exit_imsm_reshape_super
;
10367 super
->current_vol
= dev
->index
;
10368 change
= imsm_analyze_change(st
, &geo
, direction
);
10371 ret_val
= imsm_takeover(st
, &geo
);
10373 case CH_MIGRATION
: {
10374 struct imsm_update_reshape_migration
*u
= NULL
;
10376 imsm_create_metadata_update_for_migration(
10379 dprintf("imsm: Cannot prepare update\n");
10383 /* update metadata locally */
10384 imsm_update_metadata_locally(st
, u
, len
);
10385 /* and possibly remotely */
10386 if (st
->update_tail
)
10387 append_metadata_update(st
, u
, len
);
10392 case CH_ARRAY_SIZE
: {
10393 struct imsm_update_size_change
*u
= NULL
;
10395 imsm_create_metadata_update_for_size_change(
10398 dprintf("imsm: Cannot prepare update\n");
10402 /* update metadata locally */
10403 imsm_update_metadata_locally(st
, u
, len
);
10404 /* and possibly remotely */
10405 if (st
->update_tail
)
10406 append_metadata_update(st
, u
, len
);
10416 exit_imsm_reshape_super
:
10417 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10421 #define COMPLETED_OK 0
10422 #define COMPLETED_NONE 1
10423 #define COMPLETED_DELAYED 2
10425 static int read_completed(int fd
, unsigned long long *val
)
10430 ret
= sysfs_fd_get_str(fd
, buf
, 50);
10434 ret
= COMPLETED_OK
;
10435 if (strncmp(buf
, "none", 4) == 0) {
10436 ret
= COMPLETED_NONE
;
10437 } else if (strncmp(buf
, "delayed", 7) == 0) {
10438 ret
= COMPLETED_DELAYED
;
10441 *val
= strtoull(buf
, &ep
, 0);
10442 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
10448 /*******************************************************************************
10449 * Function: wait_for_reshape_imsm
10450 * Description: Function writes new sync_max value and waits until
10451 * reshape process reach new position
10453 * sra : general array info
10454 * ndata : number of disks in new array's layout
10457 * 1 : there is no reshape in progress,
10459 ******************************************************************************/
10460 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10462 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
10464 unsigned long long completed
;
10465 /* to_complete : new sync_max position */
10466 unsigned long long to_complete
= sra
->reshape_progress
;
10467 unsigned long long position_to_set
= to_complete
/ ndata
;
10470 dprintf("cannot open reshape_position\n");
10475 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10477 dprintf("cannot read reshape_position (no reshape in progres)\n");
10486 if (completed
> position_to_set
) {
10487 dprintf("wrong next position to set %llu (%llu)\n",
10488 to_complete
, position_to_set
);
10492 dprintf("Position set: %llu\n", position_to_set
);
10493 if (sysfs_set_num(sra
, NULL
, "sync_max",
10494 position_to_set
) != 0) {
10495 dprintf("cannot set reshape position to %llu\n",
10504 int timeout
= 3000;
10506 sysfs_wait(fd
, &timeout
);
10507 if (sysfs_get_str(sra
, NULL
, "sync_action",
10509 strncmp(action
, "reshape", 7) != 0) {
10510 if (strncmp(action
, "idle", 4) == 0)
10516 rc
= read_completed(fd
, &completed
);
10518 dprintf("cannot read reshape_position (in loop)\n");
10521 } else if (rc
== COMPLETED_NONE
)
10523 } while (completed
< position_to_set
);
10529 /*******************************************************************************
10530 * Function: check_degradation_change
10531 * Description: Check that array hasn't become failed.
