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__
));
240 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
243 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
245 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
247 #define MIGR_REC_BUF_SIZE 512 /* size of migr_record i/o buffer */
248 #define MIGR_REC_POSITION 512 /* migr_record position offset on disk,
249 * MIGR_REC_BUF_SIZE <= MIGR_REC_POSITION
253 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
254 * be recovered using srcMap */
255 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
256 * already been migrated and must
257 * be recovered from checkpoint area */
259 __u32 rec_status
; /* Status used to determine how to restart
260 * migration in case it aborts
262 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
263 __u32 family_num
; /* Family number of MPB
264 * containing the RaidDev
265 * that is migrating */
266 __u32 ascending_migr
; /* True if migrating in increasing
268 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
269 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
271 * advances per unit-of-operation */
272 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
273 __u32 dest_1st_member_lba
; /* First member lba on first
274 * stripe of destination */
275 __u32 num_migr_units
; /* Total num migration units-of-op */
276 __u32 post_migr_vol_cap
; /* Size of volume after
277 * migration completes */
278 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
279 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
280 * migration ckpt record was read from
281 * (for recovered migrations) */
282 } __attribute__ ((__packed__
));
287 * 2: metadata does not match
295 struct md_list
*next
;
298 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
300 static __u8
migr_type(struct imsm_dev
*dev
)
302 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
303 dev
->status
& DEV_VERIFY_AND_FIX
)
306 return dev
->vol
.migr_type
;
309 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
311 /* for compatibility with older oroms convert MIGR_REPAIR, into
312 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
314 if (migr_type
== MIGR_REPAIR
) {
315 dev
->vol
.migr_type
= MIGR_VERIFY
;
316 dev
->status
|= DEV_VERIFY_AND_FIX
;
318 dev
->vol
.migr_type
= migr_type
;
319 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
323 static unsigned int sector_count(__u32 bytes
)
325 return ROUND_UP(bytes
, 512) / 512;
328 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
330 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
334 struct imsm_dev
*dev
;
335 struct intel_dev
*next
;
340 enum sys_dev_type type
;
343 struct intel_hba
*next
;
350 /* internal representation of IMSM metadata */
353 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
354 struct imsm_super
*anchor
; /* immovable parameters */
357 void *migr_rec_buf
; /* buffer for I/O operations */
358 struct migr_record
*migr_rec
; /* migration record */
360 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
361 array, it indicates that mdmon is allowed to clean migration
363 size_t len
; /* size of the 'buf' allocation */
364 void *next_buf
; /* for realloc'ing buf from the manager */
366 int updates_pending
; /* count of pending updates for mdmon */
367 int current_vol
; /* index of raid device undergoing creation */
368 unsigned long long create_offset
; /* common start for 'current_vol' */
369 __u32 random
; /* random data for seeding new family numbers */
370 struct intel_dev
*devlist
;
374 __u8 serial
[MAX_RAID_SERIAL_LEN
];
377 struct imsm_disk disk
;
380 struct extent
*e
; /* for determining freespace @ create */
381 int raiddisk
; /* slot to fill in autolayout */
383 } *disks
, *current_disk
;
384 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
386 struct dl
*missing
; /* disks removed while we weren't looking */
387 struct bbm_log
*bbm_log
;
388 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
389 const struct imsm_orom
*orom
; /* platform firmware support */
390 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
394 struct imsm_disk disk
;
395 #define IMSM_UNKNOWN_OWNER (-1)
397 struct intel_disk
*next
;
401 unsigned long long start
, size
;
404 /* definitions of reshape process types */
405 enum imsm_reshape_type
{
411 /* definition of messages passed to imsm_process_update */
412 enum imsm_update_type
{
413 update_activate_spare
,
417 update_add_remove_disk
,
418 update_reshape_container_disks
,
419 update_reshape_migration
,
421 update_general_migration_checkpoint
,
425 struct imsm_update_activate_spare
{
426 enum imsm_update_type type
;
430 struct imsm_update_activate_spare
*next
;
436 unsigned long long size
;
443 enum takeover_direction
{
447 struct imsm_update_takeover
{
448 enum imsm_update_type type
;
450 enum takeover_direction direction
;
453 struct imsm_update_reshape
{
454 enum imsm_update_type type
;
458 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
461 struct imsm_update_reshape_migration
{
462 enum imsm_update_type type
;
465 /* fields for array migration changes
472 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
475 struct imsm_update_size_change
{
476 enum imsm_update_type type
;
481 struct imsm_update_general_migration_checkpoint
{
482 enum imsm_update_type type
;
483 __u32 curr_migr_unit
;
487 __u8 serial
[MAX_RAID_SERIAL_LEN
];
490 struct imsm_update_create_array
{
491 enum imsm_update_type type
;
496 struct imsm_update_kill_array
{
497 enum imsm_update_type type
;
501 struct imsm_update_rename_array
{
502 enum imsm_update_type type
;
503 __u8 name
[MAX_RAID_SERIAL_LEN
];
507 struct imsm_update_add_remove_disk
{
508 enum imsm_update_type type
;
512 static const char *_sys_dev_type
[] = {
513 [SYS_DEV_UNKNOWN
] = "Unknown",
514 [SYS_DEV_SAS
] = "SAS",
515 [SYS_DEV_SATA
] = "SATA"
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
=NULL
;
542 for (result
= hba
; result
; result
= result
->next
) {
543 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
549 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
551 struct intel_hba
*hba
;
553 /* check if disk attached to Intel HBA */
554 hba
= find_intel_hba(super
->hba
, device
);
557 /* Check if HBA is already attached to super */
558 if (super
->hba
== NULL
) {
559 super
->hba
= alloc_intel_hba(device
);
564 /* Intel metadata allows for all disks attached to the same type HBA.
565 * Do not sypport odf HBA types mixing
567 if (device
->type
!= hba
->type
)
573 hba
->next
= alloc_intel_hba(device
);
577 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
579 struct sys_dev
*list
, *elem
, *prev
;
582 if ((list
= find_intel_devices()) == NULL
)
586 disk_path
= (char *) devname
;
588 disk_path
= diskfd_to_devpath(fd
);
595 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
596 if (path_attached_to_hba(disk_path
, elem
->path
)) {
600 prev
->next
= elem
->next
;
602 if (disk_path
!= devname
)
608 if (disk_path
!= devname
)
616 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
619 static struct supertype
*match_metadata_desc_imsm(char *arg
)
621 struct supertype
*st
;
623 if (strcmp(arg
, "imsm") != 0 &&
624 strcmp(arg
, "default") != 0
628 st
= xcalloc(1, sizeof(*st
));
629 st
->container_dev
= NoMdDev
;
630 st
->ss
= &super_imsm
;
631 st
->max_devs
= IMSM_MAX_DEVICES
;
632 st
->minor_version
= 0;
638 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
640 return &mpb
->sig
[MPB_SIG_LEN
];
644 /* retrieve a disk directly from the anchor when the anchor is known to be
645 * up-to-date, currently only at load time
647 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
649 if (index
>= mpb
->num_disks
)
651 return &mpb
->disk
[index
];
654 /* retrieve the disk description based on a index of the disk
657 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
661 for (d
= super
->disks
; d
; d
= d
->next
)
662 if (d
->index
== index
)
667 /* retrieve a disk from the parsed metadata */
668 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
672 dl
= get_imsm_dl_disk(super
, index
);
679 /* generate a checksum directly from the anchor when the anchor is known to be
680 * up-to-date, currently only at load or write_super after coalescing
682 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
684 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
685 __u32
*p
= (__u32
*) mpb
;
689 sum
+= __le32_to_cpu(*p
);
693 return sum
- __le32_to_cpu(mpb
->check_sum
);
696 static size_t sizeof_imsm_map(struct imsm_map
*map
)
698 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
701 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
703 /* A device can have 2 maps if it is in the middle of a migration.
705 * MAP_0 - we return the first map
706 * MAP_1 - we return the second map if it exists, else NULL
707 * MAP_X - we return the second map if it exists, else the first
709 struct imsm_map
*map
= &dev
->vol
.map
[0];
710 struct imsm_map
*map2
= NULL
;
712 if (dev
->vol
.migr_state
)
713 map2
= (void *)map
+ sizeof_imsm_map(map
);
715 switch (second_map
) {
732 /* return the size of the device.
733 * migr_state increases the returned size if map[0] were to be duplicated
735 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
737 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
738 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
740 /* migrating means an additional map */
741 if (dev
->vol
.migr_state
)
742 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
744 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
750 /* retrieve disk serial number list from a metadata update */
751 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
754 struct disk_info
*inf
;
756 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
757 sizeof_imsm_dev(&update
->dev
, 0);
763 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
769 if (index
>= mpb
->num_raid_devs
)
772 /* devices start after all disks */
773 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
775 for (i
= 0; i
<= index
; i
++)
777 return _mpb
+ offset
;
779 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
784 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
786 struct intel_dev
*dv
;
788 if (index
>= super
->anchor
->num_raid_devs
)
790 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
791 if (dv
->index
== index
)
798 * == MAP_0 get first map
799 * == MAP_1 get second map
800 * == MAP_X than get map according to the current migr_state
802 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
806 struct imsm_map
*map
;
808 map
= get_imsm_map(dev
, second_map
);
810 /* top byte identifies disk under rebuild */
811 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
814 #define ord_to_idx(ord) (((ord) << 8) >> 8)
815 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
817 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
819 return ord_to_idx(ord
);
822 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
824 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
827 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
832 for (slot
= 0; slot
< map
->num_members
; slot
++) {
833 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
834 if (ord_to_idx(ord
) == idx
)
841 static int get_imsm_raid_level(struct imsm_map
*map
)
843 if (map
->raid_level
== 1) {
844 if (map
->num_members
== 2)
850 return map
->raid_level
;
853 static int cmp_extent(const void *av
, const void *bv
)
855 const struct extent
*a
= av
;
856 const struct extent
*b
= bv
;
857 if (a
->start
< b
->start
)
859 if (a
->start
> b
->start
)
864 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
869 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
870 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
871 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
873 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
880 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
882 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
884 if (lo
== 0 || hi
== 0)
886 *lo
= __le32_to_cpu((unsigned)n
);
887 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
891 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
893 return (unsigned long long)__le32_to_cpu(lo
) |
894 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
897 static unsigned long long total_blocks(struct imsm_disk
*disk
)
901 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
904 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
908 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
911 static unsigned long long blocks_per_member(struct imsm_map
*map
)
915 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
919 static unsigned long long num_data_stripes(struct imsm_map
*map
)
923 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
926 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
928 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
932 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
934 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
937 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
939 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
942 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
944 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
947 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
949 /* find a list of used extents on the given physical device */
950 struct extent
*rv
, *e
;
952 int memberships
= count_memberships(dl
, super
);
955 /* trim the reserved area for spares, so they can join any array
956 * regardless of whether the OROM has assigned sectors from the
957 * IMSM_RESERVED_SECTORS region
960 reservation
= imsm_min_reserved_sectors(super
);
962 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
964 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
967 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
968 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
969 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
971 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
972 e
->start
= pba_of_lba0(map
);
973 e
->size
= blocks_per_member(map
);
977 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
979 /* determine the start of the metadata
980 * when no raid devices are defined use the default
981 * ...otherwise allow the metadata to truncate the value
982 * as is the case with older versions of imsm
985 struct extent
*last
= &rv
[memberships
- 1];
986 unsigned long long remainder
;
988 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
989 /* round down to 1k block to satisfy precision of the kernel
993 /* make sure remainder is still sane */
994 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
995 remainder
= ROUND_UP(super
->len
, 512) >> 9;
996 if (reservation
> remainder
)
997 reservation
= remainder
;
999 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1004 /* try to determine how much space is reserved for metadata from
1005 * the last get_extents() entry, otherwise fallback to the
1008 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1014 /* for spares just return a minimal reservation which will grow
1015 * once the spare is picked up by an array
1017 if (dl
->index
== -1)
1018 return MPB_SECTOR_CNT
;
1020 e
= get_extents(super
, dl
);
1022 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1024 /* scroll to last entry */
1025 for (i
= 0; e
[i
].size
; i
++)
1028 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1035 static int is_spare(struct imsm_disk
*disk
)
1037 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1040 static int is_configured(struct imsm_disk
*disk
)
1042 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1045 static int is_failed(struct imsm_disk
*disk
)
1047 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1050 /* try to determine how much space is reserved for metadata from
1051 * the last get_extents() entry on the smallest active disk,
1052 * otherwise fallback to the default
1054 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1058 unsigned long long min_active
;
1060 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1061 struct dl
*dl
, *dl_min
= NULL
;
1067 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1070 unsigned long long blocks
= total_blocks(&dl
->disk
);
1071 if (blocks
< min_active
|| min_active
== 0) {
1073 min_active
= blocks
;
1079 /* find last lba used by subarrays on the smallest active disk */
1080 e
= get_extents(super
, dl_min
);
1083 for (i
= 0; e
[i
].size
; i
++)
1086 remainder
= min_active
- e
[i
].start
;
1089 /* to give priority to recovery we should not require full
1090 IMSM_RESERVED_SECTORS from the spare */
1091 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1093 /* if real reservation is smaller use that value */
1094 return (remainder
< rv
) ? remainder
: rv
;
1097 /* Return minimum size of a spare that can be used in this array*/
1098 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1100 struct intel_super
*super
= st
->sb
;
1104 unsigned long long rv
= 0;
1108 /* find first active disk in array */
1110 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1114 /* find last lba used by subarrays */
1115 e
= get_extents(super
, dl
);
1118 for (i
= 0; e
[i
].size
; i
++)
1121 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1124 /* add the amount of space needed for metadata */
1125 rv
= rv
+ imsm_min_reserved_sectors(super
);
1130 static int is_gen_migration(struct imsm_dev
*dev
);
1133 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1134 struct imsm_dev
*dev
);
1136 static void print_imsm_dev(struct intel_super
*super
,
1137 struct imsm_dev
*dev
,
1143 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1144 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1148 printf("[%.16s]:\n", dev
->volume
);
1149 printf(" UUID : %s\n", uuid
);
1150 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1152 printf(" <-- %d", get_imsm_raid_level(map2
));
1154 printf(" Members : %d", map
->num_members
);
1156 printf(" <-- %d", map2
->num_members
);
1158 printf(" Slots : [");
1159 for (i
= 0; i
< map
->num_members
; i
++) {
1160 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1161 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1166 for (i
= 0; i
< map2
->num_members
; i
++) {
1167 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1168 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1173 printf(" Failed disk : ");
1174 if (map
->failed_disk_num
== 0xff)
1177 printf("%i", map
->failed_disk_num
);
1179 slot
= get_imsm_disk_slot(map
, disk_idx
);
1181 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1182 printf(" This Slot : %d%s\n", slot
,
1183 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1185 printf(" This Slot : ?\n");
1186 sz
= __le32_to_cpu(dev
->size_high
);
1188 sz
+= __le32_to_cpu(dev
->size_low
);
1189 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1190 human_size(sz
* 512));
1191 sz
= blocks_per_member(map
);
1192 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1193 human_size(sz
* 512));
1194 printf(" Sector Offset : %llu\n",
1196 printf(" Num Stripes : %llu\n",
1197 num_data_stripes(map
));
1198 printf(" Chunk Size : %u KiB",
1199 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1201 printf(" <-- %u KiB",
1202 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1204 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1205 printf(" Migrate State : ");
1206 if (dev
->vol
.migr_state
) {
1207 if (migr_type(dev
) == MIGR_INIT
)
1208 printf("initialize\n");
1209 else if (migr_type(dev
) == MIGR_REBUILD
)
1210 printf("rebuild\n");
1211 else if (migr_type(dev
) == MIGR_VERIFY
)
1213 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1214 printf("general migration\n");
1215 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1216 printf("state change\n");
1217 else if (migr_type(dev
) == MIGR_REPAIR
)
1220 printf("<unknown:%d>\n", migr_type(dev
));
1223 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1224 if (dev
->vol
.migr_state
) {
1225 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1227 printf(" <-- %s", map_state_str
[map
->map_state
]);
1228 printf("\n Checkpoint : %u ",
1229 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1230 if ((is_gen_migration(dev
)) && ((slot
> 1) || (slot
< 0)))
1233 printf("(%llu)", (unsigned long long)
1234 blocks_per_migr_unit(super
, dev
));
1237 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1240 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1242 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1245 if (index
< -1 || !disk
)
1249 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1251 printf(" Disk%02d Serial : %s\n", index
, str
);
1253 printf(" Disk Serial : %s\n", str
);
1254 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1255 is_configured(disk
) ? " active" : "",
1256 is_failed(disk
) ? " failed" : "");
1257 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1258 sz
= total_blocks(disk
) - reserved
;
1259 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1260 human_size(sz
* 512));
1263 void examine_migr_rec_imsm(struct intel_super
*super
)
1265 struct migr_record
*migr_rec
= super
->migr_rec
;
1266 struct imsm_super
*mpb
= super
->anchor
;
1269 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1270 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1271 struct imsm_map
*map
;
1274 if (is_gen_migration(dev
) == 0)
1277 printf("\nMigration Record Information:");
1279 /* first map under migration */
1280 map
= get_imsm_map(dev
, MAP_0
);
1282 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1283 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1284 printf(" Empty\n ");
1285 printf("Examine one of first two disks in array\n");
1288 printf("\n Status : ");
1289 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1292 printf("Contains Data\n");
1293 printf(" Current Unit : %u\n",
1294 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1295 printf(" Family : %u\n",
1296 __le32_to_cpu(migr_rec
->family_num
));
1297 printf(" Ascending : %u\n",
1298 __le32_to_cpu(migr_rec
->ascending_migr
));
1299 printf(" Blocks Per Unit : %u\n",
1300 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1301 printf(" Dest. Depth Per Unit : %u\n",
1302 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1303 printf(" Checkpoint Area pba : %u\n",
1304 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1305 printf(" First member lba : %u\n",
1306 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1307 printf(" Total Number of Units : %u\n",
1308 __le32_to_cpu(migr_rec
->num_migr_units
));
1309 printf(" Size of volume : %u\n",
1310 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1311 printf(" Expansion space for LBA64 : %u\n",
1312 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1313 printf(" Record was read from : %u\n",
1314 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1319 #endif /* MDASSEMBLE */
1320 /*******************************************************************************
1321 * function: imsm_check_attributes
1322 * Description: Function checks if features represented by attributes flags
1323 * are supported by mdadm.
1325 * attributes - Attributes read from metadata
1327 * 0 - passed attributes contains unsupported features flags
1328 * 1 - all features are supported
1329 ******************************************************************************/
1330 static int imsm_check_attributes(__u32 attributes
)
1333 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1335 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1337 not_supported
&= attributes
;
1338 if (not_supported
) {
1339 pr_err("(IMSM): Unsupported attributes : %x\n",
1340 (unsigned)__le32_to_cpu(not_supported
));
1341 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1342 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1343 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1345 if (not_supported
& MPB_ATTRIB_2TB
) {
1346 dprintf("\t\tMPB_ATTRIB_2TB\n");
1347 not_supported
^= MPB_ATTRIB_2TB
;
1349 if (not_supported
& MPB_ATTRIB_RAID0
) {
1350 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1351 not_supported
^= MPB_ATTRIB_RAID0
;
1353 if (not_supported
& MPB_ATTRIB_RAID1
) {
1354 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1355 not_supported
^= MPB_ATTRIB_RAID1
;
1357 if (not_supported
& MPB_ATTRIB_RAID10
) {
1358 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1359 not_supported
^= MPB_ATTRIB_RAID10
;
1361 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1362 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1363 not_supported
^= MPB_ATTRIB_RAID1E
;
1365 if (not_supported
& MPB_ATTRIB_RAID5
) {
1366 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1367 not_supported
^= MPB_ATTRIB_RAID5
;
1369 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1370 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1371 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1373 if (not_supported
& MPB_ATTRIB_BBM
) {
1374 dprintf("\t\tMPB_ATTRIB_BBM\n");
1375 not_supported
^= MPB_ATTRIB_BBM
;
1377 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1378 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1379 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1381 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1382 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1383 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1385 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1386 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1387 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1389 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1390 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1391 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1393 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1394 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1395 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1399 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1408 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1410 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1412 struct intel_super
*super
= st
->sb
;
1413 struct imsm_super
*mpb
= super
->anchor
;
1414 char str
[MAX_SIGNATURE_LENGTH
];
1419 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1422 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1423 printf(" Magic : %s\n", str
);
1424 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1425 printf(" Version : %s\n", get_imsm_version(mpb
));
1426 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1427 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1428 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1429 printf(" Attributes : ");
1430 if (imsm_check_attributes(mpb
->attributes
))
1431 printf("All supported\n");
1433 printf("not supported\n");
1434 getinfo_super_imsm(st
, &info
, NULL
);
1435 fname_from_uuid(st
, &info
, nbuf
, ':');
1436 printf(" UUID : %s\n", nbuf
+ 5);
1437 sum
= __le32_to_cpu(mpb
->check_sum
);
1438 printf(" Checksum : %08x %s\n", sum
,
1439 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1440 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1441 printf(" Disks : %d\n", mpb
->num_disks
);
1442 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1443 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1444 if (super
->bbm_log
) {
1445 struct bbm_log
*log
= super
->bbm_log
;
1448 printf("Bad Block Management Log:\n");
1449 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1450 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1451 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1452 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1453 printf(" First Spare : %llx\n",
1454 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1456 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1458 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1460 super
->current_vol
= i
;
1461 getinfo_super_imsm(st
, &info
, NULL
);
1462 fname_from_uuid(st
, &info
, nbuf
, ':');
1463 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1465 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1466 if (i
== super
->disks
->index
)
1468 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1471 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1472 if (dl
->index
== -1)
1473 print_imsm_disk(&dl
->disk
, -1, reserved
);
1475 examine_migr_rec_imsm(super
);
1478 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1480 /* We just write a generic IMSM ARRAY entry */
1483 struct intel_super
*super
= st
->sb
;
1485 if (!super
->anchor
->num_raid_devs
) {
1486 printf("ARRAY metadata=imsm\n");
1490 getinfo_super_imsm(st
, &info
, NULL
);
1491 fname_from_uuid(st
, &info
, nbuf
, ':');
1492 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1495 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1497 /* We just write a generic IMSM ARRAY entry */
1501 struct intel_super
*super
= st
->sb
;
1504 if (!super
->anchor
->num_raid_devs
)
1507 getinfo_super_imsm(st
, &info
, NULL
);
1508 fname_from_uuid(st
, &info
, nbuf
, ':');
1509 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1510 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1512 super
->current_vol
= i
;
1513 getinfo_super_imsm(st
, &info
, NULL
);
1514 fname_from_uuid(st
, &info
, nbuf1
, ':');
1515 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1516 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1520 static void export_examine_super_imsm(struct supertype
*st
)
1522 struct intel_super
*super
= st
->sb
;
1523 struct imsm_super
*mpb
= super
->anchor
;
1527 getinfo_super_imsm(st
, &info
, NULL
);
1528 fname_from_uuid(st
, &info
, nbuf
, ':');
1529 printf("MD_METADATA=imsm\n");
1530 printf("MD_LEVEL=container\n");
1531 printf("MD_UUID=%s\n", nbuf
+5);
1532 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1535 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1540 getinfo_super_imsm(st
, &info
, NULL
);
1541 fname_from_uuid(st
, &info
, nbuf
, ':');
1542 printf("\n UUID : %s\n", nbuf
+ 5);
1545 static void brief_detail_super_imsm(struct supertype
*st
)
1549 getinfo_super_imsm(st
, &info
, NULL
);
1550 fname_from_uuid(st
, &info
, nbuf
, ':');
1551 printf(" UUID=%s", nbuf
+ 5);
1554 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1555 static void fd2devname(int fd
, char *name
);
1557 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1559 /* dump an unsorted list of devices attached to AHCI Intel storage
1560 * controller, as well as non-connected ports
1562 int hba_len
= strlen(hba_path
) + 1;
1567 unsigned long port_mask
= (1 << port_count
) - 1;
1569 if (port_count
> (int)sizeof(port_mask
) * 8) {
1571 pr_err("port_count %d out of range\n", port_count
);
1575 /* scroll through /sys/dev/block looking for devices attached to
1578 dir
= opendir("/sys/dev/block");
1579 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1590 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1592 path
= devt_to_devpath(makedev(major
, minor
));
1595 if (!path_attached_to_hba(path
, hba_path
)) {
1601 /* retrieve the scsi device type */
1602 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1604 pr_err("failed to allocate 'device'\n");
1608 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1609 if (load_sys(device
, buf
) != 0) {
1611 pr_err("failed to read device type for %s\n",
1617 type
= strtoul(buf
, NULL
, 10);
1619 /* if it's not a disk print the vendor and model */
1620 if (!(type
== 0 || type
== 7 || type
== 14)) {
1623 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1624 if (load_sys(device
, buf
) == 0) {
1625 strncpy(vendor
, buf
, sizeof(vendor
));
1626 vendor
[sizeof(vendor
) - 1] = '\0';
1627 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1628 while (isspace(*c
) || *c
== '\0')
1632 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1633 if (load_sys(device
, buf
) == 0) {
1634 strncpy(model
, buf
, sizeof(model
));
1635 model
[sizeof(model
) - 1] = '\0';
1636 c
= (char *) &model
[sizeof(model
) - 1];
1637 while (isspace(*c
) || *c
== '\0')
1641 if (vendor
[0] && model
[0])
1642 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1644 switch (type
) { /* numbers from hald/linux/device.c */
1645 case 1: sprintf(buf
, "tape"); break;
1646 case 2: sprintf(buf
, "printer"); break;
1647 case 3: sprintf(buf
, "processor"); break;
1649 case 5: sprintf(buf
, "cdrom"); break;
1650 case 6: sprintf(buf
, "scanner"); break;
1651 case 8: sprintf(buf
, "media_changer"); break;
1652 case 9: sprintf(buf
, "comm"); break;
1653 case 12: sprintf(buf
, "raid"); break;
1654 default: sprintf(buf
, "unknown");
1660 /* chop device path to 'host%d' and calculate the port number */
1661 c
= strchr(&path
[hba_len
], '/');
1664 pr_err("%s - invalid path name\n", path
+ hba_len
);
1669 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1673 *c
= '/'; /* repair the full string */
1674 pr_err("failed to determine port number for %s\n",
1681 /* mark this port as used */
1682 port_mask
&= ~(1 << port
);
1684 /* print out the device information */
1686 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1690 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1692 printf(" Port%d : - disk info unavailable -\n", port
);
1694 fd2devname(fd
, buf
);
1695 printf(" Port%d : %s", port
, buf
);
1696 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1697 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1712 for (i
= 0; i
< port_count
; i
++)
1713 if (port_mask
& (1 << i
))
1714 printf(" Port%d : - no device attached -\n", i
);
1720 static void print_found_intel_controllers(struct sys_dev
*elem
)
1722 for (; elem
; elem
= elem
->next
) {
1723 pr_err("found Intel(R) ");
1724 if (elem
->type
== SYS_DEV_SATA
)
1725 fprintf(stderr
, "SATA ");
1726 else if (elem
->type
== SYS_DEV_SAS
)
1727 fprintf(stderr
, "SAS ");
1728 fprintf(stderr
, "RAID controller");
1730 fprintf(stderr
, " at %s", elem
->pci_id
);
1731 fprintf(stderr
, ".\n");
1736 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1743 if ((dir
= opendir(hba_path
)) == NULL
)
1746 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1749 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1751 if (*port_count
== 0)
1753 else if (host
< host_base
)
1756 if (host
+ 1 > *port_count
+ host_base
)
1757 *port_count
= host
+ 1 - host_base
;
1763 static void print_imsm_capability(const struct imsm_orom
*orom
)
1765 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1766 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1767 orom
->hotfix_ver
, orom
->build
);
1768 printf(" RAID Levels :%s%s%s%s%s\n",
1769 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1770 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1771 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1772 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1773 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1774 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1775 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1776 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1777 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1778 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1779 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1780 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1781 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1782 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1783 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1784 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1785 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1786 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1787 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1788 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1789 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1790 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1791 printf(" 2TB volumes :%s supported\n",
1792 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
1793 printf(" 2TB disks :%s supported\n",
1794 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
1795 printf(" Max Disks : %d\n", orom
->tds
);
1796 printf(" Max Volumes : %d per array, %d per controller\n",
1797 orom
->vpa
, orom
->vphba
);
1801 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
1803 printf("MD_FIRMWARE_TYPE=imsm\n");
1804 printf("IMSM_VERSION=%d.%d.%d.%d\n",orom
->major_ver
, orom
->minor_ver
,
1805 orom
->hotfix_ver
, orom
->build
);
1806 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
1807 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
1808 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
1809 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
1810 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
1811 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
1812 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1813 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
1814 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
1815 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
1816 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
1817 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
1818 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
1819 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
1820 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
1821 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
1822 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
1823 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
1824 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
1825 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
1826 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
1827 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
1828 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
1829 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
1830 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
1831 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
1832 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
1833 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
1836 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
1838 /* There are two components to imsm platform support, the ahci SATA
1839 * controller and the option-rom. To find the SATA controller we
1840 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1841 * controller with the Intel vendor id is present. This approach
1842 * allows mdadm to leverage the kernel's ahci detection logic, with the
1843 * caveat that if ahci.ko is not loaded mdadm will not be able to
1844 * detect platform raid capabilities. The option-rom resides in a
1845 * platform "Adapter ROM". We scan for its signature to retrieve the
1846 * platform capabilities. If raid support is disabled in the BIOS the
1847 * option-rom capability structure will not be available.
