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
);
562 /* IMSM metadata disallows to attach disks to multiple
568 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
570 struct sys_dev
*list
, *elem
;
573 if ((list
= find_intel_devices()) == NULL
)
577 disk_path
= (char *) devname
;
579 disk_path
= diskfd_to_devpath(fd
);
584 for (elem
= list
; elem
; elem
= elem
->next
)
585 if (path_attached_to_hba(disk_path
, elem
->path
))
588 if (disk_path
!= devname
)
595 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
598 static struct supertype
*match_metadata_desc_imsm(char *arg
)
600 struct supertype
*st
;
602 if (strcmp(arg
, "imsm") != 0 &&
603 strcmp(arg
, "default") != 0
607 st
= xcalloc(1, sizeof(*st
));
608 st
->ss
= &super_imsm
;
609 st
->max_devs
= IMSM_MAX_DEVICES
;
610 st
->minor_version
= 0;
616 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
618 return &mpb
->sig
[MPB_SIG_LEN
];
622 /* retrieve a disk directly from the anchor when the anchor is known to be
623 * up-to-date, currently only at load time
625 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
627 if (index
>= mpb
->num_disks
)
629 return &mpb
->disk
[index
];
632 /* retrieve the disk description based on a index of the disk
635 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
639 for (d
= super
->disks
; d
; d
= d
->next
)
640 if (d
->index
== index
)
645 /* retrieve a disk from the parsed metadata */
646 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
650 dl
= get_imsm_dl_disk(super
, index
);
657 /* generate a checksum directly from the anchor when the anchor is known to be
658 * up-to-date, currently only at load or write_super after coalescing
660 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
662 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
663 __u32
*p
= (__u32
*) mpb
;
667 sum
+= __le32_to_cpu(*p
);
671 return sum
- __le32_to_cpu(mpb
->check_sum
);
674 static size_t sizeof_imsm_map(struct imsm_map
*map
)
676 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
679 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
681 /* A device can have 2 maps if it is in the middle of a migration.
683 * MAP_0 - we return the first map
684 * MAP_1 - we return the second map if it exists, else NULL
685 * MAP_X - we return the second map if it exists, else the first
687 struct imsm_map
*map
= &dev
->vol
.map
[0];
688 struct imsm_map
*map2
= NULL
;
690 if (dev
->vol
.migr_state
)
691 map2
= (void *)map
+ sizeof_imsm_map(map
);
693 switch (second_map
) {
710 /* return the size of the device.
711 * migr_state increases the returned size if map[0] were to be duplicated
713 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
715 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
716 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
718 /* migrating means an additional map */
719 if (dev
->vol
.migr_state
)
720 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
722 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
728 /* retrieve disk serial number list from a metadata update */
729 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
732 struct disk_info
*inf
;
734 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
735 sizeof_imsm_dev(&update
->dev
, 0);
741 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
747 if (index
>= mpb
->num_raid_devs
)
750 /* devices start after all disks */
751 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
753 for (i
= 0; i
<= index
; i
++)
755 return _mpb
+ offset
;
757 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
762 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
764 struct intel_dev
*dv
;
766 if (index
>= super
->anchor
->num_raid_devs
)
768 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
769 if (dv
->index
== index
)
776 * == MAP_0 get first map
777 * == MAP_1 get second map
778 * == MAP_X than get map according to the current migr_state
780 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
784 struct imsm_map
*map
;
786 map
= get_imsm_map(dev
, second_map
);
788 /* top byte identifies disk under rebuild */
789 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
792 #define ord_to_idx(ord) (((ord) << 8) >> 8)
793 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
795 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
797 return ord_to_idx(ord
);
800 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
802 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
805 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
810 for (slot
= 0; slot
< map
->num_members
; slot
++) {
811 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
812 if (ord_to_idx(ord
) == idx
)
819 static int get_imsm_raid_level(struct imsm_map
*map
)
821 if (map
->raid_level
== 1) {
822 if (map
->num_members
== 2)
828 return map
->raid_level
;
831 static int cmp_extent(const void *av
, const void *bv
)
833 const struct extent
*a
= av
;
834 const struct extent
*b
= bv
;
835 if (a
->start
< b
->start
)
837 if (a
->start
> b
->start
)
842 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
847 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
848 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
849 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
851 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
858 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
860 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
862 if (lo
== 0 || hi
== 0)
864 *lo
= __le32_to_cpu((unsigned)n
);
865 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
869 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
871 return (unsigned long long)__le32_to_cpu(lo
) |
872 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
875 static unsigned long long total_blocks(struct imsm_disk
*disk
)
879 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
882 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
886 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
889 static unsigned long long blocks_per_member(struct imsm_map
*map
)
893 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
897 static unsigned long long num_data_stripes(struct imsm_map
*map
)
901 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
904 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
906 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
910 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
912 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
915 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
917 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
920 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
922 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
925 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
927 /* find a list of used extents on the given physical device */
928 struct extent
*rv
, *e
;
930 int memberships
= count_memberships(dl
, super
);
933 /* trim the reserved area for spares, so they can join any array
934 * regardless of whether the OROM has assigned sectors from the
935 * IMSM_RESERVED_SECTORS region
938 reservation
= imsm_min_reserved_sectors(super
);
940 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
942 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
945 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
946 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
947 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
949 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
950 e
->start
= pba_of_lba0(map
);
951 e
->size
= blocks_per_member(map
);
955 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
957 /* determine the start of the metadata
958 * when no raid devices are defined use the default
959 * ...otherwise allow the metadata to truncate the value
960 * as is the case with older versions of imsm
963 struct extent
*last
= &rv
[memberships
- 1];
964 unsigned long long remainder
;
966 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
967 /* round down to 1k block to satisfy precision of the kernel
971 /* make sure remainder is still sane */
972 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
973 remainder
= ROUND_UP(super
->len
, 512) >> 9;
974 if (reservation
> remainder
)
975 reservation
= remainder
;
977 e
->start
= total_blocks(&dl
->disk
) - reservation
;
982 /* try to determine how much space is reserved for metadata from
983 * the last get_extents() entry, otherwise fallback to the
986 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
992 /* for spares just return a minimal reservation which will grow
993 * once the spare is picked up by an array
996 return MPB_SECTOR_CNT
;
998 e
= get_extents(super
, dl
);
1000 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1002 /* scroll to last entry */
1003 for (i
= 0; e
[i
].size
; i
++)
1006 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1013 static int is_spare(struct imsm_disk
*disk
)
1015 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1018 static int is_configured(struct imsm_disk
*disk
)
1020 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1023 static int is_failed(struct imsm_disk
*disk
)
1025 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1028 /* try to determine how much space is reserved for metadata from
1029 * the last get_extents() entry on the smallest active disk,
1030 * otherwise fallback to the default
1032 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1036 unsigned long long min_active
;
1038 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1039 struct dl
*dl
, *dl_min
= NULL
;
1045 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1048 unsigned long long blocks
= total_blocks(&dl
->disk
);
1049 if (blocks
< min_active
|| min_active
== 0) {
1051 min_active
= blocks
;
1057 /* find last lba used by subarrays on the smallest active disk */
1058 e
= get_extents(super
, dl_min
);
1061 for (i
= 0; e
[i
].size
; i
++)
1064 remainder
= min_active
- e
[i
].start
;
1067 /* to give priority to recovery we should not require full
1068 IMSM_RESERVED_SECTORS from the spare */
1069 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1071 /* if real reservation is smaller use that value */
1072 return (remainder
< rv
) ? remainder
: rv
;
1075 /* Return minimum size of a spare that can be used in this array*/
1076 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1078 struct intel_super
*super
= st
->sb
;
1082 unsigned long long rv
= 0;
1086 /* find first active disk in array */
1088 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1092 /* find last lba used by subarrays */
1093 e
= get_extents(super
, dl
);
1096 for (i
= 0; e
[i
].size
; i
++)
1099 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1102 /* add the amount of space needed for metadata */
1103 rv
= rv
+ imsm_min_reserved_sectors(super
);
1108 static int is_gen_migration(struct imsm_dev
*dev
);
1111 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1112 struct imsm_dev
*dev
);
1114 static void print_imsm_dev(struct intel_super
*super
,
1115 struct imsm_dev
*dev
,
1121 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1122 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1126 printf("[%.16s]:\n", dev
->volume
);
1127 printf(" UUID : %s\n", uuid
);
1128 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1130 printf(" <-- %d", get_imsm_raid_level(map2
));
1132 printf(" Members : %d", map
->num_members
);
1134 printf(" <-- %d", map2
->num_members
);
1136 printf(" Slots : [");
1137 for (i
= 0; i
< map
->num_members
; i
++) {
1138 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1139 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1144 for (i
= 0; i
< map2
->num_members
; i
++) {
1145 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1146 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1151 printf(" Failed disk : ");
1152 if (map
->failed_disk_num
== 0xff)
1155 printf("%i", map
->failed_disk_num
);
1157 slot
= get_imsm_disk_slot(map
, disk_idx
);
1159 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1160 printf(" This Slot : %d%s\n", slot
,
1161 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1163 printf(" This Slot : ?\n");
1164 sz
= __le32_to_cpu(dev
->size_high
);
1166 sz
+= __le32_to_cpu(dev
->size_low
);
1167 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1168 human_size(sz
* 512));
1169 sz
= blocks_per_member(map
);
1170 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1171 human_size(sz
* 512));
1172 printf(" Sector Offset : %llu\n",
1174 printf(" Num Stripes : %llu\n",
1175 num_data_stripes(map
));
1176 printf(" Chunk Size : %u KiB",
1177 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1179 printf(" <-- %u KiB",
1180 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1182 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1183 printf(" Migrate State : ");
1184 if (dev
->vol
.migr_state
) {
1185 if (migr_type(dev
) == MIGR_INIT
)
1186 printf("initialize\n");
1187 else if (migr_type(dev
) == MIGR_REBUILD
)
1188 printf("rebuild\n");
1189 else if (migr_type(dev
) == MIGR_VERIFY
)
1191 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1192 printf("general migration\n");
1193 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1194 printf("state change\n");
1195 else if (migr_type(dev
) == MIGR_REPAIR
)
1198 printf("<unknown:%d>\n", migr_type(dev
));
1201 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1202 if (dev
->vol
.migr_state
) {
1203 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1205 printf(" <-- %s", map_state_str
[map
->map_state
]);
1206 printf("\n Checkpoint : %u ",
1207 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1208 if ((is_gen_migration(dev
)) && ((slot
> 1) || (slot
< 0)))
1211 printf("(%llu)", (unsigned long long)
1212 blocks_per_migr_unit(super
, dev
));
1215 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1218 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1220 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1223 if (index
< -1 || !disk
)
1227 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1229 printf(" Disk%02d Serial : %s\n", index
, str
);
1231 printf(" Disk Serial : %s\n", str
);
1232 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1233 is_configured(disk
) ? " active" : "",
1234 is_failed(disk
) ? " failed" : "");
1235 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1236 sz
= total_blocks(disk
) - reserved
;
1237 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1238 human_size(sz
* 512));
1241 void examine_migr_rec_imsm(struct intel_super
*super
)
1243 struct migr_record
*migr_rec
= super
->migr_rec
;
1244 struct imsm_super
*mpb
= super
->anchor
;
1247 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1248 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1249 struct imsm_map
*map
;
1252 if (is_gen_migration(dev
) == 0)
1255 printf("\nMigration Record Information:");
1257 /* first map under migration */
1258 map
= get_imsm_map(dev
, MAP_0
);
1260 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1261 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1262 printf(" Empty\n ");
1263 printf("Examine one of first two disks in array\n");
1266 printf("\n Status : ");
1267 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1270 printf("Contains Data\n");
1271 printf(" Current Unit : %u\n",
1272 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1273 printf(" Family : %u\n",
1274 __le32_to_cpu(migr_rec
->family_num
));
1275 printf(" Ascending : %u\n",
1276 __le32_to_cpu(migr_rec
->ascending_migr
));
1277 printf(" Blocks Per Unit : %u\n",
1278 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1279 printf(" Dest. Depth Per Unit : %u\n",
1280 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1281 printf(" Checkpoint Area pba : %u\n",
1282 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1283 printf(" First member lba : %u\n",
1284 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1285 printf(" Total Number of Units : %u\n",
1286 __le32_to_cpu(migr_rec
->num_migr_units
));
1287 printf(" Size of volume : %u\n",
1288 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1289 printf(" Expansion space for LBA64 : %u\n",
1290 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1291 printf(" Record was read from : %u\n",
1292 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1297 #endif /* MDASSEMBLE */
1298 /*******************************************************************************
1299 * function: imsm_check_attributes
1300 * Description: Function checks if features represented by attributes flags
1301 * are supported by mdadm.
1303 * attributes - Attributes read from metadata
1305 * 0 - passed attributes contains unsupported features flags
1306 * 1 - all features are supported
1307 ******************************************************************************/
1308 static int imsm_check_attributes(__u32 attributes
)
1311 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1313 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1315 not_supported
&= attributes
;
1316 if (not_supported
) {
1317 pr_err("(IMSM): Unsupported attributes : %x\n",
1318 (unsigned)__le32_to_cpu(not_supported
));
1319 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1320 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1321 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1323 if (not_supported
& MPB_ATTRIB_2TB
) {
1324 dprintf("\t\tMPB_ATTRIB_2TB\n");
1325 not_supported
^= MPB_ATTRIB_2TB
;
1327 if (not_supported
& MPB_ATTRIB_RAID0
) {
1328 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1329 not_supported
^= MPB_ATTRIB_RAID0
;
1331 if (not_supported
& MPB_ATTRIB_RAID1
) {
1332 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1333 not_supported
^= MPB_ATTRIB_RAID1
;
1335 if (not_supported
& MPB_ATTRIB_RAID10
) {
1336 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1337 not_supported
^= MPB_ATTRIB_RAID10
;
1339 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1340 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1341 not_supported
^= MPB_ATTRIB_RAID1E
;
1343 if (not_supported
& MPB_ATTRIB_RAID5
) {
1344 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1345 not_supported
^= MPB_ATTRIB_RAID5
;
1347 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1348 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1349 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1351 if (not_supported
& MPB_ATTRIB_BBM
) {
1352 dprintf("\t\tMPB_ATTRIB_BBM\n");
1353 not_supported
^= MPB_ATTRIB_BBM
;
1355 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1356 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1357 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1359 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1360 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1361 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1363 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1364 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1365 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1367 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1368 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1369 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1371 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1372 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1373 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1377 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1386 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1388 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1390 struct intel_super
*super
= st
->sb
;
1391 struct imsm_super
*mpb
= super
->anchor
;
1392 char str
[MAX_SIGNATURE_LENGTH
];
1397 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1400 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1401 printf(" Magic : %s\n", str
);
1402 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1403 printf(" Version : %s\n", get_imsm_version(mpb
));
1404 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1405 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1406 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1407 printf(" Attributes : ");
1408 if (imsm_check_attributes(mpb
->attributes
))
1409 printf("All supported\n");
1411 printf("not supported\n");
1412 getinfo_super_imsm(st
, &info
, NULL
);
1413 fname_from_uuid(st
, &info
, nbuf
, ':');
1414 printf(" UUID : %s\n", nbuf
+ 5);
1415 sum
= __le32_to_cpu(mpb
->check_sum
);
1416 printf(" Checksum : %08x %s\n", sum
,
1417 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1418 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1419 printf(" Disks : %d\n", mpb
->num_disks
);
1420 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1421 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1422 if (super
->bbm_log
) {
1423 struct bbm_log
*log
= super
->bbm_log
;
1426 printf("Bad Block Management Log:\n");
1427 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1428 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1429 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1430 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1431 printf(" First Spare : %llx\n",
1432 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1434 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1436 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1438 super
->current_vol
= i
;
1439 getinfo_super_imsm(st
, &info
, NULL
);
1440 fname_from_uuid(st
, &info
, nbuf
, ':');
1441 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1443 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1444 if (i
== super
->disks
->index
)
1446 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1449 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1450 if (dl
->index
== -1)
1451 print_imsm_disk(&dl
->disk
, -1, reserved
);
1453 examine_migr_rec_imsm(super
);
1456 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1458 /* We just write a generic IMSM ARRAY entry */
1461 struct intel_super
*super
= st
->sb
;
1463 if (!super
->anchor
->num_raid_devs
) {
1464 printf("ARRAY metadata=imsm\n");
1468 getinfo_super_imsm(st
, &info
, NULL
);
1469 fname_from_uuid(st
, &info
, nbuf
, ':');
1470 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1473 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1475 /* We just write a generic IMSM ARRAY entry */
1479 struct intel_super
*super
= st
->sb
;
1482 if (!super
->anchor
->num_raid_devs
)
1485 getinfo_super_imsm(st
, &info
, NULL
);
1486 fname_from_uuid(st
, &info
, nbuf
, ':');
1487 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1488 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1490 super
->current_vol
= i
;
1491 getinfo_super_imsm(st
, &info
, NULL
);
1492 fname_from_uuid(st
, &info
, nbuf1
, ':');
1493 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1494 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1498 static void export_examine_super_imsm(struct supertype
*st
)
1500 struct intel_super
*super
= st
->sb
;
1501 struct imsm_super
*mpb
= super
->anchor
;
1505 getinfo_super_imsm(st
, &info
, NULL
);
1506 fname_from_uuid(st
, &info
, nbuf
, ':');
1507 printf("MD_METADATA=imsm\n");
1508 printf("MD_LEVEL=container\n");
1509 printf("MD_UUID=%s\n", nbuf
+5);
1510 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1513 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1518 getinfo_super_imsm(st
, &info
, NULL
);
1519 fname_from_uuid(st
, &info
, nbuf
, ':');
1520 printf("\n UUID : %s\n", nbuf
+ 5);
1523 static void brief_detail_super_imsm(struct supertype
*st
)
1527 getinfo_super_imsm(st
, &info
, NULL
);
1528 fname_from_uuid(st
, &info
, nbuf
, ':');
1529 printf(" UUID=%s", nbuf
+ 5);
1532 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1533 static void fd2devname(int fd
, char *name
);
1535 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1537 /* dump an unsorted list of devices attached to AHCI Intel storage
1538 * controller, as well as non-connected ports
1540 int hba_len
= strlen(hba_path
) + 1;
1545 unsigned long port_mask
= (1 << port_count
) - 1;
1547 if (port_count
> (int)sizeof(port_mask
) * 8) {
1549 pr_err("port_count %d out of range\n", port_count
);
1553 /* scroll through /sys/dev/block looking for devices attached to
1556 dir
= opendir("/sys/dev/block");
1557 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1568 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1570 path
= devt_to_devpath(makedev(major
, minor
));
1573 if (!path_attached_to_hba(path
, hba_path
)) {
1579 /* retrieve the scsi device type */
1580 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1582 pr_err("failed to allocate 'device'\n");
1586 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1587 if (load_sys(device
, buf
) != 0) {
1589 pr_err("failed to read device type for %s\n",
1595 type
= strtoul(buf
, NULL
, 10);
1597 /* if it's not a disk print the vendor and model */
1598 if (!(type
== 0 || type
== 7 || type
== 14)) {
1601 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1602 if (load_sys(device
, buf
) == 0) {
1603 strncpy(vendor
, buf
, sizeof(vendor
));
1604 vendor
[sizeof(vendor
) - 1] = '\0';
1605 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1606 while (isspace(*c
) || *c
== '\0')
1610 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1611 if (load_sys(device
, buf
) == 0) {
1612 strncpy(model
, buf
, sizeof(model
));
1613 model
[sizeof(model
) - 1] = '\0';
1614 c
= (char *) &model
[sizeof(model
) - 1];
1615 while (isspace(*c
) || *c
== '\0')
1619 if (vendor
[0] && model
[0])
1620 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1622 switch (type
) { /* numbers from hald/linux/device.c */
1623 case 1: sprintf(buf
, "tape"); break;
1624 case 2: sprintf(buf
, "printer"); break;
1625 case 3: sprintf(buf
, "processor"); break;
1627 case 5: sprintf(buf
, "cdrom"); break;
1628 case 6: sprintf(buf
, "scanner"); break;
1629 case 8: sprintf(buf
, "media_changer"); break;
1630 case 9: sprintf(buf
, "comm"); break;
1631 case 12: sprintf(buf
, "raid"); break;
1632 default: sprintf(buf
, "unknown");
1638 /* chop device path to 'host%d' and calculate the port number */
1639 c
= strchr(&path
[hba_len
], '/');
1642 pr_err("%s - invalid path name\n", path
+ hba_len
);
1647 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1651 *c
= '/'; /* repair the full string */
1652 pr_err("failed to determine port number for %s\n",
1659 /* mark this port as used */
1660 port_mask
&= ~(1 << port
);
1662 /* print out the device information */
1664 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1668 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1670 printf(" Port%d : - disk info unavailable -\n", port
);
1672 fd2devname(fd
, buf
);
1673 printf(" Port%d : %s", port
, buf
);
1674 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1675 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1690 for (i
= 0; i
< port_count
; i
++)
1691 if (port_mask
& (1 << i
))
1692 printf(" Port%d : - no device attached -\n", i
);
1698 static void print_found_intel_controllers(struct sys_dev
*elem
)
1700 for (; elem
; elem
= elem
->next
) {
1701 pr_err("found Intel(R) ");
1702 if (elem
->type
== SYS_DEV_SATA
)
1703 fprintf(stderr
, "SATA ");
1704 else if (elem
->type
== SYS_DEV_SAS
)
1705 fprintf(stderr
, "SAS ");
1706 fprintf(stderr
, "RAID controller");
1708 fprintf(stderr
, " at %s", elem
->pci_id
);
1709 fprintf(stderr
, ".\n");
1714 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1721 if ((dir
= opendir(hba_path
)) == NULL
)
1724 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1727 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1729 if (*port_count
== 0)
1731 else if (host
< host_base
)
1734 if (host
+ 1 > *port_count
+ host_base
)
1735 *port_count
= host
+ 1 - host_base
;
1741 static void print_imsm_capability(const struct imsm_orom
*orom
)
1743 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1744 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1745 orom
->hotfix_ver
, orom
->build
);
1746 printf(" RAID Levels :%s%s%s%s%s\n",
1747 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1748 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1749 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1750 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1751 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1752 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1753 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1754 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1755 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1756 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1757 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1758 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1759 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1760 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1761 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1762 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1763 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1764 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1765 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1766 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1767 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1768 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1769 printf(" 2TB volumes :%s supported\n",
1770 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
1771 printf(" 2TB disks :%s supported\n",
1772 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
1773 printf(" Max Disks : %d\n", orom
->tds
);
1774 printf(" Max Volumes : %d per array, %d per controller\n",
1775 orom
->vpa
, orom
->vphba
);
1779 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
1781 printf("MD_FIRMWARE_TYPE=imsm\n");
1782 printf("IMSM_VERSION=%d.%d.%d.%d\n",orom
->major_ver
, orom
->minor_ver
,
1783 orom
->hotfix_ver
, orom
->build
);
1784 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
1785 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
1786 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
1787 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
1788 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
1789 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
1790 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1791 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
1792 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
1793 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
1794 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
1795 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
1796 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
1797 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
1798 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
1799 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
1800 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
1801 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
1802 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
1803 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
1804 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
1805 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
1806 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
1807 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
1808 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
1809 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
1810 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
1811 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
1814 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
1816 /* There are two components to imsm platform support, the ahci SATA
1817 * controller and the option-rom. To find the SATA controller we
1818 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1819 * controller with the Intel vendor id is present. This approach
1820 * allows mdadm to leverage the kernel's ahci detection logic, with the
1821 * caveat that if ahci.ko is not loaded mdadm will not be able to
1822 * detect platform raid capabilities. The option-rom resides in a
1823 * platform "Adapter ROM". We scan for its signature to retrieve the
1824 * platform capabilities. If raid support is disabled in the BIOS the
1825 * option-rom capability structure will not be available.
1827 const struct imsm_orom
*orom
;
1828 struct sys_dev
*list
, *hba
;
1833 if (enumerate_only
) {
1834 if (check_env("IMSM_NO_PLATFORM"))
1836 list
= find_intel_devices();
1839 for (hba
= list
; hba
; hba
= hba
->next
) {
1840 orom
= find_imsm_capability(hba
->type
);
1851 list
= find_intel_devices();
1854 pr_err("no active Intel(R) RAID "
1855 "controller found.\n");
1857 } else if (verbose
> 0)
1858 print_found_intel_controllers(list
);
1860 for (hba
= list
; hba
; hba
= hba
->next
) {
1861 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
1863 orom
= find_imsm_capability(hba
->type
);
1865 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
1866 hba
->path
, get_sys_dev_type(hba
->type
));
1869 print_imsm_capability(orom
);
1870 printf(" I/O Controller : %s (%s)\n",
1871 hba
->path
, get_sys_dev_type(hba
->type
));
1872 if (hba
->type
== SYS_DEV_SATA
) {
1873 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1874 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1876 pr_err("failed to enumerate "
1877 "ports on SATA controller at %s.\n", hba
->pci_id
);
1884 if (controller_path
&& result
== 1)
1885 pr_err("no active Intel(R) RAID "
1886 "controller found under %s\n",controller_path
);
1891 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
1893 const struct imsm_orom
*orom
;
1894 struct sys_dev
*list
, *hba
;
1897 list
= find_intel_devices();
1900 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
1905 for (hba
= list
; hba
; hba
= hba
->next
) {
1906 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
1908 orom
= find_imsm_capability(hba
->type
);
1911 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",hba
->path
);
1914 print_imsm_capability_export(orom
);
1924 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1926 /* the imsm metadata format does not specify any host
1927 * identification information. We return -1 since we can never
1928 * confirm nor deny whether a given array is "meant" for this
1929 * host. We rely on compare_super and the 'family_num' fields to
1930 * exclude member disks that do not belong, and we rely on
1931 * mdadm.conf to specify the arrays that should be assembled.
1932 * Auto-assembly may still pick up "foreign" arrays.
1938 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1940 /* The uuid returned here is used for:
1941 * uuid to put into bitmap file (Create, Grow)
1942 * uuid for backup header when saving critical section (Grow)
1943 * comparing uuids when re-adding a device into an array
1944 * In these cases the uuid required is that of the data-array,
1945 * not the device-set.
