2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
47 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
51 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
53 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
54 /* supports RAID CNG */
55 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
56 /* supports expanded stripe sizes of 256K, 512K and 1MB */
57 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
59 /* The OROM Support RST Caching of Volumes */
60 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
61 /* The OROM supports creating disks greater than 2TB */
62 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
63 /* The OROM supports Bad Block Management */
64 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
66 /* THe OROM Supports NVM Caching of Volumes */
67 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
68 /* The OROM supports creating volumes greater than 2TB */
69 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
70 /* originally for PMP, now it's wasted b/c. Never use this bit! */
71 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
72 /* Verify MPB contents against checksum after reading MPB */
73 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
75 /* Define all supported attributes that have to be accepted by mdadm
77 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
79 MPB_ATTRIB_2TB_DISK | \
84 MPB_ATTRIB_EXP_STRIPE_SIZE)
86 /* Define attributes that are unused but not harmful */
87 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
89 #define MPB_SECTOR_CNT 2210
90 #define IMSM_RESERVED_SECTORS 4096
91 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
92 #define SECT_PER_MB_SHIFT 11
94 /* Disk configuration info. */
95 #define IMSM_MAX_DEVICES 255
97 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
98 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
99 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
100 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
101 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
102 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
103 __u32 status
; /* 0xF0 - 0xF3 */
104 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
105 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
106 #define IMSM_DISK_FILLERS 3
107 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
110 /* map selector for map managment
116 /* RAID map configuration infos. */
118 __u32 pba_of_lba0_lo
; /* start address of partition */
119 __u32 blocks_per_member_lo
;/* blocks per member */
120 __u32 num_data_stripes_lo
; /* number of data stripes */
121 __u16 blocks_per_strip
;
122 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
123 #define IMSM_T_STATE_NORMAL 0
124 #define IMSM_T_STATE_UNINITIALIZED 1
125 #define IMSM_T_STATE_DEGRADED 2
126 #define IMSM_T_STATE_FAILED 3
128 #define IMSM_T_RAID0 0
129 #define IMSM_T_RAID1 1
130 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
131 __u8 num_members
; /* number of member disks */
132 __u8 num_domains
; /* number of parity domains */
133 __u8 failed_disk_num
; /* valid only when state is degraded */
135 __u32 pba_of_lba0_hi
;
136 __u32 blocks_per_member_hi
;
137 __u32 num_data_stripes_hi
;
138 __u32 filler
[4]; /* expansion area */
139 #define IMSM_ORD_REBUILD (1 << 24)
140 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
141 * top byte contains some flags
143 } __attribute__ ((packed
));
146 __u32 curr_migr_unit
;
147 __u32 checkpoint_id
; /* id to access curr_migr_unit */
148 __u8 migr_state
; /* Normal or Migrating */
150 #define MIGR_REBUILD 1
151 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
152 #define MIGR_GEN_MIGR 3
153 #define MIGR_STATE_CHANGE 4
154 #define MIGR_REPAIR 5
155 __u8 migr_type
; /* Initializing, Rebuilding, ... */
157 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
158 __u16 verify_errors
; /* number of mismatches */
159 __u16 bad_blocks
; /* number of bad blocks during verify */
161 struct imsm_map map
[1];
162 /* here comes another one if migr_state */
163 } __attribute__ ((packed
));
166 __u8 volume
[MAX_RAID_SERIAL_LEN
];
169 #define DEV_BOOTABLE __cpu_to_le32(0x01)
170 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
171 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
172 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
173 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
174 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
175 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
176 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
177 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
178 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
179 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
180 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
181 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
182 __u32 status
; /* Persistent RaidDev status */
183 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
187 __u8 cng_master_disk
;
191 #define IMSM_DEV_FILLERS 10
192 __u32 filler
[IMSM_DEV_FILLERS
];
194 } __attribute__ ((packed
));
197 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
198 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
199 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
200 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
201 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
202 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
203 __u32 attributes
; /* 0x34 - 0x37 */
204 __u8 num_disks
; /* 0x38 Number of configured disks */
205 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
206 __u8 error_log_pos
; /* 0x3A */
207 __u8 fill
[1]; /* 0x3B */
208 __u32 cache_size
; /* 0x3c - 0x40 in mb */
209 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
210 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
211 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
212 #define IMSM_FILLERS 35
213 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
214 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
215 /* here comes imsm_dev[num_raid_devs] */
216 /* here comes BBM logs */
217 } __attribute__ ((packed
));
219 #define BBM_LOG_MAX_ENTRIES 254
221 struct bbm_log_entry
{
222 __u64 defective_block_start
;
223 #define UNREADABLE 0xFFFFFFFF
224 __u32 spare_block_offset
;
225 __u16 remapped_marked_count
;
227 } __attribute__ ((__packed__
));
230 __u32 signature
; /* 0xABADB10C */
232 __u32 reserved_spare_block_count
; /* 0 */
233 __u32 reserved
; /* 0xFFFF */
234 __u64 first_spare_lba
;
235 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
236 } __attribute__ ((__packed__
));
239 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
242 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
244 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
246 #define MIGR_REC_BUF_SIZE 512 /* size of migr_record i/o buffer */
247 #define MIGR_REC_POSITION 512 /* migr_record position offset on disk,
248 * MIGR_REC_BUF_SIZE <= MIGR_REC_POSITION
251 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
252 * be recovered using srcMap */
253 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
254 * already been migrated and must
255 * be recovered from checkpoint area */
257 __u32 rec_status
; /* Status used to determine how to restart
258 * migration in case it aborts
260 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
261 __u32 family_num
; /* Family number of MPB
262 * containing the RaidDev
263 * that is migrating */
264 __u32 ascending_migr
; /* True if migrating in increasing
266 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
267 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
269 * advances per unit-of-operation */
270 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
271 __u32 dest_1st_member_lba
; /* First member lba on first
272 * stripe of destination */
273 __u32 num_migr_units
; /* Total num migration units-of-op */
274 __u32 post_migr_vol_cap
; /* Size of volume after
275 * migration completes */
276 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
277 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
278 * migration ckpt record was read from
279 * (for recovered migrations) */
280 } __attribute__ ((__packed__
));
285 * 2: metadata does not match
293 struct md_list
*next
;
296 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
298 static __u8
migr_type(struct imsm_dev
*dev
)
300 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
301 dev
->status
& DEV_VERIFY_AND_FIX
)
304 return dev
->vol
.migr_type
;
307 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
309 /* for compatibility with older oroms convert MIGR_REPAIR, into
310 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
312 if (migr_type
== MIGR_REPAIR
) {
313 dev
->vol
.migr_type
= MIGR_VERIFY
;
314 dev
->status
|= DEV_VERIFY_AND_FIX
;
316 dev
->vol
.migr_type
= migr_type
;
317 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
321 static unsigned int sector_count(__u32 bytes
)
323 return ROUND_UP(bytes
, 512) / 512;
326 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
328 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
332 struct imsm_dev
*dev
;
333 struct intel_dev
*next
;
338 enum sys_dev_type type
;
341 struct intel_hba
*next
;
348 /* internal representation of IMSM metadata */
351 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
352 struct imsm_super
*anchor
; /* immovable parameters */
355 void *migr_rec_buf
; /* buffer for I/O operations */
356 struct migr_record
*migr_rec
; /* migration record */
358 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
359 array, it indicates that mdmon is allowed to clean migration
361 size_t len
; /* size of the 'buf' allocation */
362 void *next_buf
; /* for realloc'ing buf from the manager */
364 int updates_pending
; /* count of pending updates for mdmon */
365 int current_vol
; /* index of raid device undergoing creation */
366 unsigned long long create_offset
; /* common start for 'current_vol' */
367 __u32 random
; /* random data for seeding new family numbers */
368 struct intel_dev
*devlist
;
372 __u8 serial
[MAX_RAID_SERIAL_LEN
];
375 struct imsm_disk disk
;
378 struct extent
*e
; /* for determining freespace @ create */
379 int raiddisk
; /* slot to fill in autolayout */
381 } *disks
, *current_disk
;
382 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
384 struct dl
*missing
; /* disks removed while we weren't looking */
385 struct bbm_log
*bbm_log
;
386 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
387 const struct imsm_orom
*orom
; /* platform firmware support */
388 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
392 struct imsm_disk disk
;
393 #define IMSM_UNKNOWN_OWNER (-1)
395 struct intel_disk
*next
;
399 unsigned long long start
, size
;
402 /* definitions of reshape process types */
403 enum imsm_reshape_type
{
409 /* definition of messages passed to imsm_process_update */
410 enum imsm_update_type
{
411 update_activate_spare
,
415 update_add_remove_disk
,
416 update_reshape_container_disks
,
417 update_reshape_migration
,
419 update_general_migration_checkpoint
,
423 struct imsm_update_activate_spare
{
424 enum imsm_update_type type
;
428 struct imsm_update_activate_spare
*next
;
434 unsigned long long size
;
441 enum takeover_direction
{
445 struct imsm_update_takeover
{
446 enum imsm_update_type type
;
448 enum takeover_direction direction
;
451 struct imsm_update_reshape
{
452 enum imsm_update_type type
;
456 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
459 struct imsm_update_reshape_migration
{
460 enum imsm_update_type type
;
463 /* fields for array migration changes
470 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
473 struct imsm_update_size_change
{
474 enum imsm_update_type type
;
479 struct imsm_update_general_migration_checkpoint
{
480 enum imsm_update_type type
;
481 __u32 curr_migr_unit
;
485 __u8 serial
[MAX_RAID_SERIAL_LEN
];
488 struct imsm_update_create_array
{
489 enum imsm_update_type type
;
494 struct imsm_update_kill_array
{
495 enum imsm_update_type type
;
499 struct imsm_update_rename_array
{
500 enum imsm_update_type type
;
501 __u8 name
[MAX_RAID_SERIAL_LEN
];
505 struct imsm_update_add_remove_disk
{
506 enum imsm_update_type type
;
509 static const char *_sys_dev_type
[] = {
510 [SYS_DEV_UNKNOWN
] = "Unknown",
511 [SYS_DEV_SAS
] = "SAS",
512 [SYS_DEV_SATA
] = "SATA",
513 [SYS_DEV_NVME
] = "NVMe",
514 [SYS_DEV_VMD
] = "VMD"
517 const char *get_sys_dev_type(enum sys_dev_type type
)
519 if (type
>= SYS_DEV_MAX
)
520 type
= SYS_DEV_UNKNOWN
;
522 return _sys_dev_type
[type
];
525 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
527 struct intel_hba
*result
= xmalloc(sizeof(*result
));
529 result
->type
= device
->type
;
530 result
->path
= xstrdup(device
->path
);
532 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
538 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
540 struct intel_hba
*result
=NULL
;
541 for (result
= hba
; result
; result
= result
->next
) {
542 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
548 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
550 struct intel_hba
*hba
;
552 /* check if disk attached to Intel HBA */
553 hba
= find_intel_hba(super
->hba
, device
);
556 /* Check if HBA is already attached to super */
557 if (super
->hba
== NULL
) {
558 super
->hba
= alloc_intel_hba(device
);
563 /* Intel metadata allows for all disks attached to the same type HBA.
564 * Do not support HBA types mixing
566 if (device
->type
!= hba
->type
)
569 /* Always forbid spanning between VMD domains (seen as different controllers by mdadm) */
570 if (device
->type
== SYS_DEV_VMD
&& !path_attached_to_hba(device
->path
, hba
->path
))
573 /* Multiple same type HBAs can be used if they share the same OROM */
574 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
576 if (device_orom
!= super
->orom
)
582 hba
->next
= alloc_intel_hba(device
);
586 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
588 struct sys_dev
*list
, *elem
;
591 if ((list
= find_intel_devices()) == NULL
)
595 disk_path
= (char *) devname
;
597 disk_path
= diskfd_to_devpath(fd
);
602 for (elem
= list
; elem
; elem
= elem
->next
)
603 if (path_attached_to_hba(disk_path
, elem
->path
))
606 if (disk_path
!= devname
)
612 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
615 static struct supertype
*match_metadata_desc_imsm(char *arg
)
617 struct supertype
*st
;
619 if (strcmp(arg
, "imsm") != 0 &&
620 strcmp(arg
, "default") != 0
624 st
= xcalloc(1, sizeof(*st
));
625 st
->ss
= &super_imsm
;
626 st
->max_devs
= IMSM_MAX_DEVICES
;
627 st
->minor_version
= 0;
633 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
635 return &mpb
->sig
[MPB_SIG_LEN
];
639 /* retrieve a disk directly from the anchor when the anchor is known to be
640 * up-to-date, currently only at load time
642 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
644 if (index
>= mpb
->num_disks
)
646 return &mpb
->disk
[index
];
649 /* retrieve the disk description based on a index of the disk
652 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
656 for (d
= super
->disks
; d
; d
= d
->next
)
657 if (d
->index
== index
)
662 /* retrieve a disk from the parsed metadata */
663 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
667 dl
= get_imsm_dl_disk(super
, index
);
674 /* generate a checksum directly from the anchor when the anchor is known to be
675 * up-to-date, currently only at load or write_super after coalescing
677 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
679 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
680 __u32
*p
= (__u32
*) mpb
;
684 sum
+= __le32_to_cpu(*p
);
688 return sum
- __le32_to_cpu(mpb
->check_sum
);
691 static size_t sizeof_imsm_map(struct imsm_map
*map
)
693 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
696 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
698 /* A device can have 2 maps if it is in the middle of a migration.
700 * MAP_0 - we return the first map
701 * MAP_1 - we return the second map if it exists, else NULL
702 * MAP_X - we return the second map if it exists, else the first
704 struct imsm_map
*map
= &dev
->vol
.map
[0];
705 struct imsm_map
*map2
= NULL
;
707 if (dev
->vol
.migr_state
)
708 map2
= (void *)map
+ sizeof_imsm_map(map
);
710 switch (second_map
) {
727 /* return the size of the device.