1849 const struct imsm_orom
*orom
;
1850 struct sys_dev
*list
, *hba
;
1855 if (enumerate_only
) {
1856 if (check_env("IMSM_NO_PLATFORM"))
1858 list
= find_intel_devices();
1861 for (hba
= list
; hba
; hba
= hba
->next
) {
1862 orom
= find_imsm_capability(hba
->type
);
1870 free_sys_dev(&list
);
1874 list
= find_intel_devices();
1877 pr_err("no active Intel(R) RAID "
1878 "controller found.\n");
1879 free_sys_dev(&list
);
1881 } else if (verbose
> 0)
1882 print_found_intel_controllers(list
);
1884 for (hba
= list
; hba
; hba
= hba
->next
) {
1885 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
1887 orom
= find_imsm_capability(hba
->type
);
1889 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
1890 hba
->path
, get_sys_dev_type(hba
->type
));
1893 print_imsm_capability(orom
);
1894 printf(" I/O Controller : %s (%s)\n",
1895 hba
->path
, get_sys_dev_type(hba
->type
));
1896 if (hba
->type
== SYS_DEV_SATA
) {
1897 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1898 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1900 pr_err("failed to enumerate "
1901 "ports on SATA controller at %s.\n", hba
->pci_id
);
1908 if (controller_path
&& result
== 1)
1909 pr_err("no active Intel(R) RAID "
1910 "controller found under %s\n",controller_path
);
1912 free_sys_dev(&list
);
1916 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
1918 const struct imsm_orom
*orom
;
1919 struct sys_dev
*list
, *hba
;
1922 list
= find_intel_devices();
1925 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
1927 free_sys_dev(&list
);
1931 for (hba
= list
; hba
; hba
= hba
->next
) {
1932 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
1934 orom
= find_imsm_capability(hba
->type
);
1937 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",hba
->path
);
1940 print_imsm_capability_export(orom
);
1950 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1952 /* the imsm metadata format does not specify any host
1953 * identification information. We return -1 since we can never
1954 * confirm nor deny whether a given array is "meant" for this
1955 * host. We rely on compare_super and the 'family_num' fields to
1956 * exclude member disks that do not belong, and we rely on
1957 * mdadm.conf to specify the arrays that should be assembled.
1958 * Auto-assembly may still pick up "foreign" arrays.
1964 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1966 /* The uuid returned here is used for:
1967 * uuid to put into bitmap file (Create, Grow)
1968 * uuid for backup header when saving critical section (Grow)
1969 * comparing uuids when re-adding a device into an array
1970 * In these cases the uuid required is that of the data-array,
1971 * not the device-set.
1972 * uuid to recognise same set when adding a missing device back
1973 * to an array. This is a uuid for the device-set.
1975 * For each of these we can make do with a truncated
1976 * or hashed uuid rather than the original, as long as
1978 * In each case the uuid required is that of the data-array,
1979 * not the device-set.
1981 /* imsm does not track uuid's so we synthesis one using sha1 on
1982 * - The signature (Which is constant for all imsm array, but no matter)
1983 * - the orig_family_num of the container
1984 * - the index number of the volume
1985 * - the 'serial' number of the volume.
1986 * Hopefully these are all constant.
1988 struct intel_super
*super
= st
->sb
;
1991 struct sha1_ctx ctx
;
1992 struct imsm_dev
*dev
= NULL
;
1995 /* some mdadm versions failed to set ->orig_family_num, in which
1996 * case fall back to ->family_num. orig_family_num will be
1997 * fixed up with the first metadata update.
1999 family_num
= super
->anchor
->orig_family_num
;
2000 if (family_num
== 0)
2001 family_num
= super
->anchor
->family_num
;
2002 sha1_init_ctx(&ctx
);
2003 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2004 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2005 if (super
->current_vol
>= 0)
2006 dev
= get_imsm_dev(super
, super
->current_vol
);
2008 __u32 vol
= super
->current_vol
;
2009 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2010 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2012 sha1_finish_ctx(&ctx
, buf
);
2013 memcpy(uuid
, buf
, 4*4);
2018 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2020 __u8
*v
= get_imsm_version(mpb
);
2021 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2022 char major
[] = { 0, 0, 0 };
2023 char minor
[] = { 0 ,0, 0 };
2024 char patch
[] = { 0, 0, 0 };
2025 char *ver_parse
[] = { major
, minor
, patch
};
2029 while (*v
!= '\0' && v
< end
) {
2030 if (*v
!= '.' && j
< 2)
2031 ver_parse
[i
][j
++] = *v
;
2039 *m
= strtol(minor
, NULL
, 0);
2040 *p
= strtol(patch
, NULL
, 0);
2044 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2046 /* migr_strip_size when repairing or initializing parity */
2047 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2048 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2050 switch (get_imsm_raid_level(map
)) {
2055 return 128*1024 >> 9;
2059 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2061 /* migr_strip_size when rebuilding a degraded disk, no idea why
2062 * this is different than migr_strip_size_resync(), but it's good
2065 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2066 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2068 switch (get_imsm_raid_level(map
)) {
2071 if (map
->num_members
% map
->num_domains
== 0)
2072 return 128*1024 >> 9;
2076 return max((__u32
) 64*1024 >> 9, chunk
);
2078 return 128*1024 >> 9;
2082 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2084 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2085 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2086 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2087 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2089 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2092 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2094 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2095 int level
= get_imsm_raid_level(lo
);
2097 if (level
== 1 || level
== 10) {
2098 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2100 return hi
->num_domains
;
2102 return num_stripes_per_unit_resync(dev
);
2105 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2107 /* named 'imsm_' because raid0, raid1 and raid10
2108 * counter-intuitively have the same number of data disks
2110 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2112 switch (get_imsm_raid_level(map
)) {
2114 return map
->num_members
;
2118 return map
->num_members
/2;
2120 return map
->num_members
- 1;
2122 dprintf("%s: unsupported raid level\n", __func__
);
2127 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2129 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2130 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2132 switch(get_imsm_raid_level(map
)) {
2135 return chunk
* map
->num_domains
;
2137 return chunk
* map
->num_members
;
2143 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2145 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2146 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2147 __u32 strip
= block
/ chunk
;
2149 switch (get_imsm_raid_level(map
)) {
2152 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2153 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2155 return vol_stripe
* chunk
+ block
% chunk
;
2157 __u32 stripe
= strip
/ (map
->num_members
- 1);
2159 return stripe
* chunk
+ block
% chunk
;
2166 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2167 struct imsm_dev
*dev
)
2169 /* calculate the conversion factor between per member 'blocks'
2170 * (md/{resync,rebuild}_start) and imsm migration units, return
2171 * 0 for the 'not migrating' and 'unsupported migration' cases
2173 if (!dev
->vol
.migr_state
)
2176 switch (migr_type(dev
)) {
2177 case MIGR_GEN_MIGR
: {
2178 struct migr_record
*migr_rec
= super
->migr_rec
;
2179 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2184 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2185 __u32 stripes_per_unit
;
2186 __u32 blocks_per_unit
;
2195 /* yes, this is really the translation of migr_units to
2196 * per-member blocks in the 'resync' case
2198 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2199 migr_chunk
= migr_strip_blocks_resync(dev
);
2200 disks
= imsm_num_data_members(dev
, MAP_0
);
2201 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2202 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2203 segment
= blocks_per_unit
/ stripe
;
2204 block_rel
= blocks_per_unit
- segment
* stripe
;
2205 parity_depth
= parity_segment_depth(dev
);
2206 block_map
= map_migr_block(dev
, block_rel
);
2207 return block_map
+ parity_depth
* segment
;
2209 case MIGR_REBUILD
: {
2210 __u32 stripes_per_unit
;
2213 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2214 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2215 return migr_chunk
* stripes_per_unit
;
2217 case MIGR_STATE_CHANGE
:
2223 static int imsm_level_to_layout(int level
)
2231 return ALGORITHM_LEFT_ASYMMETRIC
;
2238 /*******************************************************************************
2239 * Function: read_imsm_migr_rec
2240 * Description: Function reads imsm migration record from last sector of disk
2242 * fd : disk descriptor
2243 * super : metadata info
2247 ******************************************************************************/
2248 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2251 unsigned long long dsize
;
2253 get_dev_size(fd
, NULL
, &dsize
);
2254 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2255 pr_err("Cannot seek to anchor block: %s\n",
2259 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2260 MIGR_REC_BUF_SIZE
) {
2261 pr_err("Cannot read migr record block: %s\n",
2271 static struct imsm_dev
*imsm_get_device_during_migration(
2272 struct intel_super
*super
)
2275 struct intel_dev
*dv
;
2277 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2278 if (is_gen_migration(dv
->dev
))
2284 /*******************************************************************************
2285 * Function: load_imsm_migr_rec
2286 * Description: Function reads imsm migration record (it is stored at the last
2289 * super : imsm internal array info
2290 * info : general array info
2294 * -2 : no migration in progress
2295 ******************************************************************************/
2296 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2299 struct dl
*dl
= NULL
;
2303 struct imsm_dev
*dev
;
2304 struct imsm_map
*map
= NULL
;
2307 /* find map under migration */
2308 dev
= imsm_get_device_during_migration(super
);
2309 /* nothing to load,no migration in progress?
2313 map
= get_imsm_map(dev
, MAP_0
);
2316 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2317 /* skip spare and failed disks
2319 if (sd
->disk
.raid_disk
< 0)
2321 /* read only from one of the first two slots */
2323 slot
= get_imsm_disk_slot(map
,
2324 sd
->disk
.raid_disk
);
2325 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2328 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2329 fd
= dev_open(nm
, O_RDONLY
);
2335 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2336 /* skip spare and failed disks
2340 /* read only from one of the first two slots */
2342 slot
= get_imsm_disk_slot(map
, dl
->index
);
2343 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2345 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2346 fd
= dev_open(nm
, O_RDONLY
);
2353 retval
= read_imsm_migr_rec(fd
, super
);
2362 /*******************************************************************************
2363 * function: imsm_create_metadata_checkpoint_update
2364 * Description: It creates update for checkpoint change.
2366 * super : imsm internal array info
2367 * u : pointer to prepared update
2370 * If length is equal to 0, input pointer u contains no update
2371 ******************************************************************************/
2372 static int imsm_create_metadata_checkpoint_update(
2373 struct intel_super
*super
,
2374 struct imsm_update_general_migration_checkpoint
**u
)
2377 int update_memory_size
= 0;
2379 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2385 /* size of all update data without anchor */
2386 update_memory_size
=
2387 sizeof(struct imsm_update_general_migration_checkpoint
);
2389 *u
= xcalloc(1, update_memory_size
);
2391 dprintf("error: cannot get memory for "
2392 "imsm_create_metadata_checkpoint_update update\n");
2395 (*u
)->type
= update_general_migration_checkpoint
;
2396 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2397 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2398 (*u
)->curr_migr_unit
);
2400 return update_memory_size
;
2404 static void imsm_update_metadata_locally(struct supertype
*st
,
2405 void *buf
, int len
);
2407 /*******************************************************************************
2408 * Function: write_imsm_migr_rec
2409 * Description: Function writes imsm migration record
2410 * (at the last sector of disk)
2412 * super : imsm internal array info
2416 ******************************************************************************/
2417 static int write_imsm_migr_rec(struct supertype
*st
)
2419 struct intel_super
*super
= st
->sb
;
2420 unsigned long long dsize
;
2426 struct imsm_update_general_migration_checkpoint
*u
;
2427 struct imsm_dev
*dev
;
2428 struct imsm_map
*map
= NULL
;
2430 /* find map under migration */
2431 dev
= imsm_get_device_during_migration(super
);
2432 /* if no migration, write buffer anyway to clear migr_record
2433 * on disk based on first available device
2436 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2437 super
->current_vol
);
2439 map
= get_imsm_map(dev
, MAP_0
);
2441 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2444 /* skip failed and spare devices */
2447 /* write to 2 first slots only */
2449 slot
= get_imsm_disk_slot(map
, sd
->index
);
2450 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2453 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2454 fd
= dev_open(nm
, O_RDWR
);
2457 get_dev_size(fd
, NULL
, &dsize
);
2458 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2459 pr_err("Cannot seek to anchor block: %s\n",
2463 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2464 MIGR_REC_BUF_SIZE
) {
2465 pr_err("Cannot write migr record block: %s\n",
2472 /* update checkpoint information in metadata */
2473 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2476 dprintf("imsm: Cannot prepare update\n");
2479 /* update metadata locally */
2480 imsm_update_metadata_locally(st
, u
, len
);
2481 /* and possibly remotely */
2482 if (st
->update_tail
) {
2483 append_metadata_update(st
, u
, len
);
2484 /* during reshape we do all work inside metadata handler
2485 * manage_reshape(), so metadata update has to be triggered
2488 flush_metadata_updates(st
);
2489 st
->update_tail
= &st
->updates
;
2499 #endif /* MDASSEMBLE */
2501 /* spare/missing disks activations are not allowe when
2502 * array/container performs reshape operation, because
2503 * all arrays in container works on the same disks set
2505 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2508 struct intel_dev
*i_dev
;
2509 struct imsm_dev
*dev
;
2511 /* check whole container
2513 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2515 if (is_gen_migration(dev
)) {
2516 /* No repair during any migration in container
2524 static unsigned long long imsm_component_size_aligment_check(int level
,
2526 unsigned long long component_size
)
2528 unsigned int component_size_alligment
;
2530 /* check component size aligment
2532 component_size_alligment
= component_size
% (chunk_size
/512);
2534 dprintf("imsm_component_size_aligment_check(Level: %i, "
2535 "chunk_size = %i, component_size = %llu), "
2536 "component_size_alligment = %u\n",
2537 level
, chunk_size
, component_size
,
2538 component_size_alligment
);
2540 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
2541 dprintf("imsm: reported component size alligned from %llu ",
2543 component_size
-= component_size_alligment
;
2544 dprintf("to %llu (%i).\n",
2545 component_size
, component_size_alligment
);
2548 return component_size
;
2551 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2553 struct intel_super
*super
= st
->sb
;
2554 struct migr_record
*migr_rec
= super
->migr_rec
;
2555 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2556 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2557 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2558 struct imsm_map
*map_to_analyse
= map
;
2561 int map_disks
= info
->array
.raid_disks
;
2563 memset(info
, 0, sizeof(*info
));
2565 map_to_analyse
= prev_map
;
2567 dl
= super
->current_disk
;
2569 info
->container_member
= super
->current_vol
;
2570 info
->array
.raid_disks
= map
->num_members
;
2571 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2572 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2573 info
->array
.md_minor
= -1;
2574 info
->array
.ctime
= 0;
2575 info
->array
.utime
= 0;
2576 info
->array
.chunk_size
=
2577 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2578 info
->array
.state
= !dev
->vol
.dirty
;
2579 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2580 info
->custom_array_size
<<= 32;
2581 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2582 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2584 if (is_gen_migration(dev
)) {
2585 info
->reshape_active
= 1;
2586 info
->new_level
= get_imsm_raid_level(map
);
2587 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2588 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2589 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2590 if (info
->delta_disks
) {
2591 /* this needs to be applied to every array
2594 info
->reshape_active
= CONTAINER_RESHAPE
;
2596 /* We shape information that we give to md might have to be
2597 * modify to cope with md's requirement for reshaping arrays.
2598 * For example, when reshaping a RAID0, md requires it to be
2599 * presented as a degraded RAID4.
2600 * Also if a RAID0 is migrating to a RAID5 we need to specify
2601 * the array as already being RAID5, but the 'before' layout
2602 * is a RAID4-like layout.
2604 switch (info
->array
.level
) {
2606 switch(info
->new_level
) {
2608 /* conversion is happening as RAID4 */
2609 info
->array
.level
= 4;
2610 info
->array
.raid_disks
+= 1;
2613 /* conversion is happening as RAID5 */
2614 info
->array
.level
= 5;
2615 info
->array
.layout
= ALGORITHM_PARITY_N
;
2616 info
->delta_disks
-= 1;
2619 /* FIXME error message */
2620 info
->array
.level
= UnSet
;
2626 info
->new_level
= UnSet
;
2627 info
->new_layout
= UnSet
;
2628 info
->new_chunk
= info
->array
.chunk_size
;
2629 info
->delta_disks
= 0;
2633 info
->disk
.major
= dl
->major
;
2634 info
->disk
.minor
= dl
->minor
;
2635 info
->disk
.number
= dl
->index
;
2636 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2640 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2641 info
->component_size
= blocks_per_member(map_to_analyse
);
2643 info
->component_size
= imsm_component_size_aligment_check(
2645 info
->array
.chunk_size
,
2646 info
->component_size
);
2648 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2649 info
->recovery_start
= MaxSector
;
2651 info
->reshape_progress
= 0;
2652 info
->resync_start
= MaxSector
;
2653 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2655 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2656 info
->resync_start
= 0;
2658 if (dev
->vol
.migr_state
) {
2659 switch (migr_type(dev
)) {
2662 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2664 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2666 info
->resync_start
= blocks_per_unit
* units
;
2669 case MIGR_GEN_MIGR
: {
2670 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2672 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2673 unsigned long long array_blocks
;
2676 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2678 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2679 (super
->migr_rec
->rec_status
==
2680 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2683 info
->reshape_progress
= blocks_per_unit
* units
;
2685 dprintf("IMSM: General Migration checkpoint : %llu "
2686 "(%llu) -> read reshape progress : %llu\n",
2687 (unsigned long long)units
,
2688 (unsigned long long)blocks_per_unit
,
2689 info
->reshape_progress
);
2691 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2692 if (used_disks
> 0) {
2693 array_blocks
= blocks_per_member(map
) *
2695 /* round array size down to closest MB
2697 info
->custom_array_size
= (array_blocks
2698 >> SECT_PER_MB_SHIFT
)
2699 << SECT_PER_MB_SHIFT
;
2703 /* we could emulate the checkpointing of
2704 * 'sync_action=check' migrations, but for now
2705 * we just immediately complete them
2708 /* this is handled by container_content_imsm() */
2709 case MIGR_STATE_CHANGE
:
2710 /* FIXME handle other migrations */
2712 /* we are not dirty, so... */
2713 info
->resync_start
= MaxSector
;
2717 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2718 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2720 info
->array
.major_version
= -1;
2721 info
->array
.minor_version
= -2;
2722 devname
= devnum2devname(st
->container_dev
);
2723 *info
->text_version
= '\0';
2725 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2727 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2728 uuid_from_super_imsm(st
, info
->uuid
);
2732 for (i
=0; i
<map_disks
; i
++) {
2734 if (i
< info
->array
.raid_disks
) {
2735 struct imsm_disk
*dsk
;
2736 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2737 dsk
= get_imsm_disk(super
, j
);
2738 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2745 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2746 int failed
, int look_in_map
);
2748 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2753 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2755 if (is_gen_migration(dev
)) {
2758 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2760 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2761 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2762 if (map2
->map_state
!= map_state
) {
2763 map2
->map_state
= map_state
;
2764 super
->updates_pending
++;
2770 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2774 for (d
= super
->missing
; d
; d
= d
->next
)
2775 if (d
->index
== index
)
2780 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2782 struct intel_super
*super
= st
->sb
;
2783 struct imsm_disk
*disk
;
2784 int map_disks
= info
->array
.raid_disks
;
2785 int max_enough
= -1;
2787 struct imsm_super
*mpb
;
2789 if (super
->current_vol
>= 0) {
2790 getinfo_super_imsm_volume(st
, info
, map
);
2793 memset(info
, 0, sizeof(*info
));
2795 /* Set raid_disks to zero so that Assemble will always pull in valid
2798 info
->array
.raid_disks
= 0;
2799 info
->array
.level
= LEVEL_CONTAINER
;
2800 info
->array
.layout
= 0;
2801 info
->array
.md_minor
= -1;
2802 info
->array
.ctime
= 0; /* N/A for imsm */
2803 info
->array
.utime
= 0;
2804 info
->array
.chunk_size
= 0;
2806 info
->disk
.major
= 0;
2807 info
->disk
.minor
= 0;
2808 info
->disk
.raid_disk
= -1;
2809 info
->reshape_active
= 0;
2810 info
->array
.major_version
= -1;
2811 info
->array
.minor_version
= -2;
2812 strcpy(info
->text_version
, "imsm");
2813 info
->safe_mode_delay
= 0;
2814 info
->disk
.number
= -1;
2815 info
->disk
.state
= 0;
2817 info
->recovery_start
= MaxSector
;
2818 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2820 /* do we have the all the insync disks that we expect? */
2821 mpb
= super
->anchor
;
2823 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2824 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2825 int failed
, enough
, j
, missing
= 0;
2826 struct imsm_map
*map
;
2829 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2830 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2831 map
= get_imsm_map(dev
, MAP_0
);
2833 /* any newly missing disks?
2834 * (catches single-degraded vs double-degraded)
2836 for (j
= 0; j
< map
->num_members
; j
++) {
2837 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2838 __u32 idx
= ord_to_idx(ord
);
2840 if (!(ord
& IMSM_ORD_REBUILD
) &&
2841 get_imsm_missing(super
, idx
)) {
2847 if (state
== IMSM_T_STATE_FAILED
)
2849 else if (state
== IMSM_T_STATE_DEGRADED
&&
2850 (state
!= map
->map_state
|| missing
))
2852 else /* we're normal, or already degraded */
2854 if (is_gen_migration(dev
) && missing
) {
2855 /* during general migration we need all disks
2856 * that process is running on.
2857 * No new missing disk is allowed.
2861 /* no more checks necessary
2865 /* in the missing/failed disk case check to see
2866 * if at least one array is runnable
2868 max_enough
= max(max_enough
, enough
);
2870 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2871 info
->container_enough
= max_enough
;
2874 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2876 disk
= &super
->disks
->disk
;
2877 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
2878 info
->component_size
= reserved
;
2879 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2880 /* we don't change info->disk.raid_disk here because
2881 * this state will be finalized in mdmon after we have
2882 * found the 'most fresh' version of the metadata
2884 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2885 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2888 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2889 * ->compare_super may have updated the 'num_raid_devs' field for spares
2891 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2892 uuid_from_super_imsm(st
, info
->uuid
);
2894 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2896 /* I don't know how to compute 'map' on imsm, so use safe default */
2899 for (i
= 0; i
< map_disks
; i
++)
2905 /* allocates memory and fills disk in mdinfo structure
2906 * for each disk in array */
2907 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2909 struct mdinfo
*mddev
= NULL
;
2910 struct intel_super
*super
= st
->sb
;
2911 struct imsm_disk
*disk
;
2914 if (!super
|| !super
->disks
)
2917 mddev
= xcalloc(1, sizeof(*mddev
));
2921 tmp
= xcalloc(1, sizeof(*tmp
));
2923 tmp
->next
= mddev
->devs
;
2925 tmp
->disk
.number
= count
++;
2926 tmp
->disk
.major
= dl
->major
;
2927 tmp
->disk
.minor
= dl
->minor
;
2928 tmp
->disk
.state
= is_configured(disk
) ?
2929 (1 << MD_DISK_ACTIVE
) : 0;
2930 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2931 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2932 tmp
->disk
.raid_disk
= -1;
2938 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2939 char *update
, char *devname
, int verbose
,
2940 int uuid_set
, char *homehost
)
2942 /* For 'assemble' and 'force' we need to return non-zero if any
2943 * change was made. For others, the return value is ignored.
2944 * Update options are:
2945 * force-one : This device looks a bit old but needs to be included,
2946 * update age info appropriately.
2947 * assemble: clear any 'faulty' flag to allow this device to
2949 * force-array: Array is degraded but being forced, mark it clean
2950 * if that will be needed to assemble it.
2952 * newdev: not used ????
2953 * grow: Array has gained a new device - this is currently for
2955 * resync: mark as dirty so a resync will happen.
2956 * name: update the name - preserving the homehost
2957 * uuid: Change the uuid of the array to match watch is given
2959 * Following are not relevant for this imsm:
2960 * sparc2.2 : update from old dodgey metadata
2961 * super-minor: change the preferred_minor number
2962 * summaries: update redundant counters.
2963 * homehost: update the recorded homehost
2964 * _reshape_progress: record new reshape_progress position.
2967 struct intel_super
*super
= st
->sb
;
2968 struct imsm_super
*mpb
;
2970 /* we can only update container info */
2971 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2974 mpb
= super
->anchor
;
2976 if (strcmp(update
, "uuid") == 0) {
2977 /* We take this to mean that the family_num should be updated.