10533 * info : for sysfs access
10534 * sources : source disks descriptors
10535 * degraded: previous degradation level
10537 * degradation level
10538 ******************************************************************************/
10539 int check_degradation_change(struct mdinfo
*info
,
10543 unsigned long long new_degraded
;
10546 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10547 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
10548 /* check each device to ensure it is still working */
10551 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10552 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10554 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10556 if (sysfs_get_str(info
,
10557 sd
, "state", sbuf
, 20) < 0 ||
10558 strstr(sbuf
, "faulty") ||
10559 strstr(sbuf
, "in_sync") == NULL
) {
10560 /* this device is dead */
10561 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10562 if (sd
->disk
.raid_disk
>= 0 &&
10563 sources
[sd
->disk
.raid_disk
] >= 0) {
10565 sd
->disk
.raid_disk
]);
10566 sources
[sd
->disk
.raid_disk
] =
10575 return new_degraded
;
10578 /*******************************************************************************
10579 * Function: imsm_manage_reshape
10580 * Description: Function finds array under reshape and it manages reshape
10581 * process. It creates stripes backups (if required) and sets
10584 * afd : Backup handle (nattive) - not used
10585 * sra : general array info
10586 * reshape : reshape parameters - not used
10587 * st : supertype structure
10588 * blocks : size of critical section [blocks]
10589 * fds : table of source device descriptor
10590 * offsets : start of array (offest per devices)
10592 * destfd : table of destination device descriptor
10593 * destoffsets : table of destination offsets (per device)
10595 * 1 : success, reshape is done
10597 ******************************************************************************/
10598 static int imsm_manage_reshape(
10599 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10600 struct supertype
*st
, unsigned long backup_blocks
,
10601 int *fds
, unsigned long long *offsets
,
10602 int dests
, int *destfd
, unsigned long long *destoffsets
)
10605 struct intel_super
*super
= st
->sb
;
10606 struct intel_dev
*dv
;
10607 struct imsm_dev
*dev
= NULL
;
10608 struct imsm_map
*map_src
;
10609 int migr_vol_qan
= 0;
10610 int ndata
, odata
; /* [bytes] */
10611 int chunk
; /* [bytes] */
10612 struct migr_record
*migr_rec
;
10614 unsigned int buf_size
; /* [bytes] */
10615 unsigned long long max_position
; /* array size [bytes] */
10616 unsigned long long next_step
; /* [blocks]/[bytes] */
10617 unsigned long long old_data_stripe_length
;
10618 unsigned long long start_src
; /* [bytes] */
10619 unsigned long long start
; /* [bytes] */
10620 unsigned long long start_buf_shift
; /* [bytes] */
10622 int source_layout
= 0;
10627 if (!fds
|| !offsets
)
10630 /* Find volume during the reshape */
10631 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10632 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10633 && dv
->dev
->vol
.migr_state
== 1) {
10638 /* Only one volume can migrate at the same time */
10639 if (migr_vol_qan
!= 1) {
10640 pr_err("%s", migr_vol_qan
?
10641 "Number of migrating volumes greater than 1\n" :
10642 "There is no volume during migrationg\n");
10646 map_src
= get_imsm_map(dev
, MAP_1
);
10647 if (map_src
== NULL
)
10650 ndata
= imsm_num_data_members(dev
, MAP_0
);
10651 odata
= imsm_num_data_members(dev
, MAP_1
);
10653 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10654 old_data_stripe_length
= odata
* chunk
;
10656 migr_rec
= super
->migr_rec
;
10658 /* initialize migration record for start condition */
10659 if (sra
->reshape_progress
== 0)
10660 init_migr_record_imsm(st
, dev
, sra
);
10662 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10663 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
10666 /* Save checkpoint to update migration record for current
10667 * reshape position (in md). It can be farther than current
10668 * reshape position in metadata.
10670 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10671 /* ignore error == 2, this can mean end of reshape here
10673 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
10678 /* size for data */
10679 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10680 /* extend buffer size for parity disk */
10681 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10682 /* add space for stripe aligment */
10683 buf_size
+= old_data_stripe_length
;
10684 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10685 dprintf("imsm: Cannot allocate checpoint buffer\n");
10689 max_position
= sra
->component_size
* ndata
;
10690 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10692 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10693 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10694 /* current reshape position [blocks] */
10695 unsigned long long current_position
=
10696 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10697 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10698 unsigned long long border
;
10700 /* Check that array hasn't become failed.