1946 * uuid to recognise same set when adding a missing device back
1947 * to an array. This is a uuid for the device-set.
1949 * For each of these we can make do with a truncated
1950 * or hashed uuid rather than the original, as long as
1952 * In each case the uuid required is that of the data-array,
1953 * not the device-set.
1955 /* imsm does not track uuid's so we synthesis one using sha1 on
1956 * - The signature (Which is constant for all imsm array, but no matter)
1957 * - the orig_family_num of the container
1958 * - the index number of the volume
1959 * - the 'serial' number of the volume.
1960 * Hopefully these are all constant.
1962 struct intel_super
*super
= st
->sb
;
1965 struct sha1_ctx ctx
;
1966 struct imsm_dev
*dev
= NULL
;
1969 /* some mdadm versions failed to set ->orig_family_num, in which
1970 * case fall back to ->family_num. orig_family_num will be
1971 * fixed up with the first metadata update.
1973 family_num
= super
->anchor
->orig_family_num
;
1974 if (family_num
== 0)
1975 family_num
= super
->anchor
->family_num
;
1976 sha1_init_ctx(&ctx
);
1977 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1978 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1979 if (super
->current_vol
>= 0)
1980 dev
= get_imsm_dev(super
, super
->current_vol
);
1982 __u32 vol
= super
->current_vol
;
1983 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1984 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1986 sha1_finish_ctx(&ctx
, buf
);
1987 memcpy(uuid
, buf
, 4*4);
1992 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1994 __u8
*v
= get_imsm_version(mpb
);
1995 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1996 char major
[] = { 0, 0, 0 };
1997 char minor
[] = { 0 ,0, 0 };
1998 char patch
[] = { 0, 0, 0 };
1999 char *ver_parse
[] = { major
, minor
, patch
};
2003 while (*v
!= '\0' && v
< end
) {
2004 if (*v
!= '.' && j
< 2)
2005 ver_parse
[i
][j
++] = *v
;
2013 *m
= strtol(minor
, NULL
, 0);
2014 *p
= strtol(patch
, NULL
, 0);
2018 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2020 /* migr_strip_size when repairing or initializing parity */
2021 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2022 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2024 switch (get_imsm_raid_level(map
)) {
2029 return 128*1024 >> 9;
2033 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2035 /* migr_strip_size when rebuilding a degraded disk, no idea why
2036 * this is different than migr_strip_size_resync(), but it's good
2039 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2040 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2042 switch (get_imsm_raid_level(map
)) {
2045 if (map
->num_members
% map
->num_domains
== 0)
2046 return 128*1024 >> 9;
2050 return max((__u32
) 64*1024 >> 9, chunk
);
2052 return 128*1024 >> 9;
2056 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2058 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2059 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2060 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2061 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2063 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2066 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2068 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2069 int level
= get_imsm_raid_level(lo
);
2071 if (level
== 1 || level
== 10) {
2072 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2074 return hi
->num_domains
;
2076 return num_stripes_per_unit_resync(dev
);
2079 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2081 /* named 'imsm_' because raid0, raid1 and raid10
2082 * counter-intuitively have the same number of data disks
2084 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2086 switch (get_imsm_raid_level(map
)) {
2088 return map
->num_members
;
2092 return map
->num_members
/2;
2094 return map
->num_members
- 1;
2096 dprintf("%s: unsupported raid level\n", __func__
);
2101 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2103 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2104 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2106 switch(get_imsm_raid_level(map
)) {
2109 return chunk
* map
->num_domains
;
2111 return chunk
* map
->num_members
;
2117 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2119 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2120 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2121 __u32 strip
= block
/ chunk
;
2123 switch (get_imsm_raid_level(map
)) {
2126 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2127 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2129 return vol_stripe
* chunk
+ block
% chunk
;
2131 __u32 stripe
= strip
/ (map
->num_members
- 1);
2133 return stripe
* chunk
+ block
% chunk
;
2140 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2141 struct imsm_dev
*dev
)
2143 /* calculate the conversion factor between per member 'blocks'
2144 * (md/{resync,rebuild}_start) and imsm migration units, return
2145 * 0 for the 'not migrating' and 'unsupported migration' cases
2147 if (!dev
->vol
.migr_state
)
2150 switch (migr_type(dev
)) {
2151 case MIGR_GEN_MIGR
: {
2152 struct migr_record
*migr_rec
= super
->migr_rec
;
2153 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2158 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2159 __u32 stripes_per_unit
;
2160 __u32 blocks_per_unit
;
2169 /* yes, this is really the translation of migr_units to
2170 * per-member blocks in the 'resync' case
2172 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2173 migr_chunk
= migr_strip_blocks_resync(dev
);
2174 disks
= imsm_num_data_members(dev
, MAP_0
);
2175 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2176 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2177 segment
= blocks_per_unit
/ stripe
;
2178 block_rel
= blocks_per_unit
- segment
* stripe
;
2179 parity_depth
= parity_segment_depth(dev
);
2180 block_map
= map_migr_block(dev
, block_rel
);
2181 return block_map
+ parity_depth
* segment
;
2183 case MIGR_REBUILD
: {
2184 __u32 stripes_per_unit
;
2187 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2188 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2189 return migr_chunk
* stripes_per_unit
;
2191 case MIGR_STATE_CHANGE
:
2197 static int imsm_level_to_layout(int level
)
2205 return ALGORITHM_LEFT_ASYMMETRIC
;
2212 /*******************************************************************************
2213 * Function: read_imsm_migr_rec
2214 * Description: Function reads imsm migration record from last sector of disk
2216 * fd : disk descriptor
2217 * super : metadata info
2221 ******************************************************************************/
2222 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2225 unsigned long long dsize
;
2227 get_dev_size(fd
, NULL
, &dsize
);
2228 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2229 pr_err("Cannot seek to anchor block: %s\n",
2233 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2234 MIGR_REC_BUF_SIZE
) {
2235 pr_err("Cannot read migr record block: %s\n",
2245 static struct imsm_dev
*imsm_get_device_during_migration(
2246 struct intel_super
*super
)
2249 struct intel_dev
*dv
;
2251 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2252 if (is_gen_migration(dv
->dev
))
2258 /*******************************************************************************
2259 * Function: load_imsm_migr_rec
2260 * Description: Function reads imsm migration record (it is stored at the last
2263 * super : imsm internal array info
2264 * info : general array info
2268 * -2 : no migration in progress
2269 ******************************************************************************/
2270 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2273 struct dl
*dl
= NULL
;
2277 struct imsm_dev
*dev
;
2278 struct imsm_map
*map
= NULL
;
2281 /* find map under migration */
2282 dev
= imsm_get_device_during_migration(super
);
2283 /* nothing to load,no migration in progress?
2287 map
= get_imsm_map(dev
, MAP_0
);
2290 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2291 /* skip spare and failed disks
2293 if (sd
->disk
.raid_disk
< 0)
2295 /* read only from one of the first two slots */
2297 slot
= get_imsm_disk_slot(map
,
2298 sd
->disk
.raid_disk
);
2299 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2302 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2303 fd
= dev_open(nm
, O_RDONLY
);
2309 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2310 /* skip spare and failed disks
2314 /* read only from one of the first two slots */
2316 slot
= get_imsm_disk_slot(map
, dl
->index
);
2317 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2319 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2320 fd
= dev_open(nm
, O_RDONLY
);
2327 retval
= read_imsm_migr_rec(fd
, super
);
2336 /*******************************************************************************
2337 * function: imsm_create_metadata_checkpoint_update
2338 * Description: It creates update for checkpoint change.
2340 * super : imsm internal array info
2341 * u : pointer to prepared update
2344 * If length is equal to 0, input pointer u contains no update
2345 ******************************************************************************/
2346 static int imsm_create_metadata_checkpoint_update(
2347 struct intel_super
*super
,
2348 struct imsm_update_general_migration_checkpoint
**u
)
2351 int update_memory_size
= 0;
2353 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2359 /* size of all update data without anchor */
2360 update_memory_size
=
2361 sizeof(struct imsm_update_general_migration_checkpoint
);
2363 *u
= xcalloc(1, update_memory_size
);
2365 dprintf("error: cannot get memory for "
2366 "imsm_create_metadata_checkpoint_update update\n");
2369 (*u
)->type
= update_general_migration_checkpoint
;
2370 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2371 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2372 (*u
)->curr_migr_unit
);
2374 return update_memory_size
;
2378 static void imsm_update_metadata_locally(struct supertype
*st
,
2379 void *buf
, int len
);
2381 /*******************************************************************************
2382 * Function: write_imsm_migr_rec
2383 * Description: Function writes imsm migration record
2384 * (at the last sector of disk)
2386 * super : imsm internal array info
2390 ******************************************************************************/
2391 static int write_imsm_migr_rec(struct supertype
*st
)
2393 struct intel_super
*super
= st
->sb
;
2394 unsigned long long dsize
;
2400 struct imsm_update_general_migration_checkpoint
*u
;
2401 struct imsm_dev
*dev
;
2402 struct imsm_map
*map
= NULL
;
2404 /* find map under migration */
2405 dev
= imsm_get_device_during_migration(super
);
2406 /* if no migration, write buffer anyway to clear migr_record
2407 * on disk based on first available device
2410 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2411 super
->current_vol
);
2413 map
= get_imsm_map(dev
, MAP_0
);
2415 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2418 /* skip failed and spare devices */
2421 /* write to 2 first slots only */
2423 slot
= get_imsm_disk_slot(map
, sd
->index
);
2424 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2427 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2428 fd
= dev_open(nm
, O_RDWR
);
2431 get_dev_size(fd
, NULL
, &dsize
);
2432 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2433 pr_err("Cannot seek to anchor block: %s\n",
2437 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2438 MIGR_REC_BUF_SIZE
) {
2439 pr_err("Cannot write migr record block: %s\n",
2446 /* update checkpoint information in metadata */
2447 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2450 dprintf("imsm: Cannot prepare update\n");
2453 /* update metadata locally */
2454 imsm_update_metadata_locally(st
, u
, len
);
2455 /* and possibly remotely */
2456 if (st
->update_tail
) {
2457 append_metadata_update(st
, u
, len
);
2458 /* during reshape we do all work inside metadata handler
2459 * manage_reshape(), so metadata update has to be triggered
2462 flush_metadata_updates(st
);
2463 st
->update_tail
= &st
->updates
;
2473 #endif /* MDASSEMBLE */
2475 /* spare/missing disks activations are not allowe when
2476 * array/container performs reshape operation, because
2477 * all arrays in container works on the same disks set
2479 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2482 struct intel_dev
*i_dev
;
2483 struct imsm_dev
*dev
;
2485 /* check whole container
2487 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2489 if (is_gen_migration(dev
)) {
2490 /* No repair during any migration in container
2498 static unsigned long long imsm_component_size_aligment_check(int level
,
2500 unsigned long long component_size
)
2502 unsigned int component_size_alligment
;
2504 /* check component size aligment
2506 component_size_alligment
= component_size
% (chunk_size
/512);
2508 dprintf("imsm_component_size_aligment_check(Level: %i, "
2509 "chunk_size = %i, component_size = %llu), "
2510 "component_size_alligment = %u\n",
2511 level
, chunk_size
, component_size
,
2512 component_size_alligment
);
2514 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
2515 dprintf("imsm: reported component size alligned from %llu ",
2517 component_size
-= component_size_alligment
;
2518 dprintf("to %llu (%i).\n",
2519 component_size
, component_size_alligment
);
2522 return component_size
;
2525 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2527 struct intel_super
*super
= st
->sb
;
2528 struct migr_record
*migr_rec
= super
->migr_rec
;
2529 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2530 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2531 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2532 struct imsm_map
*map_to_analyse
= map
;
2534 int map_disks
= info
->array
.raid_disks
;
2536 memset(info
, 0, sizeof(*info
));
2538 map_to_analyse
= prev_map
;
2540 dl
= super
->current_disk
;
2542 info
->container_member
= super
->current_vol
;
2543 info
->array
.raid_disks
= map
->num_members
;
2544 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2545 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2546 info
->array
.md_minor
= -1;
2547 info
->array
.ctime
= 0;
2548 info
->array
.utime
= 0;
2549 info
->array
.chunk_size
=
2550 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2551 info
->array
.state
= !dev
->vol
.dirty
;
2552 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2553 info
->custom_array_size
<<= 32;
2554 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2555 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2557 if (is_gen_migration(dev
)) {
2558 info
->reshape_active
= 1;
2559 info
->new_level
= get_imsm_raid_level(map
);
2560 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2561 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2562 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2563 if (info
->delta_disks
) {
2564 /* this needs to be applied to every array
2567 info
->reshape_active
= CONTAINER_RESHAPE
;
2569 /* We shape information that we give to md might have to be
2570 * modify to cope with md's requirement for reshaping arrays.
2571 * For example, when reshaping a RAID0, md requires it to be
2572 * presented as a degraded RAID4.
2573 * Also if a RAID0 is migrating to a RAID5 we need to specify
2574 * the array as already being RAID5, but the 'before' layout
2575 * is a RAID4-like layout.
2577 switch (info
->array
.level
) {
2579 switch(info
->new_level
) {
2581 /* conversion is happening as RAID4 */
2582 info
->array
.level
= 4;
2583 info
->array
.raid_disks
+= 1;
2586 /* conversion is happening as RAID5 */
2587 info
->array
.level
= 5;
2588 info
->array
.layout
= ALGORITHM_PARITY_N
;
2589 info
->delta_disks
-= 1;
2592 /* FIXME error message */
2593 info
->array
.level
= UnSet
;
2599 info
->new_level
= UnSet
;
2600 info
->new_layout
= UnSet
;
2601 info
->new_chunk
= info
->array
.chunk_size
;
2602 info
->delta_disks
= 0;
2606 info
->disk
.major
= dl
->major
;
2607 info
->disk
.minor
= dl
->minor
;
2608 info
->disk
.number
= dl
->index
;
2609 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2613 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2614 info
->component_size
= blocks_per_member(map_to_analyse
);
2616 info
->component_size
= imsm_component_size_aligment_check(
2618 info
->array
.chunk_size
,
2619 info
->component_size
);
2621 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2622 info
->recovery_start
= MaxSector
;
2624 info
->reshape_progress
= 0;
2625 info
->resync_start
= MaxSector
;
2626 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2628 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2629 info
->resync_start
= 0;
2631 if (dev
->vol
.migr_state
) {
2632 switch (migr_type(dev
)) {
2635 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2637 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2639 info
->resync_start
= blocks_per_unit
* units
;
2642 case MIGR_GEN_MIGR
: {
2643 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2645 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2646 unsigned long long array_blocks
;
2649 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2651 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2652 (super
->migr_rec
->rec_status
==
2653 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2656 info
->reshape_progress
= blocks_per_unit
* units
;
2658 dprintf("IMSM: General Migration checkpoint : %llu "
2659 "(%llu) -> read reshape progress : %llu\n",
2660 (unsigned long long)units
,
2661 (unsigned long long)blocks_per_unit
,
2662 info
->reshape_progress
);
2664 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2665 if (used_disks
> 0) {
2666 array_blocks
= blocks_per_member(map
) *
2668 /* round array size down to closest MB
2670 info
->custom_array_size
= (array_blocks
2671 >> SECT_PER_MB_SHIFT
)
2672 << SECT_PER_MB_SHIFT
;
2676 /* we could emulate the checkpointing of
2677 * 'sync_action=check' migrations, but for now
2678 * we just immediately complete them
2681 /* this is handled by container_content_imsm() */
2682 case MIGR_STATE_CHANGE
:
2683 /* FIXME handle other migrations */
2685 /* we are not dirty, so... */
2686 info
->resync_start
= MaxSector
;
2690 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2691 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2693 info
->array
.major_version
= -1;
2694 info
->array
.minor_version
= -2;
2695 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
2696 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2697 uuid_from_super_imsm(st
, info
->uuid
);
2701 for (i
=0; i
<map_disks
; i
++) {
2703 if (i
< info
->array
.raid_disks
) {
2704 struct imsm_disk
*dsk
;
2705 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2706 dsk
= get_imsm_disk(super
, j
);
2707 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2714 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2715 int failed
, int look_in_map
);
2717 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2722 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2724 if (is_gen_migration(dev
)) {
2727 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2729 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2730 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2731 if (map2
->map_state
!= map_state
) {
2732 map2
->map_state
= map_state
;
2733 super
->updates_pending
++;
2739 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2743 for (d
= super
->missing
; d
; d
= d
->next
)
2744 if (d
->index
== index
)
2749 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2751 struct intel_super
*super
= st
->sb
;
2752 struct imsm_disk
*disk
;
2753 int map_disks
= info
->array
.raid_disks
;
2754 int max_enough
= -1;
2756 struct imsm_super
*mpb
;
2758 if (super
->current_vol
>= 0) {
2759 getinfo_super_imsm_volume(st
, info
, map
);
2762 memset(info
, 0, sizeof(*info
));
2764 /* Set raid_disks to zero so that Assemble will always pull in valid
2767 info
->array
.raid_disks
= 0;
2768 info
->array
.level
= LEVEL_CONTAINER
;
2769 info
->array
.layout
= 0;
2770 info
->array
.md_minor
= -1;
2771 info
->array
.ctime
= 0; /* N/A for imsm */
2772 info
->array
.utime
= 0;
2773 info
->array
.chunk_size
= 0;
2775 info
->disk
.major
= 0;
2776 info
->disk
.minor
= 0;
2777 info
->disk
.raid_disk
= -1;
2778 info
->reshape_active
= 0;
2779 info
->array
.major_version
= -1;
2780 info
->array
.minor_version
= -2;
2781 strcpy(info
->text_version
, "imsm");
2782 info
->safe_mode_delay
= 0;
2783 info
->disk
.number
= -1;
2784 info
->disk
.state
= 0;
2786 info
->recovery_start
= MaxSector
;
2787 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2789 /* do we have the all the insync disks that we expect? */
2790 mpb
= super
->anchor
;
2792 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2793 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2794 int failed
, enough
, j
, missing
= 0;
2795 struct imsm_map
*map
;
2798 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2799 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2800 map
= get_imsm_map(dev
, MAP_0
);
2802 /* any newly missing disks?
2803 * (catches single-degraded vs double-degraded)
2805 for (j
= 0; j
< map
->num_members
; j
++) {
2806 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2807 __u32 idx
= ord_to_idx(ord
);
2809 if (!(ord
& IMSM_ORD_REBUILD
) &&
2810 get_imsm_missing(super
, idx
)) {
2816 if (state
== IMSM_T_STATE_FAILED
)
2818 else if (state
== IMSM_T_STATE_DEGRADED
&&
2819 (state
!= map
->map_state
|| missing
))
2821 else /* we're normal, or already degraded */
2823 if (is_gen_migration(dev
) && missing
) {
2824 /* during general migration we need all disks
2825 * that process is running on.
2826 * No new missing disk is allowed.
2830 /* no more checks necessary
2834 /* in the missing/failed disk case check to see
2835 * if at least one array is runnable
2837 max_enough
= max(max_enough
, enough
);
2839 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2840 info
->container_enough
= max_enough
;
2843 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2845 disk
= &super
->disks
->disk
;
2846 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
2847 info
->component_size
= reserved
;
2848 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2849 /* we don't change info->disk.raid_disk here because
2850 * this state will be finalized in mdmon after we have
2851 * found the 'most fresh' version of the metadata
2853 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2854 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2857 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2858 * ->compare_super may have updated the 'num_raid_devs' field for spares
2860 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2861 uuid_from_super_imsm(st
, info
->uuid
);
2863 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2865 /* I don't know how to compute 'map' on imsm, so use safe default */
2868 for (i
= 0; i
< map_disks
; i
++)
2874 /* allocates memory and fills disk in mdinfo structure
2875 * for each disk in array */
2876 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2878 struct mdinfo
*mddev
= NULL
;
2879 struct intel_super
*super
= st
->sb
;
2880 struct imsm_disk
*disk
;
2883 if (!super
|| !super
->disks
)
2886 mddev
= xcalloc(1, sizeof(*mddev
));
2890 tmp
= xcalloc(1, sizeof(*tmp
));
2892 tmp
->next
= mddev
->devs
;
2894 tmp
->disk
.number
= count
++;
2895 tmp
->disk
.major
= dl
->major
;
2896 tmp
->disk
.minor
= dl
->minor
;
2897 tmp
->disk
.state
= is_configured(disk
) ?
2898 (1 << MD_DISK_ACTIVE
) : 0;
2899 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2900 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2901 tmp
->disk
.raid_disk
= -1;
2907 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2908 char *update
, char *devname
, int verbose
,
2909 int uuid_set
, char *homehost
)
2911 /* For 'assemble' and 'force' we need to return non-zero if any
2912 * change was made. For others, the return value is ignored.
2913 * Update options are:
2914 * force-one : This device looks a bit old but needs to be included,
2915 * update age info appropriately.
2916 * assemble: clear any 'faulty' flag to allow this device to
2918 * force-array: Array is degraded but being forced, mark it clean
2919 * if that will be needed to assemble it.
2921 * newdev: not used ????
2922 * grow: Array has gained a new device - this is currently for
2924 * resync: mark as dirty so a resync will happen.
2925 * name: update the name - preserving the homehost
2926 * uuid: Change the uuid of the array to match watch is given
2928 * Following are not relevant for this imsm:
2929 * sparc2.2 : update from old dodgey metadata
2930 * super-minor: change the preferred_minor number
2931 * summaries: update redundant counters.
2932 * homehost: update the recorded homehost
2933 * _reshape_progress: record new reshape_progress position.
2936 struct intel_super
*super
= st
->sb
;
2937 struct imsm_super
*mpb
;
2939 /* we can only update container info */
2940 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2943 mpb
= super
->anchor
;
2945 if (strcmp(update
, "uuid") == 0) {
2946 /* We take this to mean that the family_num should be updated.
2947 * However that is much smaller than the uuid so we cannot really
2948 * allow an explicit uuid to be given. And it is hard to reliably
2950 * So if !uuid_set we know the current uuid is random and just used
2951 * the first 'int' and copy it to the other 3 positions.
2952 * Otherwise we require the 4 'int's to be the same as would be the
2953 * case if we are using a random uuid. So an explicit uuid will be
2954 * accepted as long as all for ints are the same... which shouldn't hurt
2957 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
2960 if (info
->uuid
[0] != info
->uuid
[1] ||
2961 info
->uuid
[1] != info
->uuid
[2] ||
2962 info
->uuid
[2] != info
->uuid
[3])
2968 mpb
->orig_family_num
= info
->uuid
[0];
2969 } else if (strcmp(update
, "assemble") == 0)
2974 /* successful update? recompute checksum */
2976 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2981 static size_t disks_to_mpb_size(int disks
)
2985 size
= sizeof(struct imsm_super
);
2986 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2987 size
+= 2 * sizeof(struct imsm_dev
);
2988 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2989 size
+= (4 - 2) * sizeof(struct imsm_map
);
2990 /* 4 possible disk_ord_tbl's */
2991 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2996 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
2997 unsigned long long data_offset
)
2999 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3002 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3005 static void free_devlist(struct intel_super
*super
)
3007 struct intel_dev
*dv
;
3009 while (super
->devlist
) {
3010 dv
= super
->devlist
->next
;
3011 free(super
->devlist
->dev
);
3012 free(super
->devlist
);
3013 super
->devlist
= dv
;
3017 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3019 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3022 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3026 * 0 same, or first was empty, and second was copied
3027 * 1 second had wrong number
3029 * 3 wrong other info
3031 struct intel_super
*first
= st
->sb
;
3032 struct intel_super
*sec
= tst
->sb
;
3039 /* in platform dependent environment test if the disks
3040 * use the same Intel hba
3041 * If not on Intel hba at all, allow anything.
3043 if (!check_env("IMSM_NO_PLATFORM")) {
3044 if (first
->hba
&& sec
->hba
&&
3045 strcmp(first
->hba
->path
, sec
->hba
->path
) != 0) {
3047 "HBAs of devices does not match %s != %s\n",
3048 first
->hba
? first
->hba
->path
: NULL
,
3049 sec
->hba
? sec
->hba
->path
: NULL
);
3054 /* if an anchor does not have num_raid_devs set then it is a free
3057 if (first
->anchor
->num_raid_devs
> 0 &&
3058 sec
->anchor
->num_raid_devs
> 0) {
3059 /* Determine if these disks might ever have been
3060 * related. Further disambiguation can only take place
3061 * in load_super_imsm_all
3063 __u32 first_family
= first
->anchor
->orig_family_num
;
3064 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3066 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3067 MAX_SIGNATURE_LENGTH
) != 0)
3070 if (first_family
== 0)
3071 first_family
= first
->anchor
->family_num
;
3072 if (sec_family
== 0)
3073 sec_family
= sec
->anchor
->family_num
;
3075 if (first_family
!= sec_family
)
3081 /* if 'first' is a spare promote it to a populated mpb with sec's
3084 if (first
->anchor
->num_raid_devs
== 0 &&
3085 sec
->anchor
->num_raid_devs
> 0) {
3087 struct intel_dev
*dv
;
3088 struct imsm_dev
*dev
;
3090 /* we need to copy raid device info from sec if an allocation
3091 * fails here we don't associate the spare
3093 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3094 dv
= xmalloc(sizeof(*dv
));
3095 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3098 dv
->next
= first
->devlist
;
3099 first
->devlist
= dv
;
3101 if (i
< sec
->anchor
->num_raid_devs
) {
3102 /* allocation failure */
3103 free_devlist(first
);
3104 fprintf(stderr
, "imsm: failed to associate spare\n");
3107 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3108 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3109 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3110 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3111 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3112 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3118 static void fd2devname(int fd
, char *name
)
3122 char dname
[PATH_MAX
];
3127 if (fstat(fd
, &st
) != 0)
3129 sprintf(path
, "/sys/dev/block/%d:%d",
3130 major(st
.st_rdev
), minor(st
.st_rdev
));
3132 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3137 nm
= strrchr(dname
, '/');
3140 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3144 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3146 static int imsm_read_serial(int fd
, char *devname
,
3147 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3149 unsigned char scsi_serial
[255];
3158 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3160 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3162 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3163 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3164 fd2devname(fd
, (char *) serial
);
3170 pr_err("Failed to retrieve serial for %s\n",
3175 rsp_len
= scsi_serial
[3];
3178 pr_err("Failed to retrieve serial for %s\n",
3182 rsp_buf
= (char *) &scsi_serial
[4];
3184 /* trim all whitespace and non-printable characters and convert
3187 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3190 /* ':' is reserved for use in placeholder serial
3191 * numbers for missing disks
3199 len
= dest
- rsp_buf
;
3202 /* truncate leading characters */
3203 if (len
> MAX_RAID_SERIAL_LEN
) {
3204 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3205 len
= MAX_RAID_SERIAL_LEN
;
3208 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3209 memcpy(serial
, dest
, len
);
3214 static int serialcmp(__u8
*s1
, __u8
*s2
)
3216 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3219 static void serialcpy(__u8
*dest
, __u8
*src
)
3221 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3224 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3228 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3229 if (serialcmp(dl
->serial
, serial
) == 0)
3235 static struct imsm_disk
*
3236 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3240 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3241 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3243 if (serialcmp(disk
->serial
, serial
) == 0) {
3254 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3256 struct imsm_disk
*disk
;
3261 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3263 rv
= imsm_read_serial(fd
, devname
, serial
);
3268 dl
= xcalloc(1, sizeof(*dl
));
3271 dl
->major
= major(stb
.st_rdev
);
3272 dl
->minor
= minor(stb
.st_rdev
);
3273 dl
->next
= super
->disks
;
3274 dl
->fd
= keep_fd
? fd
: -1;
3275 assert(super
->disks
== NULL
);
3277 serialcpy(dl
->serial
, serial
);
3280 fd2devname(fd
, name
);
3282 dl
->devname
= xstrdup(devname
);
3284 dl
->devname
= xstrdup(name
);
3286 /* look up this disk's index in the current anchor */
3287 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3290 /* only set index on disks that are a member of a
3291 * populated contianer, i.e. one with raid_devs
3293 if (is_failed(&dl
->disk
))
3295 else if (is_spare(&dl
->disk
))
3303 /* When migrating map0 contains the 'destination' state while map1
3304 * contains the current state. When not migrating map0 contains the
3305 * current state. This routine assumes that map[0].map_state is set to
3306 * the current array state before being called.