2978 * However that is much smaller than the uuid so we cannot really
2979 * allow an explicit uuid to be given. And it is hard to reliably
2981 * So if !uuid_set we know the current uuid is random and just used
2982 * the first 'int' and copy it to the other 3 positions.
2983 * Otherwise we require the 4 'int's to be the same as would be the
2984 * case if we are using a random uuid. So an explicit uuid will be
2985 * accepted as long as all for ints are the same... which shouldn't hurt
2988 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
2991 if (info
->uuid
[0] != info
->uuid
[1] ||
2992 info
->uuid
[1] != info
->uuid
[2] ||
2993 info
->uuid
[2] != info
->uuid
[3])
2999 mpb
->orig_family_num
= info
->uuid
[0];
3000 } else if (strcmp(update
, "assemble") == 0)
3005 /* successful update? recompute checksum */
3007 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3012 static size_t disks_to_mpb_size(int disks
)
3016 size
= sizeof(struct imsm_super
);
3017 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3018 size
+= 2 * sizeof(struct imsm_dev
);
3019 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3020 size
+= (4 - 2) * sizeof(struct imsm_map
);
3021 /* 4 possible disk_ord_tbl's */
3022 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3027 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3028 unsigned long long data_offset
)
3030 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3033 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3036 static void free_devlist(struct intel_super
*super
)
3038 struct intel_dev
*dv
;
3040 while (super
->devlist
) {
3041 dv
= super
->devlist
->next
;
3042 free(super
->devlist
->dev
);
3043 free(super
->devlist
);
3044 super
->devlist
= dv
;
3048 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3050 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3053 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3057 * 0 same, or first was empty, and second was copied
3058 * 1 second had wrong number
3060 * 3 wrong other info
3062 struct intel_super
*first
= st
->sb
;
3063 struct intel_super
*sec
= tst
->sb
;
3070 /* in platform dependent environment test if the disks
3071 * use the same Intel hba
3072 * If not on Intel hba at all, allow anything.
3074 if (!check_env("IMSM_NO_PLATFORM")) {
3075 if (first
->hba
&& sec
->hba
&&
3076 first
->hba
->type
!= sec
->hba
->type
) {
3078 "HBAs of devices does not match %s != %s\n",
3079 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
3080 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
3085 /* if an anchor does not have num_raid_devs set then it is a free
3088 if (first
->anchor
->num_raid_devs
> 0 &&
3089 sec
->anchor
->num_raid_devs
> 0) {
3090 /* Determine if these disks might ever have been
3091 * related. Further disambiguation can only take place
3092 * in load_super_imsm_all
3094 __u32 first_family
= first
->anchor
->orig_family_num
;
3095 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3097 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3098 MAX_SIGNATURE_LENGTH
) != 0)
3101 if (first_family
== 0)
3102 first_family
= first
->anchor
->family_num
;
3103 if (sec_family
== 0)
3104 sec_family
= sec
->anchor
->family_num
;
3106 if (first_family
!= sec_family
)
3112 /* if 'first' is a spare promote it to a populated mpb with sec's
3115 if (first
->anchor
->num_raid_devs
== 0 &&
3116 sec
->anchor
->num_raid_devs
> 0) {
3118 struct intel_dev
*dv
;
3119 struct imsm_dev
*dev
;
3121 /* we need to copy raid device info from sec if an allocation
3122 * fails here we don't associate the spare
3124 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3125 dv
= xmalloc(sizeof(*dv
));
3126 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3129 dv
->next
= first
->devlist
;
3130 first
->devlist
= dv
;
3132 if (i
< sec
->anchor
->num_raid_devs
) {
3133 /* allocation failure */
3134 free_devlist(first
);
3135 fprintf(stderr
, "imsm: failed to associate spare\n");
3138 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3139 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3140 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3141 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3142 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3143 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3149 static void fd2devname(int fd
, char *name
)
3153 char dname
[PATH_MAX
];
3158 if (fstat(fd
, &st
) != 0)
3160 sprintf(path
, "/sys/dev/block/%d:%d",
3161 major(st
.st_rdev
), minor(st
.st_rdev
));
3163 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3168 nm
= strrchr(dname
, '/');
3171 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3175 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3177 static int imsm_read_serial(int fd
, char *devname
,
3178 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3180 unsigned char scsi_serial
[255];
3189 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3191 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3193 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3194 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3195 fd2devname(fd
, (char *) serial
);
3201 pr_err("Failed to retrieve serial for %s\n",
3206 rsp_len
= scsi_serial
[3];
3209 pr_err("Failed to retrieve serial for %s\n",
3213 rsp_buf
= (char *) &scsi_serial
[4];
3215 /* trim all whitespace and non-printable characters and convert
3218 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3221 /* ':' is reserved for use in placeholder serial
3222 * numbers for missing disks
3230 len
= dest
- rsp_buf
;
3233 /* truncate leading characters */
3234 if (len
> MAX_RAID_SERIAL_LEN
) {
3235 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3236 len
= MAX_RAID_SERIAL_LEN
;
3239 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3240 memcpy(serial
, dest
, len
);
3245 static int serialcmp(__u8
*s1
, __u8
*s2
)
3247 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3250 static void serialcpy(__u8
*dest
, __u8
*src
)
3252 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3255 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3259 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3260 if (serialcmp(dl
->serial
, serial
) == 0)
3266 static struct imsm_disk
*
3267 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3271 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3272 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3274 if (serialcmp(disk
->serial
, serial
) == 0) {
3285 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3287 struct imsm_disk
*disk
;
3292 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3294 rv
= imsm_read_serial(fd
, devname
, serial
);
3299 dl
= xcalloc(1, sizeof(*dl
));
3302 dl
->major
= major(stb
.st_rdev
);
3303 dl
->minor
= minor(stb
.st_rdev
);
3304 dl
->next
= super
->disks
;
3305 dl
->fd
= keep_fd
? fd
: -1;
3306 assert(super
->disks
== NULL
);
3308 serialcpy(dl
->serial
, serial
);
3311 fd2devname(fd
, name
);
3313 dl
->devname
= xstrdup(devname
);
3315 dl
->devname
= xstrdup(name
);
3317 /* look up this disk's index in the current anchor */
3318 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3321 /* only set index on disks that are a member of a
3322 * populated contianer, i.e. one with raid_devs
3324 if (is_failed(&dl
->disk
))
3326 else if (is_spare(&dl
->disk
))
3334 /* When migrating map0 contains the 'destination' state while map1
3335 * contains the current state. When not migrating map0 contains the
3336 * current state. This routine assumes that map[0].map_state is set to
3337 * the current array state before being called.
3339 * Migration is indicated by one of the following states
3340 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3341 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3342 * map1state=unitialized)
3343 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3345 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3346 * map1state=degraded)
3347 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3350 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3351 __u8 to_state
, int migr_type
)
3353 struct imsm_map
*dest
;
3354 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3356 dev
->vol
.migr_state
= 1;
3357 set_migr_type(dev
, migr_type
);
3358 dev
->vol
.curr_migr_unit
= 0;
3359 dest
= get_imsm_map(dev
, MAP_1
);
3361 /* duplicate and then set the target end state in map[0] */
3362 memcpy(dest
, src
, sizeof_imsm_map(src
));
3363 if ((migr_type
== MIGR_REBUILD
) ||
3364 (migr_type
== MIGR_GEN_MIGR
)) {
3368 for (i
= 0; i
< src
->num_members
; i
++) {
3369 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3370 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3374 if (migr_type
== MIGR_GEN_MIGR
)
3375 /* Clear migration record */
3376 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3378 src
->map_state
= to_state
;
3381 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3384 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3385 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3389 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3390 * completed in the last migration.
3392 * FIXME add support for raid-level-migration
3394 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3395 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3396 /* when final map state is other than expected
3397 * merge maps (not for migration)
3401 for (i
= 0; i
< prev
->num_members
; i
++)
3402 for (j
= 0; j
< map
->num_members
; j
++)
3403 /* during online capacity expansion
3404 * disks position can be changed
3405 * if takeover is used
3407 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3408 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3409 map
->disk_ord_tbl
[j
] |=
3410 prev
->disk_ord_tbl
[i
];
3413 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3414 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3417 dev
->vol
.migr_state
= 0;
3418 set_migr_type(dev
, 0);
3419 dev
->vol
.curr_migr_unit
= 0;
3420 map
->map_state
= map_state
;
3424 static int parse_raid_devices(struct intel_super
*super
)
3427 struct imsm_dev
*dev_new
;
3428 size_t len
, len_migr
;
3430 size_t space_needed
= 0;
3431 struct imsm_super
*mpb
= super
->anchor
;
3433 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3434 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3435 struct intel_dev
*dv
;
3437 len
= sizeof_imsm_dev(dev_iter
, 0);
3438 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3440 space_needed
+= len_migr
- len
;
3442 dv
= xmalloc(sizeof(*dv
));
3443 if (max_len
< len_migr
)
3445 if (max_len
> len_migr
)
3446 space_needed
+= max_len
- len_migr
;
3447 dev_new
= xmalloc(max_len
);
3448 imsm_copy_dev(dev_new
, dev_iter
);
3451 dv
->next
= super
->devlist
;
3452 super
->devlist
= dv
;
3455 /* ensure that super->buf is large enough when all raid devices
3458 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3461 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3462 if (posix_memalign(&buf
, 512, len
) != 0)
3465 memcpy(buf
, super
->buf
, super
->len
);
3466 memset(buf
+ super
->len
, 0, len
- super
->len
);
3475 /* retrieve a pointer to the bbm log which starts after all raid devices */
3476 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3480 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3482 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3488 /*******************************************************************************
3489 * Function: check_mpb_migr_compatibility
3490 * Description: Function checks for unsupported migration features:
3491 * - migration optimization area (pba_of_lba0)
3492 * - descending reshape (ascending_migr)
3494 * super : imsm metadata information
3496 * 0 : migration is compatible
3497 * -1 : migration is not compatible
3498 ******************************************************************************/
3499 int check_mpb_migr_compatibility(struct intel_super
*super
)
3501 struct imsm_map
*map0
, *map1
;
3502 struct migr_record
*migr_rec
= super
->migr_rec
;
3505 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3506 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3509 dev_iter
->vol
.migr_state
== 1 &&
3510 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3511 /* This device is migrating */
3512 map0
= get_imsm_map(dev_iter
, MAP_0
);
3513 map1
= get_imsm_map(dev_iter
, MAP_1
);
3514 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3515 /* migration optimization area was used */
3517 if (migr_rec
->ascending_migr
== 0
3518 && migr_rec
->dest_depth_per_unit
> 0)
3519 /* descending reshape not supported yet */
3526 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3528 /* load_imsm_mpb - read matrix metadata
3529 * allocates super->mpb to be freed by free_imsm
3531 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3533 unsigned long long dsize
;
3534 unsigned long long sectors
;
3536 struct imsm_super
*anchor
;
3539 get_dev_size(fd
, NULL
, &dsize
);
3542 pr_err("%s: device to small for imsm\n",
3547 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3549 pr_err("Cannot seek to anchor block on %s: %s\n",
3550 devname
, strerror(errno
));
3554 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3556 pr_err("Failed to allocate imsm anchor buffer"
3557 " on %s\n", devname
);
3560 if (read(fd
, anchor
, 512) != 512) {
3562 pr_err("Cannot read anchor block on %s: %s\n",
3563 devname
, strerror(errno
));
3568 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3570 pr_err("no IMSM anchor on %s\n", devname
);
3575 __free_imsm(super
, 0);
3576 /* reload capability and hba */
3578 /* capability and hba must be updated with new super allocation */
3579 find_intel_hba_capability(fd
, super
, devname
);
3580 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3581 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3583 pr_err("unable to allocate %zu byte mpb buffer\n",
3588 memcpy(super
->buf
, anchor
, 512);
3590 sectors
= mpb_sectors(anchor
) - 1;
3593 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3594 pr_err("%s could not allocate migr_rec buffer\n", __func__
);
3598 super
->clean_migration_record_by_mdmon
= 0;
3601 check_sum
= __gen_imsm_checksum(super
->anchor
);
3602 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3604 pr_err("IMSM checksum %x != %x on %s\n",
3606 __le32_to_cpu(super
->anchor
->check_sum
),
3614 /* read the extended mpb */
3615 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3617 pr_err("Cannot seek to extended mpb on %s: %s\n",
3618 devname
, strerror(errno
));
3622 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3624 pr_err("Cannot read extended mpb on %s: %s\n",
3625 devname
, strerror(errno
));
3629 check_sum
= __gen_imsm_checksum(super
->anchor
);
3630 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3632 pr_err("IMSM checksum %x != %x on %s\n",
3633 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3638 /* FIXME the BBM log is disk specific so we cannot use this global
3639 * buffer for all disks. Ok for now since we only look at the global
3640 * bbm_log_size parameter to gate assembly
3642 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3647 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3649 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3650 static void clear_hi(struct intel_super
*super
)
3652 struct imsm_super
*mpb
= super
->anchor
;
3654 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3656 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3657 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3658 disk
->total_blocks_hi
= 0;
3660 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3661 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3664 for (n
= 0; n
< 2; ++n
) {
3665 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3668 map
->pba_of_lba0_hi
= 0;
3669 map
->blocks_per_member_hi
= 0;
3670 map
->num_data_stripes_hi
= 0;
3676 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3680 err
= load_imsm_mpb(fd
, super
, devname
);
3683 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3686 err
= parse_raid_devices(super
);
3691 static void __free_imsm_disk(struct dl
*d
)
3703 static void free_imsm_disks(struct intel_super
*super
)
3707 while (super
->disks
) {
3709 super
->disks
= d
->next
;
3710 __free_imsm_disk(d
);
3712 while (super
->disk_mgmt_list
) {
3713 d
= super
->disk_mgmt_list
;
3714 super
->disk_mgmt_list
= d
->next
;
3715 __free_imsm_disk(d
);
3717 while (super
->missing
) {
3719 super
->missing
= d
->next
;
3720 __free_imsm_disk(d
);
3725 /* free all the pieces hanging off of a super pointer */
3726 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3728 struct intel_hba
*elem
, *next
;
3734 /* unlink capability description */
3736 if (super
->migr_rec_buf
) {
3737 free(super
->migr_rec_buf
);
3738 super
->migr_rec_buf
= NULL
;
3741 free_imsm_disks(super
);
3742 free_devlist(super
);
3746 free((void *)elem
->path
);
3754 static void free_imsm(struct intel_super
*super
)
3756 __free_imsm(super
, 1);
3760 static void free_super_imsm(struct supertype
*st
)
3762 struct intel_super
*super
= st
->sb
;
3771 static struct intel_super
*alloc_super(void)
3773 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
3775 super
->current_vol
= -1;
3776 super
->create_offset
= ~((unsigned long long) 0);
3781 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3783 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3785 struct sys_dev
*hba_name
;
3788 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3793 hba_name
= find_disk_attached_hba(fd
, NULL
);
3796 pr_err("%s is not attached to Intel(R) RAID controller.\n",
3800 rv
= attach_hba_to_super(super
, hba_name
);
3803 struct intel_hba
*hba
= super
->hba
;
3805 pr_err("%s is attached to Intel(R) %s RAID "
3806 "controller (%s),\n"
3807 " but the container is assigned to Intel(R) "
3808 "%s RAID controller (",
3811 hba_name
->pci_id
? : "Err!",
3812 get_sys_dev_type(hba_name
->type
));
3815 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3817 fprintf(stderr
, ", ");
3821 fprintf(stderr
, ").\n"
3822 " Mixing devices attached to different controllers "
3823 "is not allowed.\n");
3825 free_sys_dev(&hba_name
);
3828 super
->orom
= find_imsm_capability(hba_name
->type
);
3829 free_sys_dev(&hba_name
);
3835 /* find_missing - helper routine for load_super_imsm_all that identifies
3836 * disks that have disappeared from the system. This routine relies on
3837 * the mpb being uptodate, which it is at load time.
3839 static int find_missing(struct intel_super
*super
)
3842 struct imsm_super
*mpb
= super
->anchor
;
3844 struct imsm_disk
*disk
;
3846 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3847 disk
= __get_imsm_disk(mpb
, i
);
3848 dl
= serial_to_dl(disk
->serial
, super
);
3852 dl
= xmalloc(sizeof(*dl
));
3856 dl
->devname
= xstrdup("missing");
3858 serialcpy(dl
->serial
, disk
->serial
);
3861 dl
->next
= super
->missing
;
3862 super
->missing
= dl
;
3869 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3871 struct intel_disk
*idisk
= disk_list
;
3874 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3876 idisk
= idisk
->next
;
3882 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3883 struct intel_super
*super
,
3884 struct intel_disk
**disk_list
)
3886 struct imsm_disk
*d
= &super
->disks
->disk
;
3887 struct imsm_super
*mpb
= super
->anchor
;
3890 for (i
= 0; i
< tbl_size
; i
++) {
3891 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3892 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3894 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3895 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3896 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3897 __func__
, super
->disks
->major
,
3898 super
->disks
->minor
,
3899 table
[i
]->disks
->major
,
3900 table
[i
]->disks
->minor
);
3904 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3905 is_configured(d
) == is_configured(tbl_d
)) &&
3906 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3907 /* current version of the mpb is a
3908 * better candidate than the one in
3909 * super_table, but copy over "cross
3910 * generational" status
3912 struct intel_disk
*idisk
;
3914 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3915 __func__
, super
->disks
->major
,
3916 super
->disks
->minor
,
3917 table
[i
]->disks
->major
,
3918 table
[i
]->disks
->minor
);
3920 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3921 if (idisk
&& is_failed(&idisk
->disk
))
3922 tbl_d
->status
|= FAILED_DISK
;
3925 struct intel_disk
*idisk
;
3926 struct imsm_disk
*disk
;
3928 /* tbl_mpb is more up to date, but copy
3929 * over cross generational status before
3932 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3933 if (disk
&& is_failed(disk
))
3934 d
->status
|= FAILED_DISK
;
3936 idisk
= disk_list_get(d
->serial
, *disk_list
);
3939 if (disk
&& is_configured(disk
))
3940 idisk
->disk
.status
|= CONFIGURED_DISK
;
3943 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3944 __func__
, super
->disks
->major
,
3945 super
->disks
->minor
,
3946 table
[i
]->disks
->major
,
3947 table
[i
]->disks
->minor
);
3955 table
[tbl_size
++] = super
;
3959 /* update/extend the merged list of imsm_disk records */
3960 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3961 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3962 struct intel_disk
*idisk
;
3964 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3966 idisk
->disk
.status
|= disk
->status
;
3967 if (is_configured(&idisk
->disk
) ||
3968 is_failed(&idisk
->disk
))
3969 idisk
->disk
.status
&= ~(SPARE_DISK
);
3971 idisk
= xcalloc(1, sizeof(*idisk
));
3972 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3973 idisk
->disk
= *disk
;
3974 idisk
->next
= *disk_list
;
3978 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3985 static struct intel_super
*
3986 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3989 struct imsm_super
*mpb
= super
->anchor
;
3993 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3994 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3995 struct intel_disk
*idisk
;
3997 idisk
= disk_list_get(disk
->serial
, disk_list
);
3999 if (idisk
->owner
== owner
||
4000 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4003 dprintf("%s: '%.16s' owner %d != %d\n",
4004 __func__
, disk
->serial
, idisk
->owner
,
4007 dprintf("%s: unknown disk %x [%d]: %.16s\n",
4008 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
4014 if (ok_count
== mpb
->num_disks
)
4019 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4021 struct intel_super
*s
;
4023 for (s
= super_list
; s
; s
= s
->next
) {
4024 if (family_num
!= s
->anchor
->family_num
)
4026 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
4027 __le32_to_cpu(family_num
), s
->disks
->devname
);
4031 static struct intel_super
*
4032 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4034 struct intel_super
*super_table
[len
];
4035 struct intel_disk
*disk_list
= NULL
;
4036 struct intel_super
*champion
, *spare
;
4037 struct intel_super
*s
, **del
;
4042 memset(super_table
, 0, sizeof(super_table
));
4043 for (s
= *super_list
; s
; s
= s
->next
)
4044 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4046 for (i
= 0; i
< tbl_size
; i
++) {
4047 struct imsm_disk
*d
;
4048 struct intel_disk
*idisk
;
4049 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4052 d
= &s
->disks
->disk
;
4054 /* 'd' must appear in merged disk list for its
4055 * configuration to be valid
4057 idisk
= disk_list_get(d
->serial
, disk_list
);
4058 if (idisk
&& idisk
->owner
== i
)
4059 s
= validate_members(s
, disk_list
, i
);
4064 dprintf("%s: marking family: %#x from %d:%d offline\n",
4065 __func__
, mpb
->family_num
,
4066 super_table
[i
]->disks
->major
,
4067 super_table
[i
]->disks
->minor
);
4071 /* This is where the mdadm implementation differs from the Windows
4072 * driver which has no strict concept of a container. We can only
4073 * assemble one family from a container, so when returning a prodigal
4074 * array member to this system the code will not be able to disambiguate
4075 * the container contents that should be assembled ("foreign" versus
4076 * "local"). It requires user intervention to set the orig_family_num
4077 * to a new value to establish a new container. The Windows driver in
4078 * this situation fixes up the volume name in place and manages the
4079 * foreign array as an independent entity.
4084 for (i
= 0; i
< tbl_size
; i
++) {
4085 struct intel_super
*tbl_ent
= super_table
[i
];
4091 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4096 if (s
&& !is_spare
) {
4097 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4099 } else if (!s
&& !is_spare
)
4112 fprintf(stderr
, "Chose family %#x on '%s', "
4113 "assemble conflicts to new container with '--update=uuid'\n",
4114 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4116 /* collect all dl's onto 'champion', and update them to
4117 * champion's version of the status
4119 for (s
= *super_list
; s
; s
= s
->next
) {
4120 struct imsm_super
*mpb
= champion
->anchor
;
4121 struct dl
*dl
= s
->disks
;
4126 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4128 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4129 struct imsm_disk
*disk
;
4131 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4134 /* only set index on disks that are a member of
4135 * a populated contianer, i.e. one with
4138 if (is_failed(&dl
->disk
))
4140 else if (is_spare(&dl
->disk
))
4146 if (i
>= mpb
->num_disks
) {
4147 struct intel_disk
*idisk
;
4149 idisk
= disk_list_get(dl
->serial
, disk_list
);
4150 if (idisk
&& is_spare(&idisk
->disk
) &&
4151 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4159 dl
->next
= champion
->disks
;
4160 champion
->disks
= dl
;
4164 /* delete 'champion' from super_list */
4165 for (del
= super_list
; *del
; ) {
4166 if (*del
== champion
) {
4167 *del
= (*del
)->next
;
4170 del
= &(*del
)->next
;
4172 champion
->next
= NULL
;
4176 struct intel_disk
*idisk
= disk_list
;
4178 disk_list
= disk_list
->next
;
4187 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4188 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4189 int major
, int minor
, int keep_fd
);
4191 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4192 int *max
, int keep_fd
);
4195 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4196 char *devname
, struct md_list
*devlist
,
4199 struct intel_super
*super_list
= NULL
;
4200 struct intel_super
*super
= NULL
;
4205 /* 'fd' is an opened container */
4206 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4208 /* get super block from devlist devices */
4209 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4212 /* all mpbs enter, maybe one leaves */
4213 super
= imsm_thunderdome(&super_list
, i
);
4219 if (find_missing(super
) != 0) {
4225 /* load migration record */
4226 err
= load_imsm_migr_rec(super
, NULL
);
4228 /* migration is in progress,
4229 * but migr_rec cannot be loaded,
4235 /* Check migration compatibility */
4236 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4237 pr_err("Unsupported migration detected");
4239 fprintf(stderr
, " on %s\n", devname
);
4241 fprintf(stderr
, " (IMSM).\n");
4250 while (super_list
) {
4251 struct intel_super
*s
= super_list
;
4253 super_list
= super_list
->next
;
4263 st
->container_dev
= fd2devnum(fd
);
4265 st
->container_dev
= NoMdDev
;
4266 if (err
== 0 && st
->ss
== NULL
) {
4267 st
->ss
= &super_imsm
;
4268 st
->minor_version
= 0;
4269 st
->max_devs
= IMSM_MAX_DEVICES
;
4276 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4277 int *max
, int keep_fd
)
4279 struct md_list
*tmpdev
;
4283 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4284 if (tmpdev
->used
!= 1)
4286 if (tmpdev
->container
== 1) {
4288 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4290 pr_err("cannot open device %s: %s\n",
4291 tmpdev
->devname
, strerror(errno
));
4295 err
= get_sra_super_block(fd
, super_list
,
4296 tmpdev
->devname
, &lmax
,
4305 int major
= major(tmpdev
->st_rdev
);
4306 int minor
= minor(tmpdev
->st_rdev
);
4307 err
= get_super_block(super_list
,
4324 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4325 int major
, int minor
, int keep_fd
)
4327 struct intel_super
*s
= NULL
;
4339 sprintf(nm
, "%d:%d", major
, minor
);
4340 dfd
= dev_open(nm
, O_RDWR
);
4346 find_intel_hba_capability(dfd
, s
, devname
);
4347 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4349 /* retry the load if we might have raced against mdmon */
4350 if (err
== 3 && (devnum
!= -1) && mdmon_running(devnum
))
4351 for (retry
= 0; retry
< 3; retry
++) {
4353 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4359 s
->next
= *super_list
;
4367 if ((dfd
>= 0) && (!keep_fd
))
4374 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4381 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4385 if (sra
->array
.major_version
!= -1 ||
4386 sra
->array
.minor_version
!= -2 ||
4387 strcmp(sra
->text_version
, "imsm") != 0) {
4392 devnum
= fd2devnum(fd
);
4393 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4394 if (get_super_block(super_list
, devnum
, devname
,
4395 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4406 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4408 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4412 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4414 struct intel_super
*super
;
4417 if (test_partition(fd
))
4418 /* IMSM not allowed on partitions */
4421 free_super_imsm(st
);
4423 super
= alloc_super();
4424 /* Load hba and capabilities if they exist.
4425 * But do not preclude loading metadata in case capabilities or hba are
4426 * non-compliant and ignore_hw_compat is set.