10702 degraded
= check_degradation_change(sra
, fds
, degraded
);
10703 if (degraded
> 1) {
10704 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
10708 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10710 if ((current_position
+ next_step
) > max_position
)
10711 next_step
= max_position
- current_position
;
10713 start
= current_position
* 512;
10715 /* align reading start to old geometry */
10716 start_buf_shift
= start
% old_data_stripe_length
;
10717 start_src
= start
- start_buf_shift
;
10719 border
= (start_src
/ odata
) - (start
/ ndata
);
10721 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10722 /* save critical stripes to buf
10723 * start - start address of current unit
10724 * to backup [bytes]
10725 * start_src - start address of current unit
10726 * to backup alligned to source array
10729 unsigned long long next_step_filler
;
10730 unsigned long long copy_length
= next_step
* 512;
10732 /* allign copy area length to stripe in old geometry */
10733 next_step_filler
= ((copy_length
+ start_buf_shift
)
10734 % old_data_stripe_length
);
10735 if (next_step_filler
)
10736 next_step_filler
= (old_data_stripe_length
10737 - next_step_filler
);
10738 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
10739 start
, start_src
, copy_length
,
10740 start_buf_shift
, next_step_filler
);
10742 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10743 chunk
, map_src
->raid_level
,
10744 source_layout
, 0, NULL
, start_src
,
10746 next_step_filler
+ start_buf_shift
,
10748 dprintf("imsm: Cannot save stripes to buffer\n");
10751 /* Convert data to destination format and store it
10752 * in backup general migration area
10754 if (save_backup_imsm(st
, dev
, sra
,
10755 buf
+ start_buf_shift
, copy_length
)) {
10756 dprintf("imsm: Cannot save stripes to target devices\n");
10759 if (save_checkpoint_imsm(st
, sra
,
10760 UNIT_SRC_IN_CP_AREA
)) {
10761 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
10765 /* set next step to use whole border area */
10766 border
/= next_step
;
10768 next_step
*= border
;
10770 /* When data backed up, checkpoint stored,
10771 * kick the kernel to reshape unit of data
10773 next_step
= next_step
+ sra
->reshape_progress
;
10774 /* limit next step to array max position */
10775 if (next_step
> max_position
)
10776 next_step
= max_position
;
10777 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10778 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10779 sra
->reshape_progress
= next_step
;
10781 /* wait until reshape finish */
10782 if (wait_for_reshape_imsm(sra
, ndata
)) {
10783 dprintf("wait_for_reshape_imsm returned error!\n");
10789 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10790 /* ignore error == 2, this can mean end of reshape here
10792 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
10798 /* clear migr_rec on disks after successful migration */
10801 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
10802 for (d
= super
->disks
; d
; d
= d
->next
) {
10803 if (d
->index
< 0 || is_failed(&d
->disk
))
10805 unsigned long long dsize
;
10807 get_dev_size(d
->fd
, NULL
, &dsize
);
10808 if (lseek64(d
->fd
, dsize
- MIGR_REC_POSITION
,
10810 if (write(d
->fd
, super
->migr_rec_buf
,
10811 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
10812 perror("Write migr_rec failed");
10816 /* return '1' if done */
10820 /* See Grow.c: abort_reshape() for further explanation */
10821 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
10822 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
10823 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
10828 #endif /* MDASSEMBLE */
10830 struct superswitch super_imsm
= {
10832 .examine_super
= examine_super_imsm
,
10833 .brief_examine_super
= brief_examine_super_imsm
,
10834 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10835 .export_examine_super
= export_examine_super_imsm
,
10836 .detail_super
= detail_super_imsm
,
10837 .brief_detail_super
= brief_detail_super_imsm
,
10838 .write_init_super
= write_init_super_imsm
,
10839 .validate_geometry
= validate_geometry_imsm
,
10840 .add_to_super
= add_to_super_imsm
,
10841 .remove_from_super
= remove_from_super_imsm
,
10842 .detail_platform
= detail_platform_imsm
,
10843 .export_detail_platform
= export_detail_platform_imsm
,
10844 .kill_subarray
= kill_subarray_imsm
,
10845 .update_subarray
= update_subarray_imsm
,
10846 .load_container
= load_container_imsm
,
10847 .default_geometry
= default_geometry_imsm
,
10848 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10849 .reshape_super
= imsm_reshape_super
,
10850 .manage_reshape
= imsm_manage_reshape
,
10851 .recover_backup
= recover_backup_imsm
,
10852 .copy_metadata
= copy_metadata_imsm
,
10854 .match_home
= match_home_imsm
,
10855 .uuid_from_super
= uuid_from_super_imsm
,
10856 .getinfo_super
= getinfo_super_imsm
,
10857 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10858 .update_super
= update_super_imsm
,
10860 .avail_size
= avail_size_imsm
,
10861 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10863 .compare_super
= compare_super_imsm
,
10865 .load_super
= load_super_imsm
,
10866 .init_super
= init_super_imsm
,
10867 .store_super
= store_super_imsm
,
10868 .free_super
= free_super_imsm
,
10869 .match_metadata_desc
= match_metadata_desc_imsm
,
10870 .container_content
= container_content_imsm
,
10871 .validate_container
= validate_container_imsm
,
10878 .open_new
= imsm_open_new
,
10879 .set_array_state
= imsm_set_array_state
,
10880 .set_disk
= imsm_set_disk
,
10881 .sync_metadata
= imsm_sync_metadata
,
10882 .activate_spare
= imsm_activate_spare
,
10883 .process_update
= imsm_process_update
,
10884 .prepare_update
= imsm_prepare_update
,
10885 #endif /* MDASSEMBLE */