728 * migr_state increases the returned size if map[0] were to be duplicated
730 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
732 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
733 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
735 /* migrating means an additional map */
736 if (dev
->vol
.migr_state
)
737 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
739 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
745 /* retrieve disk serial number list from a metadata update */
746 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
749 struct disk_info
*inf
;
751 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
752 sizeof_imsm_dev(&update
->dev
, 0);
758 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
764 if (index
>= mpb
->num_raid_devs
)
767 /* devices start after all disks */
768 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
770 for (i
= 0; i
<= index
; i
++)
772 return _mpb
+ offset
;
774 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
779 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
781 struct intel_dev
*dv
;
783 if (index
>= super
->anchor
->num_raid_devs
)
785 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
786 if (dv
->index
== index
)
793 * == MAP_0 get first map
794 * == MAP_1 get second map
795 * == MAP_X than get map according to the current migr_state
797 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
801 struct imsm_map
*map
;
803 map
= get_imsm_map(dev
, second_map
);
805 /* top byte identifies disk under rebuild */
806 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
809 #define ord_to_idx(ord) (((ord) << 8) >> 8)
810 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
812 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
814 return ord_to_idx(ord
);
817 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
819 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
822 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
827 for (slot
= 0; slot
< map
->num_members
; slot
++) {
828 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
829 if (ord_to_idx(ord
) == idx
)
836 static int get_imsm_raid_level(struct imsm_map
*map
)
838 if (map
->raid_level
== 1) {
839 if (map
->num_members
== 2)
845 return map
->raid_level
;
848 static int cmp_extent(const void *av
, const void *bv
)
850 const struct extent
*a
= av
;
851 const struct extent
*b
= bv
;
852 if (a
->start
< b
->start
)
854 if (a
->start
> b
->start
)
859 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
864 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
865 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
866 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
868 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
875 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
877 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
879 if (lo
== 0 || hi
== 0)
881 *lo
= __le32_to_cpu((unsigned)n
);
882 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
886 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
888 return (unsigned long long)__le32_to_cpu(lo
) |
889 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
892 static unsigned long long total_blocks(struct imsm_disk
*disk
)
896 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
899 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
903 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
906 static unsigned long long blocks_per_member(struct imsm_map
*map
)
910 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
914 static unsigned long long num_data_stripes(struct imsm_map
*map
)
918 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
921 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
923 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
927 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
929 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
932 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
934 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
937 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
939 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
942 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
944 /* find a list of used extents on the given physical device */
945 struct extent
*rv
, *e
;
947 int memberships
= count_memberships(dl
, super
);
950 /* trim the reserved area for spares, so they can join any array
951 * regardless of whether the OROM has assigned sectors from the
952 * IMSM_RESERVED_SECTORS region
955 reservation
= imsm_min_reserved_sectors(super
);
957 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
959 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
962 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
963 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
964 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
966 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
967 e
->start
= pba_of_lba0(map
);
968 e
->size
= blocks_per_member(map
);
972 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
974 /* determine the start of the metadata
975 * when no raid devices are defined use the default
976 * ...otherwise allow the metadata to truncate the value
977 * as is the case with older versions of imsm
980 struct extent
*last
= &rv
[memberships
- 1];
981 unsigned long long remainder
;
983 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
984 /* round down to 1k block to satisfy precision of the kernel
988 /* make sure remainder is still sane */
989 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
990 remainder
= ROUND_UP(super
->len
, 512) >> 9;
991 if (reservation
> remainder
)
992 reservation
= remainder
;
994 e
->start
= total_blocks(&dl
->disk
) - reservation
;
999 /* try to determine how much space is reserved for metadata from
1000 * the last get_extents() entry, otherwise fallback to the
1003 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1009 /* for spares just return a minimal reservation which will grow
1010 * once the spare is picked up by an array
1012 if (dl
->index
== -1)
1013 return MPB_SECTOR_CNT
;
1015 e
= get_extents(super
, dl
);
1017 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1019 /* scroll to last entry */
1020 for (i
= 0; e
[i
].size
; i
++)
1023 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1030 static int is_spare(struct imsm_disk
*disk
)
1032 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1035 static int is_configured(struct imsm_disk
*disk
)
1037 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1040 static int is_failed(struct imsm_disk
*disk
)
1042 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1045 /* try to determine how much space is reserved for metadata from
1046 * the last get_extents() entry on the smallest active disk,
1047 * otherwise fallback to the default
1049 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1053 unsigned long long min_active
;
1055 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1056 struct dl
*dl
, *dl_min
= NULL
;
1062 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1065 unsigned long long blocks
= total_blocks(&dl
->disk
);
1066 if (blocks
< min_active
|| min_active
== 0) {
1068 min_active
= blocks
;
1074 /* find last lba used by subarrays on the smallest active disk */
1075 e
= get_extents(super
, dl_min
);
1078 for (i
= 0; e
[i
].size
; i
++)
1081 remainder
= min_active
- e
[i
].start
;
1084 /* to give priority to recovery we should not require full
1085 IMSM_RESERVED_SECTORS from the spare */
1086 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1088 /* if real reservation is smaller use that value */
1089 return (remainder
< rv
) ? remainder
: rv
;
1092 /* Return minimum size of a spare that can be used in this array*/
1093 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1095 struct intel_super
*super
= st
->sb
;
1099 unsigned long long rv
= 0;
1103 /* find first active disk in array */
1105 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1109 /* find last lba used by subarrays */
1110 e
= get_extents(super
, dl
);
1113 for (i
= 0; e
[i
].size
; i
++)
1116 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1119 /* add the amount of space needed for metadata */
1120 rv
= rv
+ imsm_min_reserved_sectors(super
);
1125 static int is_gen_migration(struct imsm_dev
*dev
);
1128 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1129 struct imsm_dev
*dev
);
1131 static void print_imsm_dev(struct intel_super
*super
,
1132 struct imsm_dev
*dev
,
1138 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1139 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1143 printf("[%.16s]:\n", dev
->volume
);
1144 printf(" UUID : %s\n", uuid
);
1145 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1147 printf(" <-- %d", get_imsm_raid_level(map2
));
1149 printf(" Members : %d", map
->num_members
);
1151 printf(" <-- %d", map2
->num_members
);
1153 printf(" Slots : [");
1154 for (i
= 0; i
< map
->num_members
; i
++) {
1155 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1156 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1161 for (i
= 0; i
< map2
->num_members
; i
++) {
1162 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1163 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1168 printf(" Failed disk : ");
1169 if (map
->failed_disk_num
== 0xff)
1172 printf("%i", map
->failed_disk_num
);
1174 slot
= get_imsm_disk_slot(map
, disk_idx
);
1176 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1177 printf(" This Slot : %d%s\n", slot
,
1178 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1180 printf(" This Slot : ?\n");
1181 sz
= __le32_to_cpu(dev
->size_high
);
1183 sz
+= __le32_to_cpu(dev
->size_low
);
1184 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1185 human_size(sz
* 512));
1186 sz
= blocks_per_member(map
);
1187 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1188 human_size(sz
* 512));
1189 printf(" Sector Offset : %llu\n",
1191 printf(" Num Stripes : %llu\n",
1192 num_data_stripes(map
));
1193 printf(" Chunk Size : %u KiB",
1194 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1196 printf(" <-- %u KiB",
1197 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1199 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1200 printf(" Migrate State : ");
1201 if (dev
->vol
.migr_state
) {
1202 if (migr_type(dev
) == MIGR_INIT
)
1203 printf("initialize\n");
1204 else if (migr_type(dev
) == MIGR_REBUILD
)
1205 printf("rebuild\n");
1206 else if (migr_type(dev
) == MIGR_VERIFY
)
1208 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1209 printf("general migration\n");
1210 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1211 printf("state change\n");
1212 else if (migr_type(dev
) == MIGR_REPAIR
)
1215 printf("<unknown:%d>\n", migr_type(dev
));
1218 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1219 if (dev
->vol
.migr_state
) {
1220 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1222 printf(" <-- %s", map_state_str
[map
->map_state
]);
1223 printf("\n Checkpoint : %u ",
1224 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1225 if ((is_gen_migration(dev
)) && ((slot
> 1) || (slot
< 0)))
1228 printf("(%llu)", (unsigned long long)
1229 blocks_per_migr_unit(super
, dev
));
1232 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1235 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1237 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1240 if (index
< -1 || !disk
)
1244 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1246 printf(" Disk%02d Serial : %s\n", index
, str
);
1248 printf(" Disk Serial : %s\n", str
);
1249 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1250 is_configured(disk
) ? " active" : "",
1251 is_failed(disk
) ? " failed" : "");
1252 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1253 sz
= total_blocks(disk
) - reserved
;
1254 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1255 human_size(sz
* 512));
1258 void examine_migr_rec_imsm(struct intel_super
*super
)
1260 struct migr_record
*migr_rec
= super
->migr_rec
;
1261 struct imsm_super
*mpb
= super
->anchor
;
1264 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1265 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1266 struct imsm_map
*map
;
1269 if (is_gen_migration(dev
) == 0)
1272 printf("\nMigration Record Information:");
1274 /* first map under migration */
1275 map
= get_imsm_map(dev
, MAP_0
);
1277 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1278 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1279 printf(" Empty\n ");
1280 printf("Examine one of first two disks in array\n");
1283 printf("\n Status : ");
1284 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1287 printf("Contains Data\n");
1288 printf(" Current Unit : %u\n",
1289 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1290 printf(" Family : %u\n",
1291 __le32_to_cpu(migr_rec
->family_num
));
1292 printf(" Ascending : %u\n",
1293 __le32_to_cpu(migr_rec
->ascending_migr
));
1294 printf(" Blocks Per Unit : %u\n",
1295 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1296 printf(" Dest. Depth Per Unit : %u\n",
1297 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1298 printf(" Checkpoint Area pba : %u\n",
1299 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1300 printf(" First member lba : %u\n",
1301 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1302 printf(" Total Number of Units : %u\n",
1303 __le32_to_cpu(migr_rec
->num_migr_units
));
1304 printf(" Size of volume : %u\n",
1305 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1306 printf(" Expansion space for LBA64 : %u\n",
1307 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1308 printf(" Record was read from : %u\n",
1309 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1314 #endif /* MDASSEMBLE */
1315 /*******************************************************************************
1316 * function: imsm_check_attributes
1317 * Description: Function checks if features represented by attributes flags
1318 * are supported by mdadm.
1320 * attributes - Attributes read from metadata
1322 * 0 - passed attributes contains unsupported features flags
1323 * 1 - all features are supported
1324 ******************************************************************************/
1325 static int imsm_check_attributes(__u32 attributes
)
1328 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1330 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1332 not_supported
&= attributes
;
1333 if (not_supported
) {
1334 pr_err("(IMSM): Unsupported attributes : %x\n",
1335 (unsigned)__le32_to_cpu(not_supported
));
1336 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1337 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1338 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1340 if (not_supported
& MPB_ATTRIB_2TB
) {
1341 dprintf("\t\tMPB_ATTRIB_2TB\n");
1342 not_supported
^= MPB_ATTRIB_2TB
;
1344 if (not_supported
& MPB_ATTRIB_RAID0
) {
1345 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1346 not_supported
^= MPB_ATTRIB_RAID0
;
1348 if (not_supported
& MPB_ATTRIB_RAID1
) {
1349 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1350 not_supported
^= MPB_ATTRIB_RAID1
;
1352 if (not_supported
& MPB_ATTRIB_RAID10
) {
1353 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1354 not_supported
^= MPB_ATTRIB_RAID10
;
1356 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1357 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1358 not_supported
^= MPB_ATTRIB_RAID1E
;
1360 if (not_supported
& MPB_ATTRIB_RAID5
) {
1361 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1362 not_supported
^= MPB_ATTRIB_RAID5
;
1364 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1365 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1366 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1368 if (not_supported
& MPB_ATTRIB_BBM
) {
1369 dprintf("\t\tMPB_ATTRIB_BBM\n");
1370 not_supported
^= MPB_ATTRIB_BBM
;
1372 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1373 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1374 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1376 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1377 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1378 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1380 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1381 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1382 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1384 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1385 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1386 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1388 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1389 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1390 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1394 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1403 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1405 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1407 struct intel_super
*super
= st
->sb
;
1408 struct imsm_super
*mpb
= super
->anchor
;
1409 char str
[MAX_SIGNATURE_LENGTH
];
1414 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1417 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1418 printf(" Magic : %s\n", str
);
1419 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1420 printf(" Version : %s\n", get_imsm_version(mpb
));
1421 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1422 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1423 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1424 printf(" Attributes : ");
1425 if (imsm_check_attributes(mpb
->attributes
))
1426 printf("All supported\n");
1428 printf("not supported\n");
1429 getinfo_super_imsm(st
, &info
, NULL
);
1430 fname_from_uuid(st
, &info
, nbuf
, ':');
1431 printf(" UUID : %s\n", nbuf
+ 5);
1432 sum
= __le32_to_cpu(mpb
->check_sum
);
1433 printf(" Checksum : %08x %s\n", sum
,
1434 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1435 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1436 printf(" Disks : %d\n", mpb
->num_disks
);
1437 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1438 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1439 if (super
->bbm_log
) {
1440 struct bbm_log
*log
= super
->bbm_log
;
1443 printf("Bad Block Management Log:\n");
1444 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1445 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1446 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1447 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1448 printf(" First Spare : %llx\n",
1449 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1451 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1453 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1455 super
->current_vol
= i
;
1456 getinfo_super_imsm(st
, &info
, NULL
);
1457 fname_from_uuid(st
, &info
, nbuf
, ':');
1458 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1460 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1461 if (i
== super
->disks
->index
)
1463 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1466 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1467 if (dl
->index
== -1)
1468 print_imsm_disk(&dl
->disk
, -1, reserved
);
1470 examine_migr_rec_imsm(super
);
1473 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1475 /* We just write a generic IMSM ARRAY entry */
1478 struct intel_super
*super
= st
->sb
;
1480 if (!super
->anchor
->num_raid_devs
) {
1481 printf("ARRAY metadata=imsm\n");
1485 getinfo_super_imsm(st
, &info
, NULL
);
1486 fname_from_uuid(st
, &info
, nbuf
, ':');
1487 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1490 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1492 /* We just write a generic IMSM ARRAY entry */
1496 struct intel_super
*super
= st
->sb
;
1499 if (!super
->anchor
->num_raid_devs
)
1502 getinfo_super_imsm(st
, &info
, NULL
);
1503 fname_from_uuid(st
, &info
, nbuf
, ':');
1504 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1505 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1507 super
->current_vol
= i
;
1508 getinfo_super_imsm(st
, &info
, NULL
);
1509 fname_from_uuid(st
, &info
, nbuf1
, ':');
1510 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1511 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1515 static void export_examine_super_imsm(struct supertype
*st
)
1517 struct intel_super
*super
= st
->sb
;
1518 struct imsm_super
*mpb
= super
->anchor
;
1522 getinfo_super_imsm(st
, &info
, NULL
);
1523 fname_from_uuid(st
, &info
, nbuf
, ':');
1524 printf("MD_METADATA=imsm\n");
1525 printf("MD_LEVEL=container\n");
1526 printf("MD_UUID=%s\n", nbuf
+5);
1527 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1530 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
1532 /* The second last 512byte sector of the device contains
1533 * the "struct imsm_super" metadata.
1534 * This contains mpb_size which is the size in bytes of the
1535 * extended metadata. This is located immediately before
1537 * We want to read all that, plus the last sector which
1538 * may contain a migration record, and write it all
1542 unsigned long long dsize
, offset
;
1544 struct imsm_super
*sb
;
1547 if (posix_memalign(&buf
, 4096, 4096) != 0)
1550 if (!get_dev_size(from
, NULL
, &dsize
))
1553 if (lseek64(from
, dsize
-1024, 0) < 0)
1555 if (read(from
, buf
, 512) != 512)
1558 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
1561 sectors
= mpb_sectors(sb
) + 2;
1562 offset
= dsize
- sectors
* 512;
1563 if (lseek64(from
, offset
, 0) < 0 ||
1564 lseek64(to
, offset
, 0) < 0)
1566 while (written
< sectors
* 512) {
1567 int n
= sectors
*512 - written
;
1570 if (read(from
, buf
, n
) != n
)
1572 if (write(to
, buf
, n
) != n
)
1583 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1588 getinfo_super_imsm(st
, &info
, NULL
);
1589 fname_from_uuid(st
, &info
, nbuf
, ':');
1590 printf("\n UUID : %s\n", nbuf
+ 5);
1593 static void brief_detail_super_imsm(struct supertype
*st
)
1597 getinfo_super_imsm(st
, &info
, NULL
);
1598 fname_from_uuid(st
, &info
, nbuf
, ':');
1599 printf(" UUID=%s", nbuf
+ 5);
1602 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1603 static void fd2devname(int fd
, char *name
);
1605 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1607 /* dump an unsorted list of devices attached to AHCI Intel storage
1608 * controller, as well as non-connected ports
1610 int hba_len
= strlen(hba_path
) + 1;
1615 unsigned long port_mask
= (1 << port_count
) - 1;
1617 if (port_count
> (int)sizeof(port_mask
) * 8) {
1619 pr_err("port_count %d out of range\n", port_count
);
1623 /* scroll through /sys/dev/block looking for devices attached to
1626 dir
= opendir("/sys/dev/block");
1627 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1638 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1640 path
= devt_to_devpath(makedev(major
, minor
));
1643 if (!path_attached_to_hba(path
, hba_path
)) {
1649 /* retrieve the scsi device type */
1650 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1652 pr_err("failed to allocate 'device'\n");
1656 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1657 if (load_sys(device
, buf
) != 0) {
1659 pr_err("failed to read device type for %s\n",
1665 type
= strtoul(buf
, NULL
, 10);
1667 /* if it's not a disk print the vendor and model */
1668 if (!(type
== 0 || type
== 7 || type
== 14)) {
1671 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1672 if (load_sys(device
, buf
) == 0) {
1673 strncpy(vendor
, buf
, sizeof(vendor
));
1674 vendor
[sizeof(vendor
) - 1] = '\0';
1675 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1676 while (isspace(*c
) || *c
== '\0')
1680 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1681 if (load_sys(device
, buf
) == 0) {
1682 strncpy(model
, buf
, sizeof(model
));
1683 model
[sizeof(model
) - 1] = '\0';
1684 c
= (char *) &model
[sizeof(model
) - 1];
1685 while (isspace(*c
) || *c
== '\0')
1689 if (vendor
[0] && model
[0])
1690 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1692 switch (type
) { /* numbers from hald/linux/device.c */
1693 case 1: sprintf(buf
, "tape"); break;
1694 case 2: sprintf(buf
, "printer"); break;
1695 case 3: sprintf(buf
, "processor"); break;
1697 case 5: sprintf(buf
, "cdrom"); break;
1698 case 6: sprintf(buf
, "scanner"); break;
1699 case 8: sprintf(buf
, "media_changer"); break;
1700 case 9: sprintf(buf
, "comm"); break;
1701 case 12: sprintf(buf
, "raid"); break;
1702 default: sprintf(buf
, "unknown");
1708 /* chop device path to 'host%d' and calculate the port number */
1709 c
= strchr(&path
[hba_len
], '/');
1712 pr_err("%s - invalid path name\n", path
+ hba_len
);
1717 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
1718 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
1722 *c
= '/'; /* repair the full string */
1723 pr_err("failed to determine port number for %s\n",
1730 /* mark this port as used */
1731 port_mask
&= ~(1 << port
);
1733 /* print out the device information */
1735 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1739 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1741 printf(" Port%d : - disk info unavailable -\n", port
);
1743 fd2devname(fd
, buf
);
1744 printf(" Port%d : %s", port
, buf
);
1745 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1746 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1761 for (i
= 0; i
< port_count
; i
++)
1762 if (port_mask
& (1 << i
))
1763 printf(" Port%d : - no device attached -\n", i
);
1769 static int print_vmd_attached_devs(struct sys_dev
*hba
)
1777 if (hba
->type
!= SYS_DEV_VMD
)
1780 /* scroll through /sys/dev/block looking for devices attached to
1783 dir
= opendir("/sys/bus/pci/drivers/nvme");
1784 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1787 /* is 'ent' a device? check that the 'subsystem' link exists and
1788 * that its target matches 'bus'
1790 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
1792 n
= readlink(path
, link
, sizeof(link
));
1793 if (n
< 0 || n
>= (int)sizeof(link
))
1796 c
= strrchr(link
, '/');
1799 if (strncmp("pci", c
+1, strlen("pci")) != 0)
1802 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
1803 /* if not a intel NVMe - skip it*/
1804 if (devpath_to_vendor(path
) != 0x8086)
1807 rp
= realpath(path
, NULL
);
1811 if (path_attached_to_hba(rp
, hba
->path
)) {
1812 printf(" NVMe under VMD : %s\n", rp
);
1820 static void print_found_intel_controllers(struct sys_dev
*elem
)
1822 for (; elem
; elem
= elem
->next
) {
1823 pr_err("found Intel(R) ");
1824 if (elem
->type
== SYS_DEV_SATA
)
1825 fprintf(stderr
, "SATA ");
1826 else if (elem
->type
== SYS_DEV_SAS
)
1827 fprintf(stderr
, "SAS ");
1828 else if (elem
->type
== SYS_DEV_NVME
)
1829 fprintf(stderr
, "NVMe ");
1831 if (elem
->type
== SYS_DEV_VMD
)
1832 fprintf(stderr
, "VMD domain");
1834 fprintf(stderr
, "RAID controller");
1837 fprintf(stderr
, " at %s", elem
->pci_id
);
1838 fprintf(stderr
, ".\n");
1843 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1850 if ((dir
= opendir(hba_path
)) == NULL
)
1853 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1856 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
1857 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
1859 if (*port_count
== 0)
1861 else if (host
< host_base
)
1864 if (host
+ 1 > *port_count
+ host_base
)
1865 *port_count
= host
+ 1 - host_base
;
1871 static void print_imsm_capability(const struct imsm_orom
*orom
)
1873 printf(" Platform : Intel(R) ");
1874 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
1875 printf("Matrix Storage Manager\n");
1877 printf("Rapid Storage Technology%s\n",
1878 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
1879 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
1880 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
1881 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
1882 printf(" RAID Levels :%s%s%s%s%s\n",
1883 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1884 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1885 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1886 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1887 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1888 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1889 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1890 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1891 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1892 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1893 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1894 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1895 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1896 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1897 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1898 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1899 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1900 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1901 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1902 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1903 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1904 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1905 printf(" 2TB volumes :%s supported\n",
1906 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
1907 printf(" 2TB disks :%s supported\n",
1908 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
1909 printf(" Max Disks : %d\n", orom
->tds
);
1910 printf(" Max Volumes : %d per array, %d per %s\n",
1911 orom
->vpa
, orom
->vphba
,
1912 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
1916 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
1918 printf("MD_FIRMWARE_TYPE=imsm\n");
1919 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
1920 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1921 orom
->hotfix_ver
, orom
->build
);
1922 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
1923 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
1924 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
1925 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
1926 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
1927 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
1928 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1929 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
1930 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
1931 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
1932 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
1933 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
1934 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
1935 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
1936 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
1937 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
1938 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
1939 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
1940 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
1941 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
1942 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
1943 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
1944 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
1945 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
1946 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
1947 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
1948 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
1949 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
1952 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
1954 /* There are two components to imsm platform support, the ahci SATA
1955 * controller and the option-rom. To find the SATA controller we
1956 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1957 * controller with the Intel vendor id is present. This approach
1958 * allows mdadm to leverage the kernel's ahci detection logic, with the
1959 * caveat that if ahci.ko is not loaded mdadm will not be able to
1960 * detect platform raid capabilities. The option-rom resides in a
1961 * platform "Adapter ROM". We scan for its signature to retrieve the
1962 * platform capabilities. If raid support is disabled in the BIOS the
1963 * option-rom capability structure will not be available.