3308 * Migration is indicated by one of the following states
3309 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3310 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3311 * map1state=unitialized)
3312 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3314 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3315 * map1state=degraded)
3316 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3319 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3320 __u8 to_state
, int migr_type
)
3322 struct imsm_map
*dest
;
3323 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3325 dev
->vol
.migr_state
= 1;
3326 set_migr_type(dev
, migr_type
);
3327 dev
->vol
.curr_migr_unit
= 0;
3328 dest
= get_imsm_map(dev
, MAP_1
);
3330 /* duplicate and then set the target end state in map[0] */
3331 memcpy(dest
, src
, sizeof_imsm_map(src
));
3332 if ((migr_type
== MIGR_REBUILD
) ||
3333 (migr_type
== MIGR_GEN_MIGR
)) {
3337 for (i
= 0; i
< src
->num_members
; i
++) {
3338 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3339 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3343 if (migr_type
== MIGR_GEN_MIGR
)
3344 /* Clear migration record */
3345 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3347 src
->map_state
= to_state
;
3350 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3353 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3354 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3358 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3359 * completed in the last migration.
3361 * FIXME add support for raid-level-migration
3363 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3364 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3365 /* when final map state is other than expected
3366 * merge maps (not for migration)
3370 for (i
= 0; i
< prev
->num_members
; i
++)
3371 for (j
= 0; j
< map
->num_members
; j
++)
3372 /* during online capacity expansion
3373 * disks position can be changed
3374 * if takeover is used
3376 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3377 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3378 map
->disk_ord_tbl
[j
] |=
3379 prev
->disk_ord_tbl
[i
];
3382 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3383 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3386 dev
->vol
.migr_state
= 0;
3387 set_migr_type(dev
, 0);
3388 dev
->vol
.curr_migr_unit
= 0;
3389 map
->map_state
= map_state
;
3393 static int parse_raid_devices(struct intel_super
*super
)
3396 struct imsm_dev
*dev_new
;
3397 size_t len
, len_migr
;
3399 size_t space_needed
= 0;
3400 struct imsm_super
*mpb
= super
->anchor
;
3402 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3403 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3404 struct intel_dev
*dv
;
3406 len
= sizeof_imsm_dev(dev_iter
, 0);
3407 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3409 space_needed
+= len_migr
- len
;
3411 dv
= xmalloc(sizeof(*dv
));
3412 if (max_len
< len_migr
)
3414 if (max_len
> len_migr
)
3415 space_needed
+= max_len
- len_migr
;
3416 dev_new
= xmalloc(max_len
);
3417 imsm_copy_dev(dev_new
, dev_iter
);
3420 dv
->next
= super
->devlist
;
3421 super
->devlist
= dv
;
3424 /* ensure that super->buf is large enough when all raid devices
3427 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3430 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3431 if (posix_memalign(&buf
, 512, len
) != 0)
3434 memcpy(buf
, super
->buf
, super
->len
);
3435 memset(buf
+ super
->len
, 0, len
- super
->len
);
3444 /* retrieve a pointer to the bbm log which starts after all raid devices */
3445 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3449 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3451 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3457 /*******************************************************************************
3458 * Function: check_mpb_migr_compatibility
3459 * Description: Function checks for unsupported migration features:
3460 * - migration optimization area (pba_of_lba0)
3461 * - descending reshape (ascending_migr)
3463 * super : imsm metadata information
3465 * 0 : migration is compatible
3466 * -1 : migration is not compatible
3467 ******************************************************************************/
3468 int check_mpb_migr_compatibility(struct intel_super
*super
)
3470 struct imsm_map
*map0
, *map1
;
3471 struct migr_record
*migr_rec
= super
->migr_rec
;
3474 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3475 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3478 dev_iter
->vol
.migr_state
== 1 &&
3479 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3480 /* This device is migrating */
3481 map0
= get_imsm_map(dev_iter
, MAP_0
);
3482 map1
= get_imsm_map(dev_iter
, MAP_1
);
3483 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3484 /* migration optimization area was used */
3486 if (migr_rec
->ascending_migr
== 0
3487 && migr_rec
->dest_depth_per_unit
> 0)
3488 /* descending reshape not supported yet */
3495 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3497 /* load_imsm_mpb - read matrix metadata
3498 * allocates super->mpb to be freed by free_imsm
3500 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3502 unsigned long long dsize
;
3503 unsigned long long sectors
;
3505 struct imsm_super
*anchor
;
3508 get_dev_size(fd
, NULL
, &dsize
);
3511 pr_err("%s: device to small for imsm\n",
3516 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3518 pr_err("Cannot seek to anchor block on %s: %s\n",
3519 devname
, strerror(errno
));
3523 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3525 pr_err("Failed to allocate imsm anchor buffer"
3526 " on %s\n", devname
);
3529 if (read(fd
, anchor
, 512) != 512) {
3531 pr_err("Cannot read anchor block on %s: %s\n",
3532 devname
, strerror(errno
));
3537 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3539 pr_err("no IMSM anchor on %s\n", devname
);
3544 __free_imsm(super
, 0);
3545 /* reload capability and hba */
3547 /* capability and hba must be updated with new super allocation */
3548 find_intel_hba_capability(fd
, super
, devname
);
3549 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3550 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3552 pr_err("unable to allocate %zu byte mpb buffer\n",
3557 memcpy(super
->buf
, anchor
, 512);
3559 sectors
= mpb_sectors(anchor
) - 1;
3562 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3563 pr_err("%s could not allocate migr_rec buffer\n", __func__
);
3567 super
->clean_migration_record_by_mdmon
= 0;
3570 check_sum
= __gen_imsm_checksum(super
->anchor
);
3571 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3573 pr_err("IMSM checksum %x != %x on %s\n",
3575 __le32_to_cpu(super
->anchor
->check_sum
),
3583 /* read the extended mpb */
3584 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3586 pr_err("Cannot seek to extended mpb on %s: %s\n",
3587 devname
, strerror(errno
));
3591 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3593 pr_err("Cannot read extended mpb on %s: %s\n",
3594 devname
, strerror(errno
));
3598 check_sum
= __gen_imsm_checksum(super
->anchor
);
3599 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3601 pr_err("IMSM checksum %x != %x on %s\n",
3602 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3607 /* FIXME the BBM log is disk specific so we cannot use this global
3608 * buffer for all disks. Ok for now since we only look at the global
3609 * bbm_log_size parameter to gate assembly
3611 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3616 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3618 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3619 static void clear_hi(struct intel_super
*super
)
3621 struct imsm_super
*mpb
= super
->anchor
;
3623 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3625 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3626 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3627 disk
->total_blocks_hi
= 0;
3629 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3630 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3633 for (n
= 0; n
< 2; ++n
) {
3634 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3637 map
->pba_of_lba0_hi
= 0;
3638 map
->blocks_per_member_hi
= 0;
3639 map
->num_data_stripes_hi
= 0;
3645 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3649 err
= load_imsm_mpb(fd
, super
, devname
);
3652 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3655 err
= parse_raid_devices(super
);
3660 static void __free_imsm_disk(struct dl
*d
)
3672 static void free_imsm_disks(struct intel_super
*super
)
3676 while (super
->disks
) {
3678 super
->disks
= d
->next
;
3679 __free_imsm_disk(d
);
3681 while (super
->disk_mgmt_list
) {
3682 d
= super
->disk_mgmt_list
;
3683 super
->disk_mgmt_list
= d
->next
;
3684 __free_imsm_disk(d
);
3686 while (super
->missing
) {
3688 super
->missing
= d
->next
;
3689 __free_imsm_disk(d
);
3694 /* free all the pieces hanging off of a super pointer */
3695 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3697 struct intel_hba
*elem
, *next
;
3703 /* unlink capability description */
3705 if (super
->migr_rec_buf
) {
3706 free(super
->migr_rec_buf
);
3707 super
->migr_rec_buf
= NULL
;
3710 free_imsm_disks(super
);
3711 free_devlist(super
);
3715 free((void *)elem
->path
);
3723 static void free_imsm(struct intel_super
*super
)
3725 __free_imsm(super
, 1);
3729 static void free_super_imsm(struct supertype
*st
)
3731 struct intel_super
*super
= st
->sb
;
3740 static struct intel_super
*alloc_super(void)
3742 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
3744 super
->current_vol
= -1;
3745 super
->create_offset
= ~((unsigned long long) 0);
3750 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3752 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3754 struct sys_dev
*hba_name
;
3757 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3762 hba_name
= find_disk_attached_hba(fd
, NULL
);
3765 pr_err("%s is not attached to Intel(R) RAID controller.\n",
3769 rv
= attach_hba_to_super(super
, hba_name
);
3772 struct intel_hba
*hba
= super
->hba
;
3774 pr_err("%s is attached to Intel(R) %s RAID "
3775 "controller (%s),\n"
3776 " but the container is assigned to Intel(R) "
3777 "%s RAID controller (",
3780 hba_name
->pci_id
? : "Err!",
3781 get_sys_dev_type(hba_name
->type
));
3784 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3786 fprintf(stderr
, ", ");
3790 fprintf(stderr
, ").\n"
3791 " Mixing devices attached to multiple controllers "
3792 "is not allowed.\n");
3796 super
->orom
= find_imsm_capability(hba_name
->type
);
3802 /* find_missing - helper routine for load_super_imsm_all that identifies
3803 * disks that have disappeared from the system. This routine relies on
3804 * the mpb being uptodate, which it is at load time.
3806 static int find_missing(struct intel_super
*super
)
3809 struct imsm_super
*mpb
= super
->anchor
;
3811 struct imsm_disk
*disk
;
3813 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3814 disk
= __get_imsm_disk(mpb
, i
);
3815 dl
= serial_to_dl(disk
->serial
, super
);
3819 dl
= xmalloc(sizeof(*dl
));
3823 dl
->devname
= xstrdup("missing");
3825 serialcpy(dl
->serial
, disk
->serial
);
3828 dl
->next
= super
->missing
;
3829 super
->missing
= dl
;
3836 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3838 struct intel_disk
*idisk
= disk_list
;
3841 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3843 idisk
= idisk
->next
;
3849 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3850 struct intel_super
*super
,
3851 struct intel_disk
**disk_list
)
3853 struct imsm_disk
*d
= &super
->disks
->disk
;
3854 struct imsm_super
*mpb
= super
->anchor
;
3857 for (i
= 0; i
< tbl_size
; i
++) {
3858 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3859 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3861 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3862 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3863 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3864 __func__
, super
->disks
->major
,
3865 super
->disks
->minor
,
3866 table
[i
]->disks
->major
,
3867 table
[i
]->disks
->minor
);
3871 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3872 is_configured(d
) == is_configured(tbl_d
)) &&
3873 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3874 /* current version of the mpb is a
3875 * better candidate than the one in
3876 * super_table, but copy over "cross
3877 * generational" status
3879 struct intel_disk
*idisk
;
3881 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3882 __func__
, super
->disks
->major
,
3883 super
->disks
->minor
,
3884 table
[i
]->disks
->major
,
3885 table
[i
]->disks
->minor
);
3887 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3888 if (idisk
&& is_failed(&idisk
->disk
))
3889 tbl_d
->status
|= FAILED_DISK
;
3892 struct intel_disk
*idisk
;
3893 struct imsm_disk
*disk
;
3895 /* tbl_mpb is more up to date, but copy
3896 * over cross generational status before
3899 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3900 if (disk
&& is_failed(disk
))
3901 d
->status
|= FAILED_DISK
;
3903 idisk
= disk_list_get(d
->serial
, *disk_list
);
3906 if (disk
&& is_configured(disk
))
3907 idisk
->disk
.status
|= CONFIGURED_DISK
;
3910 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3911 __func__
, super
->disks
->major
,
3912 super
->disks
->minor
,
3913 table
[i
]->disks
->major
,
3914 table
[i
]->disks
->minor
);
3922 table
[tbl_size
++] = super
;
3926 /* update/extend the merged list of imsm_disk records */
3927 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3928 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3929 struct intel_disk
*idisk
;
3931 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3933 idisk
->disk
.status
|= disk
->status
;
3934 if (is_configured(&idisk
->disk
) ||
3935 is_failed(&idisk
->disk
))
3936 idisk
->disk
.status
&= ~(SPARE_DISK
);
3938 idisk
= xcalloc(1, sizeof(*idisk
));
3939 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3940 idisk
->disk
= *disk
;
3941 idisk
->next
= *disk_list
;
3945 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3952 static struct intel_super
*
3953 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3956 struct imsm_super
*mpb
= super
->anchor
;
3960 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3961 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3962 struct intel_disk
*idisk
;
3964 idisk
= disk_list_get(disk
->serial
, disk_list
);
3966 if (idisk
->owner
== owner
||
3967 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3970 dprintf("%s: '%.16s' owner %d != %d\n",
3971 __func__
, disk
->serial
, idisk
->owner
,
3974 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3975 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3981 if (ok_count
== mpb
->num_disks
)
3986 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3988 struct intel_super
*s
;
3990 for (s
= super_list
; s
; s
= s
->next
) {
3991 if (family_num
!= s
->anchor
->family_num
)
3993 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3994 __le32_to_cpu(family_num
), s
->disks
->devname
);
3998 static struct intel_super
*
3999 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4001 struct intel_super
*super_table
[len
];
4002 struct intel_disk
*disk_list
= NULL
;
4003 struct intel_super
*champion
, *spare
;
4004 struct intel_super
*s
, **del
;
4009 memset(super_table
, 0, sizeof(super_table
));
4010 for (s
= *super_list
; s
; s
= s
->next
)
4011 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4013 for (i
= 0; i
< tbl_size
; i
++) {
4014 struct imsm_disk
*d
;
4015 struct intel_disk
*idisk
;
4016 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4019 d
= &s
->disks
->disk
;
4021 /* 'd' must appear in merged disk list for its
4022 * configuration to be valid
4024 idisk
= disk_list_get(d
->serial
, disk_list
);
4025 if (idisk
&& idisk
->owner
== i
)
4026 s
= validate_members(s
, disk_list
, i
);
4031 dprintf("%s: marking family: %#x from %d:%d offline\n",
4032 __func__
, mpb
->family_num
,
4033 super_table
[i
]->disks
->major
,
4034 super_table
[i
]->disks
->minor
);
4038 /* This is where the mdadm implementation differs from the Windows
4039 * driver which has no strict concept of a container. We can only
4040 * assemble one family from a container, so when returning a prodigal
4041 * array member to this system the code will not be able to disambiguate
4042 * the container contents that should be assembled ("foreign" versus
4043 * "local"). It requires user intervention to set the orig_family_num
4044 * to a new value to establish a new container. The Windows driver in
4045 * this situation fixes up the volume name in place and manages the
4046 * foreign array as an independent entity.
4051 for (i
= 0; i
< tbl_size
; i
++) {
4052 struct intel_super
*tbl_ent
= super_table
[i
];
4058 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4063 if (s
&& !is_spare
) {
4064 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4066 } else if (!s
&& !is_spare
)
4079 fprintf(stderr
, "Chose family %#x on '%s', "
4080 "assemble conflicts to new container with '--update=uuid'\n",
4081 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4083 /* collect all dl's onto 'champion', and update them to
4084 * champion's version of the status
4086 for (s
= *super_list
; s
; s
= s
->next
) {
4087 struct imsm_super
*mpb
= champion
->anchor
;
4088 struct dl
*dl
= s
->disks
;
4093 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4095 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4096 struct imsm_disk
*disk
;
4098 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4101 /* only set index on disks that are a member of
4102 * a populated contianer, i.e. one with
4105 if (is_failed(&dl
->disk
))
4107 else if (is_spare(&dl
->disk
))
4113 if (i
>= mpb
->num_disks
) {
4114 struct intel_disk
*idisk
;
4116 idisk
= disk_list_get(dl
->serial
, disk_list
);
4117 if (idisk
&& is_spare(&idisk
->disk
) &&
4118 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4126 dl
->next
= champion
->disks
;
4127 champion
->disks
= dl
;
4131 /* delete 'champion' from super_list */
4132 for (del
= super_list
; *del
; ) {
4133 if (*del
== champion
) {
4134 *del
= (*del
)->next
;
4137 del
= &(*del
)->next
;
4139 champion
->next
= NULL
;
4143 struct intel_disk
*idisk
= disk_list
;
4145 disk_list
= disk_list
->next
;
4154 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4155 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4156 int major
, int minor
, int keep_fd
);
4158 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4159 int *max
, int keep_fd
);
4162 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4163 char *devname
, struct md_list
*devlist
,
4166 struct intel_super
*super_list
= NULL
;
4167 struct intel_super
*super
= NULL
;
4172 /* 'fd' is an opened container */
4173 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4175 /* get super block from devlist devices */
4176 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4179 /* all mpbs enter, maybe one leaves */
4180 super
= imsm_thunderdome(&super_list
, i
);
4186 if (find_missing(super
) != 0) {
4192 /* load migration record */
4193 err
= load_imsm_migr_rec(super
, NULL
);
4195 /* migration is in progress,
4196 * but migr_rec cannot be loaded,
4202 /* Check migration compatibility */
4203 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4204 pr_err("Unsupported migration detected");
4206 fprintf(stderr
, " on %s\n", devname
);
4208 fprintf(stderr
, " (IMSM).\n");
4217 while (super_list
) {
4218 struct intel_super
*s
= super_list
;
4220 super_list
= super_list
->next
;
4230 strcpy(st
->container_devnm
, fd2devnm(fd
));
4232 st
->container_devnm
[0] = 0;
4233 if (err
== 0 && st
->ss
== NULL
) {
4234 st
->ss
= &super_imsm
;
4235 st
->minor_version
= 0;
4236 st
->max_devs
= IMSM_MAX_DEVICES
;
4243 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4244 int *max
, int keep_fd
)
4246 struct md_list
*tmpdev
;
4250 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4251 if (tmpdev
->used
!= 1)
4253 if (tmpdev
->container
== 1) {
4255 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4257 pr_err("cannot open device %s: %s\n",
4258 tmpdev
->devname
, strerror(errno
));
4262 err
= get_sra_super_block(fd
, super_list
,
4263 tmpdev
->devname
, &lmax
,
4272 int major
= major(tmpdev
->st_rdev
);
4273 int minor
= minor(tmpdev
->st_rdev
);
4274 err
= get_super_block(super_list
,
4291 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4292 int major
, int minor
, int keep_fd
)
4294 struct intel_super
*s
= NULL
;
4306 sprintf(nm
, "%d:%d", major
, minor
);
4307 dfd
= dev_open(nm
, O_RDWR
);
4313 find_intel_hba_capability(dfd
, s
, devname
);
4314 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4316 /* retry the load if we might have raced against mdmon */
4317 if (err
== 3 && devnm
&& mdmon_running(devnm
))
4318 for (retry
= 0; retry
< 3; retry
++) {
4320 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4326 s
->next
= *super_list
;
4334 if ((dfd
>= 0) && (!keep_fd
))
4341 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4348 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4352 if (sra
->array
.major_version
!= -1 ||
4353 sra
->array
.minor_version
!= -2 ||
4354 strcmp(sra
->text_version
, "imsm") != 0) {
4359 devnm
= fd2devnm(fd
);
4360 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4361 if (get_super_block(super_list
, devnm
, devname
,
4362 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4373 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4375 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4379 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4381 struct intel_super
*super
;
4384 if (test_partition(fd
))
4385 /* IMSM not allowed on partitions */
4388 free_super_imsm(st
);
4390 super
= alloc_super();
4391 /* Load hba and capabilities if they exist.
4392 * But do not preclude loading metadata in case capabilities or hba are
4393 * non-compliant and ignore_hw_compat is set.
4395 rv
= find_intel_hba_capability(fd
, super
, devname
);
4396 /* no orom/efi or non-intel hba of the disk */
4397 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4399 pr_err("No OROM/EFI properties for %s\n", devname
);
4403 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4407 pr_err("Failed to load all information "
4408 "sections on %s\n", devname
);
4414 if (st
->ss
== NULL
) {
4415 st
->ss
= &super_imsm
;
4416 st
->minor_version
= 0;
4417 st
->max_devs
= IMSM_MAX_DEVICES
;
4420 /* load migration record */
4421 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4422 /* Check for unsupported migration features */
4423 if (check_mpb_migr_compatibility(super
) != 0) {
4424 pr_err("Unsupported migration detected");
4426 fprintf(stderr
, " on %s\n", devname
);
4428 fprintf(stderr
, " (IMSM).\n");
4436 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4438 if (info
->level
== 1)
4440 return info
->chunk_size
>> 9;
4443 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4444 unsigned long long size
)
4446 if (info
->level
== 1)
4449 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4452 static void imsm_update_version_info(struct intel_super
*super
)
4454 /* update the version and attributes */
4455 struct imsm_super
*mpb
= super
->anchor
;
4457 struct imsm_dev
*dev
;
4458 struct imsm_map
*map
;
4461 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4462 dev
= get_imsm_dev(super
, i
);
4463 map
= get_imsm_map(dev
, MAP_0
);
4464 if (__le32_to_cpu(dev
->size_high
) > 0)
4465 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4467 /* FIXME detect when an array spans a port multiplier */
4469 mpb
->attributes
|= MPB_ATTRIB_PM
;
4472 if (mpb
->num_raid_devs
> 1 ||
4473 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4474 version
= MPB_VERSION_ATTRIBS
;
4475 switch (get_imsm_raid_level(map
)) {
4476 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4477 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4478 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4479 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4482 if (map
->num_members
>= 5)
4483 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4484 else if (dev
->status
== DEV_CLONE_N_GO
)
4485 version
= MPB_VERSION_CNG
;
4486 else if (get_imsm_raid_level(map
) == 5)
4487 version
= MPB_VERSION_RAID5
;
4488 else if (map
->num_members
>= 3)
4489 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4490 else if (get_imsm_raid_level(map
) == 1)
4491 version
= MPB_VERSION_RAID1
;
4493 version
= MPB_VERSION_RAID0
;
4495 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4499 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4501 struct imsm_super
*mpb
= super
->anchor
;
4502 char *reason
= NULL
;
4505 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4506 reason
= "must be 16 characters or less";
4508 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4509 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4511 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4512 reason
= "already exists";
4517 if (reason
&& !quiet
)
4518 pr_err("imsm volume name %s\n", reason
);
4523 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4524 unsigned long long size
, char *name
,
4525 char *homehost
, int *uuid
,
4526 long long data_offset
)
4528 /* We are creating a volume inside a pre-existing container.
4529 * so st->sb is already set.