4428 rv
= find_intel_hba_capability(fd
, super
, devname
);
4429 /* no orom/efi or non-intel hba of the disk */
4430 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4432 pr_err("No OROM/EFI properties for %s\n", devname
);
4436 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4440 pr_err("Failed to load all information "
4441 "sections on %s\n", devname
);
4447 if (st
->ss
== NULL
) {
4448 st
->ss
= &super_imsm
;
4449 st
->minor_version
= 0;
4450 st
->max_devs
= IMSM_MAX_DEVICES
;
4453 /* load migration record */
4454 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4455 /* Check for unsupported migration features */
4456 if (check_mpb_migr_compatibility(super
) != 0) {
4457 pr_err("Unsupported migration detected");
4459 fprintf(stderr
, " on %s\n", devname
);
4461 fprintf(stderr
, " (IMSM).\n");
4469 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4471 if (info
->level
== 1)
4473 return info
->chunk_size
>> 9;
4476 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4477 unsigned long long size
)
4479 if (info
->level
== 1)
4482 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4485 static void imsm_update_version_info(struct intel_super
*super
)
4487 /* update the version and attributes */
4488 struct imsm_super
*mpb
= super
->anchor
;
4490 struct imsm_dev
*dev
;
4491 struct imsm_map
*map
;
4494 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4495 dev
= get_imsm_dev(super
, i
);
4496 map
= get_imsm_map(dev
, MAP_0
);
4497 if (__le32_to_cpu(dev
->size_high
) > 0)
4498 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4500 /* FIXME detect when an array spans a port multiplier */
4502 mpb
->attributes
|= MPB_ATTRIB_PM
;
4505 if (mpb
->num_raid_devs
> 1 ||
4506 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4507 version
= MPB_VERSION_ATTRIBS
;
4508 switch (get_imsm_raid_level(map
)) {
4509 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4510 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4511 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4512 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4515 if (map
->num_members
>= 5)
4516 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4517 else if (dev
->status
== DEV_CLONE_N_GO
)
4518 version
= MPB_VERSION_CNG
;
4519 else if (get_imsm_raid_level(map
) == 5)
4520 version
= MPB_VERSION_RAID5
;
4521 else if (map
->num_members
>= 3)
4522 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4523 else if (get_imsm_raid_level(map
) == 1)
4524 version
= MPB_VERSION_RAID1
;
4526 version
= MPB_VERSION_RAID0
;
4528 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4532 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4534 struct imsm_super
*mpb
= super
->anchor
;
4535 char *reason
= NULL
;
4538 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4539 reason
= "must be 16 characters or less";
4541 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4542 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4544 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4545 reason
= "already exists";
4550 if (reason
&& !quiet
)
4551 pr_err("imsm volume name %s\n", reason
);
4556 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4557 unsigned long long size
, char *name
,
4558 char *homehost
, int *uuid
,
4559 long long data_offset
)
4561 /* We are creating a volume inside a pre-existing container.
4562 * so st->sb is already set.
4564 struct intel_super
*super
= st
->sb
;
4565 struct imsm_super
*mpb
= super
->anchor
;
4566 struct intel_dev
*dv
;
4567 struct imsm_dev
*dev
;
4568 struct imsm_vol
*vol
;
4569 struct imsm_map
*map
;
4570 int idx
= mpb
->num_raid_devs
;
4572 unsigned long long array_blocks
;
4573 size_t size_old
, size_new
;
4574 unsigned long long num_data_stripes
;
4576 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4577 pr_err("This imsm-container already has the "
4578 "maximum of %d volumes\n", super
->orom
->vpa
);
4582 /* ensure the mpb is large enough for the new data */
4583 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4584 size_new
= disks_to_mpb_size(info
->nr_disks
);
4585 if (size_new
> size_old
) {
4587 size_t size_round
= ROUND_UP(size_new
, 512);
4589 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4590 pr_err("could not allocate new mpb\n");
4593 if (posix_memalign(&super
->migr_rec_buf
, 512,
4594 MIGR_REC_BUF_SIZE
) != 0) {
4595 pr_err("%s could not allocate migr_rec buffer\n",
4602 memcpy(mpb_new
, mpb
, size_old
);
4605 super
->anchor
= mpb_new
;
4606 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4607 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4609 super
->current_vol
= idx
;
4611 /* handle 'failed_disks' by either:
4612 * a) create dummy disk entries in the table if this the first
4613 * volume in the array. We add them here as this is the only
4614 * opportunity to add them. add_to_super_imsm_volume()
4615 * handles the non-failed disks and continues incrementing
4617 * b) validate that 'failed_disks' matches the current number
4618 * of missing disks if the container is populated
4620 if (super
->current_vol
== 0) {
4622 for (i
= 0; i
< info
->failed_disks
; i
++) {
4623 struct imsm_disk
*disk
;
4626 disk
= __get_imsm_disk(mpb
, i
);
4627 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4628 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4629 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4632 find_missing(super
);
4637 for (d
= super
->missing
; d
; d
= d
->next
)
4639 if (info
->failed_disks
> missing
) {
4640 pr_err("unable to add 'missing' disk to container\n");
4645 if (!check_name(super
, name
, 0))
4647 dv
= xmalloc(sizeof(*dv
));
4648 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4649 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4650 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4651 info
->layout
, info
->chunk_size
,
4653 /* round array size down to closest MB */
4654 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4656 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4657 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4658 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4660 vol
->migr_state
= 0;
4661 set_migr_type(dev
, MIGR_INIT
);
4662 vol
->dirty
= !info
->state
;
4663 vol
->curr_migr_unit
= 0;
4664 map
= get_imsm_map(dev
, MAP_0
);
4665 set_pba_of_lba0(map
, super
->create_offset
);
4666 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4667 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4668 map
->failed_disk_num
= ~0;
4669 if (info
->level
> 0)
4670 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
4671 : IMSM_T_STATE_UNINITIALIZED
);
4673 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4674 IMSM_T_STATE_NORMAL
;
4677 if (info
->level
== 1 && info
->raid_disks
> 2) {
4680 pr_err("imsm does not support more than 2 disks"
4681 "in a raid1 volume\n");
4685 map
->raid_level
= info
->level
;
4686 if (info
->level
== 10) {
4687 map
->raid_level
= 1;
4688 map
->num_domains
= info
->raid_disks
/ 2;
4689 } else if (info
->level
== 1)
4690 map
->num_domains
= info
->raid_disks
;
4692 map
->num_domains
= 1;
4694 /* info->size is only int so use the 'size' parameter instead */
4695 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4696 num_data_stripes
/= map
->num_domains
;
4697 set_num_data_stripes(map
, num_data_stripes
);
4699 map
->num_members
= info
->raid_disks
;
4700 for (i
= 0; i
< map
->num_members
; i
++) {
4701 /* initialized in add_to_super */
4702 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4704 mpb
->num_raid_devs
++;
4707 dv
->index
= super
->current_vol
;
4708 dv
->next
= super
->devlist
;
4709 super
->devlist
= dv
;
4711 imsm_update_version_info(super
);
4716 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4717 unsigned long long size
, char *name
,
4718 char *homehost
, int *uuid
,
4719 unsigned long long data_offset
)
4721 /* This is primarily called by Create when creating a new array.
4722 * We will then get add_to_super called for each component, and then
4723 * write_init_super called to write it out to each device.
4724 * For IMSM, Create can create on fresh devices or on a pre-existing
4726 * To create on a pre-existing array a different method will be called.
4727 * This one is just for fresh drives.
4729 struct intel_super
*super
;
4730 struct imsm_super
*mpb
;
4734 if (data_offset
!= INVALID_SECTORS
) {
4735 fprintf(stderr
, Name
": data-offset not supported by imsm\n");
4740 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
,
4744 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4748 super
= alloc_super();
4749 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4754 pr_err("%s could not allocate superblock\n", __func__
);
4757 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4758 pr_err("%s could not allocate migr_rec buffer\n", __func__
);
4763 memset(super
->buf
, 0, mpb_size
);
4765 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4769 /* zeroing superblock */
4773 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4775 version
= (char *) mpb
->sig
;
4776 strcpy(version
, MPB_SIGNATURE
);
4777 version
+= strlen(MPB_SIGNATURE
);
4778 strcpy(version
, MPB_VERSION_RAID0
);
4784 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4785 int fd
, char *devname
)
4787 struct intel_super
*super
= st
->sb
;
4788 struct imsm_super
*mpb
= super
->anchor
;
4789 struct imsm_disk
*_disk
;
4790 struct imsm_dev
*dev
;
4791 struct imsm_map
*map
;
4795 dev
= get_imsm_dev(super
, super
->current_vol
);
4796 map
= get_imsm_map(dev
, MAP_0
);
4798 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4799 pr_err("%s: Cannot add spare devices to IMSM volume\n",
4805 /* we're doing autolayout so grab the pre-marked (in
4806 * validate_geometry) raid_disk
4808 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4809 if (dl
->raiddisk
== dk
->raid_disk
)
4812 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4813 if (dl
->major
== dk
->major
&&
4814 dl
->minor
== dk
->minor
)
4819 pr_err("%s is not a member of the same container\n", devname
);
4823 /* add a pristine spare to the metadata */
4824 if (dl
->index
< 0) {
4825 dl
->index
= super
->anchor
->num_disks
;
4826 super
->anchor
->num_disks
++;
4828 /* Check the device has not already been added */
4829 slot
= get_imsm_disk_slot(map
, dl
->index
);
4831 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4832 pr_err("%s has been included in this array twice\n",
4836 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4837 dl
->disk
.status
= CONFIGURED_DISK
;
4839 /* update size of 'missing' disks to be at least as large as the
4840 * largest acitve member (we only have dummy missing disks when
4841 * creating the first volume)
4843 if (super
->current_vol
== 0) {
4844 for (df
= super
->missing
; df
; df
= df
->next
) {
4845 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
4846 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
4847 _disk
= __get_imsm_disk(mpb
, df
->index
);
4852 /* refresh unset/failed slots to point to valid 'missing' entries */
4853 for (df
= super
->missing
; df
; df
= df
->next
)
4854 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4855 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4857 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4859 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4860 if (is_gen_migration(dev
)) {
4861 struct imsm_map
*map2
= get_imsm_map(dev
,
4863 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4864 if ((slot2
< map2
->num_members
) &&
4866 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4869 if ((unsigned)df
->index
==
4871 set_imsm_ord_tbl_ent(map2
,
4877 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4881 /* if we are creating the first raid device update the family number */
4882 if (super
->current_vol
== 0) {
4884 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4886 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4887 if (!_dev
|| !_disk
) {
4888 pr_err("BUG mpb setup error\n");
4894 sum
+= __gen_imsm_checksum(mpb
);
4895 mpb
->family_num
= __cpu_to_le32(sum
);
4896 mpb
->orig_family_num
= mpb
->family_num
;
4898 super
->current_disk
= dl
;
4903 * Function marks disk as spare and restores disk serial
4904 * in case it was previously marked as failed by takeover operation
4906 * -1 : critical error
4907 * 0 : disk is marked as spare but serial is not set
4910 int mark_spare(struct dl
*disk
)
4912 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4919 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4920 /* Restore disk serial number, because takeover marks disk
4921 * as failed and adds to serial ':0' before it becomes
4924 serialcpy(disk
->serial
, serial
);
4925 serialcpy(disk
->disk
.serial
, serial
);
4928 disk
->disk
.status
= SPARE_DISK
;
4934 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4935 int fd
, char *devname
,
4936 unsigned long long data_offset
)
4938 struct intel_super
*super
= st
->sb
;
4940 unsigned long long size
;
4945 /* If we are on an RAID enabled platform check that the disk is
4946 * attached to the raid controller.
4947 * We do not need to test disks attachment for container based additions,
4948 * they shall be already tested when container was created/assembled.
4950 rv
= find_intel_hba_capability(fd
, super
, devname
);
4951 /* no orom/efi or non-intel hba of the disk */
4953 dprintf("capability: %p fd: %d ret: %d\n",
4954 super
->orom
, fd
, rv
);
4958 if (super
->current_vol
>= 0)
4959 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4962 dd
= xcalloc(sizeof(*dd
), 1);
4963 dd
->major
= major(stb
.st_rdev
);
4964 dd
->minor
= minor(stb
.st_rdev
);
4965 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
4968 dd
->action
= DISK_ADD
;
4969 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4971 pr_err("failed to retrieve scsi serial, aborting\n");
4976 get_dev_size(fd
, NULL
, &size
);
4978 serialcpy(dd
->disk
.serial
, dd
->serial
);
4979 set_total_blocks(&dd
->disk
, size
);
4980 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
4981 struct imsm_super
*mpb
= super
->anchor
;
4982 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
4985 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4986 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4988 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4990 if (st
->update_tail
) {
4991 dd
->next
= super
->disk_mgmt_list
;
4992 super
->disk_mgmt_list
= dd
;
4994 dd
->next
= super
->disks
;
4996 super
->updates_pending
++;
5003 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5005 struct intel_super
*super
= st
->sb
;
5008 /* remove from super works only in mdmon - for communication
5009 * manager - monitor. Check if communication memory buffer
5012 if (!st
->update_tail
) {
5013 pr_err("%s shall be used in mdmon context only"
5014 "(line %d).\n", __func__
, __LINE__
);
5017 dd
= xcalloc(1, sizeof(*dd
));
5018 dd
->major
= dk
->major
;
5019 dd
->minor
= dk
->minor
;
5022 dd
->action
= DISK_REMOVE
;
5024 dd
->next
= super
->disk_mgmt_list
;
5025 super
->disk_mgmt_list
= dd
;
5031 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5035 struct imsm_super anchor
;
5036 } spare_record
__attribute__ ((aligned(512)));
5038 /* spare records have their own family number and do not have any defined raid
5041 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5043 struct imsm_super
*mpb
= super
->anchor
;
5044 struct imsm_super
*spare
= &spare_record
.anchor
;
5048 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
5049 spare
->generation_num
= __cpu_to_le32(1UL),
5050 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5051 spare
->num_disks
= 1,
5052 spare
->num_raid_devs
= 0,
5053 spare
->cache_size
= mpb
->cache_size
,
5054 spare
->pwr_cycle_count
= __cpu_to_le32(1),
5056 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5057 MPB_SIGNATURE MPB_VERSION_RAID0
);
5059 for (d
= super
->disks
; d
; d
= d
->next
) {
5063 spare
->disk
[0] = d
->disk
;
5064 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5065 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5067 sum
= __gen_imsm_checksum(spare
);
5068 spare
->family_num
= __cpu_to_le32(sum
);
5069 spare
->orig_family_num
= 0;
5070 sum
= __gen_imsm_checksum(spare
);
5071 spare
->check_sum
= __cpu_to_le32(sum
);
5073 if (store_imsm_mpb(d
->fd
, spare
)) {
5074 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
5075 __func__
, d
->major
, d
->minor
, strerror(errno
));
5087 static int write_super_imsm(struct supertype
*st
, int doclose
)
5089 struct intel_super
*super
= st
->sb
;
5090 struct imsm_super
*mpb
= super
->anchor
;
5096 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5098 int clear_migration_record
= 1;
5100 /* 'generation' is incremented everytime the metadata is written */
5101 generation
= __le32_to_cpu(mpb
->generation_num
);
5103 mpb
->generation_num
= __cpu_to_le32(generation
);
5105 /* fix up cases where previous mdadm releases failed to set
5108 if (mpb
->orig_family_num
== 0)
5109 mpb
->orig_family_num
= mpb
->family_num
;
5111 for (d
= super
->disks
; d
; d
= d
->next
) {
5115 mpb
->disk
[d
->index
] = d
->disk
;
5119 for (d
= super
->missing
; d
; d
= d
->next
) {
5120 mpb
->disk
[d
->index
] = d
->disk
;
5123 mpb
->num_disks
= num_disks
;
5124 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5126 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5127 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5128 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5130 imsm_copy_dev(dev
, dev2
);
5131 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5133 if (is_gen_migration(dev2
))
5134 clear_migration_record
= 0;
5136 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5137 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5139 /* recalculate checksum */
5140 sum
= __gen_imsm_checksum(mpb
);
5141 mpb
->check_sum
= __cpu_to_le32(sum
);
5143 if (super
->clean_migration_record_by_mdmon
) {
5144 clear_migration_record
= 1;
5145 super
->clean_migration_record_by_mdmon
= 0;
5147 if (clear_migration_record
)
5148 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5150 /* write the mpb for disks that compose raid devices */
5151 for (d
= super
->disks
; d
; d
= d
->next
) {
5152 if (d
->index
< 0 || is_failed(&d
->disk
))
5155 if (clear_migration_record
) {
5156 unsigned long long dsize
;
5158 get_dev_size(d
->fd
, NULL
, &dsize
);
5159 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5160 if (write(d
->fd
, super
->migr_rec_buf
,
5161 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5162 perror("Write migr_rec failed");
5166 if (store_imsm_mpb(d
->fd
, mpb
))
5168 "%s: failed for device %d:%d (fd: %d)%s\n",
5169 __func__
, d
->major
, d
->minor
,
5170 d
->fd
, strerror(errno
));
5179 return write_super_imsm_spares(super
, doclose
);
5185 static int create_array(struct supertype
*st
, int dev_idx
)
5188 struct imsm_update_create_array
*u
;
5189 struct intel_super
*super
= st
->sb
;
5190 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5191 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5192 struct disk_info
*inf
;
5193 struct imsm_disk
*disk
;
5196 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5197 sizeof(*inf
) * map
->num_members
;
5199 u
->type
= update_create_array
;
5200 u
->dev_idx
= dev_idx
;
5201 imsm_copy_dev(&u
->dev
, dev
);
5202 inf
= get_disk_info(u
);
5203 for (i
= 0; i
< map
->num_members
; i
++) {
5204 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5206 disk
= get_imsm_disk(super
, idx
);
5207 serialcpy(inf
[i
].serial
, disk
->serial
);
5209 append_metadata_update(st
, u
, len
);
5214 static int mgmt_disk(struct supertype
*st
)
5216 struct intel_super
*super
= st
->sb
;
5218 struct imsm_update_add_remove_disk
*u
;
5220 if (!super
->disk_mgmt_list
)
5225 u
->type
= update_add_remove_disk
;
5226 append_metadata_update(st
, u
, len
);
5231 static int write_init_super_imsm(struct supertype
*st
)
5233 struct intel_super
*super
= st
->sb
;
5234 int current_vol
= super
->current_vol
;
5236 /* we are done with current_vol reset it to point st at the container */
5237 super
->current_vol
= -1;
5239 if (st
->update_tail
) {
5240 /* queue the recently created array / added disk
5241 * as a metadata update */
5244 /* determine if we are creating a volume or adding a disk */
5245 if (current_vol
< 0) {
5246 /* in the mgmt (add/remove) disk case we are running
5247 * in mdmon context, so don't close fd's
5249 return mgmt_disk(st
);
5251 rv
= create_array(st
, current_vol
);
5256 for (d
= super
->disks
; d
; d
= d
->next
)
5257 Kill(d
->devname
, NULL
, 0, -1, 1);
5258 return write_super_imsm(st
, 1);
5263 static int store_super_imsm(struct supertype
*st
, int fd
)
5265 struct intel_super
*super
= st
->sb
;
5266 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5272 return store_imsm_mpb(fd
, mpb
);
5278 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5280 return __le32_to_cpu(mpb
->bbm_log_size
);
5284 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5285 int layout
, int raiddisks
, int chunk
,
5286 unsigned long long size
,
5287 unsigned long long data_offset
,
5289 unsigned long long *freesize
,
5293 unsigned long long ldsize
;
5294 struct intel_super
*super
=NULL
;
5297 if (level
!= LEVEL_CONTAINER
)
5302 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5305 pr_err("imsm: Cannot open %s: %s\n",
5306 dev
, strerror(errno
));
5309 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5314 /* capabilities retrieve could be possible
5315 * note that there is no fd for the disks in array.
5317 super
= alloc_super();
5318 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
5322 fd2devname(fd
, str
);
5323 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5324 fd
, str
, super
->orom
, rv
, raiddisks
);
5326 /* no orom/efi or non-intel hba of the disk */
5333 if (raiddisks
> super
->orom
->tds
) {
5335 pr_err("%d exceeds maximum number of"
5336 " platform supported disks: %d\n",
5337 raiddisks
, super
->orom
->tds
);
5341 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5342 (ldsize
>> 9) >> 32 > 0) {
5344 pr_err("%s exceeds maximum platform supported size\n", dev
);
5350 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
5356 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5358 const unsigned long long base_start
= e
[*idx
].start
;
5359 unsigned long long end
= base_start
+ e
[*idx
].size
;
5362 if (base_start
== end
)
5366 for (i
= *idx
; i
< num_extents
; i
++) {
5367 /* extend overlapping extents */
5368 if (e
[i
].start
>= base_start
&&
5369 e
[i
].start
<= end
) {
5372 if (e
[i
].start
+ e
[i
].size
> end
)
5373 end
= e
[i
].start
+ e
[i
].size
;
5374 } else if (e
[i
].start
> end
) {
5380 return end
- base_start
;
5383 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5385 /* build a composite disk with all known extents and generate a new
5386 * 'maxsize' given the "all disks in an array must share a common start
5387 * offset" constraint
5389 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
5393 unsigned long long pos
;
5394 unsigned long long start
= 0;
5395 unsigned long long maxsize
;
5396 unsigned long reserve
;
5398 /* coalesce and sort all extents. also, check to see if we need to
5399 * reserve space between member arrays
5402 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5405 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5408 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5413 while (i
< sum_extents
) {
5414 e
[j
].start
= e
[i
].start
;
5415 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5417 if (e
[j
-1].size
== 0)
5426 unsigned long long esize
;
5428 esize
= e
[i
].start
- pos
;
5429 if (esize
>= maxsize
) {
5434 pos
= e
[i
].start
+ e
[i
].size
;
5436 } while (e
[i
-1].size
);
5442 /* FIXME assumes volume at offset 0 is the first volume in a
5445 if (start_extent
> 0)
5446 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5450 if (maxsize
< reserve
)
5453 super
->create_offset
= ~((unsigned long long) 0);
5454 if (start
+ reserve
> super
->create_offset
)
5455 return 0; /* start overflows create_offset */
5456 super
->create_offset
= start
+ reserve
;
5458 return maxsize
- reserve
;
5461 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5463 if (level
< 0 || level
== 6 || level
== 4)
5466 /* if we have an orom prevent invalid raid levels */
5469 case 0: return imsm_orom_has_raid0(orom
);
5472 return imsm_orom_has_raid1e(orom
);
5473 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5474 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5475 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5478 return 1; /* not on an Intel RAID platform so anything goes */
5485 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5486 int dpa
, int verbose
)
5488 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5489 struct mdstat_ent
*memb
= NULL
;
5492 struct md_list
*dv
= NULL
;
5495 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5496 if (memb
->metadata_version
&&
5497 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5498 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5499 !is_subarray(memb
->metadata_version
+9) &&
5501 struct dev_member
*dev
= memb
->members
;
5503 while(dev
&& (fd
< 0)) {
5504 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5505 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5507 fd
= open(path
, O_RDONLY
, 0);
5508 if ((num
<= 0) || (fd
< 0)) {
5509 pr_vrb(": Cannot open %s: %s\n",
5510 dev
->name
, strerror(errno
));
5516 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5517 struct mdstat_ent
*vol
;
5518 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5519 if ((vol
->active
> 0) &&
5520 vol
->metadata_version
&&
5521 is_container_member(vol
, memb
->dev
)) {
5526 if (*devlist
&& (found
< dpa
)) {
5527 dv
= xcalloc(1, sizeof(*dv
));
5528 dv
->devname
= xmalloc(strlen(memb
->dev
) + strlen("/dev/") + 1);
5529 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->dev
);
5532 dv
->next
= *devlist
;
5540 free_mdstat(mdstat
);
5545 static struct md_list
*
5546 get_loop_devices(void)
5549 struct md_list
*devlist
= NULL
;
5550 struct md_list
*dv
= NULL
;
5552 for(i
= 0; i
< 12; i
++) {
5553 dv
= xcalloc(1, sizeof(*dv
));
5554 dv
->devname
= xmalloc(40);
5555 sprintf(dv
->devname
, "/dev/loop%d", i
);
5563 static struct md_list
*
5564 get_devices(const char *hba_path
)
5566 struct md_list
*devlist
= NULL
;
5567 struct md_list
*dv
= NULL
;
5573 devlist
= get_loop_devices();
5576 /* scroll through /sys/dev/block looking for devices attached to
5579 dir
= opendir("/sys/dev/block");
5580 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5585 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5587 path
= devt_to_devpath(makedev(major
, minor
));
5590 if (!path_attached_to_hba(path
, hba_path
)) {
5597 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5599 fd2devname(fd
, buf
);
5602 pr_err("cannot open device: %s\n",
5608 dv
= xcalloc(1, sizeof(*dv
));
5609 dv
->devname
= xstrdup(buf
);
5616 devlist
= devlist
->next
;
5626 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5627 int verbose
, int *found
)
5629 struct md_list
*tmpdev
;
5631 struct supertype
*st
= NULL
;
5633 /* first walk the list of devices to find a consistent set
5634 * that match the criterea, if that is possible.
5635 * We flag the ones we like with 'used'.
5638 st
= match_metadata_desc_imsm("imsm");
5640 pr_vrb(": cannot allocate memory for imsm supertype\n");
5644 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5645 char *devname
= tmpdev
->devname
;
5647 struct supertype
*tst
;
5649 if (tmpdev
->used
> 1)
5651 tst
= dup_super(st
);
5653 pr_vrb(": cannot allocate memory for imsm supertype\n");
5656 tmpdev
->container
= 0;
5657 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5659 dprintf(": cannot open device %s: %s\n",
5660 devname
, strerror(errno
));
5662 } else if (fstat(dfd
, &stb
)< 0) {
5664 dprintf(": fstat failed for %s: %s\n",
5665 devname
, strerror(errno
));
5667 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5668 dprintf(": %s is not a block device.\n",
5671 } else if (must_be_container(dfd
)) {
5672 struct supertype
*cst
;
5673 cst
= super_by_fd(dfd
, NULL
);
5675 dprintf(": cannot recognize container type %s\n",
5678 } else if (tst
->ss
!= st
->ss
) {
5679 dprintf(": non-imsm container - ignore it: %s\n",
5682 } else if (!tst
->ss
->load_container
||
5683 tst
->ss
->load_container(tst
, dfd
, NULL
))
5686 tmpdev
->container
= 1;
5689 cst
->ss
->free_super(cst
);
5691 tmpdev
->st_rdev
= stb
.st_rdev
;
5692 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5693 dprintf(": no RAID superblock on %s\n",
5696 } else if (tst
->ss
->compare_super
== NULL
) {
5697 dprintf(": Cannot assemble %s metadata on %s\n",
5698 tst
->ss
->name
, devname
);
5704 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5705 /* Ignore unrecognised devices during auto-assembly */
5710 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5712 if (st
->minor_version
== -1)
5713 st
->minor_version
= tst
->minor_version
;
5715 if (memcmp(info
.uuid
, uuid_zero
,
5716 sizeof(int[4])) == 0) {
5717 /* this is a floating spare. It cannot define
5718 * an array unless there are no more arrays of
5719 * this type to be found. It can be included
5720 * in an array of this type though.
5726 if (st
->ss
!= tst
->ss
||
5727 st
->minor_version
!= tst
->minor_version
||
5728 st
->ss
->compare_super(st
, tst
) != 0) {
5729 /* Some mismatch. If exactly one array matches this host,
5730 * we can resolve on that one.