1965 struct sys_dev
*list
, *hba
;
1970 if (enumerate_only
) {
1971 if (check_env("IMSM_NO_PLATFORM"))
1973 list
= find_intel_devices();
1976 for (hba
= list
; hba
; hba
= hba
->next
) {
1977 if (find_imsm_capability(hba
)) {
1987 list
= find_intel_devices();
1990 pr_err("no active Intel(R) RAID controller found.\n");
1992 } else if (verbose
> 0)
1993 print_found_intel_controllers(list
);
1995 for (hba
= list
; hba
; hba
= hba
->next
) {
1996 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
1998 if (!find_imsm_capability(hba
)) {
2000 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2001 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2002 get_sys_dev_type(hba
->type
));
2008 if (controller_path
&& result
== 1) {
2009 pr_err("no active Intel(R) RAID controller found under %s\n",
2014 const struct orom_entry
*entry
;
2016 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2017 if (entry
->type
== SYS_DEV_VMD
) {
2018 for (hba
= list
; hba
; hba
= hba
->next
) {
2019 if (hba
->type
== SYS_DEV_VMD
) {
2021 print_imsm_capability(&entry
->orom
);
2022 printf(" I/O Controller : %s (%s)\n",
2023 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2024 print_vmd_attached_devs(hba
);
2031 print_imsm_capability(&entry
->orom
);
2032 if (entry
->type
== SYS_DEV_NVME
) {
2033 for (hba
= list
; hba
; hba
= hba
->next
) {
2034 if (hba
->type
== SYS_DEV_NVME
)
2035 printf(" NVMe Device : %s\n", hba
->path
);
2041 struct devid_list
*devid
;
2042 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2043 hba
= device_by_id(devid
->devid
);
2047 printf(" I/O Controller : %s (%s)\n",
2048 hba
->path
, get_sys_dev_type(hba
->type
));
2049 if (hba
->type
== SYS_DEV_SATA
) {
2050 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2051 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2053 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2064 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2066 struct sys_dev
*list
, *hba
;
2069 list
= find_intel_devices();
2072 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2077 for (hba
= list
; hba
; hba
= hba
->next
) {
2078 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2080 if (!find_imsm_capability(hba
) && verbose
> 0) {
2082 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2083 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2089 const struct orom_entry
*entry
;
2091 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2092 if (entry
->type
== SYS_DEV_VMD
) {
2093 for (hba
= list
; hba
; hba
= hba
->next
)
2094 print_imsm_capability_export(&entry
->orom
);
2097 print_imsm_capability_export(&entry
->orom
);
2105 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2107 /* the imsm metadata format does not specify any host
2108 * identification information. We return -1 since we can never
2109 * confirm nor deny whether a given array is "meant" for this
2110 * host. We rely on compare_super and the 'family_num' fields to
2111 * exclude member disks that do not belong, and we rely on
2112 * mdadm.conf to specify the arrays that should be assembled.
2113 * Auto-assembly may still pick up "foreign" arrays.
2119 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2121 /* The uuid returned here is used for:
2122 * uuid to put into bitmap file (Create, Grow)
2123 * uuid for backup header when saving critical section (Grow)
2124 * comparing uuids when re-adding a device into an array
2125 * In these cases the uuid required is that of the data-array,
2126 * not the device-set.
2127 * uuid to recognise same set when adding a missing device back
2128 * to an array. This is a uuid for the device-set.
2130 * For each of these we can make do with a truncated
2131 * or hashed uuid rather than the original, as long as
2133 * In each case the uuid required is that of the data-array,
2134 * not the device-set.
2136 /* imsm does not track uuid's so we synthesis one using sha1 on
2137 * - The signature (Which is constant for all imsm array, but no matter)
2138 * - the orig_family_num of the container
2139 * - the index number of the volume
2140 * - the 'serial' number of the volume.
2141 * Hopefully these are all constant.
2143 struct intel_super
*super
= st
->sb
;
2146 struct sha1_ctx ctx
;
2147 struct imsm_dev
*dev
= NULL
;
2150 /* some mdadm versions failed to set ->orig_family_num, in which
2151 * case fall back to ->family_num. orig_family_num will be
2152 * fixed up with the first metadata update.
2154 family_num
= super
->anchor
->orig_family_num
;
2155 if (family_num
== 0)
2156 family_num
= super
->anchor
->family_num
;
2157 sha1_init_ctx(&ctx
);
2158 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2159 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2160 if (super
->current_vol
>= 0)
2161 dev
= get_imsm_dev(super
, super
->current_vol
);
2163 __u32 vol
= super
->current_vol
;
2164 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2165 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2167 sha1_finish_ctx(&ctx
, buf
);
2168 memcpy(uuid
, buf
, 4*4);
2173 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2175 __u8
*v
= get_imsm_version(mpb
);
2176 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2177 char major
[] = { 0, 0, 0 };
2178 char minor
[] = { 0 ,0, 0 };
2179 char patch
[] = { 0, 0, 0 };
2180 char *ver_parse
[] = { major
, minor
, patch
};
2184 while (*v
!= '\0' && v
< end
) {
2185 if (*v
!= '.' && j
< 2)
2186 ver_parse
[i
][j
++] = *v
;
2194 *m
= strtol(minor
, NULL
, 0);
2195 *p
= strtol(patch
, NULL
, 0);
2199 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2201 /* migr_strip_size when repairing or initializing parity */
2202 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2203 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2205 switch (get_imsm_raid_level(map
)) {
2210 return 128*1024 >> 9;
2214 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2216 /* migr_strip_size when rebuilding a degraded disk, no idea why
2217 * this is different than migr_strip_size_resync(), but it's good
2220 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2221 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2223 switch (get_imsm_raid_level(map
)) {
2226 if (map
->num_members
% map
->num_domains
== 0)
2227 return 128*1024 >> 9;
2231 return max((__u32
) 64*1024 >> 9, chunk
);
2233 return 128*1024 >> 9;
2237 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2239 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2240 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2241 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2242 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2244 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2247 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2249 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2250 int level
= get_imsm_raid_level(lo
);
2252 if (level
== 1 || level
== 10) {
2253 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2255 return hi
->num_domains
;
2257 return num_stripes_per_unit_resync(dev
);
2260 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2262 /* named 'imsm_' because raid0, raid1 and raid10
2263 * counter-intuitively have the same number of data disks
2265 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2267 switch (get_imsm_raid_level(map
)) {
2269 return map
->num_members
;
2273 return map
->num_members
/2;
2275 return map
->num_members
- 1;
2277 dprintf("unsupported raid level\n");
2282 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2284 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2285 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2287 switch(get_imsm_raid_level(map
)) {
2290 return chunk
* map
->num_domains
;
2292 return chunk
* map
->num_members
;
2298 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2300 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2301 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2302 __u32 strip
= block
/ chunk
;
2304 switch (get_imsm_raid_level(map
)) {
2307 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2308 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2310 return vol_stripe
* chunk
+ block
% chunk
;
2312 __u32 stripe
= strip
/ (map
->num_members
- 1);
2314 return stripe
* chunk
+ block
% chunk
;
2321 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2322 struct imsm_dev
*dev
)
2324 /* calculate the conversion factor between per member 'blocks'
2325 * (md/{resync,rebuild}_start) and imsm migration units, return
2326 * 0 for the 'not migrating' and 'unsupported migration' cases
2328 if (!dev
->vol
.migr_state
)
2331 switch (migr_type(dev
)) {
2332 case MIGR_GEN_MIGR
: {
2333 struct migr_record
*migr_rec
= super
->migr_rec
;
2334 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2339 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2340 __u32 stripes_per_unit
;
2341 __u32 blocks_per_unit
;
2350 /* yes, this is really the translation of migr_units to
2351 * per-member blocks in the 'resync' case
2353 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2354 migr_chunk
= migr_strip_blocks_resync(dev
);
2355 disks
= imsm_num_data_members(dev
, MAP_0
);
2356 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2357 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2358 segment
= blocks_per_unit
/ stripe
;
2359 block_rel
= blocks_per_unit
- segment
* stripe
;
2360 parity_depth
= parity_segment_depth(dev
);
2361 block_map
= map_migr_block(dev
, block_rel
);
2362 return block_map
+ parity_depth
* segment
;
2364 case MIGR_REBUILD
: {
2365 __u32 stripes_per_unit
;
2368 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2369 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2370 return migr_chunk
* stripes_per_unit
;
2372 case MIGR_STATE_CHANGE
:
2378 static int imsm_level_to_layout(int level
)
2386 return ALGORITHM_LEFT_ASYMMETRIC
;
2393 /*******************************************************************************
2394 * Function: read_imsm_migr_rec
2395 * Description: Function reads imsm migration record from last sector of disk
2397 * fd : disk descriptor
2398 * super : metadata info
2402 ******************************************************************************/
2403 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2406 unsigned long long dsize
;
2408 get_dev_size(fd
, NULL
, &dsize
);
2409 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2410 pr_err("Cannot seek to anchor block: %s\n",
2414 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2415 MIGR_REC_BUF_SIZE
) {
2416 pr_err("Cannot read migr record block: %s\n",
2426 static struct imsm_dev
*imsm_get_device_during_migration(
2427 struct intel_super
*super
)
2430 struct intel_dev
*dv
;
2432 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2433 if (is_gen_migration(dv
->dev
))
2439 /*******************************************************************************
2440 * Function: load_imsm_migr_rec
2441 * Description: Function reads imsm migration record (it is stored at the last
2444 * super : imsm internal array info
2445 * info : general array info
2449 * -2 : no migration in progress
2450 ******************************************************************************/
2451 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2454 struct dl
*dl
= NULL
;
2458 struct imsm_dev
*dev
;
2459 struct imsm_map
*map
= NULL
;
2462 /* find map under migration */
2463 dev
= imsm_get_device_during_migration(super
);
2464 /* nothing to load,no migration in progress?
2468 map
= get_imsm_map(dev
, MAP_0
);
2471 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2472 /* skip spare and failed disks
2474 if (sd
->disk
.raid_disk
< 0)
2476 /* read only from one of the first two slots */
2478 slot
= get_imsm_disk_slot(map
,
2479 sd
->disk
.raid_disk
);
2480 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2483 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2484 fd
= dev_open(nm
, O_RDONLY
);
2490 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2491 /* skip spare and failed disks
2495 /* read only from one of the first two slots */
2497 slot
= get_imsm_disk_slot(map
, dl
->index
);
2498 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2500 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2501 fd
= dev_open(nm
, O_RDONLY
);
2508 retval
= read_imsm_migr_rec(fd
, super
);
2517 /*******************************************************************************
2518 * function: imsm_create_metadata_checkpoint_update
2519 * Description: It creates update for checkpoint change.
2521 * super : imsm internal array info
2522 * u : pointer to prepared update
2525 * If length is equal to 0, input pointer u contains no update
2526 ******************************************************************************/
2527 static int imsm_create_metadata_checkpoint_update(
2528 struct intel_super
*super
,
2529 struct imsm_update_general_migration_checkpoint
**u
)
2532 int update_memory_size
= 0;
2534 dprintf("(enter)\n");
2540 /* size of all update data without anchor */
2541 update_memory_size
=
2542 sizeof(struct imsm_update_general_migration_checkpoint
);
2544 *u
= xcalloc(1, update_memory_size
);
2546 dprintf("error: cannot get memory\n");
2549 (*u
)->type
= update_general_migration_checkpoint
;
2550 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2551 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
2553 return update_memory_size
;
2556 static void imsm_update_metadata_locally(struct supertype
*st
,
2557 void *buf
, int len
);
2559 /*******************************************************************************
2560 * Function: write_imsm_migr_rec
2561 * Description: Function writes imsm migration record
2562 * (at the last sector of disk)
2564 * super : imsm internal array info
2568 ******************************************************************************/
2569 static int write_imsm_migr_rec(struct supertype
*st
)
2571 struct intel_super
*super
= st
->sb
;
2572 unsigned long long dsize
;
2578 struct imsm_update_general_migration_checkpoint
*u
;
2579 struct imsm_dev
*dev
;
2580 struct imsm_map
*map
= NULL
;
2582 /* find map under migration */
2583 dev
= imsm_get_device_during_migration(super
);
2584 /* if no migration, write buffer anyway to clear migr_record
2585 * on disk based on first available device
2588 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2589 super
->current_vol
);
2591 map
= get_imsm_map(dev
, MAP_0
);
2593 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2596 /* skip failed and spare devices */
2599 /* write to 2 first slots only */
2601 slot
= get_imsm_disk_slot(map
, sd
->index
);
2602 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2605 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2606 fd
= dev_open(nm
, O_RDWR
);
2609 get_dev_size(fd
, NULL
, &dsize
);
2610 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2611 pr_err("Cannot seek to anchor block: %s\n",
2615 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2616 MIGR_REC_BUF_SIZE
) {
2617 pr_err("Cannot write migr record block: %s\n",
2624 /* update checkpoint information in metadata */
2625 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2628 dprintf("imsm: Cannot prepare update\n");
2631 /* update metadata locally */
2632 imsm_update_metadata_locally(st
, u
, len
);
2633 /* and possibly remotely */
2634 if (st
->update_tail
) {
2635 append_metadata_update(st
, u
, len
);
2636 /* during reshape we do all work inside metadata handler
2637 * manage_reshape(), so metadata update has to be triggered
2640 flush_metadata_updates(st
);
2641 st
->update_tail
= &st
->updates
;
2651 #endif /* MDASSEMBLE */
2653 /* spare/missing disks activations are not allowe when
2654 * array/container performs reshape operation, because
2655 * all arrays in container works on the same disks set
2657 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2660 struct intel_dev
*i_dev
;
2661 struct imsm_dev
*dev
;
2663 /* check whole container
2665 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2667 if (is_gen_migration(dev
)) {
2668 /* No repair during any migration in container
2676 static unsigned long long imsm_component_size_aligment_check(int level
,
2678 unsigned long long component_size
)
2680 unsigned int component_size_alligment
;
2682 /* check component size aligment
2684 component_size_alligment
= component_size
% (chunk_size
/512);
2686 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
2687 level
, chunk_size
, component_size
,
2688 component_size_alligment
);
2690 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
2691 dprintf("imsm: reported component size alligned from %llu ",
2693 component_size
-= component_size_alligment
;
2694 dprintf_cont("to %llu (%i).\n",
2695 component_size
, component_size_alligment
);
2698 return component_size
;
2701 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2703 struct intel_super
*super
= st
->sb
;
2704 struct migr_record
*migr_rec
= super
->migr_rec
;
2705 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2706 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2707 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2708 struct imsm_map
*map_to_analyse
= map
;
2710 int map_disks
= info
->array
.raid_disks
;
2712 memset(info
, 0, sizeof(*info
));
2714 map_to_analyse
= prev_map
;
2716 dl
= super
->current_disk
;
2718 info
->container_member
= super
->current_vol
;
2719 info
->array
.raid_disks
= map
->num_members
;
2720 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2721 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2722 info
->array
.md_minor
= -1;
2723 info
->array
.ctime
= 0;
2724 info
->array
.utime
= 0;
2725 info
->array
.chunk_size
=
2726 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2727 info
->array
.state
= !dev
->vol
.dirty
;
2728 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2729 info
->custom_array_size
<<= 32;
2730 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2731 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2733 if (is_gen_migration(dev
)) {
2734 info
->reshape_active
= 1;
2735 info
->new_level
= get_imsm_raid_level(map
);
2736 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2737 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2738 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2739 if (info
->delta_disks
) {
2740 /* this needs to be applied to every array
2743 info
->reshape_active
= CONTAINER_RESHAPE
;
2745 /* We shape information that we give to md might have to be
2746 * modify to cope with md's requirement for reshaping arrays.
2747 * For example, when reshaping a RAID0, md requires it to be
2748 * presented as a degraded RAID4.
2749 * Also if a RAID0 is migrating to a RAID5 we need to specify
2750 * the array as already being RAID5, but the 'before' layout
2751 * is a RAID4-like layout.
2753 switch (info
->array
.level
) {
2755 switch(info
->new_level
) {
2757 /* conversion is happening as RAID4 */
2758 info
->array
.level
= 4;
2759 info
->array
.raid_disks
+= 1;
2762 /* conversion is happening as RAID5 */
2763 info
->array
.level
= 5;
2764 info
->array
.layout
= ALGORITHM_PARITY_N
;
2765 info
->delta_disks
-= 1;
2768 /* FIXME error message */
2769 info
->array
.level
= UnSet
;
2775 info
->new_level
= UnSet
;
2776 info
->new_layout
= UnSet
;
2777 info
->new_chunk
= info
->array
.chunk_size
;
2778 info
->delta_disks
= 0;
2782 info
->disk
.major
= dl
->major
;
2783 info
->disk
.minor
= dl
->minor
;
2784 info
->disk
.number
= dl
->index
;
2785 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2789 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2790 info
->component_size
= blocks_per_member(map_to_analyse
);
2792 info
->component_size
= imsm_component_size_aligment_check(
2794 info
->array
.chunk_size
,
2795 info
->component_size
);
2797 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2798 info
->recovery_start
= MaxSector
;
2800 info
->reshape_progress
= 0;
2801 info
->resync_start
= MaxSector
;
2802 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2804 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2805 info
->resync_start
= 0;
2807 if (dev
->vol
.migr_state
) {
2808 switch (migr_type(dev
)) {
2811 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2813 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2815 info
->resync_start
= blocks_per_unit
* units
;
2818 case MIGR_GEN_MIGR
: {
2819 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2821 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2822 unsigned long long array_blocks
;
2825 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2827 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2828 (super
->migr_rec
->rec_status
==
2829 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2832 info
->reshape_progress
= blocks_per_unit
* units
;
2834 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
2835 (unsigned long long)units
,
2836 (unsigned long long)blocks_per_unit
,
2837 info
->reshape_progress
);
2839 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2840 if (used_disks
> 0) {
2841 array_blocks
= blocks_per_member(map
) *
2843 /* round array size down to closest MB
2845 info
->custom_array_size
= (array_blocks
2846 >> SECT_PER_MB_SHIFT
)
2847 << SECT_PER_MB_SHIFT
;
2851 /* we could emulate the checkpointing of
2852 * 'sync_action=check' migrations, but for now
2853 * we just immediately complete them
2856 /* this is handled by container_content_imsm() */
2857 case MIGR_STATE_CHANGE
:
2858 /* FIXME handle other migrations */
2860 /* we are not dirty, so... */
2861 info
->resync_start
= MaxSector
;
2865 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2866 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2868 info
->array
.major_version
= -1;
2869 info
->array
.minor_version
= -2;
2870 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
2871 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2872 uuid_from_super_imsm(st
, info
->uuid
);
2876 for (i
=0; i
<map_disks
; i
++) {
2878 if (i
< info
->array
.raid_disks
) {
2879 struct imsm_disk
*dsk
;
2880 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2881 dsk
= get_imsm_disk(super
, j
);
2882 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2889 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2890 int failed
, int look_in_map
);
2892 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2896 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2898 if (is_gen_migration(dev
)) {
2901 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2903 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2904 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2905 if (map2
->map_state
!= map_state
) {
2906 map2
->map_state
= map_state
;
2907 super
->updates_pending
++;
2913 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2917 for (d
= super
->missing
; d
; d
= d
->next
)
2918 if (d
->index
== index
)
2923 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2925 struct intel_super
*super
= st
->sb
;
2926 struct imsm_disk
*disk
;
2927 int map_disks
= info
->array
.raid_disks
;
2928 int max_enough
= -1;
2930 struct imsm_super
*mpb
;
2932 if (super
->current_vol
>= 0) {
2933 getinfo_super_imsm_volume(st
, info
, map
);
2936 memset(info
, 0, sizeof(*info
));
2938 /* Set raid_disks to zero so that Assemble will always pull in valid
2941 info
->array
.raid_disks
= 0;
2942 info
->array
.level
= LEVEL_CONTAINER
;
2943 info
->array
.layout
= 0;
2944 info
->array
.md_minor
= -1;
2945 info
->array
.ctime
= 0; /* N/A for imsm */
2946 info
->array
.utime
= 0;
2947 info
->array
.chunk_size
= 0;
2949 info
->disk
.major
= 0;
2950 info
->disk
.minor
= 0;
2951 info
->disk
.raid_disk
= -1;
2952 info
->reshape_active
= 0;
2953 info
->array
.major_version
= -1;
2954 info
->array
.minor_version
= -2;
2955 strcpy(info
->text_version
, "imsm");
2956 info
->safe_mode_delay
= 0;
2957 info
->disk
.number
= -1;
2958 info
->disk
.state
= 0;
2960 info
->recovery_start
= MaxSector
;
2961 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2963 /* do we have the all the insync disks that we expect? */
2964 mpb
= super
->anchor
;
2966 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2967 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2968 int failed
, enough
, j
, missing
= 0;
2969 struct imsm_map
*map
;
2972 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2973 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2974 map
= get_imsm_map(dev
, MAP_0
);
2976 /* any newly missing disks?