4531 struct intel_super
*super
= st
->sb
;
4532 struct imsm_super
*mpb
= super
->anchor
;
4533 struct intel_dev
*dv
;
4534 struct imsm_dev
*dev
;
4535 struct imsm_vol
*vol
;
4536 struct imsm_map
*map
;
4537 int idx
= mpb
->num_raid_devs
;
4539 unsigned long long array_blocks
;
4540 size_t size_old
, size_new
;
4541 unsigned long long num_data_stripes
;
4543 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4544 pr_err("This imsm-container already has the "
4545 "maximum of %d volumes\n", super
->orom
->vpa
);
4549 /* ensure the mpb is large enough for the new data */
4550 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4551 size_new
= disks_to_mpb_size(info
->nr_disks
);
4552 if (size_new
> size_old
) {
4554 size_t size_round
= ROUND_UP(size_new
, 512);
4556 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4557 pr_err("could not allocate new mpb\n");
4560 if (posix_memalign(&super
->migr_rec_buf
, 512,
4561 MIGR_REC_BUF_SIZE
) != 0) {
4562 pr_err("%s could not allocate migr_rec buffer\n",
4569 memcpy(mpb_new
, mpb
, size_old
);
4572 super
->anchor
= mpb_new
;
4573 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4574 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4576 super
->current_vol
= idx
;
4578 /* handle 'failed_disks' by either:
4579 * a) create dummy disk entries in the table if this the first
4580 * volume in the array. We add them here as this is the only
4581 * opportunity to add them. add_to_super_imsm_volume()
4582 * handles the non-failed disks and continues incrementing
4584 * b) validate that 'failed_disks' matches the current number
4585 * of missing disks if the container is populated
4587 if (super
->current_vol
== 0) {
4589 for (i
= 0; i
< info
->failed_disks
; i
++) {
4590 struct imsm_disk
*disk
;
4593 disk
= __get_imsm_disk(mpb
, i
);
4594 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4595 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4596 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4599 find_missing(super
);
4604 for (d
= super
->missing
; d
; d
= d
->next
)
4606 if (info
->failed_disks
> missing
) {
4607 pr_err("unable to add 'missing' disk to container\n");
4612 if (!check_name(super
, name
, 0))
4614 dv
= xmalloc(sizeof(*dv
));
4615 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4616 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4617 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4618 info
->layout
, info
->chunk_size
,
4620 /* round array size down to closest MB */
4621 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4623 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4624 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4625 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4627 vol
->migr_state
= 0;
4628 set_migr_type(dev
, MIGR_INIT
);
4629 vol
->dirty
= !info
->state
;
4630 vol
->curr_migr_unit
= 0;
4631 map
= get_imsm_map(dev
, MAP_0
);
4632 set_pba_of_lba0(map
, super
->create_offset
);
4633 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4634 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4635 map
->failed_disk_num
= ~0;
4636 if (info
->level
> 0)
4637 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
4638 : IMSM_T_STATE_UNINITIALIZED
);
4640 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4641 IMSM_T_STATE_NORMAL
;
4644 if (info
->level
== 1 && info
->raid_disks
> 2) {
4647 pr_err("imsm does not support more than 2 disks"
4648 "in a raid1 volume\n");
4652 map
->raid_level
= info
->level
;
4653 if (info
->level
== 10) {
4654 map
->raid_level
= 1;
4655 map
->num_domains
= info
->raid_disks
/ 2;
4656 } else if (info
->level
== 1)
4657 map
->num_domains
= info
->raid_disks
;
4659 map
->num_domains
= 1;
4661 /* info->size is only int so use the 'size' parameter instead */
4662 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4663 num_data_stripes
/= map
->num_domains
;
4664 set_num_data_stripes(map
, num_data_stripes
);
4666 map
->num_members
= info
->raid_disks
;
4667 for (i
= 0; i
< map
->num_members
; i
++) {
4668 /* initialized in add_to_super */
4669 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4671 mpb
->num_raid_devs
++;
4674 dv
->index
= super
->current_vol
;
4675 dv
->next
= super
->devlist
;
4676 super
->devlist
= dv
;
4678 imsm_update_version_info(super
);
4683 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4684 unsigned long long size
, char *name
,
4685 char *homehost
, int *uuid
,
4686 unsigned long long data_offset
)
4688 /* This is primarily called by Create when creating a new array.
4689 * We will then get add_to_super called for each component, and then
4690 * write_init_super called to write it out to each device.
4691 * For IMSM, Create can create on fresh devices or on a pre-existing
4693 * To create on a pre-existing array a different method will be called.
4694 * This one is just for fresh drives.
4696 struct intel_super
*super
;
4697 struct imsm_super
*mpb
;
4701 if (data_offset
!= INVALID_SECTORS
) {
4702 fprintf(stderr
, Name
": data-offset not supported by imsm\n");
4707 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
,
4711 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4715 super
= alloc_super();
4716 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4721 pr_err("%s could not allocate superblock\n", __func__
);
4724 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4725 pr_err("%s could not allocate migr_rec buffer\n", __func__
);
4730 memset(super
->buf
, 0, mpb_size
);
4732 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4736 /* zeroing superblock */
4740 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4742 version
= (char *) mpb
->sig
;
4743 strcpy(version
, MPB_SIGNATURE
);
4744 version
+= strlen(MPB_SIGNATURE
);
4745 strcpy(version
, MPB_VERSION_RAID0
);
4751 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4752 int fd
, char *devname
)
4754 struct intel_super
*super
= st
->sb
;
4755 struct imsm_super
*mpb
= super
->anchor
;
4756 struct imsm_disk
*_disk
;
4757 struct imsm_dev
*dev
;
4758 struct imsm_map
*map
;
4762 dev
= get_imsm_dev(super
, super
->current_vol
);
4763 map
= get_imsm_map(dev
, MAP_0
);
4765 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4766 pr_err("%s: Cannot add spare devices to IMSM volume\n",
4772 /* we're doing autolayout so grab the pre-marked (in
4773 * validate_geometry) raid_disk
4775 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4776 if (dl
->raiddisk
== dk
->raid_disk
)
4779 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4780 if (dl
->major
== dk
->major
&&
4781 dl
->minor
== dk
->minor
)
4786 pr_err("%s is not a member of the same container\n", devname
);
4790 /* add a pristine spare to the metadata */
4791 if (dl
->index
< 0) {
4792 dl
->index
= super
->anchor
->num_disks
;
4793 super
->anchor
->num_disks
++;
4795 /* Check the device has not already been added */
4796 slot
= get_imsm_disk_slot(map
, dl
->index
);
4798 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4799 pr_err("%s has been included in this array twice\n",
4803 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4804 dl
->disk
.status
= CONFIGURED_DISK
;
4806 /* update size of 'missing' disks to be at least as large as the
4807 * largest acitve member (we only have dummy missing disks when
4808 * creating the first volume)
4810 if (super
->current_vol
== 0) {
4811 for (df
= super
->missing
; df
; df
= df
->next
) {
4812 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
4813 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
4814 _disk
= __get_imsm_disk(mpb
, df
->index
);
4819 /* refresh unset/failed slots to point to valid 'missing' entries */
4820 for (df
= super
->missing
; df
; df
= df
->next
)
4821 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4822 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4824 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4826 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4827 if (is_gen_migration(dev
)) {
4828 struct imsm_map
*map2
= get_imsm_map(dev
,
4830 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4831 if ((slot2
< map2
->num_members
) &&
4833 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4836 if ((unsigned)df
->index
==
4838 set_imsm_ord_tbl_ent(map2
,
4844 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4848 /* if we are creating the first raid device update the family number */
4849 if (super
->current_vol
== 0) {
4851 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4853 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4854 if (!_dev
|| !_disk
) {
4855 pr_err("BUG mpb setup error\n");
4861 sum
+= __gen_imsm_checksum(mpb
);
4862 mpb
->family_num
= __cpu_to_le32(sum
);
4863 mpb
->orig_family_num
= mpb
->family_num
;
4865 super
->current_disk
= dl
;
4870 * Function marks disk as spare and restores disk serial
4871 * in case it was previously marked as failed by takeover operation
4873 * -1 : critical error
4874 * 0 : disk is marked as spare but serial is not set
4877 int mark_spare(struct dl
*disk
)
4879 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4886 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4887 /* Restore disk serial number, because takeover marks disk
4888 * as failed and adds to serial ':0' before it becomes
4891 serialcpy(disk
->serial
, serial
);
4892 serialcpy(disk
->disk
.serial
, serial
);
4895 disk
->disk
.status
= SPARE_DISK
;
4901 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4902 int fd
, char *devname
,
4903 unsigned long long data_offset
)
4905 struct intel_super
*super
= st
->sb
;
4907 unsigned long long size
;
4912 /* If we are on an RAID enabled platform check that the disk is
4913 * attached to the raid controller.
4914 * We do not need to test disks attachment for container based additions,
4915 * they shall be already tested when container was created/assembled.
4917 rv
= find_intel_hba_capability(fd
, super
, devname
);
4918 /* no orom/efi or non-intel hba of the disk */
4920 dprintf("capability: %p fd: %d ret: %d\n",
4921 super
->orom
, fd
, rv
);
4925 if (super
->current_vol
>= 0)
4926 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4929 dd
= xcalloc(sizeof(*dd
), 1);
4930 dd
->major
= major(stb
.st_rdev
);
4931 dd
->minor
= minor(stb
.st_rdev
);
4932 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
4935 dd
->action
= DISK_ADD
;
4936 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4938 pr_err("failed to retrieve scsi serial, aborting\n");
4943 get_dev_size(fd
, NULL
, &size
);
4945 serialcpy(dd
->disk
.serial
, dd
->serial
);
4946 set_total_blocks(&dd
->disk
, size
);
4947 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
4948 struct imsm_super
*mpb
= super
->anchor
;
4949 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
4952 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4953 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4955 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4957 if (st
->update_tail
) {
4958 dd
->next
= super
->disk_mgmt_list
;
4959 super
->disk_mgmt_list
= dd
;
4961 dd
->next
= super
->disks
;
4963 super
->updates_pending
++;
4970 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4972 struct intel_super
*super
= st
->sb
;
4975 /* remove from super works only in mdmon - for communication
4976 * manager - monitor. Check if communication memory buffer
4979 if (!st
->update_tail
) {
4980 pr_err("%s shall be used in mdmon context only"
4981 "(line %d).\n", __func__
, __LINE__
);
4984 dd
= xcalloc(1, sizeof(*dd
));
4985 dd
->major
= dk
->major
;
4986 dd
->minor
= dk
->minor
;
4989 dd
->action
= DISK_REMOVE
;
4991 dd
->next
= super
->disk_mgmt_list
;
4992 super
->disk_mgmt_list
= dd
;
4998 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5002 struct imsm_super anchor
;
5003 } spare_record
__attribute__ ((aligned(512)));
5005 /* spare records have their own family number and do not have any defined raid
5008 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5010 struct imsm_super
*mpb
= super
->anchor
;
5011 struct imsm_super
*spare
= &spare_record
.anchor
;
5015 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
5016 spare
->generation_num
= __cpu_to_le32(1UL),
5017 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5018 spare
->num_disks
= 1,
5019 spare
->num_raid_devs
= 0,
5020 spare
->cache_size
= mpb
->cache_size
,
5021 spare
->pwr_cycle_count
= __cpu_to_le32(1),
5023 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5024 MPB_SIGNATURE MPB_VERSION_RAID0
);
5026 for (d
= super
->disks
; d
; d
= d
->next
) {
5030 spare
->disk
[0] = d
->disk
;
5031 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5032 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5034 sum
= __gen_imsm_checksum(spare
);
5035 spare
->family_num
= __cpu_to_le32(sum
);
5036 spare
->orig_family_num
= 0;
5037 sum
= __gen_imsm_checksum(spare
);
5038 spare
->check_sum
= __cpu_to_le32(sum
);
5040 if (store_imsm_mpb(d
->fd
, spare
)) {
5041 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
5042 __func__
, d
->major
, d
->minor
, strerror(errno
));
5054 static int write_super_imsm(struct supertype
*st
, int doclose
)
5056 struct intel_super
*super
= st
->sb
;
5057 struct imsm_super
*mpb
= super
->anchor
;
5063 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5065 int clear_migration_record
= 1;
5067 /* 'generation' is incremented everytime the metadata is written */
5068 generation
= __le32_to_cpu(mpb
->generation_num
);
5070 mpb
->generation_num
= __cpu_to_le32(generation
);
5072 /* fix up cases where previous mdadm releases failed to set
5075 if (mpb
->orig_family_num
== 0)
5076 mpb
->orig_family_num
= mpb
->family_num
;
5078 for (d
= super
->disks
; d
; d
= d
->next
) {
5082 mpb
->disk
[d
->index
] = d
->disk
;
5086 for (d
= super
->missing
; d
; d
= d
->next
) {
5087 mpb
->disk
[d
->index
] = d
->disk
;
5090 mpb
->num_disks
= num_disks
;
5091 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5093 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5094 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5095 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5097 imsm_copy_dev(dev
, dev2
);
5098 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5100 if (is_gen_migration(dev2
))
5101 clear_migration_record
= 0;
5103 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5104 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5106 /* recalculate checksum */
5107 sum
= __gen_imsm_checksum(mpb
);
5108 mpb
->check_sum
= __cpu_to_le32(sum
);
5110 if (super
->clean_migration_record_by_mdmon
) {
5111 clear_migration_record
= 1;
5112 super
->clean_migration_record_by_mdmon
= 0;
5114 if (clear_migration_record
)
5115 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5117 /* write the mpb for disks that compose raid devices */
5118 for (d
= super
->disks
; d
; d
= d
->next
) {
5119 if (d
->index
< 0 || is_failed(&d
->disk
))
5122 if (clear_migration_record
) {
5123 unsigned long long dsize
;
5125 get_dev_size(d
->fd
, NULL
, &dsize
);
5126 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5127 if (write(d
->fd
, super
->migr_rec_buf
,
5128 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5129 perror("Write migr_rec failed");
5133 if (store_imsm_mpb(d
->fd
, mpb
))
5135 "%s: failed for device %d:%d (fd: %d)%s\n",
5136 __func__
, d
->major
, d
->minor
,
5137 d
->fd
, strerror(errno
));
5146 return write_super_imsm_spares(super
, doclose
);
5152 static int create_array(struct supertype
*st
, int dev_idx
)
5155 struct imsm_update_create_array
*u
;
5156 struct intel_super
*super
= st
->sb
;
5157 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5158 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5159 struct disk_info
*inf
;
5160 struct imsm_disk
*disk
;
5163 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5164 sizeof(*inf
) * map
->num_members
;
5166 u
->type
= update_create_array
;
5167 u
->dev_idx
= dev_idx
;
5168 imsm_copy_dev(&u
->dev
, dev
);
5169 inf
= get_disk_info(u
);
5170 for (i
= 0; i
< map
->num_members
; i
++) {
5171 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5173 disk
= get_imsm_disk(super
, idx
);
5174 serialcpy(inf
[i
].serial
, disk
->serial
);
5176 append_metadata_update(st
, u
, len
);
5181 static int mgmt_disk(struct supertype
*st
)
5183 struct intel_super
*super
= st
->sb
;
5185 struct imsm_update_add_remove_disk
*u
;
5187 if (!super
->disk_mgmt_list
)
5192 u
->type
= update_add_remove_disk
;
5193 append_metadata_update(st
, u
, len
);
5198 static int write_init_super_imsm(struct supertype
*st
)
5200 struct intel_super
*super
= st
->sb
;
5201 int current_vol
= super
->current_vol
;
5203 /* we are done with current_vol reset it to point st at the container */
5204 super
->current_vol
= -1;
5206 if (st
->update_tail
) {
5207 /* queue the recently created array / added disk
5208 * as a metadata update */
5211 /* determine if we are creating a volume or adding a disk */
5212 if (current_vol
< 0) {
5213 /* in the mgmt (add/remove) disk case we are running
5214 * in mdmon context, so don't close fd's
5216 return mgmt_disk(st
);
5218 rv
= create_array(st
, current_vol
);
5223 for (d
= super
->disks
; d
; d
= d
->next
)
5224 Kill(d
->devname
, NULL
, 0, -1, 1);
5225 return write_super_imsm(st
, 1);
5230 static int store_super_imsm(struct supertype
*st
, int fd
)
5232 struct intel_super
*super
= st
->sb
;
5233 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5239 return store_imsm_mpb(fd
, mpb
);
5245 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5247 return __le32_to_cpu(mpb
->bbm_log_size
);
5251 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5252 int layout
, int raiddisks
, int chunk
,
5253 unsigned long long size
,
5254 unsigned long long data_offset
,
5256 unsigned long long *freesize
,
5260 unsigned long long ldsize
;
5261 struct intel_super
*super
=NULL
;
5264 if (level
!= LEVEL_CONTAINER
)
5269 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5272 pr_err("imsm: Cannot open %s: %s\n",
5273 dev
, strerror(errno
));
5276 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5281 /* capabilities retrieve could be possible
5282 * note that there is no fd for the disks in array.
5284 super
= alloc_super();
5285 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
5289 fd2devname(fd
, str
);
5290 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5291 fd
, str
, super
->orom
, rv
, raiddisks
);
5293 /* no orom/efi or non-intel hba of the disk */
5300 if (raiddisks
> super
->orom
->tds
) {
5302 pr_err("%d exceeds maximum number of"
5303 " platform supported disks: %d\n",
5304 raiddisks
, super
->orom
->tds
);
5308 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5309 (ldsize
>> 9) >> 32 > 0) {
5311 pr_err("%s exceeds maximum platform supported size\n", dev
);
5317 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
5323 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5325 const unsigned long long base_start
= e
[*idx
].start
;
5326 unsigned long long end
= base_start
+ e
[*idx
].size
;
5329 if (base_start
== end
)
5333 for (i
= *idx
; i
< num_extents
; i
++) {
5334 /* extend overlapping extents */
5335 if (e
[i
].start
>= base_start
&&
5336 e
[i
].start
<= end
) {
5339 if (e
[i
].start
+ e
[i
].size
> end
)
5340 end
= e
[i
].start
+ e
[i
].size
;
5341 } else if (e
[i
].start
> end
) {
5347 return end
- base_start
;
5350 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5352 /* build a composite disk with all known extents and generate a new
5353 * 'maxsize' given the "all disks in an array must share a common start
5354 * offset" constraint
5356 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
5360 unsigned long long pos
;
5361 unsigned long long start
= 0;
5362 unsigned long long maxsize
;
5363 unsigned long reserve
;
5365 /* coalesce and sort all extents. also, check to see if we need to
5366 * reserve space between member arrays
5369 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5372 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5375 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5380 while (i
< sum_extents
) {
5381 e
[j
].start
= e
[i
].start
;
5382 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5384 if (e
[j
-1].size
== 0)
5393 unsigned long long esize
;
5395 esize
= e
[i
].start
- pos
;
5396 if (esize
>= maxsize
) {
5401 pos
= e
[i
].start
+ e
[i
].size
;
5403 } while (e
[i
-1].size
);
5409 /* FIXME assumes volume at offset 0 is the first volume in a
5412 if (start_extent
> 0)
5413 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5417 if (maxsize
< reserve
)
5420 super
->create_offset
= ~((unsigned long long) 0);
5421 if (start
+ reserve
> super
->create_offset
)
5422 return 0; /* start overflows create_offset */
5423 super
->create_offset
= start
+ reserve
;
5425 return maxsize
- reserve
;
5428 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5430 if (level
< 0 || level
== 6 || level
== 4)
5433 /* if we have an orom prevent invalid raid levels */
5436 case 0: return imsm_orom_has_raid0(orom
);
5439 return imsm_orom_has_raid1e(orom
);
5440 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5441 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5442 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5445 return 1; /* not on an Intel RAID platform so anything goes */
5452 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5453 int dpa
, int verbose
)
5455 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5456 struct mdstat_ent
*memb
= NULL
;
5459 struct md_list
*dv
= NULL
;
5462 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5463 if (memb
->metadata_version
&&
5464 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5465 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5466 !is_subarray(memb
->metadata_version
+9) &&
5468 struct dev_member
*dev
= memb
->members
;
5470 while(dev
&& (fd
< 0)) {
5471 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5472 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5474 fd
= open(path
, O_RDONLY
, 0);
5475 if ((num
<= 0) || (fd
< 0)) {
5476 pr_vrb(": Cannot open %s: %s\n",
5477 dev
->name
, strerror(errno
));
5483 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5484 struct mdstat_ent
*vol
;
5485 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5486 if ((vol
->active
> 0) &&
5487 vol
->metadata_version
&&
5488 is_container_member(vol
, memb
->dev
)) {
5493 if (*devlist
&& (found
< dpa
)) {
5494 dv
= xcalloc(1, sizeof(*dv
));
5495 dv
->devname
= xmalloc(strlen(memb
->dev
) + strlen("/dev/") + 1);
5496 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->dev
);
5499 dv
->next
= *devlist
;
5507 free_mdstat(mdstat
);
5512 static struct md_list
*
5513 get_loop_devices(void)
5516 struct md_list
*devlist
= NULL
;
5517 struct md_list
*dv
= NULL
;
5519 for(i
= 0; i
< 12; i
++) {
5520 dv
= xcalloc(1, sizeof(*dv
));
5521 dv
->devname
= xmalloc(40);
5522 sprintf(dv
->devname
, "/dev/loop%d", i
);
5530 static struct md_list
*
5531 get_devices(const char *hba_path
)
5533 struct md_list
*devlist
= NULL
;
5534 struct md_list
*dv
= NULL
;
5540 devlist
= get_loop_devices();
5543 /* scroll through /sys/dev/block looking for devices attached to
5546 dir
= opendir("/sys/dev/block");
5547 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5552 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5554 path
= devt_to_devpath(makedev(major
, minor
));
5557 if (!path_attached_to_hba(path
, hba_path
)) {
5564 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5566 fd2devname(fd
, buf
);
5569 pr_err("cannot open device: %s\n",
5575 dv
= xcalloc(1, sizeof(*dv
));
5576 dv
->devname
= xstrdup(buf
);
5583 devlist
= devlist
->next
;
5593 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5594 int verbose
, int *found
)
5596 struct md_list
*tmpdev
;
5598 struct supertype
*st
= NULL
;
5600 /* first walk the list of devices to find a consistent set
5601 * that match the criterea, if that is possible.
5602 * We flag the ones we like with 'used'.
5605 st
= match_metadata_desc_imsm("imsm");
5607 pr_vrb(": cannot allocate memory for imsm supertype\n");
5611 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5612 char *devname
= tmpdev
->devname
;
5614 struct supertype
*tst
;
5616 if (tmpdev
->used
> 1)
5618 tst
= dup_super(st
);
5620 pr_vrb(": cannot allocate memory for imsm supertype\n");
5623 tmpdev
->container
= 0;
5624 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5626 dprintf(": cannot open device %s: %s\n",
5627 devname
, strerror(errno
));
5629 } else if (fstat(dfd
, &stb
)< 0) {
5631 dprintf(": fstat failed for %s: %s\n",
5632 devname
, strerror(errno
));
5634 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5635 dprintf(": %s is not a block device.\n",
5638 } else if (must_be_container(dfd
)) {
5639 struct supertype
*cst
;
5640 cst
= super_by_fd(dfd
, NULL
);
5642 dprintf(": cannot recognize container type %s\n",
5645 } else if (tst
->ss
!= st
->ss
) {
5646 dprintf(": non-imsm container - ignore it: %s\n",
5649 } else if (!tst
->ss
->load_container
||
5650 tst
->ss
->load_container(tst
, dfd
, NULL
))
5653 tmpdev
->container
= 1;
5656 cst
->ss
->free_super(cst
);
5658 tmpdev
->st_rdev
= stb
.st_rdev
;
5659 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5660 dprintf(": no RAID superblock on %s\n",
5663 } else if (tst
->ss
->compare_super
== NULL
) {
5664 dprintf(": Cannot assemble %s metadata on %s\n",
5665 tst
->ss
->name
, devname
);
5671 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5672 /* Ignore unrecognised devices during auto-assembly */
5677 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5679 if (st
->minor_version
== -1)
5680 st
->minor_version
= tst
->minor_version
;
5682 if (memcmp(info
.uuid
, uuid_zero
,
5683 sizeof(int[4])) == 0) {
5684 /* this is a floating spare. It cannot define
5685 * an array unless there are no more arrays of
5686 * this type to be found. It can be included
5687 * in an array of this type though.
5693 if (st
->ss
!= tst
->ss
||
5694 st
->minor_version
!= tst
->minor_version
||
5695 st
->ss
->compare_super(st
, tst
) != 0) {
5696 /* Some mismatch. If exactly one array matches this host,
5697 * we can resolve on that one.
5698 * Or, if we are auto assembling, we just ignore the second
5701 dprintf(": superblock on %s doesn't match others - assembly aborted\n",
5707 dprintf("found: devname: %s\n", devname
);
5711 tst
->ss
->free_super(tst
);
5715 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5716 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5717 for (iter
= head
; iter
; iter
= iter
->next
) {
5718 dprintf("content->text_version: %s vol\n",
5719 iter
->text_version
);
5720 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5721 /* do not assemble arrays with unsupported
5723 dprintf(": Cannot activate member %s.\n",
5724 iter
->text_version
);
5731 dprintf(" no valid super block on device list: err: %d %p\n",
5735 dprintf(" no more devices to examin\n");
5738 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5739 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5741 if (count
< tmpdev
->found
)
5744 count
-= tmpdev
->found
;
5747 if (tmpdev
->used
== 1)
5752 st
->ss
->free_super(st
);
5758 count_volumes(char *hba
, int dpa
, int verbose
)
5760 struct md_list
*devlist
= NULL
;
5764 devlist
= get_devices(hba
);
5765 /* if no intel devices return zero volumes */
5766 if (devlist
== NULL
)
5769 count
= active_arrays_by_format("imsm", hba
, &devlist
, dpa
, verbose
);
5770 dprintf(" path: %s active arrays: %d\n", hba
, count
);
5771 if (devlist
== NULL
)
5775 count
+= count_volumes_list(devlist
,
5779 dprintf("found %d count: %d\n", found
, count
);
5782 dprintf("path: %s total number of volumes: %d\n", hba
, count
);
5785 struct md_list
*dv
= devlist
;
5786 devlist
= devlist
->next
;
5793 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5795 /* up to 512 if the plaform supports it, otherwise the platform max.
5796 * 128 if no platform detected
5798 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5800 return min(512, (1 << fs
));
5804 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5805 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
5807 /* check/set platform and metadata limits/defaults */
5808 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5809 pr_vrb(": platform supports a maximum of %d disks per array\n",
5814 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5815 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5816 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5817 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5821 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5822 *chunk
= imsm_default_chunk(super
->orom
);
5824 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5825 pr_vrb(": platform does not support a chunk size of: "
5830 if (layout
!= imsm_level_to_layout(level
)) {
5832 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5833 else if (level
== 10)
5834 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5836 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5841 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 && chunk
&&
5842 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
5843 pr_vrb(": platform does not support a volume size over 2TB\n");
5849 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5850 * FIX ME add ahci details
5852 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5853 int layout
, int raiddisks
, int *chunk
,
5854 unsigned long long size
,
5855 unsigned long long data_offset
,
5857 unsigned long long *freesize
,
5861 struct intel_super
*super
= st
->sb
;
5862 struct imsm_super
*mpb
;
5864 unsigned long long pos
= 0;
5865 unsigned long long maxsize
;
5869 /* We must have the container info already read in. */
5873 mpb
= super
->anchor
;
5875 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
5876 pr_err("RAID gemetry validation failed. "
5877 "Cannot proceed with the action(s).\n");
5881 /* General test: make sure there is space for
5882 * 'raiddisks' device extents of size 'size' at a given
5885 unsigned long long minsize
= size
;
5886 unsigned long long start_offset
= MaxSector
;
5889 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5890 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5895 e
= get_extents(super
, dl
);
5898 unsigned long long esize
;
5899 esize
= e
[i
].start
- pos
;
5900 if (esize
>= minsize
)
5902 if (found
&& start_offset
== MaxSector
) {
5905 } else if (found
&& pos
!= start_offset
) {
5909 pos
= e
[i
].start
+ e
[i
].size
;
5911 } while (e
[i
-1].size
);
5916 if (dcnt
< raiddisks
) {
5918 pr_err("imsm: Not enough "
5919 "devices with space for this array "
5927 /* This device must be a member of the set */
5928 if (stat(dev
, &stb
) < 0)
5930 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5932 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5933 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5934 dl
->minor
== (int)minor(stb
.st_rdev
))
5939 pr_err("%s is not in the "
5940 "same imsm set\n", dev
);
5942 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5943 /* If a volume is present then the current creation attempt
5944 * cannot incorporate new spares because the orom may not
5945 * understand this configuration (all member disks must be
5946 * members of each array in the container).