5731 * Or, if we are auto assembling, we just ignore the second
5734 dprintf(": superblock on %s doesn't match others - assembly aborted\n",
5740 dprintf("found: devname: %s\n", devname
);
5744 tst
->ss
->free_super(tst
);
5748 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5749 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5750 for (iter
= head
; iter
; iter
= iter
->next
) {
5751 dprintf("content->text_version: %s vol\n",
5752 iter
->text_version
);
5753 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5754 /* do not assemble arrays with unsupported
5756 dprintf(": Cannot activate member %s.\n",
5757 iter
->text_version
);
5764 dprintf(" no valid super block on device list: err: %d %p\n",
5768 dprintf(" no more devices to examin\n");
5771 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5772 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5774 if (count
< tmpdev
->found
)
5777 count
-= tmpdev
->found
;
5780 if (tmpdev
->used
== 1)
5785 st
->ss
->free_super(st
);
5791 count_volumes(char *hba
, int dpa
, int verbose
)
5793 struct md_list
*devlist
= NULL
;
5797 devlist
= get_devices(hba
);
5798 /* if no intel devices return zero volumes */
5799 if (devlist
== NULL
)
5802 count
= active_arrays_by_format("imsm", hba
, &devlist
, dpa
, verbose
);
5803 dprintf(" path: %s active arrays: %d\n", hba
, count
);
5804 if (devlist
== NULL
)
5808 count
+= count_volumes_list(devlist
,
5812 dprintf("found %d count: %d\n", found
, count
);
5815 dprintf("path: %s total number of volumes: %d\n", hba
, count
);
5818 struct md_list
*dv
= devlist
;
5819 devlist
= devlist
->next
;
5826 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5828 /* up to 512 if the plaform supports it, otherwise the platform max.
5829 * 128 if no platform detected
5831 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5833 return min(512, (1 << fs
));
5837 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5838 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
5840 /* check/set platform and metadata limits/defaults */
5841 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5842 pr_vrb(": platform supports a maximum of %d disks per array\n",
5847 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5848 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5849 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5850 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5854 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5855 *chunk
= imsm_default_chunk(super
->orom
);
5857 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5858 pr_vrb(": platform does not support a chunk size of: "
5863 if (layout
!= imsm_level_to_layout(level
)) {
5865 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5866 else if (level
== 10)
5867 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5869 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5874 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 && chunk
&&
5875 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
5876 pr_vrb(": platform does not support a volume size over 2TB\n");
5882 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5883 * FIX ME add ahci details
5885 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5886 int layout
, int raiddisks
, int *chunk
,
5887 unsigned long long size
,
5888 unsigned long long data_offset
,
5890 unsigned long long *freesize
,
5894 struct intel_super
*super
= st
->sb
;
5895 struct imsm_super
*mpb
;
5897 unsigned long long pos
= 0;
5898 unsigned long long maxsize
;
5902 /* We must have the container info already read in. */
5906 mpb
= super
->anchor
;
5908 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
5909 pr_err("RAID gemetry validation failed. "
5910 "Cannot proceed with the action(s).\n");
5914 /* General test: make sure there is space for
5915 * 'raiddisks' device extents of size 'size' at a given
5918 unsigned long long minsize
= size
;
5919 unsigned long long start_offset
= MaxSector
;
5922 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5923 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5928 e
= get_extents(super
, dl
);
5931 unsigned long long esize
;
5932 esize
= e
[i
].start
- pos
;
5933 if (esize
>= minsize
)
5935 if (found
&& start_offset
== MaxSector
) {
5938 } else if (found
&& pos
!= start_offset
) {
5942 pos
= e
[i
].start
+ e
[i
].size
;
5944 } while (e
[i
-1].size
);
5949 if (dcnt
< raiddisks
) {
5951 pr_err("imsm: Not enough "
5952 "devices with space for this array "
5960 /* This device must be a member of the set */
5961 if (stat(dev
, &stb
) < 0)
5963 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5965 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5966 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5967 dl
->minor
== (int)minor(stb
.st_rdev
))
5972 pr_err("%s is not in the "
5973 "same imsm set\n", dev
);
5975 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5976 /* If a volume is present then the current creation attempt
5977 * cannot incorporate new spares because the orom may not
5978 * understand this configuration (all member disks must be
5979 * members of each array in the container).
5981 pr_err("%s is a spare and a volume"
5982 " is already defined for this container\n", dev
);
5983 pr_err("The option-rom requires all member"
5984 " disks to be a member of all volumes\n");
5986 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
5987 mpb
->num_disks
!= raiddisks
) {
5988 pr_err("The option-rom requires all member"
5989 " disks to be a member of all volumes\n");
5993 /* retrieve the largest free space block */
5994 e
= get_extents(super
, dl
);
5999 unsigned long long esize
;
6001 esize
= e
[i
].start
- pos
;
6002 if (esize
>= maxsize
)
6004 pos
= e
[i
].start
+ e
[i
].size
;
6006 } while (e
[i
-1].size
);
6011 pr_err("unable to determine free space for: %s\n",
6015 if (maxsize
< size
) {
6017 pr_err("%s not enough space (%llu < %llu)\n",
6018 dev
, maxsize
, size
);
6022 /* count total number of extents for merge */
6024 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6026 i
+= dl
->extent_cnt
;
6028 maxsize
= merge_extents(super
, i
);
6030 if (!check_env("IMSM_NO_PLATFORM") &&
6031 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6032 pr_err("attempting to create a second "
6033 "volume with size less then remaining space. "
6038 if (maxsize
< size
|| maxsize
== 0) {
6041 pr_err("no free space"
6042 " left on device. Aborting...\n");
6044 pr_err("not enough space"
6045 " to create volume of given size"
6046 " (%llu < %llu). Aborting...\n",
6052 *freesize
= maxsize
;
6055 int count
= count_volumes(super
->hba
->path
,
6056 super
->orom
->dpa
, verbose
);
6057 if (super
->orom
->vphba
<= count
) {
6058 pr_vrb(": platform does not support more than %d raid volumes.\n",
6059 super
->orom
->vphba
);
6066 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6067 unsigned long long size
, int chunk
,
6068 unsigned long long *freesize
)
6070 struct intel_super
*super
= st
->sb
;
6071 struct imsm_super
*mpb
= super
->anchor
;
6076 unsigned long long maxsize
;
6077 unsigned long long minsize
;
6081 /* find the largest common start free region of the possible disks */
6085 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6091 /* don't activate new spares if we are orom constrained
6092 * and there is already a volume active in the container
6094 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6097 e
= get_extents(super
, dl
);
6100 for (i
= 1; e
[i
-1].size
; i
++)
6108 maxsize
= merge_extents(super
, extent_cnt
);
6112 minsize
= chunk
* 2;
6114 if (cnt
< raiddisks
||
6115 (super
->orom
&& used
&& used
!= raiddisks
) ||
6116 maxsize
< minsize
||
6118 pr_err("not enough devices with space to create array.\n");
6119 return 0; /* No enough free spaces large enough */
6130 if (!check_env("IMSM_NO_PLATFORM") &&
6131 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6132 pr_err("attempting to create a second "
6133 "volume with size less then remaining space. "
6138 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6140 dl
->raiddisk
= cnt
++;
6144 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6149 static int reserve_space(struct supertype
*st
, int raiddisks
,
6150 unsigned long long size
, int chunk
,
6151 unsigned long long *freesize
)
6153 struct intel_super
*super
= st
->sb
;
6158 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6161 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6163 dl
->raiddisk
= cnt
++;
6170 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6171 int raiddisks
, int *chunk
, unsigned long long size
,
6172 unsigned long long data_offset
,
6173 char *dev
, unsigned long long *freesize
,
6181 * if given unused devices create a container
6182 * if given given devices in a container create a member volume
6184 if (level
== LEVEL_CONTAINER
) {
6185 /* Must be a fresh device to add to a container */
6186 return validate_geometry_imsm_container(st
, level
, layout
,
6196 struct intel_super
*super
= st
->sb
;
6197 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6198 raiddisks
, chunk
, size
,
6201 /* we are being asked to automatically layout a
6202 * new volume based on the current contents of
6203 * the container. If the the parameters can be
6204 * satisfied reserve_space will record the disks,
6205 * start offset, and size of the volume to be
6206 * created. add_to_super and getinfo_super
6207 * detect when autolayout is in progress.
6209 /* assuming that freesize is always given when array is
6211 if (super
->orom
&& freesize
) {
6213 count
= count_volumes(super
->hba
->path
,
6214 super
->orom
->dpa
, verbose
);
6215 if (super
->orom
->vphba
<= count
) {
6216 pr_vrb(": platform does not support more"
6217 " than %d raid volumes.\n",
6218 super
->orom
->vphba
);
6223 return reserve_space(st
, raiddisks
, size
,
6224 chunk
?*chunk
:0, freesize
);
6229 /* creating in a given container */
6230 return validate_geometry_imsm_volume(st
, level
, layout
,
6231 raiddisks
, chunk
, size
,
6233 dev
, freesize
, verbose
);
6236 /* This device needs to be a device in an 'imsm' container */
6237 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6240 pr_err("Cannot create this array on device %s\n",
6245 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6247 pr_err("Cannot open %s: %s\n",
6248 dev
, strerror(errno
));
6251 /* Well, it is in use by someone, maybe an 'imsm' container. */
6252 cfd
= open_container(fd
);
6256 pr_err("Cannot use %s: It is busy\n",
6260 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
6261 if (sra
&& sra
->array
.major_version
== -1 &&
6262 strcmp(sra
->text_version
, "imsm") == 0)
6266 /* This is a member of a imsm container. Load the container
6267 * and try to create a volume
6269 struct intel_super
*super
;
6271 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6273 st
->container_dev
= fd2devnum(cfd
);
6275 return validate_geometry_imsm_volume(st
, level
, layout
,
6277 size
, data_offset
, dev
,
6284 pr_err("failed container membership check\n");
6290 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6292 struct intel_super
*super
= st
->sb
;
6294 if (level
&& *level
== UnSet
)
6295 *level
= LEVEL_CONTAINER
;
6297 if (level
&& layout
&& *layout
== UnSet
)
6298 *layout
= imsm_level_to_layout(*level
);
6300 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6301 *chunk
= imsm_default_chunk(super
->orom
);
6304 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6306 static int kill_subarray_imsm(struct supertype
*st
)
6308 /* remove the subarray currently referenced by ->current_vol */
6310 struct intel_dev
**dp
;
6311 struct intel_super
*super
= st
->sb
;
6312 __u8 current_vol
= super
->current_vol
;
6313 struct imsm_super
*mpb
= super
->anchor
;
6315 if (super
->current_vol
< 0)
6317 super
->current_vol
= -1; /* invalidate subarray cursor */
6319 /* block deletions that would change the uuid of active subarrays
6321 * FIXME when immutable ids are available, but note that we'll
6322 * also need to fixup the invalidated/active subarray indexes in
6325 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6328 if (i
< current_vol
)
6330 sprintf(subarray
, "%u", i
);
6331 if (is_subarray_active(subarray
, st
->devname
)) {
6332 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6339 if (st
->update_tail
) {
6340 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
6342 u
->type
= update_kill_array
;
6343 u
->dev_idx
= current_vol
;
6344 append_metadata_update(st
, u
, sizeof(*u
));
6349 for (dp
= &super
->devlist
; *dp
;)
6350 if ((*dp
)->index
== current_vol
) {
6353 handle_missing(super
, (*dp
)->dev
);
6354 if ((*dp
)->index
> current_vol
)
6359 /* no more raid devices, all active components are now spares,
6360 * but of course failed are still failed
6362 if (--mpb
->num_raid_devs
== 0) {
6365 for (d
= super
->disks
; d
; d
= d
->next
)
6370 super
->updates_pending
++;
6375 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6376 char *update
, struct mddev_ident
*ident
)
6378 /* update the subarray currently referenced by ->current_vol */
6379 struct intel_super
*super
= st
->sb
;
6380 struct imsm_super
*mpb
= super
->anchor
;
6382 if (strcmp(update
, "name") == 0) {
6383 char *name
= ident
->name
;
6387 if (is_subarray_active(subarray
, st
->devname
)) {
6388 pr_err("Unable to update name of active subarray\n");
6392 if (!check_name(super
, name
, 0))
6395 vol
= strtoul(subarray
, &ep
, 10);
6396 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6399 if (st
->update_tail
) {
6400 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
6402 u
->type
= update_rename_array
;
6404 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6405 append_metadata_update(st
, u
, sizeof(*u
));
6407 struct imsm_dev
*dev
;
6410 dev
= get_imsm_dev(super
, vol
);
6411 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6412 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6413 dev
= get_imsm_dev(super
, i
);
6414 handle_missing(super
, dev
);
6416 super
->updates_pending
++;
6423 #endif /* MDASSEMBLE */
6425 static int is_gen_migration(struct imsm_dev
*dev
)
6430 if (!dev
->vol
.migr_state
)
6433 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6439 static int is_rebuilding(struct imsm_dev
*dev
)
6441 struct imsm_map
*migr_map
;
6443 if (!dev
->vol
.migr_state
)
6446 if (migr_type(dev
) != MIGR_REBUILD
)
6449 migr_map
= get_imsm_map(dev
, MAP_1
);
6451 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6458 static int is_initializing(struct imsm_dev
*dev
)
6460 struct imsm_map
*migr_map
;
6462 if (!dev
->vol
.migr_state
)
6465 if (migr_type(dev
) != MIGR_INIT
)
6468 migr_map
= get_imsm_map(dev
, MAP_1
);
6470 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6477 static void update_recovery_start(struct intel_super
*super
,
6478 struct imsm_dev
*dev
,
6479 struct mdinfo
*array
)
6481 struct mdinfo
*rebuild
= NULL
;
6485 if (!is_rebuilding(dev
))
6488 /* Find the rebuild target, but punt on the dual rebuild case */
6489 for (d
= array
->devs
; d
; d
= d
->next
)
6490 if (d
->recovery_start
== 0) {
6497 /* (?) none of the disks are marked with
6498 * IMSM_ORD_REBUILD, so assume they are missing and the
6499 * disk_ord_tbl was not correctly updated
6501 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
6505 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6506 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6510 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6513 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6515 /* Given a container loaded by load_super_imsm_all,
6516 * extract information about all the arrays into
6518 * If 'subarray' is given, just extract info about that array.
6520 * For each imsm_dev create an mdinfo, fill it in,
6521 * then look for matching devices in super->disks
6522 * and create appropriate device mdinfo.
6524 struct intel_super
*super
= st
->sb
;
6525 struct imsm_super
*mpb
= super
->anchor
;
6526 struct mdinfo
*rest
= NULL
;
6530 int spare_disks
= 0;
6532 /* do not assemble arrays when not all attributes are supported */
6533 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6535 pr_err("Unsupported attributes in IMSM metadata."
6536 "Arrays activation is blocked.\n");
6539 /* check for bad blocks */
6540 if (imsm_bbm_log_size(super
->anchor
)) {
6541 pr_err("BBM log found in IMSM metadata."
6542 "Arrays activation is blocked.\n");
6547 /* count spare devices, not used in maps
6549 for (d
= super
->disks
; d
; d
= d
->next
)
6553 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6554 struct imsm_dev
*dev
;
6555 struct imsm_map
*map
;
6556 struct imsm_map
*map2
;
6557 struct mdinfo
*this;
6565 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6568 dev
= get_imsm_dev(super
, i
);
6569 map
= get_imsm_map(dev
, MAP_0
);
6570 map2
= get_imsm_map(dev
, MAP_1
);
6572 /* do not publish arrays that are in the middle of an
6573 * unsupported migration
6575 if (dev
->vol
.migr_state
&&
6576 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6577 pr_err("cannot assemble volume '%.16s':"
6578 " unsupported migration in progress\n",
6582 /* do not publish arrays that are not support by controller's
6586 this = xmalloc(sizeof(*this));
6588 super
->current_vol
= i
;
6589 getinfo_super_imsm_volume(st
, this, NULL
);
6592 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6593 /* mdadm does not support all metadata features- set the bit in all arrays state */
6594 if (!validate_geometry_imsm_orom(super
,
6595 get_imsm_raid_level(map
), /* RAID level */
6596 imsm_level_to_layout(get_imsm_raid_level(map
)),
6597 map
->num_members
, /* raid disks */
6598 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6600 pr_err("IMSM RAID geometry validation"
6601 " failed. Array %s activation is blocked.\n",
6603 this->array
.state
|=
6604 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6605 (1<<MD_SB_BLOCK_VOLUME
);
6609 /* if array has bad blocks, set suitable bit in all arrays state */
6611 this->array
.state
|=
6612 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6613 (1<<MD_SB_BLOCK_VOLUME
);
6615 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6616 unsigned long long recovery_start
;
6617 struct mdinfo
*info_d
;
6624 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6625 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6626 for (d
= super
->disks
; d
; d
= d
->next
)
6627 if (d
->index
== idx
)
6630 recovery_start
= MaxSector
;
6633 if (d
&& is_failed(&d
->disk
))
6635 if (ord
& IMSM_ORD_REBUILD
)
6639 * if we skip some disks the array will be assmebled degraded;
6640 * reset resync start to avoid a dirty-degraded
6641 * situation when performing the intial sync
6643 * FIXME handle dirty degraded
6645 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6646 this->resync_start
= MaxSector
;
6650 info_d
= xcalloc(1, sizeof(*info_d
));
6651 info_d
->next
= this->devs
;
6652 this->devs
= info_d
;
6654 info_d
->disk
.number
= d
->index
;
6655 info_d
->disk
.major
= d
->major
;
6656 info_d
->disk
.minor
= d
->minor
;
6657 info_d
->disk
.raid_disk
= slot
;
6658 info_d
->recovery_start
= recovery_start
;
6660 if (slot
< map2
->num_members
)
6661 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6663 this->array
.spare_disks
++;
6665 if (slot
< map
->num_members
)
6666 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6668 this->array
.spare_disks
++;
6670 if (info_d
->recovery_start
== MaxSector
)
6671 this->array
.working_disks
++;
6673 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6674 info_d
->data_offset
= pba_of_lba0(map
);
6675 info_d
->component_size
= blocks_per_member(map
);
6677 /* now that the disk list is up-to-date fixup recovery_start */
6678 update_recovery_start(super
, dev
, this);
6679 this->array
.spare_disks
+= spare_disks
;
6682 /* check for reshape */
6683 if (this->reshape_active
== 1)
6684 recover_backup_imsm(st
, this);
6693 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6694 int failed
, int look_in_map
)
6696 struct imsm_map
*map
;
6698 map
= get_imsm_map(dev
, look_in_map
);
6701 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6702 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6704 switch (get_imsm_raid_level(map
)) {
6706 return IMSM_T_STATE_FAILED
;
6709 if (failed
< map
->num_members
)
6710 return IMSM_T_STATE_DEGRADED
;
6712 return IMSM_T_STATE_FAILED
;
6717 * check to see if any mirrors have failed, otherwise we
6718 * are degraded. Even numbered slots are mirrored on
6722 /* gcc -Os complains that this is unused */
6723 int insync
= insync
;
6725 for (i
= 0; i
< map
->num_members
; i
++) {
6726 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6727 int idx
= ord_to_idx(ord
);
6728 struct imsm_disk
*disk
;
6730 /* reset the potential in-sync count on even-numbered
6731 * slots. num_copies is always 2 for imsm raid10
6736 disk
= get_imsm_disk(super
, idx
);
6737 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6740 /* no in-sync disks left in this mirror the
6744 return IMSM_T_STATE_FAILED
;
6747 return IMSM_T_STATE_DEGRADED
;
6751 return IMSM_T_STATE_DEGRADED
;
6753 return IMSM_T_STATE_FAILED
;
6759 return map
->map_state
;
6762 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6767 struct imsm_disk
*disk
;
6768 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6769 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6770 struct imsm_map
*map_for_loop
;
6775 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6776 * disks that are being rebuilt. New failures are recorded to
6777 * map[0]. So we look through all the disks we started with and
6778 * see if any failures are still present, or if any new ones
6782 if (prev
&& (map
->num_members
< prev
->num_members
))
6783 map_for_loop
= prev
;
6785 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6787 /* when MAP_X is passed both maps failures are counted
6790 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6791 (i
< prev
->num_members
)) {
6792 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6793 idx_1
= ord_to_idx(ord
);
6795 disk
= get_imsm_disk(super
, idx_1
);
6796 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6799 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6800 (i
< map
->num_members
)) {
6801 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6802 idx
= ord_to_idx(ord
);
6805 disk
= get_imsm_disk(super
, idx
);
6806 if (!disk
|| is_failed(disk
) ||
6807 ord
& IMSM_ORD_REBUILD
)
6817 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6820 struct intel_super
*super
= c
->sb
;
6821 struct imsm_super
*mpb
= super
->anchor
;
6823 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6824 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6825 __func__
, atoi(inst
));
6829 dprintf("imsm: open_new %s\n", inst
);
6830 a
->info
.container_member
= atoi(inst
);
6834 static int is_resyncing(struct imsm_dev
*dev
)
6836 struct imsm_map
*migr_map
;
6838 if (!dev
->vol
.migr_state
)
6841 if (migr_type(dev
) == MIGR_INIT
||
6842 migr_type(dev
) == MIGR_REPAIR
)
6845 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6848 migr_map
= get_imsm_map(dev
, MAP_1
);
6850 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6851 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6857 /* return true if we recorded new information */
6858 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6862 struct imsm_map
*map
;
6863 char buf
[MAX_RAID_SERIAL_LEN
+3];
6864 unsigned int len
, shift
= 0;
6866 /* new failures are always set in map[0] */
6867 map
= get_imsm_map(dev
, MAP_0
);
6869 slot
= get_imsm_disk_slot(map
, idx
);
6873 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6874 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6877 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6878 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6880 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6881 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6882 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6884 disk
->status
|= FAILED_DISK
;
6885 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6886 /* mark failures in second map if second map exists and this disk
6888 * This is valid for migration, initialization and rebuild
6890 if (dev
->vol
.migr_state
) {
6891 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6892 int slot2
= get_imsm_disk_slot(map2
, idx
);
6894 if ((slot2
< map2
->num_members
) &&
6896 set_imsm_ord_tbl_ent(map2
, slot2
,
6897 idx
| IMSM_ORD_REBUILD
);
6899 if (map
->failed_disk_num
== 0xff)
6900 map
->failed_disk_num
= slot
;
6904 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6906 mark_failure(dev
, disk
, idx
);
6908 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6911 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6912 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6915 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6919 if (!super
->missing
)
6922 dprintf("imsm: mark missing\n");
6923 /* end process for initialization and rebuild only
6925 if (is_gen_migration(dev
) == 0) {
6929 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6930 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6932 end_migration(dev
, super
, map_state
);
6934 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6935 mark_missing(dev
, &dl
->disk
, dl
->index
);
6936 super
->updates_pending
++;
6939 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
6942 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6943 unsigned long long array_blocks
;
6944 struct imsm_map
*map
;
6946 if (used_disks
== 0) {
6947 /* when problems occures
6948 * return current array_blocks value
6950 array_blocks
= __le32_to_cpu(dev
->size_high
);
6951 array_blocks
= array_blocks
<< 32;
6952 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6954 return array_blocks
;
6957 /* set array size in metadata
6959 if (new_size
<= 0) {
6960 /* OLCE size change is caused by added disks
6962 map
= get_imsm_map(dev
, MAP_0
);
6963 array_blocks
= blocks_per_member(map
) * used_disks
;
6965 /* Online Volume Size Change
6966 * Using available free space
6968 array_blocks
= new_size
;
6971 /* round array size down to closest MB
6973 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6974 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6975 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6977 return array_blocks
;
6980 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6982 static void imsm_progress_container_reshape(struct intel_super
*super
)
6984 /* if no device has a migr_state, but some device has a
6985 * different number of members than the previous device, start
6986 * changing the number of devices in this device to match
6989 struct imsm_super
*mpb
= super
->anchor
;
6990 int prev_disks
= -1;
6994 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6995 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6996 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6997 struct imsm_map
*map2
;
6998 int prev_num_members
;
7000 if (dev
->vol
.migr_state
)
7003 if (prev_disks
== -1)
7004 prev_disks
= map
->num_members
;
7005 if (prev_disks
== map
->num_members
)
7008 /* OK, this array needs to enter reshape mode.
7009 * i.e it needs a migr_state
7012 copy_map_size
= sizeof_imsm_map(map
);
7013 prev_num_members
= map
->num_members
;
7014 map
->num_members
= prev_disks
;
7015 dev
->vol
.migr_state
= 1;
7016 dev
->vol
.curr_migr_unit
= 0;
7017 set_migr_type(dev
, MIGR_GEN_MIGR
);
7018 for (i
= prev_num_members
;
7019 i
< map
->num_members
; i
++)
7020 set_imsm_ord_tbl_ent(map
, i
, i
);
7021 map2
= get_imsm_map(dev
, MAP_1
);
7022 /* Copy the current map */
7023 memcpy(map2
, map
, copy_map_size
);
7024 map2
->num_members
= prev_num_members
;
7026 imsm_set_array_size(dev
, -1);
7027 super
->clean_migration_record_by_mdmon
= 1;
7028 super
->updates_pending
++;
7032 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7033 * states are handled in imsm_set_disk() with one exception, when a
7034 * resync is stopped due to a new failure this routine will set the
7035 * 'degraded' state for the array.
7037 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7039 int inst
= a
->info
.container_member
;
7040 struct intel_super
*super
= a
->container
->sb
;
7041 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7042 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7043 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7044 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7045 __u32 blocks_per_unit
;
7047 if (dev
->vol
.migr_state
&&
7048 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7049 /* array state change is blocked due to reshape action
7051 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7052 * - finish the reshape (if last_checkpoint is big and action != reshape)
7053 * - update curr_migr_unit
7055 if (a
->curr_action
== reshape
) {
7056 /* still reshaping, maybe update curr_migr_unit */
7057 goto mark_checkpoint
;
7059 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7060 /* for some reason we aborted the reshape.