2977 * (catches single-degraded vs double-degraded)
2979 for (j
= 0; j
< map
->num_members
; j
++) {
2980 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2981 __u32 idx
= ord_to_idx(ord
);
2983 if (!(ord
& IMSM_ORD_REBUILD
) &&
2984 get_imsm_missing(super
, idx
)) {
2990 if (state
== IMSM_T_STATE_FAILED
)
2992 else if (state
== IMSM_T_STATE_DEGRADED
&&
2993 (state
!= map
->map_state
|| missing
))
2995 else /* we're normal, or already degraded */
2997 if (is_gen_migration(dev
) && missing
) {
2998 /* during general migration we need all disks
2999 * that process is running on.
3000 * No new missing disk is allowed.
3004 /* no more checks necessary
3008 /* in the missing/failed disk case check to see
3009 * if at least one array is runnable
3011 max_enough
= max(max_enough
, enough
);
3013 dprintf("enough: %d\n", max_enough
);
3014 info
->container_enough
= max_enough
;
3017 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3019 disk
= &super
->disks
->disk
;
3020 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3021 info
->component_size
= reserved
;
3022 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3023 /* we don't change info->disk.raid_disk here because
3024 * this state will be finalized in mdmon after we have
3025 * found the 'most fresh' version of the metadata
3027 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3028 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3031 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3032 * ->compare_super may have updated the 'num_raid_devs' field for spares
3034 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3035 uuid_from_super_imsm(st
, info
->uuid
);
3037 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3039 /* I don't know how to compute 'map' on imsm, so use safe default */
3042 for (i
= 0; i
< map_disks
; i
++)
3048 /* allocates memory and fills disk in mdinfo structure
3049 * for each disk in array */
3050 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3052 struct mdinfo
*mddev
= NULL
;
3053 struct intel_super
*super
= st
->sb
;
3054 struct imsm_disk
*disk
;
3057 if (!super
|| !super
->disks
)
3060 mddev
= xcalloc(1, sizeof(*mddev
));
3064 tmp
= xcalloc(1, sizeof(*tmp
));
3066 tmp
->next
= mddev
->devs
;
3068 tmp
->disk
.number
= count
++;
3069 tmp
->disk
.major
= dl
->major
;
3070 tmp
->disk
.minor
= dl
->minor
;
3071 tmp
->disk
.state
= is_configured(disk
) ?
3072 (1 << MD_DISK_ACTIVE
) : 0;
3073 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3074 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3075 tmp
->disk
.raid_disk
= -1;
3081 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3082 char *update
, char *devname
, int verbose
,
3083 int uuid_set
, char *homehost
)
3085 /* For 'assemble' and 'force' we need to return non-zero if any
3086 * change was made. For others, the return value is ignored.
3087 * Update options are:
3088 * force-one : This device looks a bit old but needs to be included,
3089 * update age info appropriately.
3090 * assemble: clear any 'faulty' flag to allow this device to
3092 * force-array: Array is degraded but being forced, mark it clean
3093 * if that will be needed to assemble it.
3095 * newdev: not used ????
3096 * grow: Array has gained a new device - this is currently for
3098 * resync: mark as dirty so a resync will happen.
3099 * name: update the name - preserving the homehost
3100 * uuid: Change the uuid of the array to match watch is given
3102 * Following are not relevant for this imsm:
3103 * sparc2.2 : update from old dodgey metadata
3104 * super-minor: change the preferred_minor number
3105 * summaries: update redundant counters.
3106 * homehost: update the recorded homehost
3107 * _reshape_progress: record new reshape_progress position.
3110 struct intel_super
*super
= st
->sb
;
3111 struct imsm_super
*mpb
;
3113 /* we can only update container info */
3114 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3117 mpb
= super
->anchor
;
3119 if (strcmp(update
, "uuid") == 0) {
3120 /* We take this to mean that the family_num should be updated.
3121 * However that is much smaller than the uuid so we cannot really
3122 * allow an explicit uuid to be given. And it is hard to reliably
3124 * So if !uuid_set we know the current uuid is random and just used
3125 * the first 'int' and copy it to the other 3 positions.
3126 * Otherwise we require the 4 'int's to be the same as would be the
3127 * case if we are using a random uuid. So an explicit uuid will be
3128 * accepted as long as all for ints are the same... which shouldn't hurt
3131 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3134 if (info
->uuid
[0] != info
->uuid
[1] ||
3135 info
->uuid
[1] != info
->uuid
[2] ||
3136 info
->uuid
[2] != info
->uuid
[3])
3142 mpb
->orig_family_num
= info
->uuid
[0];
3143 } else if (strcmp(update
, "assemble") == 0)
3148 /* successful update? recompute checksum */
3150 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3155 static size_t disks_to_mpb_size(int disks
)
3159 size
= sizeof(struct imsm_super
);
3160 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3161 size
+= 2 * sizeof(struct imsm_dev
);
3162 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3163 size
+= (4 - 2) * sizeof(struct imsm_map
);
3164 /* 4 possible disk_ord_tbl's */
3165 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3170 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3171 unsigned long long data_offset
)
3173 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3176 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3179 static void free_devlist(struct intel_super
*super
)
3181 struct intel_dev
*dv
;
3183 while (super
->devlist
) {
3184 dv
= super
->devlist
->next
;
3185 free(super
->devlist
->dev
);
3186 free(super
->devlist
);
3187 super
->devlist
= dv
;
3191 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3193 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3196 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3200 * 0 same, or first was empty, and second was copied
3201 * 1 second had wrong number
3203 * 3 wrong other info
3205 struct intel_super
*first
= st
->sb
;
3206 struct intel_super
*sec
= tst
->sb
;
3213 /* in platform dependent environment test if the disks
3214 * use the same Intel hba
3215 * If not on Intel hba at all, allow anything.
3217 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3218 if (first
->hba
->type
!= sec
->hba
->type
) {
3220 "HBAs of devices do not match %s != %s\n",
3221 get_sys_dev_type(first
->hba
->type
),
3222 get_sys_dev_type(sec
->hba
->type
));
3225 if (first
->orom
!= sec
->orom
) {
3227 "HBAs of devices do not match %s != %s\n",
3228 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3233 /* if an anchor does not have num_raid_devs set then it is a free
3236 if (first
->anchor
->num_raid_devs
> 0 &&
3237 sec
->anchor
->num_raid_devs
> 0) {
3238 /* Determine if these disks might ever have been
3239 * related. Further disambiguation can only take place
3240 * in load_super_imsm_all
3242 __u32 first_family
= first
->anchor
->orig_family_num
;
3243 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3245 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3246 MAX_SIGNATURE_LENGTH
) != 0)
3249 if (first_family
== 0)
3250 first_family
= first
->anchor
->family_num
;
3251 if (sec_family
== 0)
3252 sec_family
= sec
->anchor
->family_num
;
3254 if (first_family
!= sec_family
)
3259 /* if 'first' is a spare promote it to a populated mpb with sec's
3262 if (first
->anchor
->num_raid_devs
== 0 &&
3263 sec
->anchor
->num_raid_devs
> 0) {
3265 struct intel_dev
*dv
;
3266 struct imsm_dev
*dev
;
3268 /* we need to copy raid device info from sec if an allocation
3269 * fails here we don't associate the spare
3271 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3272 dv
= xmalloc(sizeof(*dv
));
3273 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3276 dv
->next
= first
->devlist
;
3277 first
->devlist
= dv
;
3279 if (i
< sec
->anchor
->num_raid_devs
) {
3280 /* allocation failure */
3281 free_devlist(first
);
3282 pr_err("imsm: failed to associate spare\n");
3285 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3286 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3287 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3288 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3289 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3290 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3296 static void fd2devname(int fd
, char *name
)
3300 char dname
[PATH_MAX
];
3305 if (fstat(fd
, &st
) != 0)
3307 sprintf(path
, "/sys/dev/block/%d:%d",
3308 major(st
.st_rdev
), minor(st
.st_rdev
));
3310 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3315 nm
= strrchr(dname
, '/');
3318 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3322 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3324 static int imsm_read_serial(int fd
, char *devname
,
3325 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3327 unsigned char scsi_serial
[255];
3336 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3338 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3340 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3341 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3342 fd2devname(fd
, (char *) serial
);
3348 pr_err("Failed to retrieve serial for %s\n",
3353 rsp_len
= scsi_serial
[3];
3356 pr_err("Failed to retrieve serial for %s\n",
3360 rsp_buf
= (char *) &scsi_serial
[4];
3362 /* trim all whitespace and non-printable characters and convert
3365 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3368 /* ':' is reserved for use in placeholder serial
3369 * numbers for missing disks
3377 len
= dest
- rsp_buf
;
3380 /* truncate leading characters */
3381 if (len
> MAX_RAID_SERIAL_LEN
) {
3382 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3383 len
= MAX_RAID_SERIAL_LEN
;
3386 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3387 memcpy(serial
, dest
, len
);
3392 static int serialcmp(__u8
*s1
, __u8
*s2
)
3394 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3397 static void serialcpy(__u8
*dest
, __u8
*src
)
3399 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3402 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3406 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3407 if (serialcmp(dl
->serial
, serial
) == 0)
3413 static struct imsm_disk
*
3414 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3418 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3419 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3421 if (serialcmp(disk
->serial
, serial
) == 0) {
3432 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3434 struct imsm_disk
*disk
;
3439 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3441 rv
= imsm_read_serial(fd
, devname
, serial
);
3446 dl
= xcalloc(1, sizeof(*dl
));
3449 dl
->major
= major(stb
.st_rdev
);
3450 dl
->minor
= minor(stb
.st_rdev
);
3451 dl
->next
= super
->disks
;
3452 dl
->fd
= keep_fd
? fd
: -1;
3453 assert(super
->disks
== NULL
);
3455 serialcpy(dl
->serial
, serial
);
3458 fd2devname(fd
, name
);
3460 dl
->devname
= xstrdup(devname
);
3462 dl
->devname
= xstrdup(name
);
3464 /* look up this disk's index in the current anchor */
3465 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3468 /* only set index on disks that are a member of a
3469 * populated contianer, i.e. one with raid_devs
3471 if (is_failed(&dl
->disk
))
3473 else if (is_spare(&dl
->disk
))
3481 /* When migrating map0 contains the 'destination' state while map1
3482 * contains the current state. When not migrating map0 contains the
3483 * current state. This routine assumes that map[0].map_state is set to
3484 * the current array state before being called.
3486 * Migration is indicated by one of the following states
3487 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3488 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3489 * map1state=unitialized)
3490 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3492 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3493 * map1state=degraded)
3494 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3497 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3498 __u8 to_state
, int migr_type
)
3500 struct imsm_map
*dest
;
3501 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3503 dev
->vol
.migr_state
= 1;
3504 set_migr_type(dev
, migr_type
);
3505 dev
->vol
.curr_migr_unit
= 0;
3506 dest
= get_imsm_map(dev
, MAP_1
);
3508 /* duplicate and then set the target end state in map[0] */
3509 memcpy(dest
, src
, sizeof_imsm_map(src
));
3510 if ((migr_type
== MIGR_REBUILD
) ||
3511 (migr_type
== MIGR_GEN_MIGR
)) {
3515 for (i
= 0; i
< src
->num_members
; i
++) {
3516 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3517 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3521 if (migr_type
== MIGR_GEN_MIGR
)
3522 /* Clear migration record */
3523 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3525 src
->map_state
= to_state
;
3528 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3531 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3532 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3536 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3537 * completed in the last migration.
3539 * FIXME add support for raid-level-migration
3541 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3542 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3543 /* when final map state is other than expected
3544 * merge maps (not for migration)
3548 for (i
= 0; i
< prev
->num_members
; i
++)
3549 for (j
= 0; j
< map
->num_members
; j
++)
3550 /* during online capacity expansion
3551 * disks position can be changed
3552 * if takeover is used
3554 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3555 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3556 map
->disk_ord_tbl
[j
] |=
3557 prev
->disk_ord_tbl
[i
];
3560 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3561 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3564 dev
->vol
.migr_state
= 0;
3565 set_migr_type(dev
, 0);
3566 dev
->vol
.curr_migr_unit
= 0;
3567 map
->map_state
= map_state
;
3571 static int parse_raid_devices(struct intel_super
*super
)
3574 struct imsm_dev
*dev_new
;
3575 size_t len
, len_migr
;
3577 size_t space_needed
= 0;
3578 struct imsm_super
*mpb
= super
->anchor
;
3580 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3581 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3582 struct intel_dev
*dv
;
3584 len
= sizeof_imsm_dev(dev_iter
, 0);
3585 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3587 space_needed
+= len_migr
- len
;
3589 dv
= xmalloc(sizeof(*dv
));
3590 if (max_len
< len_migr
)
3592 if (max_len
> len_migr
)
3593 space_needed
+= max_len
- len_migr
;
3594 dev_new
= xmalloc(max_len
);
3595 imsm_copy_dev(dev_new
, dev_iter
);
3598 dv
->next
= super
->devlist
;
3599 super
->devlist
= dv
;
3602 /* ensure that super->buf is large enough when all raid devices
3605 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3608 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3609 if (posix_memalign(&buf
, 512, len
) != 0)
3612 memcpy(buf
, super
->buf
, super
->len
);
3613 memset(buf
+ super
->len
, 0, len
- super
->len
);
3622 /* retrieve a pointer to the bbm log which starts after all raid devices */
3623 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3627 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3629 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3635 /*******************************************************************************
3636 * Function: check_mpb_migr_compatibility
3637 * Description: Function checks for unsupported migration features:
3638 * - migration optimization area (pba_of_lba0)
3639 * - descending reshape (ascending_migr)
3641 * super : imsm metadata information
3643 * 0 : migration is compatible
3644 * -1 : migration is not compatible
3645 ******************************************************************************/
3646 int check_mpb_migr_compatibility(struct intel_super
*super
)
3648 struct imsm_map
*map0
, *map1
;
3649 struct migr_record
*migr_rec
= super
->migr_rec
;
3652 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3653 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3656 dev_iter
->vol
.migr_state
== 1 &&
3657 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3658 /* This device is migrating */
3659 map0
= get_imsm_map(dev_iter
, MAP_0
);
3660 map1
= get_imsm_map(dev_iter
, MAP_1
);
3661 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3662 /* migration optimization area was used */
3664 if (migr_rec
->ascending_migr
== 0
3665 && migr_rec
->dest_depth_per_unit
> 0)
3666 /* descending reshape not supported yet */
3673 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3675 /* load_imsm_mpb - read matrix metadata
3676 * allocates super->mpb to be freed by free_imsm
3678 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3680 unsigned long long dsize
;
3681 unsigned long long sectors
;
3683 struct imsm_super
*anchor
;
3686 get_dev_size(fd
, NULL
, &dsize
);
3689 pr_err("%s: device to small for imsm\n",
3694 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3696 pr_err("Cannot seek to anchor block on %s: %s\n",
3697 devname
, strerror(errno
));
3701 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3703 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
3706 if (read(fd
, anchor
, 512) != 512) {
3708 pr_err("Cannot read anchor block on %s: %s\n",
3709 devname
, strerror(errno
));
3714 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3716 pr_err("no IMSM anchor on %s\n", devname
);
3721 __free_imsm(super
, 0);
3722 /* reload capability and hba */
3724 /* capability and hba must be updated with new super allocation */
3725 find_intel_hba_capability(fd
, super
, devname
);
3726 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3727 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3729 pr_err("unable to allocate %zu byte mpb buffer\n",
3734 memcpy(super
->buf
, anchor
, 512);
3736 sectors
= mpb_sectors(anchor
) - 1;
3739 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3740 pr_err("could not allocate migr_rec buffer\n");
3744 super
->clean_migration_record_by_mdmon
= 0;
3747 check_sum
= __gen_imsm_checksum(super
->anchor
);
3748 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3750 pr_err("IMSM checksum %x != %x on %s\n",
3752 __le32_to_cpu(super
->anchor
->check_sum
),
3760 /* read the extended mpb */
3761 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3763 pr_err("Cannot seek to extended mpb on %s: %s\n",
3764 devname
, strerror(errno
));
3768 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3770 pr_err("Cannot read extended mpb on %s: %s\n",
3771 devname
, strerror(errno
));
3775 check_sum
= __gen_imsm_checksum(super
->anchor
);
3776 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3778 pr_err("IMSM checksum %x != %x on %s\n",
3779 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3784 /* FIXME the BBM log is disk specific so we cannot use this global
3785 * buffer for all disks. Ok for now since we only look at the global
3786 * bbm_log_size parameter to gate assembly
3788 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3793 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3795 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3796 static void clear_hi(struct intel_super
*super
)
3798 struct imsm_super
*mpb
= super
->anchor
;
3800 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3802 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3803 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3804 disk
->total_blocks_hi
= 0;
3806 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3807 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3810 for (n
= 0; n
< 2; ++n
) {
3811 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3814 map
->pba_of_lba0_hi
= 0;
3815 map
->blocks_per_member_hi
= 0;
3816 map
->num_data_stripes_hi
= 0;
3822 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3826 err
= load_imsm_mpb(fd
, super
, devname
);
3829 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3832 err
= parse_raid_devices(super
);
3837 static void __free_imsm_disk(struct dl
*d
)
3849 static void free_imsm_disks(struct intel_super
*super
)
3853 while (super
->disks
) {
3855 super
->disks
= d
->next
;
3856 __free_imsm_disk(d
);
3858 while (super
->disk_mgmt_list
) {
3859 d
= super
->disk_mgmt_list
;
3860 super
->disk_mgmt_list
= d
->next
;
3861 __free_imsm_disk(d
);
3863 while (super
->missing
) {
3865 super
->missing
= d
->next
;
3866 __free_imsm_disk(d
);
3871 /* free all the pieces hanging off of a super pointer */
3872 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3874 struct intel_hba
*elem
, *next
;
3880 /* unlink capability description */
3882 if (super
->migr_rec_buf
) {
3883 free(super
->migr_rec_buf
);
3884 super
->migr_rec_buf
= NULL
;
3887 free_imsm_disks(super
);
3888 free_devlist(super
);
3892 free((void *)elem
->path
);
3900 static void free_imsm(struct intel_super
*super
)
3902 __free_imsm(super
, 1);
3906 static void free_super_imsm(struct supertype
*st
)
3908 struct intel_super
*super
= st
->sb
;
3917 static struct intel_super
*alloc_super(void)
3919 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
3921 super
->current_vol
= -1;
3922 super
->create_offset
= ~((unsigned long long) 0);
3927 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3929 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3931 struct sys_dev
*hba_name
;
3934 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3939 hba_name
= find_disk_attached_hba(fd
, NULL
);
3942 pr_err("%s is not attached to Intel(R) RAID controller.\n",
3946 rv
= attach_hba_to_super(super
, hba_name
);
3949 struct intel_hba
*hba
= super
->hba
;
3951 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
3952 " but the container is assigned to Intel(R) %s %s (",
3954 get_sys_dev_type(hba_name
->type
),
3955 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
3956 hba_name
->pci_id
? : "Err!",
3957 get_sys_dev_type(super
->hba
->type
),
3958 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
3961 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3963 fprintf(stderr
, ", ");
3966 fprintf(stderr
, ").\n"
3967 " Mixing devices attached to different %s is not allowed.\n",
3968 hba_name
->type
== SYS_DEV_VMD
? "VMD domains" : "controllers");
3972 super
->orom
= find_imsm_capability(hba_name
);
3979 /* find_missing - helper routine for load_super_imsm_all that identifies
3980 * disks that have disappeared from the system. This routine relies on
3981 * the mpb being uptodate, which it is at load time.