5948 pr_err("%s is a spare and a volume"
5949 " is already defined for this container\n", dev
);
5950 pr_err("The option-rom requires all member"
5951 " disks to be a member of all volumes\n");
5953 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
5954 mpb
->num_disks
!= raiddisks
) {
5955 pr_err("The option-rom requires all member"
5956 " disks to be a member of all volumes\n");
5960 /* retrieve the largest free space block */
5961 e
= get_extents(super
, dl
);
5966 unsigned long long esize
;
5968 esize
= e
[i
].start
- pos
;
5969 if (esize
>= maxsize
)
5971 pos
= e
[i
].start
+ e
[i
].size
;
5973 } while (e
[i
-1].size
);
5978 pr_err("unable to determine free space for: %s\n",
5982 if (maxsize
< size
) {
5984 pr_err("%s not enough space (%llu < %llu)\n",
5985 dev
, maxsize
, size
);
5989 /* count total number of extents for merge */
5991 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5993 i
+= dl
->extent_cnt
;
5995 maxsize
= merge_extents(super
, i
);
5997 if (!check_env("IMSM_NO_PLATFORM") &&
5998 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
5999 pr_err("attempting to create a second "
6000 "volume with size less then remaining space. "
6005 if (maxsize
< size
|| maxsize
== 0) {
6008 pr_err("no free space"
6009 " left on device. Aborting...\n");
6011 pr_err("not enough space"
6012 " to create volume of given size"
6013 " (%llu < %llu). Aborting...\n",
6019 *freesize
= maxsize
;
6022 int count
= count_volumes(super
->hba
->path
,
6023 super
->orom
->dpa
, verbose
);
6024 if (super
->orom
->vphba
<= count
) {
6025 pr_vrb(": platform does not support more than %d raid volumes.\n",
6026 super
->orom
->vphba
);
6033 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6034 unsigned long long size
, int chunk
,
6035 unsigned long long *freesize
)
6037 struct intel_super
*super
= st
->sb
;
6038 struct imsm_super
*mpb
= super
->anchor
;
6043 unsigned long long maxsize
;
6044 unsigned long long minsize
;
6048 /* find the largest common start free region of the possible disks */
6052 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6058 /* don't activate new spares if we are orom constrained
6059 * and there is already a volume active in the container
6061 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6064 e
= get_extents(super
, dl
);
6067 for (i
= 1; e
[i
-1].size
; i
++)
6075 maxsize
= merge_extents(super
, extent_cnt
);
6079 minsize
= chunk
* 2;
6081 if (cnt
< raiddisks
||
6082 (super
->orom
&& used
&& used
!= raiddisks
) ||
6083 maxsize
< minsize
||
6085 pr_err("not enough devices with space to create array.\n");
6086 return 0; /* No enough free spaces large enough */
6097 if (!check_env("IMSM_NO_PLATFORM") &&
6098 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6099 pr_err("attempting to create a second "
6100 "volume with size less then remaining space. "
6105 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6107 dl
->raiddisk
= cnt
++;
6111 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6116 static int reserve_space(struct supertype
*st
, int raiddisks
,
6117 unsigned long long size
, int chunk
,
6118 unsigned long long *freesize
)
6120 struct intel_super
*super
= st
->sb
;
6125 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6128 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6130 dl
->raiddisk
= cnt
++;
6137 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6138 int raiddisks
, int *chunk
, unsigned long long size
,
6139 unsigned long long data_offset
,
6140 char *dev
, unsigned long long *freesize
,
6148 * if given unused devices create a container
6149 * if given given devices in a container create a member volume
6151 if (level
== LEVEL_CONTAINER
) {
6152 /* Must be a fresh device to add to a container */
6153 return validate_geometry_imsm_container(st
, level
, layout
,
6163 struct intel_super
*super
= st
->sb
;
6164 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6165 raiddisks
, chunk
, size
,
6168 /* we are being asked to automatically layout a
6169 * new volume based on the current contents of
6170 * the container. If the the parameters can be
6171 * satisfied reserve_space will record the disks,
6172 * start offset, and size of the volume to be
6173 * created. add_to_super and getinfo_super
6174 * detect when autolayout is in progress.
6176 /* assuming that freesize is always given when array is
6178 if (super
->orom
&& freesize
) {
6180 count
= count_volumes(super
->hba
->path
,
6181 super
->orom
->dpa
, verbose
);
6182 if (super
->orom
->vphba
<= count
) {
6183 pr_vrb(": platform does not support more"
6184 " than %d raid volumes.\n",
6185 super
->orom
->vphba
);
6190 return reserve_space(st
, raiddisks
, size
,
6191 chunk
?*chunk
:0, freesize
);
6196 /* creating in a given container */
6197 return validate_geometry_imsm_volume(st
, level
, layout
,
6198 raiddisks
, chunk
, size
,
6200 dev
, freesize
, verbose
);
6203 /* This device needs to be a device in an 'imsm' container */
6204 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6207 pr_err("Cannot create this array on device %s\n",
6212 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6214 pr_err("Cannot open %s: %s\n",
6215 dev
, strerror(errno
));
6218 /* Well, it is in use by someone, maybe an 'imsm' container. */
6219 cfd
= open_container(fd
);
6223 pr_err("Cannot use %s: It is busy\n",
6227 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
6228 if (sra
&& sra
->array
.major_version
== -1 &&
6229 strcmp(sra
->text_version
, "imsm") == 0)
6233 /* This is a member of a imsm container. Load the container
6234 * and try to create a volume
6236 struct intel_super
*super
;
6238 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6240 strcpy(st
->container_devnm
, fd2devnm(cfd
));
6242 return validate_geometry_imsm_volume(st
, level
, layout
,
6244 size
, data_offset
, dev
,
6251 pr_err("failed container membership check\n");
6257 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6259 struct intel_super
*super
= st
->sb
;
6261 if (level
&& *level
== UnSet
)
6262 *level
= LEVEL_CONTAINER
;
6264 if (level
&& layout
&& *layout
== UnSet
)
6265 *layout
= imsm_level_to_layout(*level
);
6267 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6268 *chunk
= imsm_default_chunk(super
->orom
);
6271 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6273 static int kill_subarray_imsm(struct supertype
*st
)
6275 /* remove the subarray currently referenced by ->current_vol */
6277 struct intel_dev
**dp
;
6278 struct intel_super
*super
= st
->sb
;
6279 __u8 current_vol
= super
->current_vol
;
6280 struct imsm_super
*mpb
= super
->anchor
;
6282 if (super
->current_vol
< 0)
6284 super
->current_vol
= -1; /* invalidate subarray cursor */
6286 /* block deletions that would change the uuid of active subarrays
6288 * FIXME when immutable ids are available, but note that we'll
6289 * also need to fixup the invalidated/active subarray indexes in
6292 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6295 if (i
< current_vol
)
6297 sprintf(subarray
, "%u", i
);
6298 if (is_subarray_active(subarray
, st
->devnm
)) {
6299 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6306 if (st
->update_tail
) {
6307 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
6309 u
->type
= update_kill_array
;
6310 u
->dev_idx
= current_vol
;
6311 append_metadata_update(st
, u
, sizeof(*u
));
6316 for (dp
= &super
->devlist
; *dp
;)
6317 if ((*dp
)->index
== current_vol
) {
6320 handle_missing(super
, (*dp
)->dev
);
6321 if ((*dp
)->index
> current_vol
)
6326 /* no more raid devices, all active components are now spares,
6327 * but of course failed are still failed
6329 if (--mpb
->num_raid_devs
== 0) {
6332 for (d
= super
->disks
; d
; d
= d
->next
)
6337 super
->updates_pending
++;
6342 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6343 char *update
, struct mddev_ident
*ident
)
6345 /* update the subarray currently referenced by ->current_vol */
6346 struct intel_super
*super
= st
->sb
;
6347 struct imsm_super
*mpb
= super
->anchor
;
6349 if (strcmp(update
, "name") == 0) {
6350 char *name
= ident
->name
;
6354 if (is_subarray_active(subarray
, st
->devnm
)) {
6355 pr_err("Unable to update name of active subarray\n");
6359 if (!check_name(super
, name
, 0))
6362 vol
= strtoul(subarray
, &ep
, 10);
6363 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6366 if (st
->update_tail
) {
6367 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
6369 u
->type
= update_rename_array
;
6371 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6372 append_metadata_update(st
, u
, sizeof(*u
));
6374 struct imsm_dev
*dev
;
6377 dev
= get_imsm_dev(super
, vol
);
6378 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6379 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6380 dev
= get_imsm_dev(super
, i
);
6381 handle_missing(super
, dev
);
6383 super
->updates_pending
++;
6390 #endif /* MDASSEMBLE */
6392 static int is_gen_migration(struct imsm_dev
*dev
)
6397 if (!dev
->vol
.migr_state
)
6400 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6406 static int is_rebuilding(struct imsm_dev
*dev
)
6408 struct imsm_map
*migr_map
;
6410 if (!dev
->vol
.migr_state
)
6413 if (migr_type(dev
) != MIGR_REBUILD
)
6416 migr_map
= get_imsm_map(dev
, MAP_1
);
6418 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6425 static int is_initializing(struct imsm_dev
*dev
)
6427 struct imsm_map
*migr_map
;
6429 if (!dev
->vol
.migr_state
)
6432 if (migr_type(dev
) != MIGR_INIT
)
6435 migr_map
= get_imsm_map(dev
, MAP_1
);
6437 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6444 static void update_recovery_start(struct intel_super
*super
,
6445 struct imsm_dev
*dev
,
6446 struct mdinfo
*array
)
6448 struct mdinfo
*rebuild
= NULL
;
6452 if (!is_rebuilding(dev
))
6455 /* Find the rebuild target, but punt on the dual rebuild case */
6456 for (d
= array
->devs
; d
; d
= d
->next
)
6457 if (d
->recovery_start
== 0) {
6464 /* (?) none of the disks are marked with
6465 * IMSM_ORD_REBUILD, so assume they are missing and the
6466 * disk_ord_tbl was not correctly updated
6468 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
6472 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6473 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6477 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6480 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6482 /* Given a container loaded by load_super_imsm_all,
6483 * extract information about all the arrays into
6485 * If 'subarray' is given, just extract info about that array.
6487 * For each imsm_dev create an mdinfo, fill it in,
6488 * then look for matching devices in super->disks
6489 * and create appropriate device mdinfo.
6491 struct intel_super
*super
= st
->sb
;
6492 struct imsm_super
*mpb
= super
->anchor
;
6493 struct mdinfo
*rest
= NULL
;
6497 int spare_disks
= 0;
6499 /* do not assemble arrays when not all attributes are supported */
6500 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6502 pr_err("Unsupported attributes in IMSM metadata."
6503 "Arrays activation is blocked.\n");
6506 /* check for bad blocks */
6507 if (imsm_bbm_log_size(super
->anchor
)) {
6508 pr_err("BBM log found in IMSM metadata."
6509 "Arrays activation is blocked.\n");
6514 /* count spare devices, not used in maps
6516 for (d
= super
->disks
; d
; d
= d
->next
)
6520 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6521 struct imsm_dev
*dev
;
6522 struct imsm_map
*map
;
6523 struct imsm_map
*map2
;
6524 struct mdinfo
*this;
6532 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6535 dev
= get_imsm_dev(super
, i
);
6536 map
= get_imsm_map(dev
, MAP_0
);
6537 map2
= get_imsm_map(dev
, MAP_1
);
6539 /* do not publish arrays that are in the middle of an
6540 * unsupported migration
6542 if (dev
->vol
.migr_state
&&
6543 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6544 pr_err("cannot assemble volume '%.16s':"
6545 " unsupported migration in progress\n",
6549 /* do not publish arrays that are not support by controller's
6553 this = xmalloc(sizeof(*this));
6555 super
->current_vol
= i
;
6556 getinfo_super_imsm_volume(st
, this, NULL
);
6559 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6560 /* mdadm does not support all metadata features- set the bit in all arrays state */
6561 if (!validate_geometry_imsm_orom(super
,
6562 get_imsm_raid_level(map
), /* RAID level */
6563 imsm_level_to_layout(get_imsm_raid_level(map
)),
6564 map
->num_members
, /* raid disks */
6565 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6567 pr_err("IMSM RAID geometry validation"
6568 " failed. Array %s activation is blocked.\n",
6570 this->array
.state
|=
6571 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6572 (1<<MD_SB_BLOCK_VOLUME
);
6576 /* if array has bad blocks, set suitable bit in all arrays state */
6578 this->array
.state
|=
6579 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6580 (1<<MD_SB_BLOCK_VOLUME
);
6582 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6583 unsigned long long recovery_start
;
6584 struct mdinfo
*info_d
;
6591 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6592 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6593 for (d
= super
->disks
; d
; d
= d
->next
)
6594 if (d
->index
== idx
)
6597 recovery_start
= MaxSector
;
6600 if (d
&& is_failed(&d
->disk
))
6602 if (ord
& IMSM_ORD_REBUILD
)
6606 * if we skip some disks the array will be assmebled degraded;
6607 * reset resync start to avoid a dirty-degraded
6608 * situation when performing the intial sync
6610 * FIXME handle dirty degraded
6612 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6613 this->resync_start
= MaxSector
;
6617 info_d
= xcalloc(1, sizeof(*info_d
));
6618 info_d
->next
= this->devs
;
6619 this->devs
= info_d
;
6621 info_d
->disk
.number
= d
->index
;
6622 info_d
->disk
.major
= d
->major
;
6623 info_d
->disk
.minor
= d
->minor
;
6624 info_d
->disk
.raid_disk
= slot
;
6625 info_d
->recovery_start
= recovery_start
;
6627 if (slot
< map2
->num_members
)
6628 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6630 this->array
.spare_disks
++;
6632 if (slot
< map
->num_members
)
6633 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6635 this->array
.spare_disks
++;
6637 if (info_d
->recovery_start
== MaxSector
)
6638 this->array
.working_disks
++;
6640 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6641 info_d
->data_offset
= pba_of_lba0(map
);
6642 info_d
->component_size
= blocks_per_member(map
);
6644 /* now that the disk list is up-to-date fixup recovery_start */
6645 update_recovery_start(super
, dev
, this);
6646 this->array
.spare_disks
+= spare_disks
;
6649 /* check for reshape */
6650 if (this->reshape_active
== 1)
6651 recover_backup_imsm(st
, this);
6660 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6661 int failed
, int look_in_map
)
6663 struct imsm_map
*map
;
6665 map
= get_imsm_map(dev
, look_in_map
);
6668 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6669 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6671 switch (get_imsm_raid_level(map
)) {
6673 return IMSM_T_STATE_FAILED
;
6676 if (failed
< map
->num_members
)
6677 return IMSM_T_STATE_DEGRADED
;
6679 return IMSM_T_STATE_FAILED
;
6684 * check to see if any mirrors have failed, otherwise we
6685 * are degraded. Even numbered slots are mirrored on
6689 /* gcc -Os complains that this is unused */
6690 int insync
= insync
;
6692 for (i
= 0; i
< map
->num_members
; i
++) {
6693 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6694 int idx
= ord_to_idx(ord
);
6695 struct imsm_disk
*disk
;
6697 /* reset the potential in-sync count on even-numbered
6698 * slots. num_copies is always 2 for imsm raid10
6703 disk
= get_imsm_disk(super
, idx
);
6704 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6707 /* no in-sync disks left in this mirror the
6711 return IMSM_T_STATE_FAILED
;
6714 return IMSM_T_STATE_DEGRADED
;
6718 return IMSM_T_STATE_DEGRADED
;
6720 return IMSM_T_STATE_FAILED
;
6726 return map
->map_state
;
6729 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6734 struct imsm_disk
*disk
;
6735 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6736 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6737 struct imsm_map
*map_for_loop
;
6742 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6743 * disks that are being rebuilt. New failures are recorded to
6744 * map[0]. So we look through all the disks we started with and
6745 * see if any failures are still present, or if any new ones
6749 if (prev
&& (map
->num_members
< prev
->num_members
))
6750 map_for_loop
= prev
;
6752 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6754 /* when MAP_X is passed both maps failures are counted
6757 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6758 (i
< prev
->num_members
)) {
6759 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6760 idx_1
= ord_to_idx(ord
);
6762 disk
= get_imsm_disk(super
, idx_1
);
6763 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6766 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6767 (i
< map
->num_members
)) {
6768 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6769 idx
= ord_to_idx(ord
);
6772 disk
= get_imsm_disk(super
, idx
);
6773 if (!disk
|| is_failed(disk
) ||
6774 ord
& IMSM_ORD_REBUILD
)
6784 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6787 struct intel_super
*super
= c
->sb
;
6788 struct imsm_super
*mpb
= super
->anchor
;
6790 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6791 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6792 __func__
, atoi(inst
));
6796 dprintf("imsm: open_new %s\n", inst
);
6797 a
->info
.container_member
= atoi(inst
);
6801 static int is_resyncing(struct imsm_dev
*dev
)
6803 struct imsm_map
*migr_map
;
6805 if (!dev
->vol
.migr_state
)
6808 if (migr_type(dev
) == MIGR_INIT
||
6809 migr_type(dev
) == MIGR_REPAIR
)
6812 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6815 migr_map
= get_imsm_map(dev
, MAP_1
);
6817 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6818 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6824 /* return true if we recorded new information */
6825 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6829 struct imsm_map
*map
;
6830 char buf
[MAX_RAID_SERIAL_LEN
+3];
6831 unsigned int len
, shift
= 0;
6833 /* new failures are always set in map[0] */
6834 map
= get_imsm_map(dev
, MAP_0
);
6836 slot
= get_imsm_disk_slot(map
, idx
);
6840 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6841 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6844 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6845 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6847 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6848 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6849 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6851 disk
->status
|= FAILED_DISK
;
6852 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6853 /* mark failures in second map if second map exists and this disk
6855 * This is valid for migration, initialization and rebuild
6857 if (dev
->vol
.migr_state
) {
6858 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6859 int slot2
= get_imsm_disk_slot(map2
, idx
);
6861 if ((slot2
< map2
->num_members
) &&
6863 set_imsm_ord_tbl_ent(map2
, slot2
,
6864 idx
| IMSM_ORD_REBUILD
);
6866 if (map
->failed_disk_num
== 0xff)
6867 map
->failed_disk_num
= slot
;
6871 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6873 mark_failure(dev
, disk
, idx
);
6875 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6878 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6879 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6882 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6886 if (!super
->missing
)
6889 dprintf("imsm: mark missing\n");
6890 /* end process for initialization and rebuild only
6892 if (is_gen_migration(dev
) == 0) {
6896 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6897 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6899 end_migration(dev
, super
, map_state
);
6901 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6902 mark_missing(dev
, &dl
->disk
, dl
->index
);
6903 super
->updates_pending
++;
6906 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
6909 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6910 unsigned long long array_blocks
;
6911 struct imsm_map
*map
;
6913 if (used_disks
== 0) {
6914 /* when problems occures
6915 * return current array_blocks value
6917 array_blocks
= __le32_to_cpu(dev
->size_high
);
6918 array_blocks
= array_blocks
<< 32;
6919 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6921 return array_blocks
;
6924 /* set array size in metadata
6926 if (new_size
<= 0) {
6927 /* OLCE size change is caused by added disks
6929 map
= get_imsm_map(dev
, MAP_0
);
6930 array_blocks
= blocks_per_member(map
) * used_disks
;
6932 /* Online Volume Size Change
6933 * Using available free space
6935 array_blocks
= new_size
;
6938 /* round array size down to closest MB
6940 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6941 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6942 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6944 return array_blocks
;
6947 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6949 static void imsm_progress_container_reshape(struct intel_super
*super
)
6951 /* if no device has a migr_state, but some device has a
6952 * different number of members than the previous device, start
6953 * changing the number of devices in this device to match
6956 struct imsm_super
*mpb
= super
->anchor
;
6957 int prev_disks
= -1;
6961 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6962 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6963 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6964 struct imsm_map
*map2
;
6965 int prev_num_members
;
6967 if (dev
->vol
.migr_state
)
6970 if (prev_disks
== -1)
6971 prev_disks
= map
->num_members
;
6972 if (prev_disks
== map
->num_members
)
6975 /* OK, this array needs to enter reshape mode.
6976 * i.e it needs a migr_state
6979 copy_map_size
= sizeof_imsm_map(map
);
6980 prev_num_members
= map
->num_members
;
6981 map
->num_members
= prev_disks
;
6982 dev
->vol
.migr_state
= 1;
6983 dev
->vol
.curr_migr_unit
= 0;
6984 set_migr_type(dev
, MIGR_GEN_MIGR
);
6985 for (i
= prev_num_members
;
6986 i
< map
->num_members
; i
++)
6987 set_imsm_ord_tbl_ent(map
, i
, i
);
6988 map2
= get_imsm_map(dev
, MAP_1
);
6989 /* Copy the current map */
6990 memcpy(map2
, map
, copy_map_size
);
6991 map2
->num_members
= prev_num_members
;
6993 imsm_set_array_size(dev
, -1);
6994 super
->clean_migration_record_by_mdmon
= 1;
6995 super
->updates_pending
++;
6999 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7000 * states are handled in imsm_set_disk() with one exception, when a
7001 * resync is stopped due to a new failure this routine will set the
7002 * 'degraded' state for the array.
7004 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7006 int inst
= a
->info
.container_member
;
7007 struct intel_super
*super
= a
->container
->sb
;
7008 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7009 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7010 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7011 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7012 __u32 blocks_per_unit
;
7014 if (dev
->vol
.migr_state
&&
7015 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7016 /* array state change is blocked due to reshape action
7018 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7019 * - finish the reshape (if last_checkpoint is big and action != reshape)
7020 * - update curr_migr_unit
7022 if (a
->curr_action
== reshape
) {
7023 /* still reshaping, maybe update curr_migr_unit */
7024 goto mark_checkpoint
;
7026 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7027 /* for some reason we aborted the reshape.