7062 * disable automatic metadata rollback
7063 * user action is required to recover process
7066 struct imsm_map
*map2
=
7067 get_imsm_map(dev
, MAP_1
);
7068 dev
->vol
.migr_state
= 0;
7069 set_migr_type(dev
, 0);
7070 dev
->vol
.curr_migr_unit
= 0;
7072 sizeof_imsm_map(map2
));
7073 super
->updates_pending
++;
7076 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7077 unsigned long long array_blocks
;
7081 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7082 if (used_disks
> 0) {
7084 blocks_per_member(map
) *
7086 /* round array size down to closest MB
7088 array_blocks
= (array_blocks
7089 >> SECT_PER_MB_SHIFT
)
7090 << SECT_PER_MB_SHIFT
;
7091 a
->info
.custom_array_size
= array_blocks
;
7092 /* encourage manager to update array
7096 a
->check_reshape
= 1;
7098 /* finalize online capacity expansion/reshape */
7099 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7101 mdi
->disk
.raid_disk
,
7104 imsm_progress_container_reshape(super
);
7109 /* before we activate this array handle any missing disks */
7110 if (consistent
== 2)
7111 handle_missing(super
, dev
);
7113 if (consistent
== 2 &&
7114 (!is_resync_complete(&a
->info
) ||
7115 map_state
!= IMSM_T_STATE_NORMAL
||
7116 dev
->vol
.migr_state
))
7119 if (is_resync_complete(&a
->info
)) {
7120 /* complete intialization / resync,
7121 * recovery and interrupted recovery is completed in
7124 if (is_resyncing(dev
)) {
7125 dprintf("imsm: mark resync done\n");
7126 end_migration(dev
, super
, map_state
);
7127 super
->updates_pending
++;
7128 a
->last_checkpoint
= 0;
7130 } else if ((!is_resyncing(dev
) && !failed
) &&
7131 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7132 /* mark the start of the init process if nothing is failed */
7133 dprintf("imsm: mark resync start\n");
7134 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7135 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7137 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7138 super
->updates_pending
++;
7142 /* skip checkpointing for general migration,
7143 * it is controlled in mdadm
7145 if (is_gen_migration(dev
))
7146 goto skip_mark_checkpoint
;
7148 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7149 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7150 if (blocks_per_unit
) {
7154 units
= a
->last_checkpoint
/ blocks_per_unit
;
7157 /* check that we did not overflow 32-bits, and that
7158 * curr_migr_unit needs updating
7160 if (units32
== units
&&
7162 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7163 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7164 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7165 super
->updates_pending
++;
7169 skip_mark_checkpoint
:
7170 /* mark dirty / clean */
7171 if (dev
->vol
.dirty
!= !consistent
) {
7172 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7177 super
->updates_pending
++;
7183 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7185 int inst
= a
->info
.container_member
;
7186 struct intel_super
*super
= a
->container
->sb
;
7187 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7188 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7189 struct imsm_disk
*disk
;
7191 int recovery_not_finished
= 0;
7196 if (n
> map
->num_members
)
7197 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
7198 n
, map
->num_members
- 1);
7203 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7205 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7206 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7208 /* check for new failures */
7209 if (state
& DS_FAULTY
) {
7210 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7211 super
->updates_pending
++;
7214 /* check if in_sync */
7215 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7216 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7218 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7219 super
->updates_pending
++;
7222 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7223 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7225 /* check if recovery complete, newly degraded, or failed */
7226 dprintf("imsm: Detected transition to state ");
7227 switch (map_state
) {
7228 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7229 dprintf("normal: ");
7230 if (is_rebuilding(dev
)) {
7231 dprintf("while rebuilding");
7232 /* check if recovery is really finished */
7233 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7234 if (mdi
->recovery_start
!= MaxSector
) {
7235 recovery_not_finished
= 1;
7238 if (recovery_not_finished
) {
7239 dprintf("\nimsm: Rebuild has not finished yet, "
7240 "state not changed");
7241 if (a
->last_checkpoint
< mdi
->recovery_start
) {
7242 a
->last_checkpoint
= mdi
->recovery_start
;
7243 super
->updates_pending
++;
7247 end_migration(dev
, super
, map_state
);
7248 map
= get_imsm_map(dev
, MAP_0
);
7249 map
->failed_disk_num
= ~0;
7250 super
->updates_pending
++;
7251 a
->last_checkpoint
= 0;
7254 if (is_gen_migration(dev
)) {
7255 dprintf("while general migration");
7256 if (a
->last_checkpoint
>= a
->info
.component_size
)
7257 end_migration(dev
, super
, map_state
);
7259 map
->map_state
= map_state
;
7260 map
= get_imsm_map(dev
, MAP_0
);
7261 map
->failed_disk_num
= ~0;
7262 super
->updates_pending
++;
7266 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7267 dprintf("degraded: ");
7268 if ((map
->map_state
!= map_state
) &&
7269 !dev
->vol
.migr_state
) {
7270 dprintf("mark degraded");
7271 map
->map_state
= map_state
;
7272 super
->updates_pending
++;
7273 a
->last_checkpoint
= 0;
7276 if (is_rebuilding(dev
)) {
7277 dprintf("while rebuilding.");
7278 if (map
->map_state
!= map_state
) {
7279 dprintf(" Map state change");
7280 end_migration(dev
, super
, map_state
);
7281 super
->updates_pending
++;
7285 if (is_gen_migration(dev
)) {
7286 dprintf("while general migration");
7287 if (a
->last_checkpoint
>= a
->info
.component_size
)
7288 end_migration(dev
, super
, map_state
);
7290 map
->map_state
= map_state
;
7291 manage_second_map(super
, dev
);
7293 super
->updates_pending
++;
7296 if (is_initializing(dev
)) {
7297 dprintf("while initialization.");
7298 map
->map_state
= map_state
;
7299 super
->updates_pending
++;
7303 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7304 dprintf("failed: ");
7305 if (is_gen_migration(dev
)) {
7306 dprintf("while general migration");
7307 map
->map_state
= map_state
;
7308 super
->updates_pending
++;
7311 if (map
->map_state
!= map_state
) {
7312 dprintf("mark failed");
7313 end_migration(dev
, super
, map_state
);
7314 super
->updates_pending
++;
7315 a
->last_checkpoint
= 0;
7320 dprintf("state %i\n", map_state
);
7326 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7329 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7330 unsigned long long dsize
;
7331 unsigned long long sectors
;
7333 get_dev_size(fd
, NULL
, &dsize
);
7335 if (mpb_size
> 512) {
7336 /* -1 to account for anchor */
7337 sectors
= mpb_sectors(mpb
) - 1;
7339 /* write the extended mpb to the sectors preceeding the anchor */
7340 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7343 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7348 /* first block is stored on second to last sector of the disk */
7349 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7352 if (write(fd
, buf
, 512) != 512)
7358 static void imsm_sync_metadata(struct supertype
*container
)
7360 struct intel_super
*super
= container
->sb
;
7362 dprintf("sync metadata: %d\n", super
->updates_pending
);
7363 if (!super
->updates_pending
)
7366 write_super_imsm(container
, 0);
7368 super
->updates_pending
= 0;
7371 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7373 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7374 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7377 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7381 if (dl
&& is_failed(&dl
->disk
))
7385 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
7390 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7391 struct active_array
*a
, int activate_new
,
7392 struct mdinfo
*additional_test_list
)
7394 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7395 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7396 struct imsm_super
*mpb
= super
->anchor
;
7397 struct imsm_map
*map
;
7398 unsigned long long pos
;
7403 __u32 array_start
= 0;
7404 __u32 array_end
= 0;
7406 struct mdinfo
*test_list
;
7408 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7409 /* If in this array, skip */
7410 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7411 if (d
->state_fd
>= 0 &&
7412 d
->disk
.major
== dl
->major
&&
7413 d
->disk
.minor
== dl
->minor
) {
7414 dprintf("%x:%x already in array\n",
7415 dl
->major
, dl
->minor
);
7420 test_list
= additional_test_list
;
7422 if (test_list
->disk
.major
== dl
->major
&&
7423 test_list
->disk
.minor
== dl
->minor
) {
7424 dprintf("%x:%x already in additional test list\n",
7425 dl
->major
, dl
->minor
);
7428 test_list
= test_list
->next
;
7433 /* skip in use or failed drives */
7434 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7436 dprintf("%x:%x status (failed: %d index: %d)\n",
7437 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7441 /* skip pure spares when we are looking for partially
7442 * assimilated drives
7444 if (dl
->index
== -1 && !activate_new
)
7447 /* Does this unused device have the requisite free space?
7448 * It needs to be able to cover all member volumes
7450 ex
= get_extents(super
, dl
);
7452 dprintf("cannot get extents\n");
7455 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7456 dev
= get_imsm_dev(super
, i
);
7457 map
= get_imsm_map(dev
, MAP_0
);
7459 /* check if this disk is already a member of
7462 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7468 array_start
= pba_of_lba0(map
);
7469 array_end
= array_start
+
7470 blocks_per_member(map
) - 1;
7473 /* check that we can start at pba_of_lba0 with
7474 * blocks_per_member of space
7476 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7480 pos
= ex
[j
].start
+ ex
[j
].size
;
7482 } while (ex
[j
-1].size
);
7489 if (i
< mpb
->num_raid_devs
) {
7490 dprintf("%x:%x does not have %u to %u available\n",
7491 dl
->major
, dl
->minor
, array_start
, array_end
);
7502 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7504 struct imsm_dev
*dev2
;
7505 struct imsm_map
*map
;
7511 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7513 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7514 if (state
== IMSM_T_STATE_FAILED
) {
7515 map
= get_imsm_map(dev2
, MAP_0
);
7518 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7520 * Check if failed disks are deleted from intel
7521 * disk list or are marked to be deleted
7523 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7524 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7526 * Do not rebuild the array if failed disks
7527 * from failed sub-array are not removed from
7531 is_failed(&idisk
->disk
) &&
7532 (idisk
->action
!= DISK_REMOVE
))
7540 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7541 struct metadata_update
**updates
)
7544 * Find a device with unused free space and use it to replace a
7545 * failed/vacant region in an array. We replace failed regions one a
7546 * array at a time. The result is that a new spare disk will be added
7547 * to the first failed array and after the monitor has finished
7548 * propagating failures the remainder will be consumed.
7550 * FIXME add a capability for mdmon to request spares from another
7554 struct intel_super
*super
= a
->container
->sb
;
7555 int inst
= a
->info
.container_member
;
7556 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7557 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7558 int failed
= a
->info
.array
.raid_disks
;
7559 struct mdinfo
*rv
= NULL
;
7562 struct metadata_update
*mu
;
7564 struct imsm_update_activate_spare
*u
;
7569 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7570 if ((d
->curr_state
& DS_FAULTY
) &&
7572 /* wait for Removal to happen */
7574 if (d
->state_fd
>= 0)
7578 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7579 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7581 if (imsm_reshape_blocks_arrays_changes(super
))
7584 /* Cannot activate another spare if rebuild is in progress already
7586 if (is_rebuilding(dev
)) {
7587 dprintf("imsm: No spare activation allowed. "
7588 "Rebuild in progress already.\n");
7592 if (a
->info
.array
.level
== 4)
7593 /* No repair for takeovered array
7594 * imsm doesn't support raid4
7598 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7599 IMSM_T_STATE_DEGRADED
)
7603 * If there are any failed disks check state of the other volume.
7604 * Block rebuild if the another one is failed until failed disks
7605 * are removed from container.
7608 dprintf("found failed disks in %.*s, check if there another"
7609 "failed sub-array.\n",
7610 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7611 /* check if states of the other volumes allow for rebuild */
7612 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7614 allowed
= imsm_rebuild_allowed(a
->container
,
7622 /* For each slot, if it is not working, find a spare */
7623 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7624 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7625 if (d
->disk
.raid_disk
== i
)
7627 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7628 if (d
&& (d
->state_fd
>= 0))
7632 * OK, this device needs recovery. Try to re-add the
7633 * previous occupant of this slot, if this fails see if
7634 * we can continue the assimilation of a spare that was
7635 * partially assimilated, finally try to activate a new
7638 dl
= imsm_readd(super
, i
, a
);
7640 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7642 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7646 /* found a usable disk with enough space */
7647 di
= xcalloc(1, sizeof(*di
));
7649 /* dl->index will be -1 in the case we are activating a
7650 * pristine spare. imsm_process_update() will create a
7651 * new index in this case. Once a disk is found to be
7652 * failed in all member arrays it is kicked from the
7655 di
->disk
.number
= dl
->index
;
7657 /* (ab)use di->devs to store a pointer to the device
7660 di
->devs
= (struct mdinfo
*) dl
;
7662 di
->disk
.raid_disk
= i
;
7663 di
->disk
.major
= dl
->major
;
7664 di
->disk
.minor
= dl
->minor
;
7666 di
->recovery_start
= 0;
7667 di
->data_offset
= pba_of_lba0(map
);
7668 di
->component_size
= a
->info
.component_size
;
7669 di
->container_member
= inst
;
7670 super
->random
= random32();
7674 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7675 i
, di
->data_offset
);
7679 /* No spares found */
7681 /* Now 'rv' has a list of devices to return.
7682 * Create a metadata_update record to update the
7683 * disk_ord_tbl for the array
7685 mu
= xmalloc(sizeof(*mu
));
7686 mu
->buf
= xcalloc(num_spares
,
7687 sizeof(struct imsm_update_activate_spare
));
7689 mu
->space_list
= NULL
;
7690 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7691 mu
->next
= *updates
;
7692 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7694 for (di
= rv
; di
; di
= di
->next
) {
7695 u
->type
= update_activate_spare
;
7696 u
->dl
= (struct dl
*) di
->devs
;
7698 u
->slot
= di
->disk
.raid_disk
;
7709 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7711 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7712 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7713 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7714 struct disk_info
*inf
= get_disk_info(u
);
7715 struct imsm_disk
*disk
;
7719 for (i
= 0; i
< map
->num_members
; i
++) {
7720 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7721 for (j
= 0; j
< new_map
->num_members
; j
++)
7722 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7730 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7732 struct dl
*dl
= NULL
;
7733 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7734 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7739 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7741 struct dl
*prev
= NULL
;
7745 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7746 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7749 prev
->next
= dl
->next
;
7751 super
->disks
= dl
->next
;
7753 __free_imsm_disk(dl
);
7754 dprintf("%s: removed %x:%x\n",
7755 __func__
, major
, minor
);
7763 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7765 static int add_remove_disk_update(struct intel_super
*super
)
7767 int check_degraded
= 0;
7768 struct dl
*disk
= NULL
;
7769 /* add/remove some spares to/from the metadata/contrainer */
7770 while (super
->disk_mgmt_list
) {
7771 struct dl
*disk_cfg
;
7773 disk_cfg
= super
->disk_mgmt_list
;
7774 super
->disk_mgmt_list
= disk_cfg
->next
;
7775 disk_cfg
->next
= NULL
;
7777 if (disk_cfg
->action
== DISK_ADD
) {
7778 disk_cfg
->next
= super
->disks
;
7779 super
->disks
= disk_cfg
;
7781 dprintf("%s: added %x:%x\n",
7782 __func__
, disk_cfg
->major
,
7784 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7785 dprintf("Disk remove action processed: %x.%x\n",
7786 disk_cfg
->major
, disk_cfg
->minor
);
7787 disk
= get_disk_super(super
,
7791 /* store action status */
7792 disk
->action
= DISK_REMOVE
;
7793 /* remove spare disks only */
7794 if (disk
->index
== -1) {
7795 remove_disk_super(super
,
7800 /* release allocate disk structure */
7801 __free_imsm_disk(disk_cfg
);
7804 return check_degraded
;
7808 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7809 struct intel_super
*super
,
7812 struct intel_dev
*id
;
7813 void **tofree
= NULL
;
7816 dprintf("apply_reshape_migration_update()\n");
7817 if ((u
->subdev
< 0) ||
7819 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7822 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7823 dprintf("imsm: Error: Memory is not allocated\n");
7827 for (id
= super
->devlist
; id
; id
= id
->next
) {
7828 if (id
->index
== (unsigned)u
->subdev
) {
7829 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7830 struct imsm_map
*map
;
7831 struct imsm_dev
*new_dev
=
7832 (struct imsm_dev
*)*space_list
;
7833 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7835 struct dl
*new_disk
;
7837 if (new_dev
== NULL
)
7839 *space_list
= **space_list
;
7840 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7841 map
= get_imsm_map(new_dev
, MAP_0
);
7843 dprintf("imsm: Error: migration in progress");
7847 to_state
= map
->map_state
;
7848 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7850 /* this should not happen */
7851 if (u
->new_disks
[0] < 0) {
7852 map
->failed_disk_num
=
7853 map
->num_members
- 1;
7854 to_state
= IMSM_T_STATE_DEGRADED
;
7856 to_state
= IMSM_T_STATE_NORMAL
;
7858 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7859 if (u
->new_level
> -1)
7860 map
->raid_level
= u
->new_level
;
7861 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7862 if ((u
->new_level
== 5) &&
7863 (migr_map
->raid_level
== 0)) {
7864 int ord
= map
->num_members
- 1;
7865 migr_map
->num_members
--;
7866 if (u
->new_disks
[0] < 0)
7867 ord
|= IMSM_ORD_REBUILD
;
7868 set_imsm_ord_tbl_ent(map
,
7869 map
->num_members
- 1,
7873 tofree
= (void **)dev
;
7875 /* update chunk size
7877 if (u
->new_chunksize
> 0)
7878 map
->blocks_per_strip
=
7879 __cpu_to_le16(u
->new_chunksize
* 2);
7883 if ((u
->new_level
!= 5) ||
7884 (migr_map
->raid_level
!= 0) ||
7885 (migr_map
->raid_level
== map
->raid_level
))
7888 if (u
->new_disks
[0] >= 0) {
7891 new_disk
= get_disk_super(super
,
7892 major(u
->new_disks
[0]),
7893 minor(u
->new_disks
[0]));
7894 dprintf("imsm: new disk for reshape is: %i:%i "
7895 "(%p, index = %i)\n",
7896 major(u
->new_disks
[0]),
7897 minor(u
->new_disks
[0]),
7898 new_disk
, new_disk
->index
);
7899 if (new_disk
== NULL
)
7900 goto error_disk_add
;
7902 new_disk
->index
= map
->num_members
- 1;
7903 /* slot to fill in autolayout
7905 new_disk
->raiddisk
= new_disk
->index
;
7906 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7907 new_disk
->disk
.status
&= ~SPARE_DISK
;
7909 goto error_disk_add
;
7912 *tofree
= *space_list
;
7913 /* calculate new size
7915 imsm_set_array_size(new_dev
, -1);
7922 *space_list
= tofree
;
7926 dprintf("Error: imsm: Cannot find disk.\n");
7930 static int apply_size_change_update(struct imsm_update_size_change
*u
,
7931 struct intel_super
*super
)
7933 struct intel_dev
*id
;
7936 dprintf("apply_size_change_update()\n");
7937 if ((u
->subdev
< 0) ||
7939 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7943 for (id
= super
->devlist
; id
; id
= id
->next
) {
7944 if (id
->index
== (unsigned)u
->subdev
) {
7945 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7946 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7947 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7948 unsigned long long blocks_per_member
;
7950 /* calculate new size
7952 blocks_per_member
= u
->new_size
/ used_disks
;
7953 dprintf("imsm: apply_size_change_update(size: %llu, "
7954 "blocks per member: %llu)\n",
7955 u
->new_size
, blocks_per_member
);
7956 set_blocks_per_member(map
, blocks_per_member
);
7957 imsm_set_array_size(dev
, u
->new_size
);
7968 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7969 struct intel_super
*super
,
7970 struct active_array
*active_array
)
7972 struct imsm_super
*mpb
= super
->anchor
;
7973 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7974 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7975 struct imsm_map
*migr_map
;
7976 struct active_array
*a
;
7977 struct imsm_disk
*disk
;
7984 int second_map_created
= 0;
7986 for (; u
; u
= u
->next
) {
7987 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
7992 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7997 fprintf(stderr
, "error: imsm_activate_spare passed "
7998 "an unknown disk (index: %d)\n",
8003 /* count failures (excluding rebuilds and the victim)
8004 * to determine map[0] state
8007 for (i
= 0; i
< map
->num_members
; i
++) {
8010 disk
= get_imsm_disk(super
,
8011 get_imsm_disk_idx(dev
, i
, MAP_X
));
8012 if (!disk
|| is_failed(disk
))
8016 /* adding a pristine spare, assign a new index */
8017 if (dl
->index
< 0) {
8018 dl
->index
= super
->anchor
->num_disks
;
8019 super
->anchor
->num_disks
++;
8022 disk
->status
|= CONFIGURED_DISK
;
8023 disk
->status
&= ~SPARE_DISK
;
8026 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8027 if (!second_map_created
) {
8028 second_map_created
= 1;
8029 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8030 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8032 map
->map_state
= to_state
;
8033 migr_map
= get_imsm_map(dev
, MAP_1
);
8034 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8035 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8036 dl
->index
| IMSM_ORD_REBUILD
);
8038 /* update the family_num to mark a new container
8039 * generation, being careful to record the existing
8040 * family_num in orig_family_num to clean up after
8041 * earlier mdadm versions that neglected to set it.
8043 if (mpb
->orig_family_num
== 0)
8044 mpb
->orig_family_num
= mpb
->family_num
;
8045 mpb
->family_num
+= super
->random
;
8047 /* count arrays using the victim in the metadata */
8049 for (a
= active_array
; a
; a
= a
->next
) {
8050 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8051 map
= get_imsm_map(dev
, MAP_0
);
8053 if (get_imsm_disk_slot(map
, victim
) >= 0)
8057 /* delete the victim if it is no longer being
8063 /* We know that 'manager' isn't touching anything,
8064 * so it is safe to delete
8066 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8067 if ((*dlp
)->index
== victim
)
8070 /* victim may be on the missing list */
8072 for (dlp
= &super
->missing
; *dlp
;
8073 dlp
= &(*dlp
)->next
)
8074 if ((*dlp
)->index
== victim
)
8076 imsm_delete(super
, dlp
, victim
);
8083 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8084 struct intel_super
*super
,
8087 struct dl
*new_disk
;
8088 struct intel_dev
*id
;
8090 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8091 int disk_count
= u
->old_raid_disks
;
8092 void **tofree
= NULL
;
8093 int devices_to_reshape
= 1;
8094 struct imsm_super
*mpb
= super
->anchor
;
8096 unsigned int dev_id
;
8098 dprintf("imsm: apply_reshape_container_disks_update()\n");
8100 /* enable spares to use in array */
8101 for (i
= 0; i
< delta_disks
; i
++) {
8102 new_disk
= get_disk_super(super
,
8103 major(u
->new_disks
[i
]),
8104 minor(u
->new_disks
[i
]));
8105 dprintf("imsm: new disk for reshape is: %i:%i "
8106 "(%p, index = %i)\n",
8107 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8108 new_disk
, new_disk
->index
);
8109 if ((new_disk
== NULL
) ||
8110 ((new_disk
->index
>= 0) &&
8111 (new_disk
->index
< u
->old_raid_disks
)))
8112 goto update_reshape_exit
;
8113 new_disk
->index
= disk_count
++;
8114 /* slot to fill in autolayout
8116 new_disk
->raiddisk
= new_disk
->index
;
8117 new_disk
->disk
.status
|=
8119 new_disk
->disk
.status
&= ~SPARE_DISK
;
8122 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8123 mpb
->num_raid_devs
);
8124 /* manage changes in volume
8126 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8127 void **sp
= *space_list
;
8128 struct imsm_dev
*newdev
;
8129 struct imsm_map
*newmap
, *oldmap
;
8131 for (id
= super
->devlist
; id
; id
= id
->next
) {
8132 if (id
->index
== dev_id
)
8141 /* Copy the dev, but not (all of) the map */
8142 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8143 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8144 newmap
= get_imsm_map(newdev
, MAP_0
);
8145 /* Copy the current map */
8146 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8147 /* update one device only
8149 if (devices_to_reshape
) {
8150 dprintf("imsm: modifying subdev: %i\n",
8152 devices_to_reshape
--;
8153 newdev
->vol
.migr_state
= 1;
8154 newdev
->vol
.curr_migr_unit
= 0;
8155 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8156 newmap
->num_members
= u
->new_raid_disks
;
8157 for (i
= 0; i
< delta_disks
; i
++) {
8158 set_imsm_ord_tbl_ent(newmap
,
8159 u
->old_raid_disks
+ i
,
8160 u
->old_raid_disks
+ i
);
8162 /* New map is correct, now need to save old map
8164 newmap
= get_imsm_map(newdev
, MAP_1
);
8165 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8167 imsm_set_array_size(newdev
, -1);
8170 sp
= (void **)id
->dev
;
8175 /* Clear migration record */
8176 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8179 *space_list
= tofree
;
8182 update_reshape_exit
:
8187 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8188 struct intel_super
*super
,
8191 struct imsm_dev
*dev
= NULL
;
8192 struct intel_dev
*dv
;
8193 struct imsm_dev
*dev_new
;
8194 struct imsm_map
*map
;
8198 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8199 if (dv
->index
== (unsigned int)u
->subarray
) {
8207 map
= get_imsm_map(dev
, MAP_0
);
8209 if (u
->direction
== R10_TO_R0
) {
8210 /* Number of failed disks must be half of initial disk number */
8211 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8212 (map
->num_members
/ 2))
8215 /* iterate through devices to mark removed disks as spare */
8216 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8217 if (dm
->disk
.status
& FAILED_DISK
) {
8218 int idx
= dm
->index
;
8219 /* update indexes on the disk list */
8220 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8221 the index values will end up being correct.... NB */
8222 for (du
= super
->disks
; du
; du
= du
->next
)
8223 if (du
->index
> idx
)
8225 /* mark as spare disk */
8230 map
->num_members
= map
->num_members
/ 2;
8231 map
->map_state
= IMSM_T_STATE_NORMAL
;
8232 map
->num_domains
= 1;
8233 map
->raid_level
= 0;
8234 map
->failed_disk_num
= -1;
8237 if (u
->direction
== R0_TO_R10
) {
8239 /* update slots in current disk list */
8240 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8244 /* create new *missing* disks */
8245 for (i
= 0; i
< map
->num_members
; i
++) {
8246 space
= *space_list
;
8249 *space_list
= *space
;
8251 memcpy(du
, super
->disks
, sizeof(*du
));
8255 du
->index
= (i
* 2) + 1;
8256 sprintf((char *)du
->disk
.serial
,
8257 " MISSING_%d", du
->index
);
8258 sprintf((char *)du
->serial
,
8259 "MISSING_%d", du
->index
);
8260 du
->next
= super
->missing
;
8261 super
->missing
= du
;
8263 /* create new dev and map */
8264 space
= *space_list
;
8267 *space_list
= *space
;
8268 dev_new
= (void *)space
;
8269 memcpy(dev_new
, dev
, sizeof(*dev
));
8270 /* update new map */
8271 map
= get_imsm_map(dev_new
, MAP_0
);
8272 map
->num_members
= map
->num_members
* 2;
8273 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8274 map
->num_domains
= 2;
8275 map
->raid_level
= 1;
8276 /* replace dev<->dev_new */
8279 /* update disk order table */
8280 for (du
= super
->disks
; du
; du
= du
->next
)
8282 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8283 for (du
= super
->missing
; du
; du
= du
->next
)
8284 if (du
->index
>= 0) {
8285 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8286 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8292 static void imsm_process_update(struct supertype
*st
,
8293 struct metadata_update
*update
)
8296 * crack open the metadata_update envelope to find the update record
8297 * update can be one of:
8298 * update_reshape_container_disks - all the arrays in the container
8299 * are being reshaped to have more devices. We need to mark
8300 * the arrays for general migration and convert selected spares
8301 * into active devices.