3983 static int find_missing(struct intel_super
*super
)
3986 struct imsm_super
*mpb
= super
->anchor
;
3988 struct imsm_disk
*disk
;
3990 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3991 disk
= __get_imsm_disk(mpb
, i
);
3992 dl
= serial_to_dl(disk
->serial
, super
);
3996 dl
= xmalloc(sizeof(*dl
));
4000 dl
->devname
= xstrdup("missing");
4002 serialcpy(dl
->serial
, disk
->serial
);
4005 dl
->next
= super
->missing
;
4006 super
->missing
= dl
;
4013 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4015 struct intel_disk
*idisk
= disk_list
;
4018 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4020 idisk
= idisk
->next
;
4026 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4027 struct intel_super
*super
,
4028 struct intel_disk
**disk_list
)
4030 struct imsm_disk
*d
= &super
->disks
->disk
;
4031 struct imsm_super
*mpb
= super
->anchor
;
4034 for (i
= 0; i
< tbl_size
; i
++) {
4035 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4036 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4038 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4039 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4040 dprintf("mpb from %d:%d matches %d:%d\n",
4041 super
->disks
->major
,
4042 super
->disks
->minor
,
4043 table
[i
]->disks
->major
,
4044 table
[i
]->disks
->minor
);
4048 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4049 is_configured(d
) == is_configured(tbl_d
)) &&
4050 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4051 /* current version of the mpb is a
4052 * better candidate than the one in
4053 * super_table, but copy over "cross
4054 * generational" status
4056 struct intel_disk
*idisk
;
4058 dprintf("mpb from %d:%d replaces %d:%d\n",
4059 super
->disks
->major
,
4060 super
->disks
->minor
,
4061 table
[i
]->disks
->major
,
4062 table
[i
]->disks
->minor
);
4064 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4065 if (idisk
&& is_failed(&idisk
->disk
))
4066 tbl_d
->status
|= FAILED_DISK
;
4069 struct intel_disk
*idisk
;
4070 struct imsm_disk
*disk
;
4072 /* tbl_mpb is more up to date, but copy
4073 * over cross generational status before
4076 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4077 if (disk
&& is_failed(disk
))
4078 d
->status
|= FAILED_DISK
;
4080 idisk
= disk_list_get(d
->serial
, *disk_list
);
4083 if (disk
&& is_configured(disk
))
4084 idisk
->disk
.status
|= CONFIGURED_DISK
;
4087 dprintf("mpb from %d:%d prefer %d:%d\n",
4088 super
->disks
->major
,
4089 super
->disks
->minor
,
4090 table
[i
]->disks
->major
,
4091 table
[i
]->disks
->minor
);
4099 table
[tbl_size
++] = super
;
4103 /* update/extend the merged list of imsm_disk records */
4104 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4105 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4106 struct intel_disk
*idisk
;
4108 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4110 idisk
->disk
.status
|= disk
->status
;
4111 if (is_configured(&idisk
->disk
) ||
4112 is_failed(&idisk
->disk
))
4113 idisk
->disk
.status
&= ~(SPARE_DISK
);
4115 idisk
= xcalloc(1, sizeof(*idisk
));
4116 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4117 idisk
->disk
= *disk
;
4118 idisk
->next
= *disk_list
;
4122 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4129 static struct intel_super
*
4130 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4133 struct imsm_super
*mpb
= super
->anchor
;
4137 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4138 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4139 struct intel_disk
*idisk
;
4141 idisk
= disk_list_get(disk
->serial
, disk_list
);
4143 if (idisk
->owner
== owner
||
4144 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4147 dprintf("'%.16s' owner %d != %d\n",
4148 disk
->serial
, idisk
->owner
,
4151 dprintf("unknown disk %x [%d]: %.16s\n",
4152 __le32_to_cpu(mpb
->family_num
), i
,
4158 if (ok_count
== mpb
->num_disks
)
4163 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4165 struct intel_super
*s
;
4167 for (s
= super_list
; s
; s
= s
->next
) {
4168 if (family_num
!= s
->anchor
->family_num
)
4170 pr_err("Conflict, offlining family %#x on '%s'\n",
4171 __le32_to_cpu(family_num
), s
->disks
->devname
);
4175 static struct intel_super
*
4176 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4178 struct intel_super
*super_table
[len
];
4179 struct intel_disk
*disk_list
= NULL
;
4180 struct intel_super
*champion
, *spare
;
4181 struct intel_super
*s
, **del
;
4186 memset(super_table
, 0, sizeof(super_table
));
4187 for (s
= *super_list
; s
; s
= s
->next
)
4188 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4190 for (i
= 0; i
< tbl_size
; i
++) {
4191 struct imsm_disk
*d
;
4192 struct intel_disk
*idisk
;
4193 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4196 d
= &s
->disks
->disk
;
4198 /* 'd' must appear in merged disk list for its
4199 * configuration to be valid
4201 idisk
= disk_list_get(d
->serial
, disk_list
);
4202 if (idisk
&& idisk
->owner
== i
)
4203 s
= validate_members(s
, disk_list
, i
);
4208 dprintf("marking family: %#x from %d:%d offline\n",
4210 super_table
[i
]->disks
->major
,
4211 super_table
[i
]->disks
->minor
);
4215 /* This is where the mdadm implementation differs from the Windows
4216 * driver which has no strict concept of a container. We can only
4217 * assemble one family from a container, so when returning a prodigal
4218 * array member to this system the code will not be able to disambiguate
4219 * the container contents that should be assembled ("foreign" versus
4220 * "local"). It requires user intervention to set the orig_family_num
4221 * to a new value to establish a new container. The Windows driver in
4222 * this situation fixes up the volume name in place and manages the
4223 * foreign array as an independent entity.
4228 for (i
= 0; i
< tbl_size
; i
++) {
4229 struct intel_super
*tbl_ent
= super_table
[i
];
4235 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4240 if (s
&& !is_spare
) {
4241 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4243 } else if (!s
&& !is_spare
)
4256 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4257 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4259 /* collect all dl's onto 'champion', and update them to
4260 * champion's version of the status
4262 for (s
= *super_list
; s
; s
= s
->next
) {
4263 struct imsm_super
*mpb
= champion
->anchor
;
4264 struct dl
*dl
= s
->disks
;
4269 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4271 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4272 struct imsm_disk
*disk
;
4274 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4277 /* only set index on disks that are a member of
4278 * a populated contianer, i.e. one with
4281 if (is_failed(&dl
->disk
))
4283 else if (is_spare(&dl
->disk
))
4289 if (i
>= mpb
->num_disks
) {
4290 struct intel_disk
*idisk
;
4292 idisk
= disk_list_get(dl
->serial
, disk_list
);
4293 if (idisk
&& is_spare(&idisk
->disk
) &&
4294 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4302 dl
->next
= champion
->disks
;
4303 champion
->disks
= dl
;
4307 /* delete 'champion' from super_list */
4308 for (del
= super_list
; *del
; ) {
4309 if (*del
== champion
) {
4310 *del
= (*del
)->next
;
4313 del
= &(*del
)->next
;
4315 champion
->next
= NULL
;
4319 struct intel_disk
*idisk
= disk_list
;
4321 disk_list
= disk_list
->next
;
4329 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4330 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4331 int major
, int minor
, int keep_fd
);
4333 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4334 int *max
, int keep_fd
);
4336 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4337 char *devname
, struct md_list
*devlist
,
4340 struct intel_super
*super_list
= NULL
;
4341 struct intel_super
*super
= NULL
;
4346 /* 'fd' is an opened container */
4347 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4349 /* get super block from devlist devices */
4350 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4353 /* all mpbs enter, maybe one leaves */
4354 super
= imsm_thunderdome(&super_list
, i
);
4360 if (find_missing(super
) != 0) {
4366 /* load migration record */
4367 err
= load_imsm_migr_rec(super
, NULL
);
4369 /* migration is in progress,
4370 * but migr_rec cannot be loaded,
4376 /* Check migration compatibility */
4377 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4378 pr_err("Unsupported migration detected");
4380 fprintf(stderr
, " on %s\n", devname
);
4382 fprintf(stderr
, " (IMSM).\n");
4391 while (super_list
) {
4392 struct intel_super
*s
= super_list
;
4394 super_list
= super_list
->next
;
4403 strcpy(st
->container_devnm
, fd2devnm(fd
));
4405 st
->container_devnm
[0] = 0;
4406 if (err
== 0 && st
->ss
== NULL
) {
4407 st
->ss
= &super_imsm
;
4408 st
->minor_version
= 0;
4409 st
->max_devs
= IMSM_MAX_DEVICES
;
4415 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4416 int *max
, int keep_fd
)
4418 struct md_list
*tmpdev
;
4422 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4423 if (tmpdev
->used
!= 1)
4425 if (tmpdev
->container
== 1) {
4427 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4429 pr_err("cannot open device %s: %s\n",
4430 tmpdev
->devname
, strerror(errno
));
4434 err
= get_sra_super_block(fd
, super_list
,
4435 tmpdev
->devname
, &lmax
,
4444 int major
= major(tmpdev
->st_rdev
);
4445 int minor
= minor(tmpdev
->st_rdev
);
4446 err
= get_super_block(super_list
,
4463 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4464 int major
, int minor
, int keep_fd
)
4466 struct intel_super
*s
= NULL
;
4478 sprintf(nm
, "%d:%d", major
, minor
);
4479 dfd
= dev_open(nm
, O_RDWR
);
4485 find_intel_hba_capability(dfd
, s
, devname
);
4486 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4488 /* retry the load if we might have raced against mdmon */
4489 if (err
== 3 && devnm
&& mdmon_running(devnm
))
4490 for (retry
= 0; retry
< 3; retry
++) {
4492 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4498 s
->next
= *super_list
;
4506 if ((dfd
>= 0) && (!keep_fd
))
4513 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4520 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4524 if (sra
->array
.major_version
!= -1 ||
4525 sra
->array
.minor_version
!= -2 ||
4526 strcmp(sra
->text_version
, "imsm") != 0) {
4531 devnm
= fd2devnm(fd
);
4532 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4533 if (get_super_block(super_list
, devnm
, devname
,
4534 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4545 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4547 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4551 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4553 struct intel_super
*super
;
4557 if (test_partition(fd
))
4558 /* IMSM not allowed on partitions */
4561 free_super_imsm(st
);
4563 super
= alloc_super();
4564 /* Load hba and capabilities if they exist.
4565 * But do not preclude loading metadata in case capabilities or hba are
4566 * non-compliant and ignore_hw_compat is set.
4568 rv
= find_intel_hba_capability(fd
, super
, devname
);
4569 /* no orom/efi or non-intel hba of the disk */
4570 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4572 pr_err("No OROM/EFI properties for %s\n", devname
);
4576 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4578 /* retry the load if we might have raced against mdmon */
4580 struct mdstat_ent
*mdstat
= mdstat_by_component(fd2devnm(fd
));
4582 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
4583 for (retry
= 0; retry
< 3; retry
++) {
4585 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4591 free_mdstat(mdstat
);
4596 pr_err("Failed to load all information sections on %s\n", devname
);
4602 if (st
->ss
== NULL
) {
4603 st
->ss
= &super_imsm
;
4604 st
->minor_version
= 0;
4605 st
->max_devs
= IMSM_MAX_DEVICES
;
4608 /* load migration record */
4609 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4610 /* Check for unsupported migration features */
4611 if (check_mpb_migr_compatibility(super
) != 0) {
4612 pr_err("Unsupported migration detected");
4614 fprintf(stderr
, " on %s\n", devname
);
4616 fprintf(stderr
, " (IMSM).\n");
4624 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4626 if (info
->level
== 1)
4628 return info
->chunk_size
>> 9;
4631 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4632 unsigned long long size
)
4634 if (info
->level
== 1)
4637 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4640 static void imsm_update_version_info(struct intel_super
*super
)
4642 /* update the version and attributes */
4643 struct imsm_super
*mpb
= super
->anchor
;
4645 struct imsm_dev
*dev
;
4646 struct imsm_map
*map
;
4649 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4650 dev
= get_imsm_dev(super
, i
);
4651 map
= get_imsm_map(dev
, MAP_0
);
4652 if (__le32_to_cpu(dev
->size_high
) > 0)
4653 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4655 /* FIXME detect when an array spans a port multiplier */
4657 mpb
->attributes
|= MPB_ATTRIB_PM
;
4660 if (mpb
->num_raid_devs
> 1 ||
4661 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4662 version
= MPB_VERSION_ATTRIBS
;
4663 switch (get_imsm_raid_level(map
)) {
4664 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4665 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4666 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4667 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4670 if (map
->num_members
>= 5)
4671 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4672 else if (dev
->status
== DEV_CLONE_N_GO
)
4673 version
= MPB_VERSION_CNG
;
4674 else if (get_imsm_raid_level(map
) == 5)
4675 version
= MPB_VERSION_RAID5
;
4676 else if (map
->num_members
>= 3)
4677 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4678 else if (get_imsm_raid_level(map
) == 1)
4679 version
= MPB_VERSION_RAID1
;
4681 version
= MPB_VERSION_RAID0
;
4683 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4687 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4689 struct imsm_super
*mpb
= super
->anchor
;
4690 char *reason
= NULL
;
4693 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4694 reason
= "must be 16 characters or less";
4696 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4697 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4699 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4700 reason
= "already exists";
4705 if (reason
&& !quiet
)
4706 pr_err("imsm volume name %s\n", reason
);
4711 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4712 unsigned long long size
, char *name
,
4713 char *homehost
, int *uuid
,
4714 long long data_offset
)
4716 /* We are creating a volume inside a pre-existing container.
4717 * so st->sb is already set.