7029 * disable automatic metadata rollback
7030 * user action is required to recover process
7033 struct imsm_map
*map2
=
7034 get_imsm_map(dev
, MAP_1
);
7035 dev
->vol
.migr_state
= 0;
7036 set_migr_type(dev
, 0);
7037 dev
->vol
.curr_migr_unit
= 0;
7039 sizeof_imsm_map(map2
));
7040 super
->updates_pending
++;
7043 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7044 unsigned long long array_blocks
;
7048 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7049 if (used_disks
> 0) {
7051 blocks_per_member(map
) *
7053 /* round array size down to closest MB
7055 array_blocks
= (array_blocks
7056 >> SECT_PER_MB_SHIFT
)
7057 << SECT_PER_MB_SHIFT
;
7058 a
->info
.custom_array_size
= array_blocks
;
7059 /* encourage manager to update array
7063 a
->check_reshape
= 1;
7065 /* finalize online capacity expansion/reshape */
7066 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7068 mdi
->disk
.raid_disk
,
7071 imsm_progress_container_reshape(super
);
7076 /* before we activate this array handle any missing disks */
7077 if (consistent
== 2)
7078 handle_missing(super
, dev
);
7080 if (consistent
== 2 &&
7081 (!is_resync_complete(&a
->info
) ||
7082 map_state
!= IMSM_T_STATE_NORMAL
||
7083 dev
->vol
.migr_state
))
7086 if (is_resync_complete(&a
->info
)) {
7087 /* complete intialization / resync,
7088 * recovery and interrupted recovery is completed in
7091 if (is_resyncing(dev
)) {
7092 dprintf("imsm: mark resync done\n");
7093 end_migration(dev
, super
, map_state
);
7094 super
->updates_pending
++;
7095 a
->last_checkpoint
= 0;
7097 } else if ((!is_resyncing(dev
) && !failed
) &&
7098 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7099 /* mark the start of the init process if nothing is failed */
7100 dprintf("imsm: mark resync start\n");
7101 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7102 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7104 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7105 super
->updates_pending
++;
7109 /* skip checkpointing for general migration,
7110 * it is controlled in mdadm
7112 if (is_gen_migration(dev
))
7113 goto skip_mark_checkpoint
;
7115 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7116 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7117 if (blocks_per_unit
) {
7121 units
= a
->last_checkpoint
/ blocks_per_unit
;
7124 /* check that we did not overflow 32-bits, and that
7125 * curr_migr_unit needs updating
7127 if (units32
== units
&&
7129 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7130 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7131 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7132 super
->updates_pending
++;
7136 skip_mark_checkpoint
:
7137 /* mark dirty / clean */
7138 if (dev
->vol
.dirty
!= !consistent
) {
7139 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7144 super
->updates_pending
++;
7150 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7152 int inst
= a
->info
.container_member
;
7153 struct intel_super
*super
= a
->container
->sb
;
7154 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7155 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7156 struct imsm_disk
*disk
;
7158 int recovery_not_finished
= 0;
7163 if (n
> map
->num_members
)
7164 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
7165 n
, map
->num_members
- 1);
7170 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7172 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7173 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7175 /* check for new failures */
7176 if (state
& DS_FAULTY
) {
7177 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7178 super
->updates_pending
++;
7181 /* check if in_sync */
7182 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7183 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7185 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7186 super
->updates_pending
++;
7189 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7190 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7192 /* check if recovery complete, newly degraded, or failed */
7193 dprintf("imsm: Detected transition to state ");
7194 switch (map_state
) {
7195 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7196 dprintf("normal: ");
7197 if (is_rebuilding(dev
)) {
7198 dprintf("while rebuilding");
7199 /* check if recovery is really finished */
7200 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7201 if (mdi
->recovery_start
!= MaxSector
) {
7202 recovery_not_finished
= 1;
7205 if (recovery_not_finished
) {
7206 dprintf("\nimsm: Rebuild has not finished yet, "
7207 "state not changed");
7208 if (a
->last_checkpoint
< mdi
->recovery_start
) {
7209 a
->last_checkpoint
= mdi
->recovery_start
;
7210 super
->updates_pending
++;
7214 end_migration(dev
, super
, map_state
);
7215 map
= get_imsm_map(dev
, MAP_0
);
7216 map
->failed_disk_num
= ~0;
7217 super
->updates_pending
++;
7218 a
->last_checkpoint
= 0;
7221 if (is_gen_migration(dev
)) {
7222 dprintf("while general migration");
7223 if (a
->last_checkpoint
>= a
->info
.component_size
)
7224 end_migration(dev
, super
, map_state
);
7226 map
->map_state
= map_state
;
7227 map
= get_imsm_map(dev
, MAP_0
);
7228 map
->failed_disk_num
= ~0;
7229 super
->updates_pending
++;
7233 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7234 dprintf("degraded: ");
7235 if ((map
->map_state
!= map_state
) &&
7236 !dev
->vol
.migr_state
) {
7237 dprintf("mark degraded");
7238 map
->map_state
= map_state
;
7239 super
->updates_pending
++;
7240 a
->last_checkpoint
= 0;
7243 if (is_rebuilding(dev
)) {
7244 dprintf("while rebuilding.");
7245 if (map
->map_state
!= map_state
) {
7246 dprintf(" Map state change");
7247 end_migration(dev
, super
, map_state
);
7248 super
->updates_pending
++;
7252 if (is_gen_migration(dev
)) {
7253 dprintf("while general migration");
7254 if (a
->last_checkpoint
>= a
->info
.component_size
)
7255 end_migration(dev
, super
, map_state
);
7257 map
->map_state
= map_state
;
7258 manage_second_map(super
, dev
);
7260 super
->updates_pending
++;
7263 if (is_initializing(dev
)) {
7264 dprintf("while initialization.");
7265 map
->map_state
= map_state
;
7266 super
->updates_pending
++;
7270 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7271 dprintf("failed: ");
7272 if (is_gen_migration(dev
)) {
7273 dprintf("while general migration");
7274 map
->map_state
= map_state
;
7275 super
->updates_pending
++;
7278 if (map
->map_state
!= map_state
) {
7279 dprintf("mark failed");
7280 end_migration(dev
, super
, map_state
);
7281 super
->updates_pending
++;
7282 a
->last_checkpoint
= 0;
7287 dprintf("state %i\n", map_state
);
7293 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7296 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7297 unsigned long long dsize
;
7298 unsigned long long sectors
;
7300 get_dev_size(fd
, NULL
, &dsize
);
7302 if (mpb_size
> 512) {
7303 /* -1 to account for anchor */
7304 sectors
= mpb_sectors(mpb
) - 1;
7306 /* write the extended mpb to the sectors preceeding the anchor */
7307 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7310 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7315 /* first block is stored on second to last sector of the disk */
7316 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7319 if (write(fd
, buf
, 512) != 512)
7325 static void imsm_sync_metadata(struct supertype
*container
)
7327 struct intel_super
*super
= container
->sb
;
7329 dprintf("sync metadata: %d\n", super
->updates_pending
);
7330 if (!super
->updates_pending
)
7333 write_super_imsm(container
, 0);
7335 super
->updates_pending
= 0;
7338 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7340 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7341 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7344 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7348 if (dl
&& is_failed(&dl
->disk
))
7352 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
7357 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7358 struct active_array
*a
, int activate_new
,
7359 struct mdinfo
*additional_test_list
)
7361 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7362 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7363 struct imsm_super
*mpb
= super
->anchor
;
7364 struct imsm_map
*map
;
7365 unsigned long long pos
;
7370 __u32 array_start
= 0;
7371 __u32 array_end
= 0;
7373 struct mdinfo
*test_list
;
7375 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7376 /* If in this array, skip */
7377 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7378 if (d
->state_fd
>= 0 &&
7379 d
->disk
.major
== dl
->major
&&
7380 d
->disk
.minor
== dl
->minor
) {
7381 dprintf("%x:%x already in array\n",
7382 dl
->major
, dl
->minor
);
7387 test_list
= additional_test_list
;
7389 if (test_list
->disk
.major
== dl
->major
&&
7390 test_list
->disk
.minor
== dl
->minor
) {
7391 dprintf("%x:%x already in additional test list\n",
7392 dl
->major
, dl
->minor
);
7395 test_list
= test_list
->next
;
7400 /* skip in use or failed drives */
7401 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7403 dprintf("%x:%x status (failed: %d index: %d)\n",
7404 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7408 /* skip pure spares when we are looking for partially
7409 * assimilated drives
7411 if (dl
->index
== -1 && !activate_new
)
7414 /* Does this unused device have the requisite free space?
7415 * It needs to be able to cover all member volumes
7417 ex
= get_extents(super
, dl
);
7419 dprintf("cannot get extents\n");
7422 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7423 dev
= get_imsm_dev(super
, i
);
7424 map
= get_imsm_map(dev
, MAP_0
);
7426 /* check if this disk is already a member of
7429 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7435 array_start
= pba_of_lba0(map
);
7436 array_end
= array_start
+
7437 blocks_per_member(map
) - 1;
7440 /* check that we can start at pba_of_lba0 with
7441 * blocks_per_member of space
7443 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7447 pos
= ex
[j
].start
+ ex
[j
].size
;
7449 } while (ex
[j
-1].size
);
7456 if (i
< mpb
->num_raid_devs
) {
7457 dprintf("%x:%x does not have %u to %u available\n",
7458 dl
->major
, dl
->minor
, array_start
, array_end
);
7469 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7471 struct imsm_dev
*dev2
;
7472 struct imsm_map
*map
;
7478 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7480 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7481 if (state
== IMSM_T_STATE_FAILED
) {
7482 map
= get_imsm_map(dev2
, MAP_0
);
7485 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7487 * Check if failed disks are deleted from intel
7488 * disk list or are marked to be deleted
7490 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7491 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7493 * Do not rebuild the array if failed disks
7494 * from failed sub-array are not removed from
7498 is_failed(&idisk
->disk
) &&
7499 (idisk
->action
!= DISK_REMOVE
))
7507 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7508 struct metadata_update
**updates
)
7511 * Find a device with unused free space and use it to replace a
7512 * failed/vacant region in an array. We replace failed regions one a
7513 * array at a time. The result is that a new spare disk will be added
7514 * to the first failed array and after the monitor has finished
7515 * propagating failures the remainder will be consumed.
7517 * FIXME add a capability for mdmon to request spares from another
7521 struct intel_super
*super
= a
->container
->sb
;
7522 int inst
= a
->info
.container_member
;
7523 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7524 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7525 int failed
= a
->info
.array
.raid_disks
;
7526 struct mdinfo
*rv
= NULL
;
7529 struct metadata_update
*mu
;
7531 struct imsm_update_activate_spare
*u
;
7536 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7537 if ((d
->curr_state
& DS_FAULTY
) &&
7539 /* wait for Removal to happen */
7541 if (d
->state_fd
>= 0)
7545 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7546 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7548 if (imsm_reshape_blocks_arrays_changes(super
))
7551 /* Cannot activate another spare if rebuild is in progress already
7553 if (is_rebuilding(dev
)) {
7554 dprintf("imsm: No spare activation allowed. "
7555 "Rebuild in progress already.\n");
7559 if (a
->info
.array
.level
== 4)
7560 /* No repair for takeovered array
7561 * imsm doesn't support raid4
7565 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7566 IMSM_T_STATE_DEGRADED
)
7570 * If there are any failed disks check state of the other volume.
7571 * Block rebuild if the another one is failed until failed disks
7572 * are removed from container.
7575 dprintf("found failed disks in %.*s, check if there another"
7576 "failed sub-array.\n",
7577 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7578 /* check if states of the other volumes allow for rebuild */
7579 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7581 allowed
= imsm_rebuild_allowed(a
->container
,
7589 /* For each slot, if it is not working, find a spare */
7590 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7591 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7592 if (d
->disk
.raid_disk
== i
)
7594 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7595 if (d
&& (d
->state_fd
>= 0))
7599 * OK, this device needs recovery. Try to re-add the
7600 * previous occupant of this slot, if this fails see if
7601 * we can continue the assimilation of a spare that was
7602 * partially assimilated, finally try to activate a new
7605 dl
= imsm_readd(super
, i
, a
);
7607 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7609 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7613 /* found a usable disk with enough space */
7614 di
= xcalloc(1, sizeof(*di
));
7616 /* dl->index will be -1 in the case we are activating a
7617 * pristine spare. imsm_process_update() will create a
7618 * new index in this case. Once a disk is found to be
7619 * failed in all member arrays it is kicked from the
7622 di
->disk
.number
= dl
->index
;
7624 /* (ab)use di->devs to store a pointer to the device
7627 di
->devs
= (struct mdinfo
*) dl
;
7629 di
->disk
.raid_disk
= i
;
7630 di
->disk
.major
= dl
->major
;
7631 di
->disk
.minor
= dl
->minor
;
7633 di
->recovery_start
= 0;
7634 di
->data_offset
= pba_of_lba0(map
);
7635 di
->component_size
= a
->info
.component_size
;
7636 di
->container_member
= inst
;
7637 super
->random
= random32();
7641 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7642 i
, di
->data_offset
);
7646 /* No spares found */
7648 /* Now 'rv' has a list of devices to return.
7649 * Create a metadata_update record to update the
7650 * disk_ord_tbl for the array
7652 mu
= xmalloc(sizeof(*mu
));
7653 mu
->buf
= xcalloc(num_spares
,
7654 sizeof(struct imsm_update_activate_spare
));
7656 mu
->space_list
= NULL
;
7657 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7658 mu
->next
= *updates
;
7659 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7661 for (di
= rv
; di
; di
= di
->next
) {
7662 u
->type
= update_activate_spare
;
7663 u
->dl
= (struct dl
*) di
->devs
;
7665 u
->slot
= di
->disk
.raid_disk
;
7676 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7678 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7679 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7680 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7681 struct disk_info
*inf
= get_disk_info(u
);
7682 struct imsm_disk
*disk
;
7686 for (i
= 0; i
< map
->num_members
; i
++) {
7687 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7688 for (j
= 0; j
< new_map
->num_members
; j
++)
7689 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7697 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7699 struct dl
*dl
= NULL
;
7700 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7701 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7706 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7708 struct dl
*prev
= NULL
;
7712 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7713 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7716 prev
->next
= dl
->next
;
7718 super
->disks
= dl
->next
;
7720 __free_imsm_disk(dl
);
7721 dprintf("%s: removed %x:%x\n",
7722 __func__
, major
, minor
);
7730 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7732 static int add_remove_disk_update(struct intel_super
*super
)
7734 int check_degraded
= 0;
7735 struct dl
*disk
= NULL
;
7736 /* add/remove some spares to/from the metadata/contrainer */
7737 while (super
->disk_mgmt_list
) {
7738 struct dl
*disk_cfg
;
7740 disk_cfg
= super
->disk_mgmt_list
;
7741 super
->disk_mgmt_list
= disk_cfg
->next
;
7742 disk_cfg
->next
= NULL
;
7744 if (disk_cfg
->action
== DISK_ADD
) {
7745 disk_cfg
->next
= super
->disks
;
7746 super
->disks
= disk_cfg
;
7748 dprintf("%s: added %x:%x\n",
7749 __func__
, disk_cfg
->major
,
7751 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7752 dprintf("Disk remove action processed: %x.%x\n",
7753 disk_cfg
->major
, disk_cfg
->minor
);
7754 disk
= get_disk_super(super
,
7758 /* store action status */
7759 disk
->action
= DISK_REMOVE
;
7760 /* remove spare disks only */
7761 if (disk
->index
== -1) {
7762 remove_disk_super(super
,
7767 /* release allocate disk structure */
7768 __free_imsm_disk(disk_cfg
);
7771 return check_degraded
;
7775 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7776 struct intel_super
*super
,
7779 struct intel_dev
*id
;
7780 void **tofree
= NULL
;
7783 dprintf("apply_reshape_migration_update()\n");
7784 if ((u
->subdev
< 0) ||
7786 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7789 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7790 dprintf("imsm: Error: Memory is not allocated\n");
7794 for (id
= super
->devlist
; id
; id
= id
->next
) {
7795 if (id
->index
== (unsigned)u
->subdev
) {
7796 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7797 struct imsm_map
*map
;
7798 struct imsm_dev
*new_dev
=
7799 (struct imsm_dev
*)*space_list
;
7800 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7802 struct dl
*new_disk
;
7804 if (new_dev
== NULL
)
7806 *space_list
= **space_list
;
7807 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7808 map
= get_imsm_map(new_dev
, MAP_0
);
7810 dprintf("imsm: Error: migration in progress");
7814 to_state
= map
->map_state
;
7815 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7817 /* this should not happen */
7818 if (u
->new_disks
[0] < 0) {
7819 map
->failed_disk_num
=
7820 map
->num_members
- 1;
7821 to_state
= IMSM_T_STATE_DEGRADED
;
7823 to_state
= IMSM_T_STATE_NORMAL
;
7825 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7826 if (u
->new_level
> -1)
7827 map
->raid_level
= u
->new_level
;
7828 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7829 if ((u
->new_level
== 5) &&
7830 (migr_map
->raid_level
== 0)) {
7831 int ord
= map
->num_members
- 1;
7832 migr_map
->num_members
--;
7833 if (u
->new_disks
[0] < 0)
7834 ord
|= IMSM_ORD_REBUILD
;
7835 set_imsm_ord_tbl_ent(map
,
7836 map
->num_members
- 1,
7840 tofree
= (void **)dev
;
7842 /* update chunk size
7844 if (u
->new_chunksize
> 0)
7845 map
->blocks_per_strip
=
7846 __cpu_to_le16(u
->new_chunksize
* 2);
7850 if ((u
->new_level
!= 5) ||
7851 (migr_map
->raid_level
!= 0) ||
7852 (migr_map
->raid_level
== map
->raid_level
))
7855 if (u
->new_disks
[0] >= 0) {
7858 new_disk
= get_disk_super(super
,
7859 major(u
->new_disks
[0]),
7860 minor(u
->new_disks
[0]));
7861 dprintf("imsm: new disk for reshape is: %i:%i "
7862 "(%p, index = %i)\n",
7863 major(u
->new_disks
[0]),
7864 minor(u
->new_disks
[0]),
7865 new_disk
, new_disk
->index
);
7866 if (new_disk
== NULL
)
7867 goto error_disk_add
;
7869 new_disk
->index
= map
->num_members
- 1;
7870 /* slot to fill in autolayout
7872 new_disk
->raiddisk
= new_disk
->index
;
7873 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7874 new_disk
->disk
.status
&= ~SPARE_DISK
;
7876 goto error_disk_add
;
7879 *tofree
= *space_list
;
7880 /* calculate new size
7882 imsm_set_array_size(new_dev
, -1);
7889 *space_list
= tofree
;
7893 dprintf("Error: imsm: Cannot find disk.\n");
7897 static int apply_size_change_update(struct imsm_update_size_change
*u
,
7898 struct intel_super
*super
)
7900 struct intel_dev
*id
;
7903 dprintf("apply_size_change_update()\n");
7904 if ((u
->subdev
< 0) ||
7906 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7910 for (id
= super
->devlist
; id
; id
= id
->next
) {
7911 if (id
->index
== (unsigned)u
->subdev
) {
7912 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7913 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7914 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7915 unsigned long long blocks_per_member
;
7917 /* calculate new size
7919 blocks_per_member
= u
->new_size
/ used_disks
;
7920 dprintf("imsm: apply_size_change_update(size: %llu, "
7921 "blocks per member: %llu)\n",
7922 u
->new_size
, blocks_per_member
);
7923 set_blocks_per_member(map
, blocks_per_member
);
7924 imsm_set_array_size(dev
, u
->new_size
);
7935 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7936 struct intel_super
*super
,
7937 struct active_array
*active_array
)
7939 struct imsm_super
*mpb
= super
->anchor
;
7940 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7941 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7942 struct imsm_map
*migr_map
;
7943 struct active_array
*a
;
7944 struct imsm_disk
*disk
;
7951 int second_map_created
= 0;
7953 for (; u
; u
= u
->next
) {
7954 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
7959 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7964 fprintf(stderr
, "error: imsm_activate_spare passed "
7965 "an unknown disk (index: %d)\n",
7970 /* count failures (excluding rebuilds and the victim)
7971 * to determine map[0] state
7974 for (i
= 0; i
< map
->num_members
; i
++) {
7977 disk
= get_imsm_disk(super
,
7978 get_imsm_disk_idx(dev
, i
, MAP_X
));
7979 if (!disk
|| is_failed(disk
))
7983 /* adding a pristine spare, assign a new index */
7984 if (dl
->index
< 0) {
7985 dl
->index
= super
->anchor
->num_disks
;
7986 super
->anchor
->num_disks
++;
7989 disk
->status
|= CONFIGURED_DISK
;
7990 disk
->status
&= ~SPARE_DISK
;
7993 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7994 if (!second_map_created
) {
7995 second_map_created
= 1;
7996 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7997 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7999 map
->map_state
= to_state
;
8000 migr_map
= get_imsm_map(dev
, MAP_1
);
8001 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8002 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8003 dl
->index
| IMSM_ORD_REBUILD
);
8005 /* update the family_num to mark a new container
8006 * generation, being careful to record the existing
8007 * family_num in orig_family_num to clean up after
8008 * earlier mdadm versions that neglected to set it.
8010 if (mpb
->orig_family_num
== 0)
8011 mpb
->orig_family_num
= mpb
->family_num
;
8012 mpb
->family_num
+= super
->random
;
8014 /* count arrays using the victim in the metadata */
8016 for (a
= active_array
; a
; a
= a
->next
) {
8017 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8018 map
= get_imsm_map(dev
, MAP_0
);
8020 if (get_imsm_disk_slot(map
, victim
) >= 0)
8024 /* delete the victim if it is no longer being
8030 /* We know that 'manager' isn't touching anything,
8031 * so it is safe to delete
8033 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8034 if ((*dlp
)->index
== victim
)
8037 /* victim may be on the missing list */
8039 for (dlp
= &super
->missing
; *dlp
;
8040 dlp
= &(*dlp
)->next
)
8041 if ((*dlp
)->index
== victim
)
8043 imsm_delete(super
, dlp
, victim
);
8050 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8051 struct intel_super
*super
,
8054 struct dl
*new_disk
;
8055 struct intel_dev
*id
;
8057 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8058 int disk_count
= u
->old_raid_disks
;
8059 void **tofree
= NULL
;
8060 int devices_to_reshape
= 1;
8061 struct imsm_super
*mpb
= super
->anchor
;
8063 unsigned int dev_id
;
8065 dprintf("imsm: apply_reshape_container_disks_update()\n");
8067 /* enable spares to use in array */
8068 for (i
= 0; i
< delta_disks
; i
++) {
8069 new_disk
= get_disk_super(super
,
8070 major(u
->new_disks
[i
]),
8071 minor(u
->new_disks
[i
]));
8072 dprintf("imsm: new disk for reshape is: %i:%i "
8073 "(%p, index = %i)\n",
8074 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8075 new_disk
, new_disk
->index
);
8076 if ((new_disk
== NULL
) ||
8077 ((new_disk
->index
>= 0) &&
8078 (new_disk
->index
< u
->old_raid_disks
)))
8079 goto update_reshape_exit
;
8080 new_disk
->index
= disk_count
++;
8081 /* slot to fill in autolayout
8083 new_disk
->raiddisk
= new_disk
->index
;
8084 new_disk
->disk
.status
|=
8086 new_disk
->disk
.status
&= ~SPARE_DISK
;
8089 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8090 mpb
->num_raid_devs
);
8091 /* manage changes in volume
8093 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8094 void **sp
= *space_list
;
8095 struct imsm_dev
*newdev
;
8096 struct imsm_map
*newmap
, *oldmap
;
8098 for (id
= super
->devlist
; id
; id
= id
->next
) {
8099 if (id
->index
== dev_id
)
8108 /* Copy the dev, but not (all of) the map */
8109 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8110 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8111 newmap
= get_imsm_map(newdev
, MAP_0
);
8112 /* Copy the current map */
8113 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8114 /* update one device only
8116 if (devices_to_reshape
) {
8117 dprintf("imsm: modifying subdev: %i\n",
8119 devices_to_reshape
--;
8120 newdev
->vol
.migr_state
= 1;
8121 newdev
->vol
.curr_migr_unit
= 0;
8122 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8123 newmap
->num_members
= u
->new_raid_disks
;
8124 for (i
= 0; i
< delta_disks
; i
++) {
8125 set_imsm_ord_tbl_ent(newmap
,
8126 u
->old_raid_disks
+ i
,
8127 u
->old_raid_disks
+ i
);
8129 /* New map is correct, now need to save old map
8131 newmap
= get_imsm_map(newdev
, MAP_1
);
8132 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8134 imsm_set_array_size(newdev
, -1);
8137 sp
= (void **)id
->dev
;
8142 /* Clear migration record */
8143 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8146 *space_list
= tofree
;
8149 update_reshape_exit
:
8154 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8155 struct intel_super
*super
,
8158 struct imsm_dev
*dev
= NULL
;
8159 struct intel_dev
*dv
;
8160 struct imsm_dev
*dev_new
;
8161 struct imsm_map
*map
;
8165 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8166 if (dv
->index
== (unsigned int)u
->subarray
) {
8174 map
= get_imsm_map(dev
, MAP_0
);
8176 if (u
->direction
== R10_TO_R0
) {
8177 /* Number of failed disks must be half of initial disk number */
8178 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8179 (map
->num_members
/ 2))
8182 /* iterate through devices to mark removed disks as spare */
8183 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8184 if (dm
->disk
.status
& FAILED_DISK
) {
8185 int idx
= dm
->index
;
8186 /* update indexes on the disk list */
8187 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8188 the index values will end up being correct.... NB */
8189 for (du
= super
->disks
; du
; du
= du
->next
)
8190 if (du
->index
> idx
)
8192 /* mark as spare disk */
8197 map
->num_members
= map
->num_members
/ 2;
8198 map
->map_state
= IMSM_T_STATE_NORMAL
;
8199 map
->num_domains
= 1;
8200 map
->raid_level
= 0;
8201 map
->failed_disk_num
= -1;
8204 if (u
->direction
== R0_TO_R10
) {
8206 /* update slots in current disk list */
8207 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8211 /* create new *missing* disks */
8212 for (i
= 0; i
< map
->num_members
; i
++) {
8213 space
= *space_list
;
8216 *space_list
= *space
;
8218 memcpy(du
, super
->disks
, sizeof(*du
));
8222 du
->index
= (i
* 2) + 1;
8223 sprintf((char *)du
->disk
.serial
,
8224 " MISSING_%d", du
->index
);
8225 sprintf((char *)du
->serial
,
8226 "MISSING_%d", du
->index
);
8227 du
->next
= super
->missing
;
8228 super
->missing
= du
;
8230 /* create new dev and map */
8231 space
= *space_list
;
8234 *space_list
= *space
;
8235 dev_new
= (void *)space
;
8236 memcpy(dev_new
, dev
, sizeof(*dev
));
8237 /* update new map */
8238 map
= get_imsm_map(dev_new
, MAP_0
);
8239 map
->num_members
= map
->num_members
* 2;
8240 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8241 map
->num_domains
= 2;
8242 map
->raid_level
= 1;
8243 /* replace dev<->dev_new */
8246 /* update disk order table */
8247 for (du
= super
->disks
; du
; du
= du
->next
)
8249 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8250 for (du
= super
->missing
; du
; du
= du
->next
)
8251 if (du
->index
>= 0) {
8252 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8253 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8259 static void imsm_process_update(struct supertype
*st
,
8260 struct metadata_update
*update
)
8263 * crack open the metadata_update envelope to find the update record
8264 * update can be one of:
8265 * update_reshape_container_disks - all the arrays in the container
8266 * are being reshaped to have more devices. We need to mark
8267 * the arrays for general migration and convert selected spares
8268 * into active devices.
8269 * update_activate_spare - a spare device has replaced a failed
8270 * device in an array, update the disk_ord_tbl. If this disk is
8271 * present in all member arrays then also clear the SPARE_DISK
8273 * update_create_array
8275 * update_rename_array
8276 * update_add_remove_disk
8278 struct intel_super
*super
= st
->sb
;
8279 struct imsm_super
*mpb
;
8280 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8282 /* update requires a larger buf but the allocation failed */
8283 if (super
->next_len
&& !super
->next_buf
) {
8284 super
->next_len
= 0;
8288 if (super
->next_buf
) {
8289 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8291 super
->len
= super
->next_len
;
8292 super
->buf
= super
->next_buf
;
8294 super
->next_len
= 0;
8295 super
->next_buf
= NULL
;
8298 mpb
= super
->anchor
;
8301 case update_general_migration_checkpoint
: {
8302 struct intel_dev
*id
;
8303 struct imsm_update_general_migration_checkpoint
*u
=
8304 (void *)update
->buf
;
8306 dprintf("imsm: process_update() "
8307 "for update_general_migration_checkpoint called\n");
8309 /* find device under general migration */
8310 for (id
= super
->devlist
; id
; id
= id
->next
) {
8311 if (is_gen_migration(id
->dev
)) {
8312 id
->dev
->vol
.curr_migr_unit
=
8313 __cpu_to_le32(u
->curr_migr_unit
);
8314 super
->updates_pending
++;
8319 case update_takeover
: {
8320 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8321 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8322 imsm_update_version_info(super
);
8323 super
->updates_pending
++;
8328 case update_reshape_container_disks
: {
8329 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8330 if (apply_reshape_container_disks_update(
8331 u
, super
, &update
->space_list
))
8332 super
->updates_pending
++;
8335 case update_reshape_migration
: {
8336 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8337 if (apply_reshape_migration_update(
8338 u
, super
, &update
->space_list
))
8339 super
->updates_pending
++;
8342 case update_size_change
: {
8343 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8344 if (apply_size_change_update(u
, super
))
8345 super
->updates_pending
++;
8348 case update_activate_spare
: {
8349 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8350 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8351 super
->updates_pending
++;
8354 case update_create_array
: {
8355 /* someone wants to create a new array, we need to be aware of
8356 * a few races/collisions:
8357 * 1/ 'Create' called by two separate instances of mdadm
8358 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8359 * devices that have since been assimilated via
8361 * In the event this update can not be carried out mdadm will
8362 * (FIX ME) notice that its update did not take hold.