8302 * update_activate_spare - a spare device has replaced a failed
8303 * device in an array, update the disk_ord_tbl. If this disk is
8304 * present in all member arrays then also clear the SPARE_DISK
8306 * update_create_array
8308 * update_rename_array
8309 * update_add_remove_disk
8311 struct intel_super
*super
= st
->sb
;
8312 struct imsm_super
*mpb
;
8313 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8315 /* update requires a larger buf but the allocation failed */
8316 if (super
->next_len
&& !super
->next_buf
) {
8317 super
->next_len
= 0;
8321 if (super
->next_buf
) {
8322 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8324 super
->len
= super
->next_len
;
8325 super
->buf
= super
->next_buf
;
8327 super
->next_len
= 0;
8328 super
->next_buf
= NULL
;
8331 mpb
= super
->anchor
;
8334 case update_general_migration_checkpoint
: {
8335 struct intel_dev
*id
;
8336 struct imsm_update_general_migration_checkpoint
*u
=
8337 (void *)update
->buf
;
8339 dprintf("imsm: process_update() "
8340 "for update_general_migration_checkpoint called\n");
8342 /* find device under general migration */
8343 for (id
= super
->devlist
; id
; id
= id
->next
) {
8344 if (is_gen_migration(id
->dev
)) {
8345 id
->dev
->vol
.curr_migr_unit
=
8346 __cpu_to_le32(u
->curr_migr_unit
);
8347 super
->updates_pending
++;
8352 case update_takeover
: {
8353 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8354 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8355 imsm_update_version_info(super
);
8356 super
->updates_pending
++;
8361 case update_reshape_container_disks
: {
8362 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8363 if (apply_reshape_container_disks_update(
8364 u
, super
, &update
->space_list
))
8365 super
->updates_pending
++;
8368 case update_reshape_migration
: {
8369 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8370 if (apply_reshape_migration_update(
8371 u
, super
, &update
->space_list
))
8372 super
->updates_pending
++;
8375 case update_size_change
: {
8376 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8377 if (apply_size_change_update(u
, super
))
8378 super
->updates_pending
++;
8381 case update_activate_spare
: {
8382 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8383 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8384 super
->updates_pending
++;
8387 case update_create_array
: {
8388 /* someone wants to create a new array, we need to be aware of
8389 * a few races/collisions:
8390 * 1/ 'Create' called by two separate instances of mdadm
8391 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8392 * devices that have since been assimilated via
8394 * In the event this update can not be carried out mdadm will
8395 * (FIX ME) notice that its update did not take hold.
8397 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8398 struct intel_dev
*dv
;
8399 struct imsm_dev
*dev
;
8400 struct imsm_map
*map
, *new_map
;
8401 unsigned long long start
, end
;
8402 unsigned long long new_start
, new_end
;
8404 struct disk_info
*inf
;
8407 /* handle racing creates: first come first serve */
8408 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8409 dprintf("%s: subarray %d already defined\n",
8410 __func__
, u
->dev_idx
);
8414 /* check update is next in sequence */
8415 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8416 dprintf("%s: can not create array %d expected index %d\n",
8417 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
8421 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8422 new_start
= pba_of_lba0(new_map
);
8423 new_end
= new_start
+ blocks_per_member(new_map
);
8424 inf
= get_disk_info(u
);
8426 /* handle activate_spare versus create race:
8427 * check to make sure that overlapping arrays do not include
8430 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8431 dev
= get_imsm_dev(super
, i
);
8432 map
= get_imsm_map(dev
, MAP_0
);
8433 start
= pba_of_lba0(map
);
8434 end
= start
+ blocks_per_member(map
);
8435 if ((new_start
>= start
&& new_start
<= end
) ||
8436 (start
>= new_start
&& start
<= new_end
))
8441 if (disks_overlap(super
, i
, u
)) {
8442 dprintf("%s: arrays overlap\n", __func__
);
8447 /* check that prepare update was successful */
8448 if (!update
->space
) {
8449 dprintf("%s: prepare update failed\n", __func__
);
8453 /* check that all disks are still active before committing
8454 * changes. FIXME: could we instead handle this by creating a
8455 * degraded array? That's probably not what the user expects,
8456 * so better to drop this update on the floor.
8458 for (i
= 0; i
< new_map
->num_members
; i
++) {
8459 dl
= serial_to_dl(inf
[i
].serial
, super
);
8461 dprintf("%s: disk disappeared\n", __func__
);
8466 super
->updates_pending
++;
8468 /* convert spares to members and fixup ord_tbl */
8469 for (i
= 0; i
< new_map
->num_members
; i
++) {
8470 dl
= serial_to_dl(inf
[i
].serial
, super
);
8471 if (dl
->index
== -1) {
8472 dl
->index
= mpb
->num_disks
;
8474 dl
->disk
.status
|= CONFIGURED_DISK
;
8475 dl
->disk
.status
&= ~SPARE_DISK
;
8477 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8482 update
->space
= NULL
;
8483 imsm_copy_dev(dev
, &u
->dev
);
8484 dv
->index
= u
->dev_idx
;
8485 dv
->next
= super
->devlist
;
8486 super
->devlist
= dv
;
8487 mpb
->num_raid_devs
++;
8489 imsm_update_version_info(super
);
8492 /* mdmon knows how to release update->space, but not
8493 * ((struct intel_dev *) update->space)->dev
8495 if (update
->space
) {
8501 case update_kill_array
: {
8502 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8503 int victim
= u
->dev_idx
;
8504 struct active_array
*a
;
8505 struct intel_dev
**dp
;
8506 struct imsm_dev
*dev
;
8508 /* sanity check that we are not affecting the uuid of
8509 * active arrays, or deleting an active array
8511 * FIXME when immutable ids are available, but note that
8512 * we'll also need to fixup the invalidated/active
8513 * subarray indexes in mdstat
8515 for (a
= st
->arrays
; a
; a
= a
->next
)
8516 if (a
->info
.container_member
>= victim
)
8518 /* by definition if mdmon is running at least one array
8519 * is active in the container, so checking
8520 * mpb->num_raid_devs is just extra paranoia
8522 dev
= get_imsm_dev(super
, victim
);
8523 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8524 dprintf("failed to delete subarray-%d\n", victim
);
8528 for (dp
= &super
->devlist
; *dp
;)
8529 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8532 if ((*dp
)->index
> (unsigned)victim
)
8536 mpb
->num_raid_devs
--;
8537 super
->updates_pending
++;
8540 case update_rename_array
: {
8541 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8542 char name
[MAX_RAID_SERIAL_LEN
+1];
8543 int target
= u
->dev_idx
;
8544 struct active_array
*a
;
8545 struct imsm_dev
*dev
;
8547 /* sanity check that we are not affecting the uuid of
8550 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8551 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8552 for (a
= st
->arrays
; a
; a
= a
->next
)
8553 if (a
->info
.container_member
== target
)
8555 dev
= get_imsm_dev(super
, u
->dev_idx
);
8556 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8557 dprintf("failed to rename subarray-%d\n", target
);
8561 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8562 super
->updates_pending
++;
8565 case update_add_remove_disk
: {
8566 /* we may be able to repair some arrays if disks are
8567 * being added, check teh status of add_remove_disk
8568 * if discs has been added.
8570 if (add_remove_disk_update(super
)) {
8571 struct active_array
*a
;
8573 super
->updates_pending
++;
8574 for (a
= st
->arrays
; a
; a
= a
->next
)
8575 a
->check_degraded
= 1;
8580 fprintf(stderr
, "error: unsuported process update type:"
8581 "(type: %d)\n", type
);
8585 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8587 static void imsm_prepare_update(struct supertype
*st
,
8588 struct metadata_update
*update
)
8591 * Allocate space to hold new disk entries, raid-device entries or a new
8592 * mpb if necessary. The manager synchronously waits for updates to
8593 * complete in the monitor, so new mpb buffers allocated here can be
8594 * integrated by the monitor thread without worrying about live pointers
8595 * in the manager thread.
8597 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8598 struct intel_super
*super
= st
->sb
;
8599 struct imsm_super
*mpb
= super
->anchor
;
8604 case update_general_migration_checkpoint
:
8605 dprintf("imsm: prepare_update() "
8606 "for update_general_migration_checkpoint called\n");
8608 case update_takeover
: {
8609 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8610 if (u
->direction
== R0_TO_R10
) {
8611 void **tail
= (void **)&update
->space_list
;
8612 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8613 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8614 int num_members
= map
->num_members
;
8617 /* allocate memory for added disks */
8618 for (i
= 0; i
< num_members
; i
++) {
8619 size
= sizeof(struct dl
);
8620 space
= xmalloc(size
);
8625 /* allocate memory for new device */
8626 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8627 (num_members
* sizeof(__u32
));
8628 space
= xmalloc(size
);
8632 len
= disks_to_mpb_size(num_members
* 2);
8637 case update_reshape_container_disks
: {
8638 /* Every raid device in the container is about to
8639 * gain some more devices, and we will enter a
8641 * So each 'imsm_map' will be bigger, and the imsm_vol
8642 * will now hold 2 of them.
8643 * Thus we need new 'struct imsm_dev' allocations sized
8644 * as sizeof_imsm_dev but with more devices in both maps.
8646 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8647 struct intel_dev
*dl
;
8648 void **space_tail
= (void**)&update
->space_list
;
8650 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8652 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8653 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8655 if (u
->new_raid_disks
> u
->old_raid_disks
)
8656 size
+= sizeof(__u32
)*2*
8657 (u
->new_raid_disks
- u
->old_raid_disks
);
8664 len
= disks_to_mpb_size(u
->new_raid_disks
);
8665 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8668 case update_reshape_migration
: {
8669 /* for migration level 0->5 we need to add disks
8670 * so the same as for container operation we will copy
8671 * device to the bigger location.
8672 * in memory prepared device and new disk area are prepared
8673 * for usage in process update
8675 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8676 struct intel_dev
*id
;
8677 void **space_tail
= (void **)&update
->space_list
;
8680 int current_level
= -1;
8682 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8684 /* add space for bigger array in update
8686 for (id
= super
->devlist
; id
; id
= id
->next
) {
8687 if (id
->index
== (unsigned)u
->subdev
) {
8688 size
= sizeof_imsm_dev(id
->dev
, 1);
8689 if (u
->new_raid_disks
> u
->old_raid_disks
)
8690 size
+= sizeof(__u32
)*2*
8691 (u
->new_raid_disks
- u
->old_raid_disks
);
8699 if (update
->space_list
== NULL
)
8702 /* add space for disk in update
8704 size
= sizeof(struct dl
);
8710 /* add spare device to update
8712 for (id
= super
->devlist
; id
; id
= id
->next
)
8713 if (id
->index
== (unsigned)u
->subdev
) {
8714 struct imsm_dev
*dev
;
8715 struct imsm_map
*map
;
8717 dev
= get_imsm_dev(super
, u
->subdev
);
8718 map
= get_imsm_map(dev
, MAP_0
);
8719 current_level
= map
->raid_level
;
8722 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8723 struct mdinfo
*spares
;
8725 spares
= get_spares_for_grow(st
);
8733 makedev(dev
->disk
.major
,
8735 dl
= get_disk_super(super
,
8738 dl
->index
= u
->old_raid_disks
;
8744 len
= disks_to_mpb_size(u
->new_raid_disks
);
8745 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8748 case update_size_change
: {
8751 case update_create_array
: {
8752 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8753 struct intel_dev
*dv
;
8754 struct imsm_dev
*dev
= &u
->dev
;
8755 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8757 struct disk_info
*inf
;
8761 inf
= get_disk_info(u
);
8762 len
= sizeof_imsm_dev(dev
, 1);
8763 /* allocate a new super->devlist entry */
8764 dv
= xmalloc(sizeof(*dv
));
8765 dv
->dev
= xmalloc(len
);
8768 /* count how many spares will be converted to members */
8769 for (i
= 0; i
< map
->num_members
; i
++) {
8770 dl
= serial_to_dl(inf
[i
].serial
, super
);
8772 /* hmm maybe it failed?, nothing we can do about
8777 if (count_memberships(dl
, super
) == 0)
8780 len
+= activate
* sizeof(struct imsm_disk
);
8787 /* check if we need a larger metadata buffer */
8788 if (super
->next_buf
)
8789 buf_len
= super
->next_len
;
8791 buf_len
= super
->len
;
8793 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8794 /* ok we need a larger buf than what is currently allocated
8795 * if this allocation fails process_update will notice that
8796 * ->next_len is set and ->next_buf is NULL
8798 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8799 if (super
->next_buf
)
8800 free(super
->next_buf
);
8802 super
->next_len
= buf_len
;
8803 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8804 memset(super
->next_buf
, 0, buf_len
);
8806 super
->next_buf
= NULL
;
8810 /* must be called while manager is quiesced */
8811 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8813 struct imsm_super
*mpb
= super
->anchor
;
8815 struct imsm_dev
*dev
;
8816 struct imsm_map
*map
;
8817 int i
, j
, num_members
;
8820 dprintf("%s: deleting device[%d] from imsm_super\n",
8823 /* shift all indexes down one */
8824 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8825 if (iter
->index
> (int)index
)
8827 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8828 if (iter
->index
> (int)index
)
8831 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8832 dev
= get_imsm_dev(super
, i
);
8833 map
= get_imsm_map(dev
, MAP_0
);
8834 num_members
= map
->num_members
;
8835 for (j
= 0; j
< num_members
; j
++) {
8836 /* update ord entries being careful not to propagate
8837 * ord-flags to the first map
8839 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8841 if (ord_to_idx(ord
) <= index
)
8844 map
= get_imsm_map(dev
, MAP_0
);
8845 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8846 map
= get_imsm_map(dev
, MAP_1
);
8848 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8853 super
->updates_pending
++;
8855 struct dl
*dl
= *dlp
;
8857 *dlp
= (*dlp
)->next
;
8858 __free_imsm_disk(dl
);
8861 #endif /* MDASSEMBLE */
8863 static void close_targets(int *targets
, int new_disks
)
8870 for (i
= 0; i
< new_disks
; i
++) {
8871 if (targets
[i
] >= 0) {
8878 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8879 struct intel_super
*super
,
8880 struct imsm_dev
*dev
)
8886 struct imsm_map
*map
;
8889 ret_val
= raid_disks
/2;
8890 /* check map if all disks pairs not failed
8893 map
= get_imsm_map(dev
, MAP_0
);
8894 for (i
= 0; i
< ret_val
; i
++) {
8895 int degradation
= 0;
8896 if (get_imsm_disk(super
, i
) == NULL
)
8898 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8900 if (degradation
== 2)
8903 map
= get_imsm_map(dev
, MAP_1
);
8904 /* if there is no second map
8905 * result can be returned
8909 /* check degradation in second map
8911 for (i
= 0; i
< ret_val
; i
++) {
8912 int degradation
= 0;
8913 if (get_imsm_disk(super
, i
) == NULL
)
8915 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8917 if (degradation
== 2)
8932 /*******************************************************************************
8933 * Function: open_backup_targets
8934 * Description: Function opens file descriptors for all devices given in
8937 * info : general array info
8938 * raid_disks : number of disks
8939 * raid_fds : table of device's file descriptors
8940 * super : intel super for raid10 degradation check
8941 * dev : intel device for raid10 degradation check
8945 ******************************************************************************/
8946 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
8947 struct intel_super
*super
, struct imsm_dev
*dev
)
8953 for (i
= 0; i
< raid_disks
; i
++)
8956 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8959 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
8960 dprintf("disk is faulty!!\n");
8964 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
8965 (sd
->disk
.raid_disk
< 0))
8968 dn
= map_dev(sd
->disk
.major
,
8970 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
8971 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
8972 fprintf(stderr
, "cannot open component\n");
8977 /* check if maximum array degradation level is not exceeded
8979 if ((raid_disks
- opened
) >
8980 imsm_get_allowed_degradation(info
->new_level
,
8983 fprintf(stderr
, "Not enough disks can be opened.\n");
8984 close_targets(raid_fds
, raid_disks
);
8991 /*******************************************************************************
8992 * Function: init_migr_record_imsm
8993 * Description: Function inits imsm migration record
8995 * super : imsm internal array info
8996 * dev : device under migration
8997 * info : general array info to find the smallest device
9000 ******************************************************************************/
9001 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9002 struct mdinfo
*info
)
9004 struct intel_super
*super
= st
->sb
;
9005 struct migr_record
*migr_rec
= super
->migr_rec
;
9007 unsigned long long dsize
, dev_sectors
;
9008 long long unsigned min_dev_sectors
= -1LLU;
9012 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9013 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9014 unsigned long long num_migr_units
;
9015 unsigned long long array_blocks
;
9017 memset(migr_rec
, 0, sizeof(struct migr_record
));
9018 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9020 /* only ascending reshape supported now */
9021 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9023 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9024 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9025 migr_rec
->dest_depth_per_unit
*=
9026 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9027 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9028 migr_rec
->blocks_per_unit
=
9029 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9030 migr_rec
->dest_depth_per_unit
=
9031 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9032 array_blocks
= info
->component_size
* new_data_disks
;
9034 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9036 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9038 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9040 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9041 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9044 /* Find the smallest dev */
9045 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9046 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9047 fd
= dev_open(nm
, O_RDONLY
);
9050 get_dev_size(fd
, NULL
, &dsize
);
9051 dev_sectors
= dsize
/ 512;
9052 if (dev_sectors
< min_dev_sectors
)
9053 min_dev_sectors
= dev_sectors
;
9056 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9057 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9059 write_imsm_migr_rec(st
);
9064 /*******************************************************************************
9065 * Function: save_backup_imsm
9066 * Description: Function saves critical data stripes to Migration Copy Area
9067 * and updates the current migration unit status.
9068 * Use restore_stripes() to form a destination stripe,
9069 * and to write it to the Copy Area.
9071 * st : supertype information
9072 * dev : imsm device that backup is saved for
9073 * info : general array info
9074 * buf : input buffer
9075 * length : length of data to backup (blocks_per_unit)
9079 ******************************************************************************/
9080 int save_backup_imsm(struct supertype
*st
,
9081 struct imsm_dev
*dev
,
9082 struct mdinfo
*info
,
9087 struct intel_super
*super
= st
->sb
;
9088 unsigned long long *target_offsets
= NULL
;
9089 int *targets
= NULL
;
9091 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9092 int new_disks
= map_dest
->num_members
;
9093 int dest_layout
= 0;
9095 unsigned long long start
;
9096 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9098 targets
= xmalloc(new_disks
* sizeof(int));
9100 for (i
= 0; i
< new_disks
; i
++)
9103 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
9105 start
= info
->reshape_progress
* 512;
9106 for (i
= 0; i
< new_disks
; i
++) {
9107 target_offsets
[i
] = (unsigned long long)
9108 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9109 /* move back copy area adderss, it will be moved forward
9110 * in restore_stripes() using start input variable
9112 target_offsets
[i
] -= start
/data_disks
;
9115 if (open_backup_targets(info
, new_disks
, targets
,
9119 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9120 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9122 if (restore_stripes(targets
, /* list of dest devices */
9123 target_offsets
, /* migration record offsets */
9126 map_dest
->raid_level
,
9128 -1, /* source backup file descriptor */
9129 0, /* input buf offset
9130 * always 0 buf is already offseted */
9134 pr_err("Error restoring stripes\n");
9142 close_targets(targets
, new_disks
);
9145 free(target_offsets
);
9150 /*******************************************************************************
9151 * Function: save_checkpoint_imsm
9152 * Description: Function called for current unit status update
9153 * in the migration record. It writes it to disk.
9155 * super : imsm internal array info
9156 * info : general array info
9160 * 2: failure, means no valid migration record
9161 * / no general migration in progress /
9162 ******************************************************************************/
9163 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9165 struct intel_super
*super
= st
->sb
;
9166 unsigned long long blocks_per_unit
;
9167 unsigned long long curr_migr_unit
;
9169 if (load_imsm_migr_rec(super
, info
) != 0) {
9170 dprintf("imsm: ERROR: Cannot read migration record "
9171 "for checkpoint save.\n");
9175 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9176 if (blocks_per_unit
== 0) {
9177 dprintf("imsm: no migration in progress.\n");
9180 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9181 /* check if array is alligned to copy area
9182 * if it is not alligned, add one to current migration unit value
9183 * this can happend on array reshape finish only
9185 if (info
->reshape_progress
% blocks_per_unit
)
9188 super
->migr_rec
->curr_migr_unit
=
9189 __cpu_to_le32(curr_migr_unit
);
9190 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9191 super
->migr_rec
->dest_1st_member_lba
=
9192 __cpu_to_le32(curr_migr_unit
*
9193 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9194 if (write_imsm_migr_rec(st
) < 0) {
9195 dprintf("imsm: Cannot write migration record "
9196 "outside backup area\n");
9203 /*******************************************************************************
9204 * Function: recover_backup_imsm
9205 * Description: Function recovers critical data from the Migration Copy Area
9206 * while assembling an array.
9208 * super : imsm internal array info
9209 * info : general array info
9211 * 0 : success (or there is no data to recover)
9213 ******************************************************************************/
9214 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9216 struct intel_super
*super
= st
->sb
;
9217 struct migr_record
*migr_rec
= super
->migr_rec
;
9218 struct imsm_map
*map_dest
= NULL
;
9219 struct intel_dev
*id
= NULL
;
9220 unsigned long long read_offset
;
9221 unsigned long long write_offset
;
9223 int *targets
= NULL
;
9224 int new_disks
, i
, err
;
9227 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9228 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9230 int skipped_disks
= 0;
9232 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9236 /* recover data only during assemblation */
9237 if (strncmp(buffer
, "inactive", 8) != 0)
9239 /* no data to recover */
9240 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9242 if (curr_migr_unit
>= num_migr_units
)
9245 /* find device during reshape */
9246 for (id
= super
->devlist
; id
; id
= id
->next
)
9247 if (is_gen_migration(id
->dev
))
9252 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9253 new_disks
= map_dest
->num_members
;
9255 read_offset
= (unsigned long long)
9256 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9258 write_offset
= ((unsigned long long)
9259 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9260 pba_of_lba0(map_dest
)) * 512;
9262 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9263 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9265 targets
= xcalloc(new_disks
, sizeof(int));
9267 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9268 pr_err("Cannot open some devices belonging to array.\n");
9272 for (i
= 0; i
< new_disks
; i
++) {
9273 if (targets
[i
] < 0) {
9277 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9278 pr_err("Cannot seek to block: %s\n",
9283 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9284 pr_err("Cannot read copy area block: %s\n",
9289 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9290 pr_err("Cannot seek to block: %s\n",
9295 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9296 pr_err("Cannot restore block: %s\n",
9303 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9307 pr_err("Cannot restore data from backup."
9308 " Too many failed disks\n");
9312 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9313 /* ignore error == 2, this can mean end of reshape here
9315 dprintf("imsm: Cannot write checkpoint to "
9316 "migration record (UNIT_SRC_NORMAL) during restart\n");
9322 for (i
= 0; i
< new_disks
; i
++)
9331 static char disk_by_path
[] = "/dev/disk/by-path/";
9333 static const char *imsm_get_disk_controller_domain(const char *path
)
9335 char disk_path
[PATH_MAX
];
9339 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
9340 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9341 if (stat(disk_path
, &st
) == 0) {
9342 struct sys_dev
* hba
;
9345 path
= devt_to_devpath(st
.st_rdev
);
9348 hba
= find_disk_attached_hba(-1, path
);
9349 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9351 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9355 dprintf("path: %s hba: %s attached: %s\n",
9356 path
, (hba
) ? hba
->path
: "NULL", drv
);
9364 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
9366 char subdev_name
[20];
9367 struct mdstat_ent
*mdstat
;
9369 sprintf(subdev_name
, "%d", subdev
);
9370 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9374 *minor
= mdstat
->devnum
;
9375 free_mdstat(mdstat
);
9379 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9380 struct geo_params
*geo
,
9381 int *old_raid_disks
,
9384 /* currently we only support increasing the number of devices
9385 * for a container. This increases the number of device for each
9386 * member array. They must all be RAID0 or RAID5.
9389 struct mdinfo
*info
, *member
;
9390 int devices_that_can_grow
= 0;
9392 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
9393 "st->devnum = (%i)\n",
9396 if (geo
->size
> 0 ||
9397 geo
->level
!= UnSet
||
9398 geo
->layout
!= UnSet
||
9399 geo
->chunksize
!= 0 ||
9400 geo
->raid_disks
== UnSet
) {
9401 dprintf("imsm: Container operation is allowed for "
9402 "raid disks number change only.\n");
9406 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9407 dprintf("imsm: Metadata changes rollback is not supported for "
9408 "container operation.\n");
9412 info
= container_content_imsm(st
, NULL
);
9413 for (member
= info
; member
; member
= member
->next
) {
9417 dprintf("imsm: checking device_num: %i\n",
9418 member
->container_member
);
9420 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9421 /* we work on container for Online Capacity Expansion
9422 * only so raid_disks has to grow
9424 dprintf("imsm: for container operation raid disks "
9425 "increase is required\n");
9429 if ((info
->array
.level
!= 0) &&
9430 (info
->array
.level
!= 5)) {
9431 /* we cannot use this container with other raid level
9433 dprintf("imsm: for container operation wrong"
9434 " raid level (%i) detected\n",
9438 /* check for platform support
9439 * for this raid level configuration
9441 struct intel_super
*super
= st
->sb
;
9442 if (!is_raid_level_supported(super
->orom
,
9443 member
->array
.level
,
9445 dprintf("platform does not support raid%d with"
9449 geo
->raid_disks
> 1 ? "s" : "");
9452 /* check if component size is aligned to chunk size
9454 if (info
->component_size
%
9455 (info
->array
.chunk_size
/512)) {
9456 dprintf("Component size is not aligned to "
9462 if (*old_raid_disks
&&
9463 info
->array
.raid_disks
!= *old_raid_disks
)
9465 *old_raid_disks
= info
->array
.raid_disks
;
9467 /* All raid5 and raid0 volumes in container
9468 * have to be ready for Online Capacity Expansion
9469 * so they need to be assembled. We have already
9470 * checked that no recovery etc is happening.
9472 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
9476 dprintf("imsm: cannot find array\n");
9479 devices_that_can_grow
++;
9482 if (!member
&& devices_that_can_grow
)
9486 dprintf("\tContainer operation allowed\n");
9488 dprintf("\tError: %i\n", ret_val
);
9493 /* Function: get_spares_for_grow
9494 * Description: Allocates memory and creates list of spare devices
9495 * avaliable in container. Checks if spare drive size is acceptable.
9496 * Parameters: Pointer to the supertype structure
9497 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9500 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9502 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9503 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9506 /******************************************************************************
9507 * function: imsm_create_metadata_update_for_reshape
9508 * Function creates update for whole IMSM container.