4719 struct intel_super
*super
= st
->sb
;
4720 struct imsm_super
*mpb
= super
->anchor
;
4721 struct intel_dev
*dv
;
4722 struct imsm_dev
*dev
;
4723 struct imsm_vol
*vol
;
4724 struct imsm_map
*map
;
4725 int idx
= mpb
->num_raid_devs
;
4727 unsigned long long array_blocks
;
4728 size_t size_old
, size_new
;
4729 unsigned long long num_data_stripes
;
4731 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4732 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
4736 /* ensure the mpb is large enough for the new data */
4737 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4738 size_new
= disks_to_mpb_size(info
->nr_disks
);
4739 if (size_new
> size_old
) {
4741 size_t size_round
= ROUND_UP(size_new
, 512);
4743 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4744 pr_err("could not allocate new mpb\n");
4747 if (posix_memalign(&super
->migr_rec_buf
, 512,
4748 MIGR_REC_BUF_SIZE
) != 0) {
4749 pr_err("could not allocate migr_rec buffer\n");
4755 memcpy(mpb_new
, mpb
, size_old
);
4758 super
->anchor
= mpb_new
;
4759 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4760 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4762 super
->current_vol
= idx
;
4764 /* handle 'failed_disks' by either:
4765 * a) create dummy disk entries in the table if this the first
4766 * volume in the array. We add them here as this is the only
4767 * opportunity to add them. add_to_super_imsm_volume()
4768 * handles the non-failed disks and continues incrementing
4770 * b) validate that 'failed_disks' matches the current number
4771 * of missing disks if the container is populated
4773 if (super
->current_vol
== 0) {
4775 for (i
= 0; i
< info
->failed_disks
; i
++) {
4776 struct imsm_disk
*disk
;
4779 disk
= __get_imsm_disk(mpb
, i
);
4780 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4781 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4782 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4785 find_missing(super
);
4790 for (d
= super
->missing
; d
; d
= d
->next
)
4792 if (info
->failed_disks
> missing
) {
4793 pr_err("unable to add 'missing' disk to container\n");
4798 if (!check_name(super
, name
, 0))
4800 dv
= xmalloc(sizeof(*dv
));
4801 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4802 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4803 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4804 info
->layout
, info
->chunk_size
,
4806 /* round array size down to closest MB */
4807 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4809 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4810 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4811 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4813 vol
->migr_state
= 0;
4814 set_migr_type(dev
, MIGR_INIT
);
4815 vol
->dirty
= !info
->state
;
4816 vol
->curr_migr_unit
= 0;
4817 map
= get_imsm_map(dev
, MAP_0
);
4818 set_pba_of_lba0(map
, super
->create_offset
);
4819 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4820 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4821 map
->failed_disk_num
= ~0;
4822 if (info
->level
> 0)
4823 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
4824 : IMSM_T_STATE_UNINITIALIZED
);
4826 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4827 IMSM_T_STATE_NORMAL
;
4830 if (info
->level
== 1 && info
->raid_disks
> 2) {
4833 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
4837 map
->raid_level
= info
->level
;
4838 if (info
->level
== 10) {
4839 map
->raid_level
= 1;
4840 map
->num_domains
= info
->raid_disks
/ 2;
4841 } else if (info
->level
== 1)
4842 map
->num_domains
= info
->raid_disks
;
4844 map
->num_domains
= 1;
4846 /* info->size is only int so use the 'size' parameter instead */
4847 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4848 num_data_stripes
/= map
->num_domains
;
4849 set_num_data_stripes(map
, num_data_stripes
);
4851 map
->num_members
= info
->raid_disks
;
4852 for (i
= 0; i
< map
->num_members
; i
++) {
4853 /* initialized in add_to_super */
4854 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4856 mpb
->num_raid_devs
++;
4859 dv
->index
= super
->current_vol
;
4860 dv
->next
= super
->devlist
;
4861 super
->devlist
= dv
;
4863 imsm_update_version_info(super
);
4868 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4869 unsigned long long size
, char *name
,
4870 char *homehost
, int *uuid
,
4871 unsigned long long data_offset
)
4873 /* This is primarily called by Create when creating a new array.
4874 * We will then get add_to_super called for each component, and then
4875 * write_init_super called to write it out to each device.
4876 * For IMSM, Create can create on fresh devices or on a pre-existing
4878 * To create on a pre-existing array a different method will be called.
4879 * This one is just for fresh drives.
4881 struct intel_super
*super
;
4882 struct imsm_super
*mpb
;
4886 if (data_offset
!= INVALID_SECTORS
) {
4887 pr_err("data-offset not supported by imsm\n");
4892 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
,
4896 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4900 super
= alloc_super();
4901 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4906 pr_err("could not allocate superblock\n");
4909 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4910 pr_err("could not allocate migr_rec buffer\n");
4915 memset(super
->buf
, 0, mpb_size
);
4917 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4921 /* zeroing superblock */
4925 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4927 version
= (char *) mpb
->sig
;
4928 strcpy(version
, MPB_SIGNATURE
);
4929 version
+= strlen(MPB_SIGNATURE
);
4930 strcpy(version
, MPB_VERSION_RAID0
);
4936 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4937 int fd
, char *devname
)
4939 struct intel_super
*super
= st
->sb
;
4940 struct imsm_super
*mpb
= super
->anchor
;
4941 struct imsm_disk
*_disk
;
4942 struct imsm_dev
*dev
;
4943 struct imsm_map
*map
;
4947 dev
= get_imsm_dev(super
, super
->current_vol
);
4948 map
= get_imsm_map(dev
, MAP_0
);
4950 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4951 pr_err("%s: Cannot add spare devices to IMSM volume\n",
4957 /* we're doing autolayout so grab the pre-marked (in
4958 * validate_geometry) raid_disk
4960 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4961 if (dl
->raiddisk
== dk
->raid_disk
)
4964 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4965 if (dl
->major
== dk
->major
&&
4966 dl
->minor
== dk
->minor
)
4971 pr_err("%s is not a member of the same container\n", devname
);
4975 /* add a pristine spare to the metadata */
4976 if (dl
->index
< 0) {
4977 dl
->index
= super
->anchor
->num_disks
;
4978 super
->anchor
->num_disks
++;
4980 /* Check the device has not already been added */
4981 slot
= get_imsm_disk_slot(map
, dl
->index
);
4983 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4984 pr_err("%s has been included in this array twice\n",
4988 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4989 dl
->disk
.status
= CONFIGURED_DISK
;
4991 /* update size of 'missing' disks to be at least as large as the
4992 * largest acitve member (we only have dummy missing disks when
4993 * creating the first volume)
4995 if (super
->current_vol
== 0) {
4996 for (df
= super
->missing
; df
; df
= df
->next
) {
4997 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
4998 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
4999 _disk
= __get_imsm_disk(mpb
, df
->index
);
5004 /* refresh unset/failed slots to point to valid 'missing' entries */
5005 for (df
= super
->missing
; df
; df
= df
->next
)
5006 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5007 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5009 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5011 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5012 if (is_gen_migration(dev
)) {
5013 struct imsm_map
*map2
= get_imsm_map(dev
,
5015 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5016 if ((slot2
< map2
->num_members
) &&
5018 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5021 if ((unsigned)df
->index
==
5023 set_imsm_ord_tbl_ent(map2
,
5029 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5033 /* if we are creating the first raid device update the family number */
5034 if (super
->current_vol
== 0) {
5036 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5038 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5039 if (!_dev
|| !_disk
) {
5040 pr_err("BUG mpb setup error\n");
5046 sum
+= __gen_imsm_checksum(mpb
);
5047 mpb
->family_num
= __cpu_to_le32(sum
);
5048 mpb
->orig_family_num
= mpb
->family_num
;
5050 super
->current_disk
= dl
;
5055 * Function marks disk as spare and restores disk serial
5056 * in case it was previously marked as failed by takeover operation
5058 * -1 : critical error
5059 * 0 : disk is marked as spare but serial is not set
5062 int mark_spare(struct dl
*disk
)
5064 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5071 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5072 /* Restore disk serial number, because takeover marks disk
5073 * as failed and adds to serial ':0' before it becomes
5076 serialcpy(disk
->serial
, serial
);
5077 serialcpy(disk
->disk
.serial
, serial
);
5080 disk
->disk
.status
= SPARE_DISK
;
5086 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5087 int fd
, char *devname
,
5088 unsigned long long data_offset
)
5090 struct intel_super
*super
= st
->sb
;
5092 unsigned long long size
;
5097 /* If we are on an RAID enabled platform check that the disk is
5098 * attached to the raid controller.
5099 * We do not need to test disks attachment for container based additions,
5100 * they shall be already tested when container was created/assembled.
5102 rv
= find_intel_hba_capability(fd
, super
, devname
);
5103 /* no orom/efi or non-intel hba of the disk */
5105 dprintf("capability: %p fd: %d ret: %d\n",
5106 super
->orom
, fd
, rv
);
5110 if (super
->current_vol
>= 0)
5111 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5114 dd
= xcalloc(sizeof(*dd
), 1);
5115 dd
->major
= major(stb
.st_rdev
);
5116 dd
->minor
= minor(stb
.st_rdev
);
5117 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5120 dd
->action
= DISK_ADD
;
5121 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5123 pr_err("failed to retrieve scsi serial, aborting\n");
5128 get_dev_size(fd
, NULL
, &size
);
5129 /* clear migr_rec when adding disk to container */
5130 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5131 if (lseek64(fd
, size
- MIGR_REC_POSITION
, SEEK_SET
) >= 0) {
5132 if (write(fd
, super
->migr_rec_buf
,
5133 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5134 perror("Write migr_rec failed");
5138 serialcpy(dd
->disk
.serial
, dd
->serial
);
5139 set_total_blocks(&dd
->disk
, size
);
5140 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5141 struct imsm_super
*mpb
= super
->anchor
;
5142 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5145 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5146 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5148 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5150 if (st
->update_tail
) {
5151 dd
->next
= super
->disk_mgmt_list
;
5152 super
->disk_mgmt_list
= dd
;
5154 dd
->next
= super
->disks
;
5156 super
->updates_pending
++;
5162 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5164 struct intel_super
*super
= st
->sb
;
5167 /* remove from super works only in mdmon - for communication
5168 * manager - monitor. Check if communication memory buffer
5171 if (!st
->update_tail
) {
5172 pr_err("shall be used in mdmon context only\n");
5175 dd
= xcalloc(1, sizeof(*dd
));
5176 dd
->major
= dk
->major
;
5177 dd
->minor
= dk
->minor
;
5180 dd
->action
= DISK_REMOVE
;
5182 dd
->next
= super
->disk_mgmt_list
;
5183 super
->disk_mgmt_list
= dd
;
5188 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5192 struct imsm_super anchor
;
5193 } spare_record
__attribute__ ((aligned(512)));
5195 /* spare records have their own family number and do not have any defined raid
5198 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5200 struct imsm_super
*mpb
= super
->anchor
;
5201 struct imsm_super
*spare
= &spare_record
.anchor
;
5205 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5206 spare
->generation_num
= __cpu_to_le32(1UL);
5207 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5208 spare
->num_disks
= 1;
5209 spare
->num_raid_devs
= 0;
5210 spare
->cache_size
= mpb
->cache_size
;
5211 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5213 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5214 MPB_SIGNATURE MPB_VERSION_RAID0
);
5216 for (d
= super
->disks
; d
; d
= d
->next
) {
5220 spare
->disk
[0] = d
->disk
;
5221 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5222 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5224 sum
= __gen_imsm_checksum(spare
);
5225 spare
->family_num
= __cpu_to_le32(sum
);
5226 spare
->orig_family_num
= 0;
5227 sum
= __gen_imsm_checksum(spare
);
5228 spare
->check_sum
= __cpu_to_le32(sum
);
5230 if (store_imsm_mpb(d
->fd
, spare
)) {
5231 pr_err("failed for device %d:%d %s\n",
5232 d
->major
, d
->minor
, strerror(errno
));
5244 static int write_super_imsm(struct supertype
*st
, int doclose
)
5246 struct intel_super
*super
= st
->sb
;
5247 struct imsm_super
*mpb
= super
->anchor
;
5253 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5255 int clear_migration_record
= 1;
5257 /* 'generation' is incremented everytime the metadata is written */
5258 generation
= __le32_to_cpu(mpb
->generation_num
);
5260 mpb
->generation_num
= __cpu_to_le32(generation
);
5262 /* fix up cases where previous mdadm releases failed to set
5265 if (mpb
->orig_family_num
== 0)
5266 mpb
->orig_family_num
= mpb
->family_num
;
5268 for (d
= super
->disks
; d
; d
= d
->next
) {
5272 mpb
->disk
[d
->index
] = d
->disk
;
5276 for (d
= super
->missing
; d
; d
= d
->next
) {
5277 mpb
->disk
[d
->index
] = d
->disk
;
5280 mpb
->num_disks
= num_disks
;
5281 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5283 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5284 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5285 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5287 imsm_copy_dev(dev
, dev2
);
5288 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5290 if (is_gen_migration(dev2
))
5291 clear_migration_record
= 0;
5293 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5294 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5296 /* recalculate checksum */
5297 sum
= __gen_imsm_checksum(mpb
);
5298 mpb
->check_sum
= __cpu_to_le32(sum
);
5300 if (super
->clean_migration_record_by_mdmon
) {
5301 clear_migration_record
= 1;
5302 super
->clean_migration_record_by_mdmon
= 0;
5304 if (clear_migration_record
)
5305 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5307 /* write the mpb for disks that compose raid devices */
5308 for (d
= super
->disks
; d
; d
= d
->next
) {
5309 if (d
->index
< 0 || is_failed(&d
->disk
))
5312 if (clear_migration_record
) {
5313 unsigned long long dsize
;
5315 get_dev_size(d
->fd
, NULL
, &dsize
);
5316 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5317 if (write(d
->fd
, super
->migr_rec_buf
,
5318 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5319 perror("Write migr_rec failed");
5323 if (store_imsm_mpb(d
->fd
, mpb
))
5325 "failed for device %d:%d (fd: %d)%s\n",
5327 d
->fd
, strerror(errno
));
5336 return write_super_imsm_spares(super
, doclose
);
5341 static int create_array(struct supertype
*st
, int dev_idx
)
5344 struct imsm_update_create_array
*u
;
5345 struct intel_super
*super
= st
->sb
;
5346 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5347 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5348 struct disk_info
*inf
;
5349 struct imsm_disk
*disk
;
5352 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5353 sizeof(*inf
) * map
->num_members
;
5355 u
->type
= update_create_array
;
5356 u
->dev_idx
= dev_idx
;
5357 imsm_copy_dev(&u
->dev
, dev
);
5358 inf
= get_disk_info(u
);
5359 for (i
= 0; i
< map
->num_members
; i
++) {
5360 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5362 disk
= get_imsm_disk(super
, idx
);
5364 disk
= get_imsm_missing(super
, idx
);
5365 serialcpy(inf
[i
].serial
, disk
->serial
);
5367 append_metadata_update(st
, u
, len
);
5372 static int mgmt_disk(struct supertype
*st
)
5374 struct intel_super
*super
= st
->sb
;
5376 struct imsm_update_add_remove_disk
*u
;
5378 if (!super
->disk_mgmt_list
)
5383 u
->type
= update_add_remove_disk
;
5384 append_metadata_update(st
, u
, len
);
5389 static int write_init_super_imsm(struct supertype
*st
)
5391 struct intel_super
*super
= st
->sb
;
5392 int current_vol
= super
->current_vol
;
5394 /* we are done with current_vol reset it to point st at the container */
5395 super
->current_vol
= -1;
5397 if (st
->update_tail
) {
5398 /* queue the recently created array / added disk
5399 * as a metadata update */
5402 /* determine if we are creating a volume or adding a disk */
5403 if (current_vol
< 0) {
5404 /* in the mgmt (add/remove) disk case we are running
5405 * in mdmon context, so don't close fd's
5407 return mgmt_disk(st
);
5409 rv
= create_array(st
, current_vol
);
5414 for (d
= super
->disks
; d
; d
= d
->next
)
5415 Kill(d
->devname
, NULL
, 0, -1, 1);
5416 return write_super_imsm(st
, 1);
5421 static int store_super_imsm(struct supertype
*st
, int fd
)
5423 struct intel_super
*super
= st
->sb
;
5424 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5430 return store_imsm_mpb(fd
, mpb
);
5436 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5438 return __le32_to_cpu(mpb
->bbm_log_size
);
5442 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5443 int layout
, int raiddisks
, int chunk
,
5444 unsigned long long size
,
5445 unsigned long long data_offset
,
5447 unsigned long long *freesize
,
5451 unsigned long long ldsize
;
5452 struct intel_super
*super
=NULL
;
5455 if (level
!= LEVEL_CONTAINER
)
5460 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5463 pr_err("imsm: Cannot open %s: %s\n",
5464 dev
, strerror(errno
));
5467 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5472 /* capabilities retrieve could be possible
5473 * note that there is no fd for the disks in array.
5475 super
= alloc_super();
5476 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
5480 fd2devname(fd
, str
);
5481 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5482 fd
, str
, super
->orom
, rv
, raiddisks
);
5484 /* no orom/efi or non-intel hba of the disk */
5491 if (raiddisks
> super
->orom
->tds
) {
5493 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
5494 raiddisks
, super
->orom
->tds
);
5498 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5499 (ldsize
>> 9) >> 32 > 0) {
5501 pr_err("%s exceeds maximum platform supported size\n", dev
);
5507 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
5513 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5515 const unsigned long long base_start
= e
[*idx
].start
;
5516 unsigned long long end
= base_start
+ e
[*idx
].size
;
5519 if (base_start
== end
)
5523 for (i
= *idx
; i
< num_extents
; i
++) {
5524 /* extend overlapping extents */
5525 if (e
[i
].start
>= base_start
&&
5526 e
[i
].start
<= end
) {
5529 if (e
[i
].start
+ e
[i
].size
> end
)
5530 end
= e
[i
].start
+ e
[i
].size
;
5531 } else if (e
[i
].start
> end
) {
5537 return end
- base_start
;
5540 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5542 /* build a composite disk with all known extents and generate a new
5543 * 'maxsize' given the "all disks in an array must share a common start
5544 * offset" constraint
5546 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
5550 unsigned long long pos
;
5551 unsigned long long start
= 0;
5552 unsigned long long maxsize
;
5553 unsigned long reserve
;
5555 /* coalesce and sort all extents. also, check to see if we need to
5556 * reserve space between member arrays
5559 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5562 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5565 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5570 while (i
< sum_extents
) {
5571 e
[j
].start
= e
[i
].start
;
5572 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5574 if (e
[j
-1].size
== 0)
5583 unsigned long long esize
;
5585 esize
= e
[i
].start
- pos
;
5586 if (esize
>= maxsize
) {
5591 pos
= e
[i
].start
+ e
[i
].size
;
5593 } while (e
[i
-1].size
);
5599 /* FIXME assumes volume at offset 0 is the first volume in a
5602 if (start_extent
> 0)
5603 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5607 if (maxsize
< reserve
)
5610 super
->create_offset
= ~((unsigned long long) 0);
5611 if (start
+ reserve
> super
->create_offset
)
5612 return 0; /* start overflows create_offset */
5613 super
->create_offset
= start
+ reserve
;
5615 return maxsize
- reserve
;
5618 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5620 if (level
< 0 || level
== 6 || level
== 4)
5623 /* if we have an orom prevent invalid raid levels */
5626 case 0: return imsm_orom_has_raid0(orom
);
5629 return imsm_orom_has_raid1e(orom
);
5630 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5631 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5632 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5635 return 1; /* not on an Intel RAID platform so anything goes */
5641 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5642 int dpa
, int verbose
)
5644 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5645 struct mdstat_ent
*memb
= NULL
;
5648 struct md_list
*dv
= NULL
;
5651 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5652 if (memb
->metadata_version
&&
5653 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5654 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5655 !is_subarray(memb
->metadata_version
+9) &&
5657 struct dev_member
*dev
= memb
->members
;
5659 while(dev
&& (fd
< 0)) {
5660 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5661 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5663 fd
= open(path
, O_RDONLY
, 0);
5664 if ((num
<= 0) || (fd
< 0)) {
5665 pr_vrb(": Cannot open %s: %s\n",
5666 dev
->name
, strerror(errno
));
5672 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5673 struct mdstat_ent
*vol
;
5674 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5675 if ((vol
->active
> 0) &&
5676 vol
->metadata_version
&&
5677 is_container_member(vol
, memb
->devnm
)) {
5682 if (*devlist
&& (found
< dpa
)) {
5683 dv
= xcalloc(1, sizeof(*dv
));
5684 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
5685 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
5688 dv
->next
= *devlist
;
5696 free_mdstat(mdstat
);
5701 static struct md_list
*
5702 get_loop_devices(void)
5705 struct md_list
*devlist
= NULL
;
5706 struct md_list
*dv
= NULL
;
5708 for(i
= 0; i
< 12; i
++) {
5709 dv
= xcalloc(1, sizeof(*dv
));
5710 dv
->devname
= xmalloc(40);
5711 sprintf(dv
->devname
, "/dev/loop%d", i
);
5719 static struct md_list
*
5720 get_devices(const char *hba_path
)
5722 struct md_list
*devlist
= NULL
;
5723 struct md_list
*dv
= NULL
;
5729 devlist
= get_loop_devices();
5732 /* scroll through /sys/dev/block looking for devices attached to
5735 dir
= opendir("/sys/dev/block");
5736 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5741 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5743 path
= devt_to_devpath(makedev(major
, minor
));
5746 if (!path_attached_to_hba(path
, hba_path
)) {
5753 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5755 fd2devname(fd
, buf
);
5758 pr_err("cannot open device: %s\n",
5763 dv
= xcalloc(1, sizeof(*dv
));
5764 dv
->devname
= xstrdup(buf
);
5771 devlist
= devlist
->next
;
5781 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5782 int verbose
, int *found
)
5784 struct md_list
*tmpdev
;
5786 struct supertype
*st
= NULL
;
5788 /* first walk the list of devices to find a consistent set
5789 * that match the criterea, if that is possible.