8364 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8365 struct intel_dev
*dv
;
8366 struct imsm_dev
*dev
;
8367 struct imsm_map
*map
, *new_map
;
8368 unsigned long long start
, end
;
8369 unsigned long long new_start
, new_end
;
8371 struct disk_info
*inf
;
8374 /* handle racing creates: first come first serve */
8375 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8376 dprintf("%s: subarray %d already defined\n",
8377 __func__
, u
->dev_idx
);
8381 /* check update is next in sequence */
8382 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8383 dprintf("%s: can not create array %d expected index %d\n",
8384 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
8388 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8389 new_start
= pba_of_lba0(new_map
);
8390 new_end
= new_start
+ blocks_per_member(new_map
);
8391 inf
= get_disk_info(u
);
8393 /* handle activate_spare versus create race:
8394 * check to make sure that overlapping arrays do not include
8397 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8398 dev
= get_imsm_dev(super
, i
);
8399 map
= get_imsm_map(dev
, MAP_0
);
8400 start
= pba_of_lba0(map
);
8401 end
= start
+ blocks_per_member(map
);
8402 if ((new_start
>= start
&& new_start
<= end
) ||
8403 (start
>= new_start
&& start
<= new_end
))
8408 if (disks_overlap(super
, i
, u
)) {
8409 dprintf("%s: arrays overlap\n", __func__
);
8414 /* check that prepare update was successful */
8415 if (!update
->space
) {
8416 dprintf("%s: prepare update failed\n", __func__
);
8420 /* check that all disks are still active before committing
8421 * changes. FIXME: could we instead handle this by creating a
8422 * degraded array? That's probably not what the user expects,
8423 * so better to drop this update on the floor.
8425 for (i
= 0; i
< new_map
->num_members
; i
++) {
8426 dl
= serial_to_dl(inf
[i
].serial
, super
);
8428 dprintf("%s: disk disappeared\n", __func__
);
8433 super
->updates_pending
++;
8435 /* convert spares to members and fixup ord_tbl */
8436 for (i
= 0; i
< new_map
->num_members
; i
++) {
8437 dl
= serial_to_dl(inf
[i
].serial
, super
);
8438 if (dl
->index
== -1) {
8439 dl
->index
= mpb
->num_disks
;
8441 dl
->disk
.status
|= CONFIGURED_DISK
;
8442 dl
->disk
.status
&= ~SPARE_DISK
;
8444 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8449 update
->space
= NULL
;
8450 imsm_copy_dev(dev
, &u
->dev
);
8451 dv
->index
= u
->dev_idx
;
8452 dv
->next
= super
->devlist
;
8453 super
->devlist
= dv
;
8454 mpb
->num_raid_devs
++;
8456 imsm_update_version_info(super
);
8459 /* mdmon knows how to release update->space, but not
8460 * ((struct intel_dev *) update->space)->dev
8462 if (update
->space
) {
8468 case update_kill_array
: {
8469 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8470 int victim
= u
->dev_idx
;
8471 struct active_array
*a
;
8472 struct intel_dev
**dp
;
8473 struct imsm_dev
*dev
;
8475 /* sanity check that we are not affecting the uuid of
8476 * active arrays, or deleting an active array
8478 * FIXME when immutable ids are available, but note that
8479 * we'll also need to fixup the invalidated/active
8480 * subarray indexes in mdstat
8482 for (a
= st
->arrays
; a
; a
= a
->next
)
8483 if (a
->info
.container_member
>= victim
)
8485 /* by definition if mdmon is running at least one array
8486 * is active in the container, so checking
8487 * mpb->num_raid_devs is just extra paranoia
8489 dev
= get_imsm_dev(super
, victim
);
8490 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8491 dprintf("failed to delete subarray-%d\n", victim
);
8495 for (dp
= &super
->devlist
; *dp
;)
8496 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8499 if ((*dp
)->index
> (unsigned)victim
)
8503 mpb
->num_raid_devs
--;
8504 super
->updates_pending
++;
8507 case update_rename_array
: {
8508 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8509 char name
[MAX_RAID_SERIAL_LEN
+1];
8510 int target
= u
->dev_idx
;
8511 struct active_array
*a
;
8512 struct imsm_dev
*dev
;
8514 /* sanity check that we are not affecting the uuid of
8517 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8518 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8519 for (a
= st
->arrays
; a
; a
= a
->next
)
8520 if (a
->info
.container_member
== target
)
8522 dev
= get_imsm_dev(super
, u
->dev_idx
);
8523 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8524 dprintf("failed to rename subarray-%d\n", target
);
8528 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8529 super
->updates_pending
++;
8532 case update_add_remove_disk
: {
8533 /* we may be able to repair some arrays if disks are
8534 * being added, check teh status of add_remove_disk
8535 * if discs has been added.
8537 if (add_remove_disk_update(super
)) {
8538 struct active_array
*a
;
8540 super
->updates_pending
++;
8541 for (a
= st
->arrays
; a
; a
= a
->next
)
8542 a
->check_degraded
= 1;
8547 fprintf(stderr
, "error: unsuported process update type:"
8548 "(type: %d)\n", type
);
8552 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8554 static void imsm_prepare_update(struct supertype
*st
,
8555 struct metadata_update
*update
)
8558 * Allocate space to hold new disk entries, raid-device entries or a new
8559 * mpb if necessary. The manager synchronously waits for updates to
8560 * complete in the monitor, so new mpb buffers allocated here can be
8561 * integrated by the monitor thread without worrying about live pointers
8562 * in the manager thread.
8564 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8565 struct intel_super
*super
= st
->sb
;
8566 struct imsm_super
*mpb
= super
->anchor
;
8571 case update_general_migration_checkpoint
:
8572 dprintf("imsm: prepare_update() "
8573 "for update_general_migration_checkpoint called\n");
8575 case update_takeover
: {
8576 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8577 if (u
->direction
== R0_TO_R10
) {
8578 void **tail
= (void **)&update
->space_list
;
8579 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8580 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8581 int num_members
= map
->num_members
;
8584 /* allocate memory for added disks */
8585 for (i
= 0; i
< num_members
; i
++) {
8586 size
= sizeof(struct dl
);
8587 space
= xmalloc(size
);
8592 /* allocate memory for new device */
8593 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8594 (num_members
* sizeof(__u32
));
8595 space
= xmalloc(size
);
8599 len
= disks_to_mpb_size(num_members
* 2);
8604 case update_reshape_container_disks
: {
8605 /* Every raid device in the container is about to
8606 * gain some more devices, and we will enter a
8608 * So each 'imsm_map' will be bigger, and the imsm_vol
8609 * will now hold 2 of them.
8610 * Thus we need new 'struct imsm_dev' allocations sized
8611 * as sizeof_imsm_dev but with more devices in both maps.
8613 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8614 struct intel_dev
*dl
;
8615 void **space_tail
= (void**)&update
->space_list
;
8617 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8619 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8620 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8622 if (u
->new_raid_disks
> u
->old_raid_disks
)
8623 size
+= sizeof(__u32
)*2*
8624 (u
->new_raid_disks
- u
->old_raid_disks
);
8631 len
= disks_to_mpb_size(u
->new_raid_disks
);
8632 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8635 case update_reshape_migration
: {
8636 /* for migration level 0->5 we need to add disks
8637 * so the same as for container operation we will copy
8638 * device to the bigger location.
8639 * in memory prepared device and new disk area are prepared
8640 * for usage in process update
8642 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8643 struct intel_dev
*id
;
8644 void **space_tail
= (void **)&update
->space_list
;
8647 int current_level
= -1;
8649 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8651 /* add space for bigger array in update
8653 for (id
= super
->devlist
; id
; id
= id
->next
) {
8654 if (id
->index
== (unsigned)u
->subdev
) {
8655 size
= sizeof_imsm_dev(id
->dev
, 1);
8656 if (u
->new_raid_disks
> u
->old_raid_disks
)
8657 size
+= sizeof(__u32
)*2*
8658 (u
->new_raid_disks
- u
->old_raid_disks
);
8666 if (update
->space_list
== NULL
)
8669 /* add space for disk in update
8671 size
= sizeof(struct dl
);
8677 /* add spare device to update
8679 for (id
= super
->devlist
; id
; id
= id
->next
)
8680 if (id
->index
== (unsigned)u
->subdev
) {
8681 struct imsm_dev
*dev
;
8682 struct imsm_map
*map
;
8684 dev
= get_imsm_dev(super
, u
->subdev
);
8685 map
= get_imsm_map(dev
, MAP_0
);
8686 current_level
= map
->raid_level
;
8689 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8690 struct mdinfo
*spares
;
8692 spares
= get_spares_for_grow(st
);
8700 makedev(dev
->disk
.major
,
8702 dl
= get_disk_super(super
,
8705 dl
->index
= u
->old_raid_disks
;
8711 len
= disks_to_mpb_size(u
->new_raid_disks
);
8712 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8715 case update_size_change
: {
8718 case update_create_array
: {
8719 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8720 struct intel_dev
*dv
;
8721 struct imsm_dev
*dev
= &u
->dev
;
8722 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8724 struct disk_info
*inf
;
8728 inf
= get_disk_info(u
);
8729 len
= sizeof_imsm_dev(dev
, 1);
8730 /* allocate a new super->devlist entry */
8731 dv
= xmalloc(sizeof(*dv
));
8732 dv
->dev
= xmalloc(len
);
8735 /* count how many spares will be converted to members */
8736 for (i
= 0; i
< map
->num_members
; i
++) {
8737 dl
= serial_to_dl(inf
[i
].serial
, super
);
8739 /* hmm maybe it failed?, nothing we can do about
8744 if (count_memberships(dl
, super
) == 0)
8747 len
+= activate
* sizeof(struct imsm_disk
);
8754 /* check if we need a larger metadata buffer */
8755 if (super
->next_buf
)
8756 buf_len
= super
->next_len
;
8758 buf_len
= super
->len
;
8760 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8761 /* ok we need a larger buf than what is currently allocated
8762 * if this allocation fails process_update will notice that
8763 * ->next_len is set and ->next_buf is NULL
8765 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8766 if (super
->next_buf
)
8767 free(super
->next_buf
);
8769 super
->next_len
= buf_len
;
8770 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8771 memset(super
->next_buf
, 0, buf_len
);
8773 super
->next_buf
= NULL
;
8777 /* must be called while manager is quiesced */
8778 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8780 struct imsm_super
*mpb
= super
->anchor
;
8782 struct imsm_dev
*dev
;
8783 struct imsm_map
*map
;
8784 int i
, j
, num_members
;
8787 dprintf("%s: deleting device[%d] from imsm_super\n",
8790 /* shift all indexes down one */
8791 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8792 if (iter
->index
> (int)index
)
8794 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8795 if (iter
->index
> (int)index
)
8798 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8799 dev
= get_imsm_dev(super
, i
);
8800 map
= get_imsm_map(dev
, MAP_0
);
8801 num_members
= map
->num_members
;
8802 for (j
= 0; j
< num_members
; j
++) {
8803 /* update ord entries being careful not to propagate
8804 * ord-flags to the first map
8806 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8808 if (ord_to_idx(ord
) <= index
)
8811 map
= get_imsm_map(dev
, MAP_0
);
8812 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8813 map
= get_imsm_map(dev
, MAP_1
);
8815 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8820 super
->updates_pending
++;
8822 struct dl
*dl
= *dlp
;
8824 *dlp
= (*dlp
)->next
;
8825 __free_imsm_disk(dl
);
8828 #endif /* MDASSEMBLE */
8830 static void close_targets(int *targets
, int new_disks
)
8837 for (i
= 0; i
< new_disks
; i
++) {
8838 if (targets
[i
] >= 0) {
8845 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8846 struct intel_super
*super
,
8847 struct imsm_dev
*dev
)
8853 struct imsm_map
*map
;
8856 ret_val
= raid_disks
/2;
8857 /* check map if all disks pairs not failed
8860 map
= get_imsm_map(dev
, MAP_0
);
8861 for (i
= 0; i
< ret_val
; i
++) {
8862 int degradation
= 0;
8863 if (get_imsm_disk(super
, i
) == NULL
)
8865 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8867 if (degradation
== 2)
8870 map
= get_imsm_map(dev
, MAP_1
);
8871 /* if there is no second map
8872 * result can be returned
8876 /* check degradation in second map
8878 for (i
= 0; i
< ret_val
; i
++) {
8879 int degradation
= 0;
8880 if (get_imsm_disk(super
, i
) == NULL
)
8882 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8884 if (degradation
== 2)
8899 /*******************************************************************************
8900 * Function: open_backup_targets
8901 * Description: Function opens file descriptors for all devices given in
8904 * info : general array info
8905 * raid_disks : number of disks
8906 * raid_fds : table of device's file descriptors
8907 * super : intel super for raid10 degradation check
8908 * dev : intel device for raid10 degradation check
8912 ******************************************************************************/
8913 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
8914 struct intel_super
*super
, struct imsm_dev
*dev
)
8920 for (i
= 0; i
< raid_disks
; i
++)
8923 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8926 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
8927 dprintf("disk is faulty!!\n");
8931 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
8932 (sd
->disk
.raid_disk
< 0))
8935 dn
= map_dev(sd
->disk
.major
,
8937 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
8938 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
8939 fprintf(stderr
, "cannot open component\n");
8944 /* check if maximum array degradation level is not exceeded
8946 if ((raid_disks
- opened
) >
8947 imsm_get_allowed_degradation(info
->new_level
,
8950 fprintf(stderr
, "Not enough disks can be opened.\n");
8951 close_targets(raid_fds
, raid_disks
);
8958 /*******************************************************************************
8959 * Function: init_migr_record_imsm
8960 * Description: Function inits imsm migration record
8962 * super : imsm internal array info
8963 * dev : device under migration
8964 * info : general array info to find the smallest device
8967 ******************************************************************************/
8968 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
8969 struct mdinfo
*info
)
8971 struct intel_super
*super
= st
->sb
;
8972 struct migr_record
*migr_rec
= super
->migr_rec
;
8974 unsigned long long dsize
, dev_sectors
;
8975 long long unsigned min_dev_sectors
= -1LLU;
8979 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8980 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
8981 unsigned long long num_migr_units
;
8982 unsigned long long array_blocks
;
8984 memset(migr_rec
, 0, sizeof(struct migr_record
));
8985 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8987 /* only ascending reshape supported now */
8988 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8990 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8991 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8992 migr_rec
->dest_depth_per_unit
*=
8993 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8994 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
8995 migr_rec
->blocks_per_unit
=
8996 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8997 migr_rec
->dest_depth_per_unit
=
8998 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
8999 array_blocks
= info
->component_size
* new_data_disks
;
9001 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9003 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9005 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9007 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9008 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9011 /* Find the smallest dev */
9012 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9013 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9014 fd
= dev_open(nm
, O_RDONLY
);
9017 get_dev_size(fd
, NULL
, &dsize
);
9018 dev_sectors
= dsize
/ 512;
9019 if (dev_sectors
< min_dev_sectors
)
9020 min_dev_sectors
= dev_sectors
;
9023 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9024 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9026 write_imsm_migr_rec(st
);
9031 /*******************************************************************************
9032 * Function: save_backup_imsm
9033 * Description: Function saves critical data stripes to Migration Copy Area
9034 * and updates the current migration unit status.
9035 * Use restore_stripes() to form a destination stripe,
9036 * and to write it to the Copy Area.
9038 * st : supertype information
9039 * dev : imsm device that backup is saved for
9040 * info : general array info
9041 * buf : input buffer
9042 * length : length of data to backup (blocks_per_unit)
9046 ******************************************************************************/
9047 int save_backup_imsm(struct supertype
*st
,
9048 struct imsm_dev
*dev
,
9049 struct mdinfo
*info
,
9054 struct intel_super
*super
= st
->sb
;
9055 unsigned long long *target_offsets
= NULL
;
9056 int *targets
= NULL
;
9058 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9059 int new_disks
= map_dest
->num_members
;
9060 int dest_layout
= 0;
9062 unsigned long long start
;
9063 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9065 targets
= xmalloc(new_disks
* sizeof(int));
9067 for (i
= 0; i
< new_disks
; i
++)
9070 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
9072 start
= info
->reshape_progress
* 512;
9073 for (i
= 0; i
< new_disks
; i
++) {
9074 target_offsets
[i
] = (unsigned long long)
9075 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9076 /* move back copy area adderss, it will be moved forward
9077 * in restore_stripes() using start input variable
9079 target_offsets
[i
] -= start
/data_disks
;
9082 if (open_backup_targets(info
, new_disks
, targets
,
9086 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9087 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9089 if (restore_stripes(targets
, /* list of dest devices */
9090 target_offsets
, /* migration record offsets */
9093 map_dest
->raid_level
,
9095 -1, /* source backup file descriptor */
9096 0, /* input buf offset
9097 * always 0 buf is already offseted */
9101 pr_err("Error restoring stripes\n");
9109 close_targets(targets
, new_disks
);
9112 free(target_offsets
);
9117 /*******************************************************************************
9118 * Function: save_checkpoint_imsm
9119 * Description: Function called for current unit status update
9120 * in the migration record. It writes it to disk.
9122 * super : imsm internal array info
9123 * info : general array info
9127 * 2: failure, means no valid migration record
9128 * / no general migration in progress /
9129 ******************************************************************************/
9130 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9132 struct intel_super
*super
= st
->sb
;
9133 unsigned long long blocks_per_unit
;
9134 unsigned long long curr_migr_unit
;
9136 if (load_imsm_migr_rec(super
, info
) != 0) {
9137 dprintf("imsm: ERROR: Cannot read migration record "
9138 "for checkpoint save.\n");
9142 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9143 if (blocks_per_unit
== 0) {
9144 dprintf("imsm: no migration in progress.\n");
9147 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9148 /* check if array is alligned to copy area
9149 * if it is not alligned, add one to current migration unit value
9150 * this can happend on array reshape finish only
9152 if (info
->reshape_progress
% blocks_per_unit
)
9155 super
->migr_rec
->curr_migr_unit
=
9156 __cpu_to_le32(curr_migr_unit
);
9157 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9158 super
->migr_rec
->dest_1st_member_lba
=
9159 __cpu_to_le32(curr_migr_unit
*
9160 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9161 if (write_imsm_migr_rec(st
) < 0) {
9162 dprintf("imsm: Cannot write migration record "
9163 "outside backup area\n");
9170 /*******************************************************************************
9171 * Function: recover_backup_imsm
9172 * Description: Function recovers critical data from the Migration Copy Area
9173 * while assembling an array.
9175 * super : imsm internal array info
9176 * info : general array info
9178 * 0 : success (or there is no data to recover)
9180 ******************************************************************************/
9181 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9183 struct intel_super
*super
= st
->sb
;
9184 struct migr_record
*migr_rec
= super
->migr_rec
;
9185 struct imsm_map
*map_dest
= NULL
;
9186 struct intel_dev
*id
= NULL
;
9187 unsigned long long read_offset
;
9188 unsigned long long write_offset
;
9190 int *targets
= NULL
;
9191 int new_disks
, i
, err
;
9194 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9195 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9197 int skipped_disks
= 0;
9199 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9203 /* recover data only during assemblation */
9204 if (strncmp(buffer
, "inactive", 8) != 0)
9206 /* no data to recover */
9207 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9209 if (curr_migr_unit
>= num_migr_units
)
9212 /* find device during reshape */
9213 for (id
= super
->devlist
; id
; id
= id
->next
)
9214 if (is_gen_migration(id
->dev
))
9219 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9220 new_disks
= map_dest
->num_members
;
9222 read_offset
= (unsigned long long)
9223 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9225 write_offset
= ((unsigned long long)
9226 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9227 pba_of_lba0(map_dest
)) * 512;
9229 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9230 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9232 targets
= xcalloc(new_disks
, sizeof(int));
9234 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9235 pr_err("Cannot open some devices belonging to array.\n");
9239 for (i
= 0; i
< new_disks
; i
++) {
9240 if (targets
[i
] < 0) {
9244 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9245 pr_err("Cannot seek to block: %s\n",
9250 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9251 pr_err("Cannot read copy area block: %s\n",
9256 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9257 pr_err("Cannot seek to block: %s\n",
9262 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9263 pr_err("Cannot restore block: %s\n",
9270 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9274 pr_err("Cannot restore data from backup."
9275 " Too many failed disks\n");
9279 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9280 /* ignore error == 2, this can mean end of reshape here
9282 dprintf("imsm: Cannot write checkpoint to "
9283 "migration record (UNIT_SRC_NORMAL) during restart\n");
9289 for (i
= 0; i
< new_disks
; i
++)
9298 static char disk_by_path
[] = "/dev/disk/by-path/";
9300 static const char *imsm_get_disk_controller_domain(const char *path
)
9302 char disk_path
[PATH_MAX
];
9306 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
9307 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9308 if (stat(disk_path
, &st
) == 0) {
9309 struct sys_dev
* hba
;
9312 path
= devt_to_devpath(st
.st_rdev
);
9315 hba
= find_disk_attached_hba(-1, path
);
9316 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9318 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9322 dprintf("path: %s hba: %s attached: %s\n",
9323 path
, (hba
) ? hba
->path
: "NULL", drv
);
9329 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
9331 static char devnm
[32];
9332 char subdev_name
[20];
9333 struct mdstat_ent
*mdstat
;
9335 sprintf(subdev_name
, "%d", subdev
);
9336 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9340 strcpy(devnm
, mdstat
->devnm
);
9341 free_mdstat(mdstat
);
9345 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9346 struct geo_params
*geo
,
9347 int *old_raid_disks
,
9350 /* currently we only support increasing the number of devices
9351 * for a container. This increases the number of device for each
9352 * member array. They must all be RAID0 or RAID5.
9355 struct mdinfo
*info
, *member
;
9356 int devices_that_can_grow
= 0;
9358 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
9359 "st->devnm = (%s)\n", st
->devnm
);
9361 if (geo
->size
> 0 ||
9362 geo
->level
!= UnSet
||
9363 geo
->layout
!= UnSet
||
9364 geo
->chunksize
!= 0 ||
9365 geo
->raid_disks
== UnSet
) {
9366 dprintf("imsm: Container operation is allowed for "
9367 "raid disks number change only.\n");
9371 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9372 dprintf("imsm: Metadata changes rollback is not supported for "
9373 "container operation.\n");
9377 info
= container_content_imsm(st
, NULL
);
9378 for (member
= info
; member
; member
= member
->next
) {
9381 dprintf("imsm: checking device_num: %i\n",
9382 member
->container_member
);
9384 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9385 /* we work on container for Online Capacity Expansion
9386 * only so raid_disks has to grow
9388 dprintf("imsm: for container operation raid disks "
9389 "increase is required\n");
9393 if ((info
->array
.level
!= 0) &&
9394 (info
->array
.level
!= 5)) {
9395 /* we cannot use this container with other raid level
9397 dprintf("imsm: for container operation wrong"
9398 " raid level (%i) detected\n",
9402 /* check for platform support
9403 * for this raid level configuration
9405 struct intel_super
*super
= st
->sb
;
9406 if (!is_raid_level_supported(super
->orom
,
9407 member
->array
.level
,
9409 dprintf("platform does not support raid%d with"
9413 geo
->raid_disks
> 1 ? "s" : "");
9416 /* check if component size is aligned to chunk size
9418 if (info
->component_size
%
9419 (info
->array
.chunk_size
/512)) {
9420 dprintf("Component size is not aligned to "
9426 if (*old_raid_disks
&&
9427 info
->array
.raid_disks
!= *old_raid_disks
)
9429 *old_raid_disks
= info
->array
.raid_disks
;
9431 /* All raid5 and raid0 volumes in container
9432 * have to be ready for Online Capacity Expansion
9433 * so they need to be assembled. We have already
9434 * checked that no recovery etc is happening.
9436 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
9437 st
->container_devnm
);
9438 if (result
== NULL
) {
9439 dprintf("imsm: cannot find array\n");
9442 devices_that_can_grow
++;
9445 if (!member
&& devices_that_can_grow
)
9449 dprintf("\tContainer operation allowed\n");
9451 dprintf("\tError: %i\n", ret_val
);
9456 /* Function: get_spares_for_grow
9457 * Description: Allocates memory and creates list of spare devices
9458 * avaliable in container. Checks if spare drive size is acceptable.
9459 * Parameters: Pointer to the supertype structure
9460 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9463 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9465 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9466 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9469 /******************************************************************************
9470 * function: imsm_create_metadata_update_for_reshape
9471 * Function creates update for whole IMSM container.
9473 ******************************************************************************/
9474 static int imsm_create_metadata_update_for_reshape(
9475 struct supertype
*st
,
9476 struct geo_params
*geo
,
9478 struct imsm_update_reshape
**updatep
)
9480 struct intel_super
*super
= st
->sb
;
9481 struct imsm_super
*mpb
= super
->anchor
;
9482 int update_memory_size
= 0;
9483 struct imsm_update_reshape
*u
= NULL
;
9484 struct mdinfo
*spares
= NULL
;
9486 int delta_disks
= 0;
9489 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
9492 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9494 /* size of all update data without anchor */
9495 update_memory_size
= sizeof(struct imsm_update_reshape
);
9497 /* now add space for spare disks that we need to add. */
9498 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9500 u
= xcalloc(1, update_memory_size
);
9501 u
->type
= update_reshape_container_disks
;
9502 u
->old_raid_disks
= old_raid_disks
;
9503 u
->new_raid_disks
= geo
->raid_disks
;
9505 /* now get spare disks list
9507 spares
= get_spares_for_grow(st
);
9510 || delta_disks
> spares
->array
.spare_disks
) {
9511 pr_err("imsm: ERROR: Cannot get spare devices "
9512 "for %s.\n", geo
->dev_name
);
9517 /* we have got spares
9518 * update disk list in imsm_disk list table in anchor
9520 dprintf("imsm: %i spares are available.\n\n",
9521 spares
->array
.spare_disks
);
9524 for (i
= 0; i
< delta_disks
; i
++) {
9529 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9531 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9532 dl
->index
= mpb
->num_disks
;
9542 dprintf("imsm: reshape update preparation :");
9543 if (i
== delta_disks
) {
9546 return update_memory_size
;
9549 dprintf(" Error\n");
9555 /******************************************************************************
9556 * function: imsm_create_metadata_update_for_size_change()
9557 * Creates update for IMSM array for array size change.