9510 ******************************************************************************/
9511 static int imsm_create_metadata_update_for_reshape(
9512 struct supertype
*st
,
9513 struct geo_params
*geo
,
9515 struct imsm_update_reshape
**updatep
)
9517 struct intel_super
*super
= st
->sb
;
9518 struct imsm_super
*mpb
= super
->anchor
;
9519 int update_memory_size
= 0;
9520 struct imsm_update_reshape
*u
= NULL
;
9521 struct mdinfo
*spares
= NULL
;
9523 int delta_disks
= 0;
9526 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
9529 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9531 /* size of all update data without anchor */
9532 update_memory_size
= sizeof(struct imsm_update_reshape
);
9534 /* now add space for spare disks that we need to add. */
9535 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9537 u
= xcalloc(1, update_memory_size
);
9538 u
->type
= update_reshape_container_disks
;
9539 u
->old_raid_disks
= old_raid_disks
;
9540 u
->new_raid_disks
= geo
->raid_disks
;
9542 /* now get spare disks list
9544 spares
= get_spares_for_grow(st
);
9547 || delta_disks
> spares
->array
.spare_disks
) {
9548 pr_err("imsm: ERROR: Cannot get spare devices "
9549 "for %s.\n", geo
->dev_name
);
9554 /* we have got spares
9555 * update disk list in imsm_disk list table in anchor
9557 dprintf("imsm: %i spares are available.\n\n",
9558 spares
->array
.spare_disks
);
9561 for (i
= 0; i
< delta_disks
; i
++) {
9566 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9568 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9569 dl
->index
= mpb
->num_disks
;
9579 dprintf("imsm: reshape update preparation :");
9580 if (i
== delta_disks
) {
9583 return update_memory_size
;
9586 dprintf(" Error\n");
9592 /******************************************************************************
9593 * function: imsm_create_metadata_update_for_size_change()
9594 * Creates update for IMSM array for array size change.
9596 ******************************************************************************/
9597 static int imsm_create_metadata_update_for_size_change(
9598 struct supertype
*st
,
9599 struct geo_params
*geo
,
9600 struct imsm_update_size_change
**updatep
)
9602 struct intel_super
*super
= st
->sb
;
9603 int update_memory_size
= 0;
9604 struct imsm_update_size_change
*u
= NULL
;
9606 dprintf("imsm_create_metadata_update_for_size_change(enter)"
9607 " New size = %llu\n", geo
->size
);
9609 /* size of all update data without anchor */
9610 update_memory_size
= sizeof(struct imsm_update_size_change
);
9612 u
= xcalloc(1, update_memory_size
);
9613 u
->type
= update_size_change
;
9614 u
->subdev
= super
->current_vol
;
9615 u
->new_size
= geo
->size
;
9617 dprintf("imsm: reshape update preparation : OK\n");
9620 return update_memory_size
;
9623 /******************************************************************************
9624 * function: imsm_create_metadata_update_for_migration()
9625 * Creates update for IMSM array.
9627 ******************************************************************************/
9628 static int imsm_create_metadata_update_for_migration(
9629 struct supertype
*st
,
9630 struct geo_params
*geo
,
9631 struct imsm_update_reshape_migration
**updatep
)
9633 struct intel_super
*super
= st
->sb
;
9634 int update_memory_size
= 0;
9635 struct imsm_update_reshape_migration
*u
= NULL
;
9636 struct imsm_dev
*dev
;
9637 int previous_level
= -1;
9639 dprintf("imsm_create_metadata_update_for_migration(enter)"
9640 " New Level = %i\n", geo
->level
);
9642 /* size of all update data without anchor */
9643 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9645 u
= xcalloc(1, update_memory_size
);
9646 u
->type
= update_reshape_migration
;
9647 u
->subdev
= super
->current_vol
;
9648 u
->new_level
= geo
->level
;
9649 u
->new_layout
= geo
->layout
;
9650 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9651 u
->new_disks
[0] = -1;
9652 u
->new_chunksize
= -1;
9654 dev
= get_imsm_dev(super
, u
->subdev
);
9656 struct imsm_map
*map
;
9658 map
= get_imsm_map(dev
, MAP_0
);
9660 int current_chunk_size
=
9661 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9663 if (geo
->chunksize
!= current_chunk_size
) {
9664 u
->new_chunksize
= geo
->chunksize
/ 1024;
9666 "chunk size change from %i to %i\n",
9667 current_chunk_size
, u
->new_chunksize
);
9669 previous_level
= map
->raid_level
;
9672 if ((geo
->level
== 5) && (previous_level
== 0)) {
9673 struct mdinfo
*spares
= NULL
;
9675 u
->new_raid_disks
++;
9676 spares
= get_spares_for_grow(st
);
9677 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9680 update_memory_size
= 0;
9681 dprintf("error: cannot get spare device "
9682 "for requested migration");
9687 dprintf("imsm: reshape update preparation : OK\n");
9690 return update_memory_size
;
9693 static void imsm_update_metadata_locally(struct supertype
*st
,
9696 struct metadata_update mu
;
9701 mu
.space_list
= NULL
;
9703 imsm_prepare_update(st
, &mu
);
9704 imsm_process_update(st
, &mu
);
9706 while (mu
.space_list
) {
9707 void **space
= mu
.space_list
;
9708 mu
.space_list
= *space
;
9713 /***************************************************************************
9714 * Function: imsm_analyze_change
9715 * Description: Function analyze change for single volume
9716 * and validate if transition is supported
9717 * Parameters: Geometry parameters, supertype structure,
9718 * metadata change direction (apply/rollback)
9719 * Returns: Operation type code on success, -1 if fail
9720 ****************************************************************************/
9721 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9722 struct geo_params
*geo
,
9729 /* number of added/removed disks in operation result */
9730 int devNumChange
= 0;
9731 /* imsm compatible layout value for array geometry verification */
9732 int imsm_layout
= -1;
9734 struct imsm_dev
*dev
;
9735 struct intel_super
*super
;
9736 unsigned long long current_size
;
9737 unsigned long long free_size
;
9738 unsigned long long max_size
;
9741 getinfo_super_imsm_volume(st
, &info
, NULL
);
9742 if ((geo
->level
!= info
.array
.level
) &&
9743 (geo
->level
>= 0) &&
9744 (geo
->level
!= UnSet
)) {
9745 switch (info
.array
.level
) {
9747 if (geo
->level
== 5) {
9748 change
= CH_MIGRATION
;
9749 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9750 pr_err("Error. Requested Layout "
9751 "not supported (left-asymmetric layout "
9752 "is supported only)!\n");
9754 goto analyse_change_exit
;
9756 imsm_layout
= geo
->layout
;
9758 devNumChange
= 1; /* parity disk added */
9759 } else if (geo
->level
== 10) {
9760 change
= CH_TAKEOVER
;
9762 devNumChange
= 2; /* two mirrors added */
9763 imsm_layout
= 0x102; /* imsm supported layout */
9768 if (geo
->level
== 0) {
9769 change
= CH_TAKEOVER
;
9771 devNumChange
= -(geo
->raid_disks
/2);
9772 imsm_layout
= 0; /* imsm raid0 layout */
9777 pr_err("Error. Level Migration from %d to %d "
9779 info
.array
.level
, geo
->level
);
9780 goto analyse_change_exit
;
9783 geo
->level
= info
.array
.level
;
9785 if ((geo
->layout
!= info
.array
.layout
)
9786 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9787 change
= CH_MIGRATION
;
9788 if ((info
.array
.layout
== 0)
9789 && (info
.array
.level
== 5)
9790 && (geo
->layout
== 5)) {
9791 /* reshape 5 -> 4 */
9792 } else if ((info
.array
.layout
== 5)
9793 && (info
.array
.level
== 5)
9794 && (geo
->layout
== 0)) {
9795 /* reshape 4 -> 5 */
9799 pr_err("Error. Layout Migration from %d to %d "
9801 info
.array
.layout
, geo
->layout
);
9803 goto analyse_change_exit
;
9806 geo
->layout
= info
.array
.layout
;
9807 if (imsm_layout
== -1)
9808 imsm_layout
= info
.array
.layout
;
9811 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9812 && (geo
->chunksize
!= info
.array
.chunk_size
))
9813 change
= CH_MIGRATION
;
9815 geo
->chunksize
= info
.array
.chunk_size
;
9817 chunk
= geo
->chunksize
/ 1024;
9820 dev
= get_imsm_dev(super
, super
->current_vol
);
9821 data_disks
= imsm_num_data_members(dev
, MAP_0
);
9822 /* compute current size per disk member
9824 current_size
= info
.custom_array_size
/ data_disks
;
9826 if ((geo
->size
> 0) && (geo
->size
!= MAX_SIZE
)) {
9827 /* align component size
9829 geo
->size
= imsm_component_size_aligment_check(
9830 get_imsm_raid_level(dev
->vol
.map
),
9833 if (geo
->size
== 0) {
9834 pr_err("Error. Size expansion is " \
9835 "supported only (current size is %llu, " \
9836 "requested size /rounded/ is 0).\n",
9838 goto analyse_change_exit
;
9842 if ((current_size
!= geo
->size
) && (geo
->size
> 0)) {
9844 pr_err("Error. Size change should be the only "
9845 "one at a time.\n");
9847 goto analyse_change_exit
;
9849 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
9850 pr_err("Error. The last volume in container "
9851 "can be expanded only (%i/%i).\n",
9852 super
->current_vol
, st
->devnum
);
9853 goto analyse_change_exit
;
9855 /* check the maximum available size
9857 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
9858 0, chunk
, &free_size
);
9860 /* Cannot find maximum available space
9864 max_size
= free_size
+ current_size
;
9865 /* align component size
9867 max_size
= imsm_component_size_aligment_check(
9868 get_imsm_raid_level(dev
->vol
.map
),
9872 if (geo
->size
== MAX_SIZE
) {
9873 /* requested size change to the maximum available size
9875 if (max_size
== 0) {
9876 pr_err("Error. Cannot find "
9877 "maximum available space.\n");
9879 goto analyse_change_exit
;
9881 geo
->size
= max_size
;
9884 if ((direction
== ROLLBACK_METADATA_CHANGES
)) {
9885 /* accept size for rollback only
9888 /* round size due to metadata compatibility
9890 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
9891 << SECT_PER_MB_SHIFT
;
9892 dprintf("Prepare update for size change to %llu\n",
9894 if (current_size
>= geo
->size
) {
9895 pr_err("Error. Size expansion is "
9896 "supported only (current size is %llu, "
9897 "requested size /rounded/ is %llu).\n",
9898 current_size
, geo
->size
);
9899 goto analyse_change_exit
;
9901 if (max_size
&& geo
->size
> max_size
) {
9902 pr_err("Error. Requested size is larger "
9903 "than maximum available size (maximum "
9904 "available size is %llu, "
9905 "requested size /rounded/ is %llu).\n",
9906 max_size
, geo
->size
);
9907 goto analyse_change_exit
;
9910 geo
->size
*= data_disks
;
9911 geo
->raid_disks
= dev
->vol
.map
->num_members
;
9912 change
= CH_ARRAY_SIZE
;
9914 if (!validate_geometry_imsm(st
,
9917 geo
->raid_disks
+ devNumChange
,
9919 geo
->size
, INVALID_SECTORS
,
9924 struct intel_super
*super
= st
->sb
;
9925 struct imsm_super
*mpb
= super
->anchor
;
9927 if (mpb
->num_raid_devs
> 1) {
9928 pr_err("Error. Cannot perform operation on %s"
9929 "- for this operation it MUST be single "
9930 "array in container\n",
9936 analyse_change_exit
:
9937 if ((direction
== ROLLBACK_METADATA_CHANGES
) &&
9938 ((change
== CH_MIGRATION
) || (change
== CH_TAKEOVER
))) {
9939 dprintf("imsm: Metadata changes rollback is not supported for "
9940 "migration and takeover operations.\n");
9946 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
9948 struct intel_super
*super
= st
->sb
;
9949 struct imsm_update_takeover
*u
;
9951 u
= xmalloc(sizeof(struct imsm_update_takeover
));
9953 u
->type
= update_takeover
;
9954 u
->subarray
= super
->current_vol
;
9956 /* 10->0 transition */
9957 if (geo
->level
== 0)
9958 u
->direction
= R10_TO_R0
;
9960 /* 0->10 transition */
9961 if (geo
->level
== 10)
9962 u
->direction
= R0_TO_R10
;
9964 /* update metadata locally */
9965 imsm_update_metadata_locally(st
, u
,
9966 sizeof(struct imsm_update_takeover
));
9967 /* and possibly remotely */
9968 if (st
->update_tail
)
9969 append_metadata_update(st
, u
,
9970 sizeof(struct imsm_update_takeover
));
9977 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
9979 int layout
, int chunksize
, int raid_disks
,
9980 int delta_disks
, char *backup
, char *dev
,
9981 int direction
, int verbose
)
9984 struct geo_params geo
;
9986 dprintf("imsm: reshape_super called.\n");
9988 memset(&geo
, 0, sizeof(struct geo_params
));
9991 geo
.dev_id
= st
->devnum
;
9994 geo
.layout
= layout
;
9995 geo
.chunksize
= chunksize
;
9996 geo
.raid_disks
= raid_disks
;
9997 if (delta_disks
!= UnSet
)
9998 geo
.raid_disks
+= delta_disks
;
10000 dprintf("\tfor level : %i\n", geo
.level
);
10001 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
10003 if (experimental() == 0)
10006 if (st
->container_dev
== st
->devnum
) {
10007 /* On container level we can only increase number of devices. */
10008 dprintf("imsm: info: Container operation\n");
10009 int old_raid_disks
= 0;
10011 if (imsm_reshape_is_allowed_on_container(
10012 st
, &geo
, &old_raid_disks
, direction
)) {
10013 struct imsm_update_reshape
*u
= NULL
;
10016 len
= imsm_create_metadata_update_for_reshape(
10017 st
, &geo
, old_raid_disks
, &u
);
10020 dprintf("imsm: Cannot prepare update\n");
10021 goto exit_imsm_reshape_super
;
10025 /* update metadata locally */
10026 imsm_update_metadata_locally(st
, u
, len
);
10027 /* and possibly remotely */
10028 if (st
->update_tail
)
10029 append_metadata_update(st
, u
, len
);
10034 pr_err("(imsm) Operation "
10035 "is not allowed on this container\n");
10038 /* On volume level we support following operations
10039 * - takeover: raid10 -> raid0; raid0 -> raid10
10040 * - chunk size migration
10041 * - migration: raid5 -> raid0; raid0 -> raid5
10043 struct intel_super
*super
= st
->sb
;
10044 struct intel_dev
*dev
= super
->devlist
;
10045 int change
, devnum
;
10046 dprintf("imsm: info: Volume operation\n");
10047 /* find requested device */
10049 if (imsm_find_array_minor_by_subdev(
10050 dev
->index
, st
->container_dev
, &devnum
) == 0
10051 && devnum
== geo
.dev_id
)
10056 pr_err("Cannot find %s (%i) subarray\n",
10057 geo
.dev_name
, geo
.dev_id
);
10058 goto exit_imsm_reshape_super
;
10060 super
->current_vol
= dev
->index
;
10061 change
= imsm_analyze_change(st
, &geo
, direction
);
10064 ret_val
= imsm_takeover(st
, &geo
);
10066 case CH_MIGRATION
: {
10067 struct imsm_update_reshape_migration
*u
= NULL
;
10069 imsm_create_metadata_update_for_migration(
10073 "Cannot prepare update\n");
10077 /* update metadata locally */
10078 imsm_update_metadata_locally(st
, u
, len
);
10079 /* and possibly remotely */
10080 if (st
->update_tail
)
10081 append_metadata_update(st
, u
, len
);
10086 case CH_ARRAY_SIZE
: {
10087 struct imsm_update_size_change
*u
= NULL
;
10089 imsm_create_metadata_update_for_size_change(
10093 "Cannot prepare update\n");
10097 /* update metadata locally */
10098 imsm_update_metadata_locally(st
, u
, len
);
10099 /* and possibly remotely */
10100 if (st
->update_tail
)
10101 append_metadata_update(st
, u
, len
);
10111 exit_imsm_reshape_super
:
10112 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10116 /*******************************************************************************
10117 * Function: wait_for_reshape_imsm
10118 * Description: Function writes new sync_max value and waits until
10119 * reshape process reach new position
10121 * sra : general array info
10122 * ndata : number of disks in new array's layout
10125 * 1 : there is no reshape in progress,
10127 ******************************************************************************/
10128 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10130 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
10131 unsigned long long completed
;
10132 /* to_complete : new sync_max position */
10133 unsigned long long to_complete
= sra
->reshape_progress
;
10134 unsigned long long position_to_set
= to_complete
/ ndata
;
10137 dprintf("imsm: wait_for_reshape_imsm() "
10138 "cannot open reshape_position\n");
10142 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10143 dprintf("imsm: wait_for_reshape_imsm() "
10144 "cannot read reshape_position (no reshape in progres)\n");
10149 if (completed
> to_complete
) {
10150 dprintf("imsm: wait_for_reshape_imsm() "
10151 "wrong next position to set %llu (%llu)\n",
10152 to_complete
, completed
);
10156 dprintf("Position set: %llu\n", position_to_set
);
10157 if (sysfs_set_num(sra
, NULL
, "sync_max",
10158 position_to_set
) != 0) {
10159 dprintf("imsm: wait_for_reshape_imsm() "
10160 "cannot set reshape position to %llu\n",
10171 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
10172 if (sysfs_get_str(sra
, NULL
, "sync_action",
10174 strncmp(action
, "reshape", 7) != 0)
10176 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10177 dprintf("imsm: wait_for_reshape_imsm() "
10178 "cannot read reshape_position (in loop)\n");
10182 } while (completed
< to_complete
);
10188 /*******************************************************************************
10189 * Function: check_degradation_change
10190 * Description: Check that array hasn't become failed.
10192 * info : for sysfs access
10193 * sources : source disks descriptors
10194 * degraded: previous degradation level
10196 * degradation level
10197 ******************************************************************************/
10198 int check_degradation_change(struct mdinfo
*info
,
10202 unsigned long long new_degraded
;
10205 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10206 if ((rv
== -1) || (new_degraded
!= (unsigned long long)degraded
)) {
10207 /* check each device to ensure it is still working */
10210 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10211 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10213 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10215 if (sysfs_get_str(info
,
10216 sd
, "state", sbuf
, 20) < 0 ||
10217 strstr(sbuf
, "faulty") ||
10218 strstr(sbuf
, "in_sync") == NULL
) {
10219 /* this device is dead */
10220 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10221 if (sd
->disk
.raid_disk
>= 0 &&
10222 sources
[sd
->disk
.raid_disk
] >= 0) {
10224 sd
->disk
.raid_disk
]);
10225 sources
[sd
->disk
.raid_disk
] =
10234 return new_degraded
;
10237 /*******************************************************************************
10238 * Function: imsm_manage_reshape
10239 * Description: Function finds array under reshape and it manages reshape
10240 * process. It creates stripes backups (if required) and sets
10243 * afd : Backup handle (nattive) - not used
10244 * sra : general array info
10245 * reshape : reshape parameters - not used
10246 * st : supertype structure
10247 * blocks : size of critical section [blocks]
10248 * fds : table of source device descriptor
10249 * offsets : start of array (offest per devices)
10251 * destfd : table of destination device descriptor
10252 * destoffsets : table of destination offsets (per device)
10254 * 1 : success, reshape is done
10256 ******************************************************************************/
10257 static int imsm_manage_reshape(
10258 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10259 struct supertype
*st
, unsigned long backup_blocks
,
10260 int *fds
, unsigned long long *offsets
,
10261 int dests
, int *destfd
, unsigned long long *destoffsets
)
10264 struct intel_super
*super
= st
->sb
;
10265 struct intel_dev
*dv
= NULL
;
10266 struct imsm_dev
*dev
= NULL
;
10267 struct imsm_map
*map_src
;
10268 int migr_vol_qan
= 0;
10269 int ndata
, odata
; /* [bytes] */
10270 int chunk
; /* [bytes] */
10271 struct migr_record
*migr_rec
;
10273 unsigned int buf_size
; /* [bytes] */
10274 unsigned long long max_position
; /* array size [bytes] */
10275 unsigned long long next_step
; /* [blocks]/[bytes] */
10276 unsigned long long old_data_stripe_length
;
10277 unsigned long long start_src
; /* [bytes] */
10278 unsigned long long start
; /* [bytes] */
10279 unsigned long long start_buf_shift
; /* [bytes] */
10281 int source_layout
= 0;
10283 if (!fds
|| !offsets
|| !sra
)
10286 /* Find volume during the reshape */
10287 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10288 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10289 && dv
->dev
->vol
.migr_state
== 1) {
10294 /* Only one volume can migrate at the same time */
10295 if (migr_vol_qan
!= 1) {
10296 pr_err(": %s", migr_vol_qan
?
10297 "Number of migrating volumes greater than 1\n" :
10298 "There is no volume during migrationg\n");
10302 map_src
= get_imsm_map(dev
, MAP_1
);
10303 if (map_src
== NULL
)
10306 ndata
= imsm_num_data_members(dev
, MAP_0
);
10307 odata
= imsm_num_data_members(dev
, MAP_1
);
10309 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10310 old_data_stripe_length
= odata
* chunk
;
10312 migr_rec
= super
->migr_rec
;
10314 /* initialize migration record for start condition */
10315 if (sra
->reshape_progress
== 0)
10316 init_migr_record_imsm(st
, dev
, sra
);
10318 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10319 dprintf("imsm: cannot restart migration when data "
10320 "are present in copy area.\n");
10323 /* Save checkpoint to update migration record for current
10324 * reshape position (in md). It can be farther than current
10325 * reshape position in metadata.
10327 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10328 /* ignore error == 2, this can mean end of reshape here
10330 dprintf("imsm: Cannot write checkpoint to "
10331 "migration record (UNIT_SRC_NORMAL, "
10332 "initial save)\n");
10337 /* size for data */
10338 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10339 /* extend buffer size for parity disk */
10340 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10341 /* add space for stripe aligment */
10342 buf_size
+= old_data_stripe_length
;
10343 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10344 dprintf("imsm: Cannot allocate checpoint buffer\n");
10348 max_position
= sra
->component_size
* ndata
;
10349 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10351 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10352 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10353 /* current reshape position [blocks] */
10354 unsigned long long current_position
=
10355 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10356 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10357 unsigned long long border
;
10359 /* Check that array hasn't become failed.
10361 degraded
= check_degradation_change(sra
, fds
, degraded
);
10362 if (degraded
> 1) {
10363 dprintf("imsm: Abort reshape due to degradation"
10364 " level (%i)\n", degraded
);
10368 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10370 if ((current_position
+ next_step
) > max_position
)
10371 next_step
= max_position
- current_position
;
10373 start
= current_position
* 512;
10375 /* allign reading start to old geometry */
10376 start_buf_shift
= start
% old_data_stripe_length
;
10377 start_src
= start
- start_buf_shift
;
10379 border
= (start_src
/ odata
) - (start
/ ndata
);
10381 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10382 /* save critical stripes to buf
10383 * start - start address of current unit
10384 * to backup [bytes]
10385 * start_src - start address of current unit
10386 * to backup alligned to source array
10389 unsigned long long next_step_filler
= 0;
10390 unsigned long long copy_length
= next_step
* 512;
10392 /* allign copy area length to stripe in old geometry */
10393 next_step_filler
= ((copy_length
+ start_buf_shift
)
10394 % old_data_stripe_length
);
10395 if (next_step_filler
)
10396 next_step_filler
= (old_data_stripe_length
10397 - next_step_filler
);
10398 dprintf("save_stripes() parameters: start = %llu,"
10399 "\tstart_src = %llu,\tnext_step*512 = %llu,"
10400 "\tstart_in_buf_shift = %llu,"
10401 "\tnext_step_filler = %llu\n",
10402 start
, start_src
, copy_length
,
10403 start_buf_shift
, next_step_filler
);
10405 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10406 chunk
, map_src
->raid_level
,
10407 source_layout
, 0, NULL
, start_src
,
10409 next_step_filler
+ start_buf_shift
,
10411 dprintf("imsm: Cannot save stripes"
10415 /* Convert data to destination format and store it
10416 * in backup general migration area
10418 if (save_backup_imsm(st
, dev
, sra
,
10419 buf
+ start_buf_shift
, copy_length
)) {
10420 dprintf("imsm: Cannot save stripes to "
10421 "target devices\n");
10424 if (save_checkpoint_imsm(st
, sra
,
10425 UNIT_SRC_IN_CP_AREA
)) {
10426 dprintf("imsm: Cannot write checkpoint to "
10427 "migration record (UNIT_SRC_IN_CP_AREA)\n");
10431 /* set next step to use whole border area */
10432 border
/= next_step
;
10434 next_step
*= border
;
10436 /* When data backed up, checkpoint stored,
10437 * kick the kernel to reshape unit of data
10439 next_step
= next_step
+ sra
->reshape_progress
;
10440 /* limit next step to array max position */
10441 if (next_step
> max_position
)
10442 next_step
= max_position
;
10443 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10444 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10445 sra
->reshape_progress
= next_step
;
10447 /* wait until reshape finish */
10448 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
10449 dprintf("wait_for_reshape_imsm returned error!\n");
10453 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10454 /* ignore error == 2, this can mean end of reshape here
10456 dprintf("imsm: Cannot write checkpoint to "
10457 "migration record (UNIT_SRC_NORMAL)\n");
10463 /* return '1' if done */
10467 abort_reshape(sra
);
10471 #endif /* MDASSEMBLE */
10473 struct superswitch super_imsm
= {
10475 .examine_super
= examine_super_imsm
,
10476 .brief_examine_super
= brief_examine_super_imsm
,
10477 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10478 .export_examine_super
= export_examine_super_imsm
,
10479 .detail_super
= detail_super_imsm
,
10480 .brief_detail_super
= brief_detail_super_imsm
,
10481 .write_init_super
= write_init_super_imsm
,
10482 .validate_geometry
= validate_geometry_imsm
,
10483 .add_to_super
= add_to_super_imsm
,
10484 .remove_from_super
= remove_from_super_imsm
,
10485 .detail_platform
= detail_platform_imsm
,
10486 .export_detail_platform
= export_detail_platform_imsm
,
10487 .kill_subarray
= kill_subarray_imsm
,
10488 .update_subarray
= update_subarray_imsm
,
10489 .load_container
= load_container_imsm
,
10490 .default_geometry
= default_geometry_imsm
,
10491 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10492 .reshape_super
= imsm_reshape_super
,
10493 .manage_reshape
= imsm_manage_reshape
,
10494 .recover_backup
= recover_backup_imsm
,
10496 .match_home
= match_home_imsm
,
10497 .uuid_from_super
= uuid_from_super_imsm
,
10498 .getinfo_super
= getinfo_super_imsm
,
10499 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10500 .update_super
= update_super_imsm
,
10502 .avail_size
= avail_size_imsm
,
10503 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10505 .compare_super
= compare_super_imsm
,
10507 .load_super
= load_super_imsm
,
10508 .init_super
= init_super_imsm
,
10509 .store_super
= store_super_imsm
,
10510 .free_super
= free_super_imsm
,
10511 .match_metadata_desc
= match_metadata_desc_imsm
,
10512 .container_content
= container_content_imsm
,
10520 .open_new
= imsm_open_new
,
10521 .set_array_state
= imsm_set_array_state
,
10522 .set_disk
= imsm_set_disk
,
10523 .sync_metadata
= imsm_sync_metadata
,
10524 .activate_spare
= imsm_activate_spare
,
10525 .process_update
= imsm_process_update
,
10526 .prepare_update
= imsm_prepare_update
,
10527 #endif /* MDASSEMBLE */