5790 * We flag the ones we like with 'used'.
5793 st
= match_metadata_desc_imsm("imsm");
5795 pr_vrb(": cannot allocate memory for imsm supertype\n");
5799 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5800 char *devname
= tmpdev
->devname
;
5802 struct supertype
*tst
;
5804 if (tmpdev
->used
> 1)
5806 tst
= dup_super(st
);
5808 pr_vrb(": cannot allocate memory for imsm supertype\n");
5811 tmpdev
->container
= 0;
5812 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5814 dprintf("cannot open device %s: %s\n",
5815 devname
, strerror(errno
));
5817 } else if (fstat(dfd
, &stb
)< 0) {
5819 dprintf("fstat failed for %s: %s\n",
5820 devname
, strerror(errno
));
5822 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5823 dprintf("%s is not a block device.\n",
5826 } else if (must_be_container(dfd
)) {
5827 struct supertype
*cst
;
5828 cst
= super_by_fd(dfd
, NULL
);
5830 dprintf("cannot recognize container type %s\n",
5833 } else if (tst
->ss
!= st
->ss
) {
5834 dprintf("non-imsm container - ignore it: %s\n",
5837 } else if (!tst
->ss
->load_container
||
5838 tst
->ss
->load_container(tst
, dfd
, NULL
))
5841 tmpdev
->container
= 1;
5844 cst
->ss
->free_super(cst
);
5846 tmpdev
->st_rdev
= stb
.st_rdev
;
5847 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5848 dprintf("no RAID superblock on %s\n",
5851 } else if (tst
->ss
->compare_super
== NULL
) {
5852 dprintf("Cannot assemble %s metadata on %s\n",
5853 tst
->ss
->name
, devname
);
5859 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5860 /* Ignore unrecognised devices during auto-assembly */
5865 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5867 if (st
->minor_version
== -1)
5868 st
->minor_version
= tst
->minor_version
;
5870 if (memcmp(info
.uuid
, uuid_zero
,
5871 sizeof(int[4])) == 0) {
5872 /* this is a floating spare. It cannot define
5873 * an array unless there are no more arrays of
5874 * this type to be found. It can be included
5875 * in an array of this type though.
5881 if (st
->ss
!= tst
->ss
||
5882 st
->minor_version
!= tst
->minor_version
||
5883 st
->ss
->compare_super(st
, tst
) != 0) {
5884 /* Some mismatch. If exactly one array matches this host,
5885 * we can resolve on that one.
5886 * Or, if we are auto assembling, we just ignore the second
5889 dprintf("superblock on %s doesn't match others - assembly aborted\n",
5895 dprintf("found: devname: %s\n", devname
);
5899 tst
->ss
->free_super(tst
);
5903 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5904 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5905 for (iter
= head
; iter
; iter
= iter
->next
) {
5906 dprintf("content->text_version: %s vol\n",
5907 iter
->text_version
);
5908 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5909 /* do not assemble arrays with unsupported
5911 dprintf("Cannot activate member %s.\n",
5912 iter
->text_version
);
5919 dprintf("No valid super block on device list: err: %d %p\n",
5923 dprintf("no more devices to examine\n");
5926 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5927 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5929 if (count
< tmpdev
->found
)
5932 count
-= tmpdev
->found
;
5935 if (tmpdev
->used
== 1)
5940 st
->ss
->free_super(st
);
5945 count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
5947 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
5949 const struct orom_entry
*entry
;
5950 struct devid_list
*dv
, *devid_list
;
5952 if (!hba
|| !hba
->path
)
5955 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
5956 if (strstr(idev
->path
, hba
->path
))
5960 if (!idev
|| !idev
->dev_id
)
5963 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
5965 if (!entry
|| !entry
->devid_list
)
5968 devid_list
= entry
->devid_list
;
5969 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
5970 struct md_list
*devlist
= NULL
;
5971 struct sys_dev
*device
= device_by_id(dv
->devid
);
5976 hba_path
= device
->path
;
5980 /* VMD has one orom entry for all domain, but spanning is not allowed.
5981 * VMD arrays should be counted per domain (controller), so skip
5982 * domains that are not the given one.
5984 if ((hba
->type
== SYS_DEV_VMD
) &&
5985 (strncmp(device
->path
, hba
->path
, strlen(device
->path
)) != 0))
5988 devlist
= get_devices(hba_path
);
5989 /* if no intel devices return zero volumes */
5990 if (devlist
== NULL
)
5993 count
+= active_arrays_by_format("imsm", hba_path
, &devlist
, dpa
, verbose
);
5994 dprintf("path: %s active arrays: %d\n", hba_path
, count
);
5995 if (devlist
== NULL
)
5999 count
+= count_volumes_list(devlist
,
6003 dprintf("found %d count: %d\n", found
, count
);
6006 dprintf("path: %s total number of volumes: %d\n", hba_path
, count
);
6009 struct md_list
*dv
= devlist
;
6010 devlist
= devlist
->next
;
6018 static int imsm_default_chunk(const struct imsm_orom
*orom
)
6020 /* up to 512 if the plaform supports it, otherwise the platform max.
6021 * 128 if no platform detected
6023 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
6025 return min(512, (1 << fs
));
6029 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
6030 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
6032 /* check/set platform and metadata limits/defaults */
6033 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
6034 pr_vrb(": platform supports a maximum of %d disks per array\n",
6039 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6040 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
6041 pr_vrb(": platform does not support raid%d with %d disk%s\n",
6042 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
6046 if (*chunk
== 0 || *chunk
== UnSet
)
6047 *chunk
= imsm_default_chunk(super
->orom
);
6049 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
6050 pr_vrb(": platform does not support a chunk size of: %d\n", *chunk
);
6054 if (layout
!= imsm_level_to_layout(level
)) {
6056 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
6057 else if (level
== 10)
6058 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
6060 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
6065 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
6066 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
6067 pr_vrb(": platform does not support a volume size over 2TB\n");
6074 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
6075 * FIX ME add ahci details
6077 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
6078 int layout
, int raiddisks
, int *chunk
,
6079 unsigned long long size
,
6080 unsigned long long data_offset
,
6082 unsigned long long *freesize
,
6086 struct intel_super
*super
= st
->sb
;
6087 struct imsm_super
*mpb
;
6089 unsigned long long pos
= 0;
6090 unsigned long long maxsize
;
6094 /* We must have the container info already read in. */
6098 mpb
= super
->anchor
;
6100 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
6101 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6105 /* General test: make sure there is space for
6106 * 'raiddisks' device extents of size 'size' at a given
6109 unsigned long long minsize
= size
;
6110 unsigned long long start_offset
= MaxSector
;
6113 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
6114 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6119 e
= get_extents(super
, dl
);
6122 unsigned long long esize
;
6123 esize
= e
[i
].start
- pos
;
6124 if (esize
>= minsize
)
6126 if (found
&& start_offset
== MaxSector
) {
6129 } else if (found
&& pos
!= start_offset
) {
6133 pos
= e
[i
].start
+ e
[i
].size
;
6135 } while (e
[i
-1].size
);
6140 if (dcnt
< raiddisks
) {
6142 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6149 /* This device must be a member of the set */
6150 if (stat(dev
, &stb
) < 0)
6152 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6154 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6155 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6156 dl
->minor
== (int)minor(stb
.st_rdev
))
6161 pr_err("%s is not in the same imsm set\n", dev
);
6163 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6164 /* If a volume is present then the current creation attempt
6165 * cannot incorporate new spares because the orom may not
6166 * understand this configuration (all member disks must be
6167 * members of each array in the container).
6169 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6170 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6172 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6173 mpb
->num_disks
!= raiddisks
) {
6174 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6178 /* retrieve the largest free space block */
6179 e
= get_extents(super
, dl
);
6184 unsigned long long esize
;
6186 esize
= e
[i
].start
- pos
;
6187 if (esize
>= maxsize
)
6189 pos
= e
[i
].start
+ e
[i
].size
;
6191 } while (e
[i
-1].size
);
6196 pr_err("unable to determine free space for: %s\n",
6200 if (maxsize
< size
) {
6202 pr_err("%s not enough space (%llu < %llu)\n",
6203 dev
, maxsize
, size
);
6207 /* count total number of extents for merge */
6209 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6211 i
+= dl
->extent_cnt
;
6213 maxsize
= merge_extents(super
, i
);
6215 if (!check_env("IMSM_NO_PLATFORM") &&
6216 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6217 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6221 if (maxsize
< size
|| maxsize
== 0) {
6224 pr_err("no free space left on device. Aborting...\n");
6226 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
6232 *freesize
= maxsize
;
6235 int count
= count_volumes(super
->hba
,
6236 super
->orom
->dpa
, verbose
);
6237 if (super
->orom
->vphba
<= count
) {
6238 pr_vrb(": platform does not support more than %d raid volumes.\n",
6239 super
->orom
->vphba
);
6246 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6247 unsigned long long size
, int chunk
,
6248 unsigned long long *freesize
)
6250 struct intel_super
*super
= st
->sb
;
6251 struct imsm_super
*mpb
= super
->anchor
;
6256 unsigned long long maxsize
;
6257 unsigned long long minsize
;
6261 /* find the largest common start free region of the possible disks */
6265 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6271 /* don't activate new spares if we are orom constrained
6272 * and there is already a volume active in the container
6274 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6277 e
= get_extents(super
, dl
);
6280 for (i
= 1; e
[i
-1].size
; i
++)
6288 maxsize
= merge_extents(super
, extent_cnt
);
6292 minsize
= chunk
* 2;
6294 if (cnt
< raiddisks
||
6295 (super
->orom
&& used
&& used
!= raiddisks
) ||
6296 maxsize
< minsize
||
6298 pr_err("not enough devices with space to create array.\n");
6299 return 0; /* No enough free spaces large enough */
6310 if (!check_env("IMSM_NO_PLATFORM") &&
6311 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6312 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6316 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6318 dl
->raiddisk
= cnt
++;
6322 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6327 static int reserve_space(struct supertype
*st
, int raiddisks
,
6328 unsigned long long size
, int chunk
,
6329 unsigned long long *freesize
)
6331 struct intel_super
*super
= st
->sb
;
6336 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6339 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6341 dl
->raiddisk
= cnt
++;
6348 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6349 int raiddisks
, int *chunk
, unsigned long long size
,
6350 unsigned long long data_offset
,
6351 char *dev
, unsigned long long *freesize
,
6359 * if given unused devices create a container
6360 * if given given devices in a container create a member volume
6362 if (level
== LEVEL_CONTAINER
) {
6363 /* Must be a fresh device to add to a container */
6364 return validate_geometry_imsm_container(st
, level
, layout
,
6374 struct intel_super
*super
= st
->sb
;
6375 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6376 raiddisks
, chunk
, size
,
6379 /* we are being asked to automatically layout a
6380 * new volume based on the current contents of
6381 * the container. If the the parameters can be
6382 * satisfied reserve_space will record the disks,
6383 * start offset, and size of the volume to be
6384 * created. add_to_super and getinfo_super
6385 * detect when autolayout is in progress.
6387 /* assuming that freesize is always given when array is
6389 if (super
->orom
&& freesize
) {
6391 count
= count_volumes(super
->hba
,
6392 super
->orom
->dpa
, verbose
);
6393 if (super
->orom
->vphba
<= count
) {
6394 pr_vrb(": platform does not support more than %d raid volumes.\n",
6395 super
->orom
->vphba
);
6400 return reserve_space(st
, raiddisks
, size
,
6406 /* creating in a given container */
6407 return validate_geometry_imsm_volume(st
, level
, layout
,
6408 raiddisks
, chunk
, size
,
6410 dev
, freesize
, verbose
);
6413 /* This device needs to be a device in an 'imsm' container */
6414 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6417 pr_err("Cannot create this array on device %s\n",
6422 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6424 pr_err("Cannot open %s: %s\n",
6425 dev
, strerror(errno
));
6428 /* Well, it is in use by someone, maybe an 'imsm' container. */
6429 cfd
= open_container(fd
);
6433 pr_err("Cannot use %s: It is busy\n",
6437 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
6438 if (sra
&& sra
->array
.major_version
== -1 &&
6439 strcmp(sra
->text_version
, "imsm") == 0)
6443 /* This is a member of a imsm container. Load the container
6444 * and try to create a volume
6446 struct intel_super
*super
;
6448 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6450 strcpy(st
->container_devnm
, fd2devnm(cfd
));
6452 return validate_geometry_imsm_volume(st
, level
, layout
,
6454 size
, data_offset
, dev
,
6461 pr_err("failed container membership check\n");
6467 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6469 struct intel_super
*super
= st
->sb
;
6471 if (level
&& *level
== UnSet
)
6472 *level
= LEVEL_CONTAINER
;
6474 if (level
&& layout
&& *layout
== UnSet
)
6475 *layout
= imsm_level_to_layout(*level
);
6477 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6478 *chunk
= imsm_default_chunk(super
->orom
);
6481 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6483 static int kill_subarray_imsm(struct supertype
*st
)
6485 /* remove the subarray currently referenced by ->current_vol */
6487 struct intel_dev
**dp
;
6488 struct intel_super
*super
= st
->sb
;
6489 __u8 current_vol
= super
->current_vol
;
6490 struct imsm_super
*mpb
= super
->anchor
;
6492 if (super
->current_vol
< 0)
6494 super
->current_vol
= -1; /* invalidate subarray cursor */
6496 /* block deletions that would change the uuid of active subarrays
6498 * FIXME when immutable ids are available, but note that we'll
6499 * also need to fixup the invalidated/active subarray indexes in
6502 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6505 if (i
< current_vol
)
6507 sprintf(subarray
, "%u", i
);
6508 if (is_subarray_active(subarray
, st
->devnm
)) {
6509 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6516 if (st
->update_tail
) {
6517 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
6519 u
->type
= update_kill_array
;
6520 u
->dev_idx
= current_vol
;
6521 append_metadata_update(st
, u
, sizeof(*u
));
6526 for (dp
= &super
->devlist
; *dp
;)
6527 if ((*dp
)->index
== current_vol
) {
6530 handle_missing(super
, (*dp
)->dev
);
6531 if ((*dp
)->index
> current_vol
)
6536 /* no more raid devices, all active components are now spares,
6537 * but of course failed are still failed
6539 if (--mpb
->num_raid_devs
== 0) {
6542 for (d
= super
->disks
; d
; d
= d
->next
)
6547 super
->updates_pending
++;
6552 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6553 char *update
, struct mddev_ident
*ident
)
6555 /* update the subarray currently referenced by ->current_vol */
6556 struct intel_super
*super
= st
->sb
;
6557 struct imsm_super
*mpb
= super
->anchor
;
6559 if (strcmp(update
, "name") == 0) {
6560 char *name
= ident
->name
;
6564 if (is_subarray_active(subarray
, st
->devnm
)) {
6565 pr_err("Unable to update name of active subarray\n");
6569 if (!check_name(super
, name
, 0))
6572 vol
= strtoul(subarray
, &ep
, 10);
6573 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6576 if (st
->update_tail
) {
6577 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
6579 u
->type
= update_rename_array
;
6581 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6582 append_metadata_update(st
, u
, sizeof(*u
));
6584 struct imsm_dev
*dev
;
6587 dev
= get_imsm_dev(super
, vol
);
6588 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6589 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6590 dev
= get_imsm_dev(super
, i
);
6591 handle_missing(super
, dev
);
6593 super
->updates_pending
++;
6600 #endif /* MDASSEMBLE */
6602 static int is_gen_migration(struct imsm_dev
*dev
)
6607 if (!dev
->vol
.migr_state
)
6610 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6616 static int is_rebuilding(struct imsm_dev
*dev
)
6618 struct imsm_map
*migr_map
;
6620 if (!dev
->vol
.migr_state
)
6623 if (migr_type(dev
) != MIGR_REBUILD
)
6626 migr_map
= get_imsm_map(dev
, MAP_1
);
6628 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6635 static int is_initializing(struct imsm_dev
*dev
)
6637 struct imsm_map
*migr_map
;
6639 if (!dev
->vol
.migr_state
)
6642 if (migr_type(dev
) != MIGR_INIT
)
6645 migr_map
= get_imsm_map(dev
, MAP_1
);
6647 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6654 static void update_recovery_start(struct intel_super
*super
,
6655 struct imsm_dev
*dev
,
6656 struct mdinfo
*array
)
6658 struct mdinfo
*rebuild
= NULL
;
6662 if (!is_rebuilding(dev
))
6665 /* Find the rebuild target, but punt on the dual rebuild case */
6666 for (d
= array
->devs
; d
; d
= d
->next
)
6667 if (d
->recovery_start
== 0) {
6674 /* (?) none of the disks are marked with
6675 * IMSM_ORD_REBUILD, so assume they are missing and the
6676 * disk_ord_tbl was not correctly updated
6678 dprintf("failed to locate out-of-sync disk\n");
6682 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6683 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6687 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6690 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6692 /* Given a container loaded by load_super_imsm_all,
6693 * extract information about all the arrays into
6695 * If 'subarray' is given, just extract info about that array.
6697 * For each imsm_dev create an mdinfo, fill it in,
6698 * then look for matching devices in super->disks
6699 * and create appropriate device mdinfo.