9559 ******************************************************************************/
9560 static int imsm_create_metadata_update_for_size_change(
9561 struct supertype
*st
,
9562 struct geo_params
*geo
,
9563 struct imsm_update_size_change
**updatep
)
9565 struct intel_super
*super
= st
->sb
;
9566 int update_memory_size
= 0;
9567 struct imsm_update_size_change
*u
= NULL
;
9569 dprintf("imsm_create_metadata_update_for_size_change(enter)"
9570 " New size = %llu\n", geo
->size
);
9572 /* size of all update data without anchor */
9573 update_memory_size
= sizeof(struct imsm_update_size_change
);
9575 u
= xcalloc(1, update_memory_size
);
9576 u
->type
= update_size_change
;
9577 u
->subdev
= super
->current_vol
;
9578 u
->new_size
= geo
->size
;
9580 dprintf("imsm: reshape update preparation : OK\n");
9583 return update_memory_size
;
9586 /******************************************************************************
9587 * function: imsm_create_metadata_update_for_migration()
9588 * Creates update for IMSM array.
9590 ******************************************************************************/
9591 static int imsm_create_metadata_update_for_migration(
9592 struct supertype
*st
,
9593 struct geo_params
*geo
,
9594 struct imsm_update_reshape_migration
**updatep
)
9596 struct intel_super
*super
= st
->sb
;
9597 int update_memory_size
= 0;
9598 struct imsm_update_reshape_migration
*u
= NULL
;
9599 struct imsm_dev
*dev
;
9600 int previous_level
= -1;
9602 dprintf("imsm_create_metadata_update_for_migration(enter)"
9603 " New Level = %i\n", geo
->level
);
9605 /* size of all update data without anchor */
9606 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9608 u
= xcalloc(1, update_memory_size
);
9609 u
->type
= update_reshape_migration
;
9610 u
->subdev
= super
->current_vol
;
9611 u
->new_level
= geo
->level
;
9612 u
->new_layout
= geo
->layout
;
9613 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9614 u
->new_disks
[0] = -1;
9615 u
->new_chunksize
= -1;
9617 dev
= get_imsm_dev(super
, u
->subdev
);
9619 struct imsm_map
*map
;
9621 map
= get_imsm_map(dev
, MAP_0
);
9623 int current_chunk_size
=
9624 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9626 if (geo
->chunksize
!= current_chunk_size
) {
9627 u
->new_chunksize
= geo
->chunksize
/ 1024;
9629 "chunk size change from %i to %i\n",
9630 current_chunk_size
, u
->new_chunksize
);
9632 previous_level
= map
->raid_level
;
9635 if ((geo
->level
== 5) && (previous_level
== 0)) {
9636 struct mdinfo
*spares
= NULL
;
9638 u
->new_raid_disks
++;
9639 spares
= get_spares_for_grow(st
);
9640 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9643 update_memory_size
= 0;
9644 dprintf("error: cannot get spare device "
9645 "for requested migration");
9650 dprintf("imsm: reshape update preparation : OK\n");
9653 return update_memory_size
;
9656 static void imsm_update_metadata_locally(struct supertype
*st
,
9659 struct metadata_update mu
;
9664 mu
.space_list
= NULL
;
9666 imsm_prepare_update(st
, &mu
);
9667 imsm_process_update(st
, &mu
);
9669 while (mu
.space_list
) {
9670 void **space
= mu
.space_list
;
9671 mu
.space_list
= *space
;
9676 /***************************************************************************
9677 * Function: imsm_analyze_change
9678 * Description: Function analyze change for single volume
9679 * and validate if transition is supported
9680 * Parameters: Geometry parameters, supertype structure,
9681 * metadata change direction (apply/rollback)
9682 * Returns: Operation type code on success, -1 if fail
9683 ****************************************************************************/
9684 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9685 struct geo_params
*geo
,
9692 /* number of added/removed disks in operation result */
9693 int devNumChange
= 0;
9694 /* imsm compatible layout value for array geometry verification */
9695 int imsm_layout
= -1;
9697 struct imsm_dev
*dev
;
9698 struct intel_super
*super
;
9699 unsigned long long current_size
;
9700 unsigned long long free_size
;
9701 unsigned long long max_size
;
9704 getinfo_super_imsm_volume(st
, &info
, NULL
);
9705 if ((geo
->level
!= info
.array
.level
) &&
9706 (geo
->level
>= 0) &&
9707 (geo
->level
!= UnSet
)) {
9708 switch (info
.array
.level
) {
9710 if (geo
->level
== 5) {
9711 change
= CH_MIGRATION
;
9712 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9713 pr_err("Error. Requested Layout "
9714 "not supported (left-asymmetric layout "
9715 "is supported only)!\n");
9717 goto analyse_change_exit
;
9719 imsm_layout
= geo
->layout
;
9721 devNumChange
= 1; /* parity disk added */
9722 } else if (geo
->level
== 10) {
9723 change
= CH_TAKEOVER
;
9725 devNumChange
= 2; /* two mirrors added */
9726 imsm_layout
= 0x102; /* imsm supported layout */
9731 if (geo
->level
== 0) {
9732 change
= CH_TAKEOVER
;
9734 devNumChange
= -(geo
->raid_disks
/2);
9735 imsm_layout
= 0; /* imsm raid0 layout */
9740 pr_err("Error. Level Migration from %d to %d "
9742 info
.array
.level
, geo
->level
);
9743 goto analyse_change_exit
;
9746 geo
->level
= info
.array
.level
;
9748 if ((geo
->layout
!= info
.array
.layout
)
9749 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9750 change
= CH_MIGRATION
;
9751 if ((info
.array
.layout
== 0)
9752 && (info
.array
.level
== 5)
9753 && (geo
->layout
== 5)) {
9754 /* reshape 5 -> 4 */
9755 } else if ((info
.array
.layout
== 5)
9756 && (info
.array
.level
== 5)
9757 && (geo
->layout
== 0)) {
9758 /* reshape 4 -> 5 */
9762 pr_err("Error. Layout Migration from %d to %d "
9764 info
.array
.layout
, geo
->layout
);
9766 goto analyse_change_exit
;
9769 geo
->layout
= info
.array
.layout
;
9770 if (imsm_layout
== -1)
9771 imsm_layout
= info
.array
.layout
;
9774 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9775 && (geo
->chunksize
!= info
.array
.chunk_size
))
9776 change
= CH_MIGRATION
;
9778 geo
->chunksize
= info
.array
.chunk_size
;
9780 chunk
= geo
->chunksize
/ 1024;
9783 dev
= get_imsm_dev(super
, super
->current_vol
);
9784 data_disks
= imsm_num_data_members(dev
, MAP_0
);
9785 /* compute current size per disk member
9787 current_size
= info
.custom_array_size
/ data_disks
;
9789 if ((geo
->size
> 0) && (geo
->size
!= MAX_SIZE
)) {
9790 /* align component size
9792 geo
->size
= imsm_component_size_aligment_check(
9793 get_imsm_raid_level(dev
->vol
.map
),
9796 if (geo
->size
== 0) {
9797 pr_err("Error. Size expansion is " \
9798 "supported only (current size is %llu, " \
9799 "requested size /rounded/ is 0).\n",
9801 goto analyse_change_exit
;
9805 if ((current_size
!= geo
->size
) && (geo
->size
> 0)) {
9807 pr_err("Error. Size change should be the only "
9808 "one at a time.\n");
9810 goto analyse_change_exit
;
9812 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
9813 pr_err("Error. The last volume in container "
9814 "can be expanded only (%i/%s).\n",
9815 super
->current_vol
, st
->devnm
);
9816 goto analyse_change_exit
;
9818 /* check the maximum available size
9820 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
9821 0, chunk
, &free_size
);
9823 /* Cannot find maximum available space
9827 max_size
= free_size
+ current_size
;
9828 /* align component size
9830 max_size
= imsm_component_size_aligment_check(
9831 get_imsm_raid_level(dev
->vol
.map
),
9835 if (geo
->size
== MAX_SIZE
) {
9836 /* requested size change to the maximum available size
9838 if (max_size
== 0) {
9839 pr_err("Error. Cannot find "
9840 "maximum available space.\n");
9842 goto analyse_change_exit
;
9844 geo
->size
= max_size
;
9847 if ((direction
== ROLLBACK_METADATA_CHANGES
)) {
9848 /* accept size for rollback only
9851 /* round size due to metadata compatibility
9853 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
9854 << SECT_PER_MB_SHIFT
;
9855 dprintf("Prepare update for size change to %llu\n",
9857 if (current_size
>= geo
->size
) {
9858 pr_err("Error. Size expansion is "
9859 "supported only (current size is %llu, "
9860 "requested size /rounded/ is %llu).\n",
9861 current_size
, geo
->size
);
9862 goto analyse_change_exit
;
9864 if (max_size
&& geo
->size
> max_size
) {
9865 pr_err("Error. Requested size is larger "
9866 "than maximum available size (maximum "
9867 "available size is %llu, "
9868 "requested size /rounded/ is %llu).\n",
9869 max_size
, geo
->size
);
9870 goto analyse_change_exit
;
9873 geo
->size
*= data_disks
;
9874 geo
->raid_disks
= dev
->vol
.map
->num_members
;
9875 change
= CH_ARRAY_SIZE
;
9877 if (!validate_geometry_imsm(st
,
9880 geo
->raid_disks
+ devNumChange
,
9882 geo
->size
, INVALID_SECTORS
,
9887 struct intel_super
*super
= st
->sb
;
9888 struct imsm_super
*mpb
= super
->anchor
;
9890 if (mpb
->num_raid_devs
> 1) {
9891 pr_err("Error. Cannot perform operation on %s"
9892 "- for this operation it MUST be single "
9893 "array in container\n",
9899 analyse_change_exit
:
9900 if ((direction
== ROLLBACK_METADATA_CHANGES
) &&
9901 ((change
== CH_MIGRATION
) || (change
== CH_TAKEOVER
))) {
9902 dprintf("imsm: Metadata changes rollback is not supported for "
9903 "migration and takeover operations.\n");
9909 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
9911 struct intel_super
*super
= st
->sb
;
9912 struct imsm_update_takeover
*u
;
9914 u
= xmalloc(sizeof(struct imsm_update_takeover
));
9916 u
->type
= update_takeover
;
9917 u
->subarray
= super
->current_vol
;
9919 /* 10->0 transition */
9920 if (geo
->level
== 0)
9921 u
->direction
= R10_TO_R0
;
9923 /* 0->10 transition */
9924 if (geo
->level
== 10)
9925 u
->direction
= R0_TO_R10
;
9927 /* update metadata locally */
9928 imsm_update_metadata_locally(st
, u
,
9929 sizeof(struct imsm_update_takeover
));
9930 /* and possibly remotely */
9931 if (st
->update_tail
)
9932 append_metadata_update(st
, u
,
9933 sizeof(struct imsm_update_takeover
));
9940 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
9942 int layout
, int chunksize
, int raid_disks
,
9943 int delta_disks
, char *backup
, char *dev
,
9944 int direction
, int verbose
)
9947 struct geo_params geo
;
9949 dprintf("imsm: reshape_super called.\n");
9951 memset(&geo
, 0, sizeof(struct geo_params
));
9954 strcpy(geo
.devnm
, st
->devnm
);
9957 geo
.layout
= layout
;
9958 geo
.chunksize
= chunksize
;
9959 geo
.raid_disks
= raid_disks
;
9960 if (delta_disks
!= UnSet
)
9961 geo
.raid_disks
+= delta_disks
;
9963 dprintf("\tfor level : %i\n", geo
.level
);
9964 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
9966 if (experimental() == 0)
9969 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
9970 /* On container level we can only increase number of devices. */
9971 dprintf("imsm: info: Container operation\n");
9972 int old_raid_disks
= 0;
9974 if (imsm_reshape_is_allowed_on_container(
9975 st
, &geo
, &old_raid_disks
, direction
)) {
9976 struct imsm_update_reshape
*u
= NULL
;
9979 len
= imsm_create_metadata_update_for_reshape(
9980 st
, &geo
, old_raid_disks
, &u
);
9983 dprintf("imsm: Cannot prepare update\n");
9984 goto exit_imsm_reshape_super
;
9988 /* update metadata locally */
9989 imsm_update_metadata_locally(st
, u
, len
);
9990 /* and possibly remotely */
9991 if (st
->update_tail
)
9992 append_metadata_update(st
, u
, len
);
9997 pr_err("(imsm) Operation "
9998 "is not allowed on this container\n");
10001 /* On volume level we support following operations
10002 * - takeover: raid10 -> raid0; raid0 -> raid10
10003 * - chunk size migration
10004 * - migration: raid5 -> raid0; raid0 -> raid5
10006 struct intel_super
*super
= st
->sb
;
10007 struct intel_dev
*dev
= super
->devlist
;
10009 dprintf("imsm: info: Volume operation\n");
10010 /* find requested device */
10013 imsm_find_array_devnm_by_subdev(
10014 dev
->index
, st
->container_devnm
);
10015 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
10020 pr_err("Cannot find %s (%s) subarray\n",
10021 geo
.dev_name
, geo
.devnm
);
10022 goto exit_imsm_reshape_super
;
10024 super
->current_vol
= dev
->index
;
10025 change
= imsm_analyze_change(st
, &geo
, direction
);
10028 ret_val
= imsm_takeover(st
, &geo
);
10030 case CH_MIGRATION
: {
10031 struct imsm_update_reshape_migration
*u
= NULL
;
10033 imsm_create_metadata_update_for_migration(
10037 "Cannot prepare update\n");
10041 /* update metadata locally */
10042 imsm_update_metadata_locally(st
, u
, len
);
10043 /* and possibly remotely */
10044 if (st
->update_tail
)
10045 append_metadata_update(st
, u
, len
);
10050 case CH_ARRAY_SIZE
: {
10051 struct imsm_update_size_change
*u
= NULL
;
10053 imsm_create_metadata_update_for_size_change(
10057 "Cannot prepare update\n");
10061 /* update metadata locally */
10062 imsm_update_metadata_locally(st
, u
, len
);
10063 /* and possibly remotely */
10064 if (st
->update_tail
)
10065 append_metadata_update(st
, u
, len
);
10075 exit_imsm_reshape_super
:
10076 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10080 /*******************************************************************************
10081 * Function: wait_for_reshape_imsm
10082 * Description: Function writes new sync_max value and waits until
10083 * reshape process reach new position
10085 * sra : general array info
10086 * ndata : number of disks in new array's layout
10089 * 1 : there is no reshape in progress,
10091 ******************************************************************************/
10092 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10094 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
10095 unsigned long long completed
;
10096 /* to_complete : new sync_max position */
10097 unsigned long long to_complete
= sra
->reshape_progress
;
10098 unsigned long long position_to_set
= to_complete
/ ndata
;
10101 dprintf("imsm: wait_for_reshape_imsm() "
10102 "cannot open reshape_position\n");
10106 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10107 dprintf("imsm: wait_for_reshape_imsm() "
10108 "cannot read reshape_position (no reshape in progres)\n");
10113 if (completed
> to_complete
) {
10114 dprintf("imsm: wait_for_reshape_imsm() "
10115 "wrong next position to set %llu (%llu)\n",
10116 to_complete
, completed
);
10120 dprintf("Position set: %llu\n", position_to_set
);
10121 if (sysfs_set_num(sra
, NULL
, "sync_max",
10122 position_to_set
) != 0) {
10123 dprintf("imsm: wait_for_reshape_imsm() "
10124 "cannot set reshape position to %llu\n",
10135 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
10136 if (sysfs_get_str(sra
, NULL
, "sync_action",
10138 strncmp(action
, "reshape", 7) != 0)
10140 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10141 dprintf("imsm: wait_for_reshape_imsm() "
10142 "cannot read reshape_position (in loop)\n");
10146 } while (completed
< to_complete
);
10152 /*******************************************************************************
10153 * Function: check_degradation_change
10154 * Description: Check that array hasn't become failed.
10156 * info : for sysfs access
10157 * sources : source disks descriptors
10158 * degraded: previous degradation level
10160 * degradation level
10161 ******************************************************************************/
10162 int check_degradation_change(struct mdinfo
*info
,
10166 unsigned long long new_degraded
;
10169 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10170 if ((rv
== -1) || (new_degraded
!= (unsigned long long)degraded
)) {
10171 /* check each device to ensure it is still working */
10174 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10175 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10177 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10179 if (sysfs_get_str(info
,
10180 sd
, "state", sbuf
, 20) < 0 ||
10181 strstr(sbuf
, "faulty") ||
10182 strstr(sbuf
, "in_sync") == NULL
) {
10183 /* this device is dead */
10184 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10185 if (sd
->disk
.raid_disk
>= 0 &&
10186 sources
[sd
->disk
.raid_disk
] >= 0) {
10188 sd
->disk
.raid_disk
]);
10189 sources
[sd
->disk
.raid_disk
] =
10198 return new_degraded
;
10201 /*******************************************************************************
10202 * Function: imsm_manage_reshape
10203 * Description: Function finds array under reshape and it manages reshape
10204 * process. It creates stripes backups (if required) and sets
10207 * afd : Backup handle (nattive) - not used
10208 * sra : general array info
10209 * reshape : reshape parameters - not used
10210 * st : supertype structure
10211 * blocks : size of critical section [blocks]
10212 * fds : table of source device descriptor
10213 * offsets : start of array (offest per devices)
10215 * destfd : table of destination device descriptor
10216 * destoffsets : table of destination offsets (per device)
10218 * 1 : success, reshape is done
10220 ******************************************************************************/
10221 static int imsm_manage_reshape(
10222 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10223 struct supertype
*st
, unsigned long backup_blocks
,
10224 int *fds
, unsigned long long *offsets
,
10225 int dests
, int *destfd
, unsigned long long *destoffsets
)
10228 struct intel_super
*super
= st
->sb
;
10229 struct intel_dev
*dv
= NULL
;
10230 struct imsm_dev
*dev
= NULL
;
10231 struct imsm_map
*map_src
;
10232 int migr_vol_qan
= 0;
10233 int ndata
, odata
; /* [bytes] */
10234 int chunk
; /* [bytes] */
10235 struct migr_record
*migr_rec
;
10237 unsigned int buf_size
; /* [bytes] */
10238 unsigned long long max_position
; /* array size [bytes] */
10239 unsigned long long next_step
; /* [blocks]/[bytes] */
10240 unsigned long long old_data_stripe_length
;
10241 unsigned long long start_src
; /* [bytes] */
10242 unsigned long long start
; /* [bytes] */
10243 unsigned long long start_buf_shift
; /* [bytes] */
10245 int source_layout
= 0;
10247 if (!fds
|| !offsets
|| !sra
)
10250 /* Find volume during the reshape */
10251 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10252 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10253 && dv
->dev
->vol
.migr_state
== 1) {
10258 /* Only one volume can migrate at the same time */
10259 if (migr_vol_qan
!= 1) {
10260 pr_err(": %s", migr_vol_qan
?
10261 "Number of migrating volumes greater than 1\n" :
10262 "There is no volume during migrationg\n");
10266 map_src
= get_imsm_map(dev
, MAP_1
);
10267 if (map_src
== NULL
)
10270 ndata
= imsm_num_data_members(dev
, MAP_0
);
10271 odata
= imsm_num_data_members(dev
, MAP_1
);
10273 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10274 old_data_stripe_length
= odata
* chunk
;
10276 migr_rec
= super
->migr_rec
;
10278 /* initialize migration record for start condition */
10279 if (sra
->reshape_progress
== 0)
10280 init_migr_record_imsm(st
, dev
, sra
);
10282 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10283 dprintf("imsm: cannot restart migration when data "
10284 "are present in copy area.\n");
10287 /* Save checkpoint to update migration record for current
10288 * reshape position (in md). It can be farther than current
10289 * reshape position in metadata.
10291 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10292 /* ignore error == 2, this can mean end of reshape here
10294 dprintf("imsm: Cannot write checkpoint to "
10295 "migration record (UNIT_SRC_NORMAL, "
10296 "initial save)\n");
10301 /* size for data */
10302 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10303 /* extend buffer size for parity disk */
10304 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10305 /* add space for stripe aligment */
10306 buf_size
+= old_data_stripe_length
;
10307 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10308 dprintf("imsm: Cannot allocate checpoint buffer\n");
10312 max_position
= sra
->component_size
* ndata
;
10313 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10315 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10316 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10317 /* current reshape position [blocks] */
10318 unsigned long long current_position
=
10319 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10320 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10321 unsigned long long border
;
10323 /* Check that array hasn't become failed.
10325 degraded
= check_degradation_change(sra
, fds
, degraded
);
10326 if (degraded
> 1) {
10327 dprintf("imsm: Abort reshape due to degradation"
10328 " level (%i)\n", degraded
);
10332 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10334 if ((current_position
+ next_step
) > max_position
)
10335 next_step
= max_position
- current_position
;
10337 start
= current_position
* 512;
10339 /* allign reading start to old geometry */
10340 start_buf_shift
= start
% old_data_stripe_length
;
10341 start_src
= start
- start_buf_shift
;
10343 border
= (start_src
/ odata
) - (start
/ ndata
);
10345 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10346 /* save critical stripes to buf
10347 * start - start address of current unit
10348 * to backup [bytes]
10349 * start_src - start address of current unit
10350 * to backup alligned to source array
10353 unsigned long long next_step_filler
= 0;
10354 unsigned long long copy_length
= next_step
* 512;
10356 /* allign copy area length to stripe in old geometry */
10357 next_step_filler
= ((copy_length
+ start_buf_shift
)
10358 % old_data_stripe_length
);
10359 if (next_step_filler
)
10360 next_step_filler
= (old_data_stripe_length
10361 - next_step_filler
);
10362 dprintf("save_stripes() parameters: start = %llu,"
10363 "\tstart_src = %llu,\tnext_step*512 = %llu,"
10364 "\tstart_in_buf_shift = %llu,"
10365 "\tnext_step_filler = %llu\n",
10366 start
, start_src
, copy_length
,
10367 start_buf_shift
, next_step_filler
);
10369 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10370 chunk
, map_src
->raid_level
,
10371 source_layout
, 0, NULL
, start_src
,
10373 next_step_filler
+ start_buf_shift
,
10375 dprintf("imsm: Cannot save stripes"
10379 /* Convert data to destination format and store it
10380 * in backup general migration area
10382 if (save_backup_imsm(st
, dev
, sra
,
10383 buf
+ start_buf_shift
, copy_length
)) {
10384 dprintf("imsm: Cannot save stripes to "
10385 "target devices\n");
10388 if (save_checkpoint_imsm(st
, sra
,
10389 UNIT_SRC_IN_CP_AREA
)) {
10390 dprintf("imsm: Cannot write checkpoint to "
10391 "migration record (UNIT_SRC_IN_CP_AREA)\n");
10395 /* set next step to use whole border area */
10396 border
/= next_step
;
10398 next_step
*= border
;
10400 /* When data backed up, checkpoint stored,
10401 * kick the kernel to reshape unit of data
10403 next_step
= next_step
+ sra
->reshape_progress
;
10404 /* limit next step to array max position */
10405 if (next_step
> max_position
)
10406 next_step
= max_position
;
10407 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10408 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10409 sra
->reshape_progress
= next_step
;
10411 /* wait until reshape finish */
10412 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
10413 dprintf("wait_for_reshape_imsm returned error!\n");
10417 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10418 /* ignore error == 2, this can mean end of reshape here
10420 dprintf("imsm: Cannot write checkpoint to "
10421 "migration record (UNIT_SRC_NORMAL)\n");
10427 /* return '1' if done */
10431 abort_reshape(sra
);
10435 #endif /* MDASSEMBLE */
10437 struct superswitch super_imsm
= {
10439 .examine_super
= examine_super_imsm
,
10440 .brief_examine_super
= brief_examine_super_imsm
,
10441 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10442 .export_examine_super
= export_examine_super_imsm
,
10443 .detail_super
= detail_super_imsm
,
10444 .brief_detail_super
= brief_detail_super_imsm
,
10445 .write_init_super
= write_init_super_imsm
,
10446 .validate_geometry
= validate_geometry_imsm
,
10447 .add_to_super
= add_to_super_imsm
,
10448 .remove_from_super
= remove_from_super_imsm
,
10449 .detail_platform
= detail_platform_imsm
,
10450 .export_detail_platform
= export_detail_platform_imsm
,
10451 .kill_subarray
= kill_subarray_imsm
,
10452 .update_subarray
= update_subarray_imsm
,
10453 .load_container
= load_container_imsm
,
10454 .default_geometry
= default_geometry_imsm
,
10455 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10456 .reshape_super
= imsm_reshape_super
,
10457 .manage_reshape
= imsm_manage_reshape
,
10458 .recover_backup
= recover_backup_imsm
,
10460 .match_home
= match_home_imsm
,
10461 .uuid_from_super
= uuid_from_super_imsm
,
10462 .getinfo_super
= getinfo_super_imsm
,
10463 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10464 .update_super
= update_super_imsm
,
10466 .avail_size
= avail_size_imsm
,
10467 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10469 .compare_super
= compare_super_imsm
,
10471 .load_super
= load_super_imsm
,
10472 .init_super
= init_super_imsm
,
10473 .store_super
= store_super_imsm
,
10474 .free_super
= free_super_imsm
,
10475 .match_metadata_desc
= match_metadata_desc_imsm
,
10476 .container_content
= container_content_imsm
,
10484 .open_new
= imsm_open_new
,
10485 .set_array_state
= imsm_set_array_state
,
10486 .set_disk
= imsm_set_disk
,
10487 .sync_metadata
= imsm_sync_metadata
,
10488 .activate_spare
= imsm_activate_spare
,
10489 .process_update
= imsm_process_update
,
10490 .prepare_update
= imsm_prepare_update
,
10491 #endif /* MDASSEMBLE */