6701 struct intel_super
*super
= st
->sb
;
6702 struct imsm_super
*mpb
= super
->anchor
;
6703 struct mdinfo
*rest
= NULL
;
6707 int spare_disks
= 0;
6709 /* do not assemble arrays when not all attributes are supported */
6710 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6712 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
6715 /* check for bad blocks */
6716 if (imsm_bbm_log_size(super
->anchor
)) {
6717 pr_err("BBM log found in IMSM metadata.Arrays activation is blocked.\n");
6721 /* count spare devices, not used in maps
6723 for (d
= super
->disks
; d
; d
= d
->next
)
6727 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6728 struct imsm_dev
*dev
;
6729 struct imsm_map
*map
;
6730 struct imsm_map
*map2
;
6731 struct mdinfo
*this;
6739 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6742 dev
= get_imsm_dev(super
, i
);
6743 map
= get_imsm_map(dev
, MAP_0
);
6744 map2
= get_imsm_map(dev
, MAP_1
);
6746 /* do not publish arrays that are in the middle of an
6747 * unsupported migration
6749 if (dev
->vol
.migr_state
&&
6750 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6751 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
6755 /* do not publish arrays that are not support by controller's
6759 this = xmalloc(sizeof(*this));
6761 super
->current_vol
= i
;
6762 getinfo_super_imsm_volume(st
, this, NULL
);
6765 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6766 /* mdadm does not support all metadata features- set the bit in all arrays state */
6767 if (!validate_geometry_imsm_orom(super
,
6768 get_imsm_raid_level(map
), /* RAID level */
6769 imsm_level_to_layout(get_imsm_raid_level(map
)),
6770 map
->num_members
, /* raid disks */
6771 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6773 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
6775 this->array
.state
|=
6776 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6777 (1<<MD_SB_BLOCK_VOLUME
);
6781 /* if array has bad blocks, set suitable bit in all arrays state */
6783 this->array
.state
|=
6784 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6785 (1<<MD_SB_BLOCK_VOLUME
);
6787 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6788 unsigned long long recovery_start
;
6789 struct mdinfo
*info_d
;
6796 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6797 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6798 for (d
= super
->disks
; d
; d
= d
->next
)
6799 if (d
->index
== idx
)
6802 recovery_start
= MaxSector
;
6805 if (d
&& is_failed(&d
->disk
))
6807 if (ord
& IMSM_ORD_REBUILD
)
6811 * if we skip some disks the array will be assmebled degraded;
6812 * reset resync start to avoid a dirty-degraded
6813 * situation when performing the intial sync
6815 * FIXME handle dirty degraded
6817 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6818 this->resync_start
= MaxSector
;
6822 info_d
= xcalloc(1, sizeof(*info_d
));
6823 info_d
->next
= this->devs
;
6824 this->devs
= info_d
;
6826 info_d
->disk
.number
= d
->index
;
6827 info_d
->disk
.major
= d
->major
;
6828 info_d
->disk
.minor
= d
->minor
;
6829 info_d
->disk
.raid_disk
= slot
;
6830 info_d
->recovery_start
= recovery_start
;
6832 if (slot
< map2
->num_members
)
6833 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6835 this->array
.spare_disks
++;
6837 if (slot
< map
->num_members
)
6838 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6840 this->array
.spare_disks
++;
6842 if (info_d
->recovery_start
== MaxSector
)
6843 this->array
.working_disks
++;
6845 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6846 info_d
->data_offset
= pba_of_lba0(map
);
6847 info_d
->component_size
= blocks_per_member(map
);
6849 /* now that the disk list is up-to-date fixup recovery_start */
6850 update_recovery_start(super
, dev
, this);
6851 this->array
.spare_disks
+= spare_disks
;
6854 /* check for reshape */
6855 if (this->reshape_active
== 1)
6856 recover_backup_imsm(st
, this);
6864 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6865 int failed
, int look_in_map
)
6867 struct imsm_map
*map
;
6869 map
= get_imsm_map(dev
, look_in_map
);
6872 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6873 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6875 switch (get_imsm_raid_level(map
)) {
6877 return IMSM_T_STATE_FAILED
;
6880 if (failed
< map
->num_members
)
6881 return IMSM_T_STATE_DEGRADED
;
6883 return IMSM_T_STATE_FAILED
;
6888 * check to see if any mirrors have failed, otherwise we
6889 * are degraded. Even numbered slots are mirrored on
6893 /* gcc -Os complains that this is unused */
6894 int insync
= insync
;
6896 for (i
= 0; i
< map
->num_members
; i
++) {
6897 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6898 int idx
= ord_to_idx(ord
);
6899 struct imsm_disk
*disk
;
6901 /* reset the potential in-sync count on even-numbered
6902 * slots. num_copies is always 2 for imsm raid10
6907 disk
= get_imsm_disk(super
, idx
);
6908 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6911 /* no in-sync disks left in this mirror the
6915 return IMSM_T_STATE_FAILED
;
6918 return IMSM_T_STATE_DEGRADED
;
6922 return IMSM_T_STATE_DEGRADED
;
6924 return IMSM_T_STATE_FAILED
;
6930 return map
->map_state
;
6933 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6938 struct imsm_disk
*disk
;
6939 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6940 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6941 struct imsm_map
*map_for_loop
;
6946 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6947 * disks that are being rebuilt. New failures are recorded to
6948 * map[0]. So we look through all the disks we started with and
6949 * see if any failures are still present, or if any new ones
6953 if (prev
&& (map
->num_members
< prev
->num_members
))
6954 map_for_loop
= prev
;
6956 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6958 /* when MAP_X is passed both maps failures are counted
6961 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6962 (i
< prev
->num_members
)) {
6963 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6964 idx_1
= ord_to_idx(ord
);
6966 disk
= get_imsm_disk(super
, idx_1
);
6967 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6970 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6971 (i
< map
->num_members
)) {
6972 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6973 idx
= ord_to_idx(ord
);
6976 disk
= get_imsm_disk(super
, idx
);
6977 if (!disk
|| is_failed(disk
) ||
6978 ord
& IMSM_ORD_REBUILD
)
6988 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6991 struct intel_super
*super
= c
->sb
;
6992 struct imsm_super
*mpb
= super
->anchor
;
6994 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6995 pr_err("subarry index %d, out of range\n", atoi(inst
));
6999 dprintf("imsm: open_new %s\n", inst
);
7000 a
->info
.container_member
= atoi(inst
);
7004 static int is_resyncing(struct imsm_dev
*dev
)
7006 struct imsm_map
*migr_map
;
7008 if (!dev
->vol
.migr_state
)
7011 if (migr_type(dev
) == MIGR_INIT
||
7012 migr_type(dev
) == MIGR_REPAIR
)
7015 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7018 migr_map
= get_imsm_map(dev
, MAP_1
);
7020 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
7021 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
7027 /* return true if we recorded new information */
7028 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7032 struct imsm_map
*map
;
7033 char buf
[MAX_RAID_SERIAL_LEN
+3];
7034 unsigned int len
, shift
= 0;
7036 /* new failures are always set in map[0] */
7037 map
= get_imsm_map(dev
, MAP_0
);
7039 slot
= get_imsm_disk_slot(map
, idx
);
7043 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
7044 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
7047 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
7048 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
7050 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
7051 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
7052 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
7054 disk
->status
|= FAILED_DISK
;
7055 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
7056 /* mark failures in second map if second map exists and this disk
7058 * This is valid for migration, initialization and rebuild
7060 if (dev
->vol
.migr_state
) {
7061 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
7062 int slot2
= get_imsm_disk_slot(map2
, idx
);
7064 if ((slot2
< map2
->num_members
) &&
7066 set_imsm_ord_tbl_ent(map2
, slot2
,
7067 idx
| IMSM_ORD_REBUILD
);
7069 if (map
->failed_disk_num
== 0xff)
7070 map
->failed_disk_num
= slot
;
7074 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7076 mark_failure(dev
, disk
, idx
);
7078 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
7081 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
7082 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
7085 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7089 if (!super
->missing
)
7092 /* When orom adds replacement for missing disk it does
7093 * not remove entry of missing disk, but just updates map with
7094 * new added disk. So it is not enough just to test if there is
7095 * any missing disk, we have to look if there are any failed disks
7096 * in map to stop migration */
7098 dprintf("imsm: mark missing\n");
7099 /* end process for initialization and rebuild only
7101 if (is_gen_migration(dev
) == 0) {
7105 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7106 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7109 end_migration(dev
, super
, map_state
);
7111 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7112 mark_missing(dev
, &dl
->disk
, dl
->index
);
7113 super
->updates_pending
++;
7116 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7119 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7120 unsigned long long array_blocks
;
7121 struct imsm_map
*map
;
7123 if (used_disks
== 0) {
7124 /* when problems occures
7125 * return current array_blocks value
7127 array_blocks
= __le32_to_cpu(dev
->size_high
);
7128 array_blocks
= array_blocks
<< 32;
7129 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7131 return array_blocks
;
7134 /* set array size in metadata
7136 if (new_size
<= 0) {
7137 /* OLCE size change is caused by added disks
7139 map
= get_imsm_map(dev
, MAP_0
);
7140 array_blocks
= blocks_per_member(map
) * used_disks
;
7142 /* Online Volume Size Change
7143 * Using available free space
7145 array_blocks
= new_size
;
7148 /* round array size down to closest MB
7150 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7151 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7152 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7154 return array_blocks
;
7157 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7159 static void imsm_progress_container_reshape(struct intel_super
*super
)
7161 /* if no device has a migr_state, but some device has a
7162 * different number of members than the previous device, start
7163 * changing the number of devices in this device to match
7166 struct imsm_super
*mpb
= super
->anchor
;
7167 int prev_disks
= -1;
7171 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7172 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7173 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7174 struct imsm_map
*map2
;
7175 int prev_num_members
;
7177 if (dev
->vol
.migr_state
)
7180 if (prev_disks
== -1)
7181 prev_disks
= map
->num_members
;
7182 if (prev_disks
== map
->num_members
)
7185 /* OK, this array needs to enter reshape mode.
7186 * i.e it needs a migr_state
7189 copy_map_size
= sizeof_imsm_map(map
);
7190 prev_num_members
= map
->num_members
;
7191 map
->num_members
= prev_disks
;
7192 dev
->vol
.migr_state
= 1;
7193 dev
->vol
.curr_migr_unit
= 0;
7194 set_migr_type(dev
, MIGR_GEN_MIGR
);
7195 for (i
= prev_num_members
;
7196 i
< map
->num_members
; i
++)
7197 set_imsm_ord_tbl_ent(map
, i
, i
);
7198 map2
= get_imsm_map(dev
, MAP_1
);
7199 /* Copy the current map */
7200 memcpy(map2
, map
, copy_map_size
);
7201 map2
->num_members
= prev_num_members
;
7203 imsm_set_array_size(dev
, -1);
7204 super
->clean_migration_record_by_mdmon
= 1;
7205 super
->updates_pending
++;
7209 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7210 * states are handled in imsm_set_disk() with one exception, when a
7211 * resync is stopped due to a new failure this routine will set the
7212 * 'degraded' state for the array.
7214 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7216 int inst
= a
->info
.container_member
;
7217 struct intel_super
*super
= a
->container
->sb
;
7218 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7219 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7220 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7221 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7222 __u32 blocks_per_unit
;
7224 if (dev
->vol
.migr_state
&&
7225 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7226 /* array state change is blocked due to reshape action
7228 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7229 * - finish the reshape (if last_checkpoint is big and action != reshape)
7230 * - update curr_migr_unit
7232 if (a
->curr_action
== reshape
) {
7233 /* still reshaping, maybe update curr_migr_unit */
7234 goto mark_checkpoint
;
7236 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7237 /* for some reason we aborted the reshape.
7239 * disable automatic metadata rollback
7240 * user action is required to recover process
7243 struct imsm_map
*map2
=
7244 get_imsm_map(dev
, MAP_1
);
7245 dev
->vol
.migr_state
= 0;
7246 set_migr_type(dev
, 0);
7247 dev
->vol
.curr_migr_unit
= 0;
7249 sizeof_imsm_map(map2
));
7250 super
->updates_pending
++;
7253 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7254 unsigned long long array_blocks
;
7258 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7259 if (used_disks
> 0) {
7261 blocks_per_member(map
) *
7263 /* round array size down to closest MB
7265 array_blocks
= (array_blocks
7266 >> SECT_PER_MB_SHIFT
)
7267 << SECT_PER_MB_SHIFT
;
7268 a
->info
.custom_array_size
= array_blocks
;
7269 /* encourage manager to update array
7273 a
->check_reshape
= 1;
7275 /* finalize online capacity expansion/reshape */
7276 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7278 mdi
->disk
.raid_disk
,
7281 imsm_progress_container_reshape(super
);
7286 /* before we activate this array handle any missing disks */
7287 if (consistent
== 2)
7288 handle_missing(super
, dev
);
7290 if (consistent
== 2 &&
7291 (!is_resync_complete(&a
->info
) ||
7292 map_state
!= IMSM_T_STATE_NORMAL
||
7293 dev
->vol
.migr_state
))
7296 if (is_resync_complete(&a
->info
)) {
7297 /* complete intialization / resync,
7298 * recovery and interrupted recovery is completed in
7301 if (is_resyncing(dev
)) {
7302 dprintf("imsm: mark resync done\n");
7303 end_migration(dev
, super
, map_state
);
7304 super
->updates_pending
++;
7305 a
->last_checkpoint
= 0;
7307 } else if ((!is_resyncing(dev
) && !failed
) &&
7308 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7309 /* mark the start of the init process if nothing is failed */
7310 dprintf("imsm: mark resync start\n");
7311 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7312 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7314 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7315 super
->updates_pending
++;
7319 /* skip checkpointing for general migration,
7320 * it is controlled in mdadm
7322 if (is_gen_migration(dev
))
7323 goto skip_mark_checkpoint
;
7325 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7326 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7327 if (blocks_per_unit
) {
7331 units
= a
->last_checkpoint
/ blocks_per_unit
;
7334 /* check that we did not overflow 32-bits, and that
7335 * curr_migr_unit needs updating
7337 if (units32
== units
&&
7339 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7340 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7341 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7342 super
->updates_pending
++;
7346 skip_mark_checkpoint
:
7347 /* mark dirty / clean */
7348 if (dev
->vol
.dirty
!= !consistent
) {
7349 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7354 super
->updates_pending
++;
7360 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7362 int inst
= a
->info
.container_member
;
7363 struct intel_super
*super
= a
->container
->sb
;
7364 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7365 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7366 struct imsm_disk
*disk
;
7368 int recovery_not_finished
= 0;
7373 if (n
> map
->num_members
)
7374 pr_err("imsm: set_disk %d out of range 0..%d\n",
7375 n
, map
->num_members
- 1);
7380 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7382 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7383 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7385 /* check for new failures */
7386 if (state
& DS_FAULTY
) {
7387 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7388 super
->updates_pending
++;
7391 /* check if in_sync */
7392 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7393 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7395 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7396 super
->updates_pending
++;
7399 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7400 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7402 /* check if recovery complete, newly degraded, or failed */
7403 dprintf("imsm: Detected transition to state ");
7404 switch (map_state
) {
7405 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7406 dprintf("normal: ");
7407 if (is_rebuilding(dev
)) {
7408 dprintf_cont("while rebuilding");
7409 /* check if recovery is really finished */
7410 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7411 if (mdi
->recovery_start
!= MaxSector
) {
7412 recovery_not_finished
= 1;
7415 if (recovery_not_finished
) {
7417 dprintf("Rebuild has not finished yet, state not changed");
7418 if (a
->last_checkpoint
< mdi
->recovery_start
) {
7419 a
->last_checkpoint
= mdi
->recovery_start
;
7420 super
->updates_pending
++;
7424 end_migration(dev
, super
, map_state
);
7425 map
= get_imsm_map(dev
, MAP_0
);
7426 map
->failed_disk_num
= ~0;
7427 super
->updates_pending
++;
7428 a
->last_checkpoint
= 0;
7431 if (is_gen_migration(dev
)) {
7432 dprintf_cont("while general migration");
7433 if (a
->last_checkpoint
>= a
->info
.component_size
)
7434 end_migration(dev
, super
, map_state
);
7436 map
->map_state
= map_state
;
7437 map
= get_imsm_map(dev
, MAP_0
);
7438 map
->failed_disk_num
= ~0;
7439 super
->updates_pending
++;
7443 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7444 dprintf_cont("degraded: ");
7445 if ((map
->map_state
!= map_state
) &&
7446 !dev
->vol
.migr_state
) {
7447 dprintf_cont("mark degraded");
7448 map
->map_state
= map_state
;
7449 super
->updates_pending
++;
7450 a
->last_checkpoint
= 0;
7453 if (is_rebuilding(dev
)) {
7454 dprintf_cont("while rebuilding.");
7455 if (map
->map_state
!= map_state
) {
7456 dprintf_cont(" Map state change");
7457 end_migration(dev
, super
, map_state
);
7458 super
->updates_pending
++;
7462 if (is_gen_migration(dev
)) {
7463 dprintf_cont("while general migration");
7464 if (a
->last_checkpoint
>= a
->info
.component_size
)
7465 end_migration(dev
, super
, map_state
);
7467 map
->map_state
= map_state
;
7468 manage_second_map(super
, dev
);
7470 super
->updates_pending
++;
7473 if (is_initializing(dev
)) {
7474 dprintf_cont("while initialization.");
7475 map
->map_state
= map_state
;
7476 super
->updates_pending
++;
7480 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7481 dprintf_cont("failed: ");
7482 if (is_gen_migration(dev
)) {
7483 dprintf_cont("while general migration");
7484 map
->map_state
= map_state
;
7485 super
->updates_pending
++;
7488 if (map
->map_state
!= map_state
) {
7489 dprintf_cont("mark failed");
7490 end_migration(dev
, super
, map_state
);
7491 super
->updates_pending
++;
7492 a
->last_checkpoint
= 0;
7497 dprintf_cont("state %i\n", map_state
);
7502 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7505 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7506 unsigned long long dsize
;
7507 unsigned long long sectors
;
7509 get_dev_size(fd
, NULL
, &dsize
);
7511 if (mpb_size
> 512) {
7512 /* -1 to account for anchor */
7513 sectors
= mpb_sectors(mpb
) - 1;
7515 /* write the extended mpb to the sectors preceeding the anchor */
7516 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7519 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7524 /* first block is stored on second to last sector of the disk */
7525 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7528 if (write(fd
, buf
, 512) != 512)
7534 static void imsm_sync_metadata(struct supertype
*container
)
7536 struct intel_super
*super
= container
->sb
;
7538 dprintf("sync metadata: %d\n", super
->updates_pending
);
7539 if (!super
->updates_pending
)
7542 write_super_imsm(container
, 0);
7544 super
->updates_pending
= 0;
7547 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7549 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7550 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7553 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7557 if (dl
&& is_failed(&dl
->disk
))
7561 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
7566 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7567 struct active_array
*a
, int activate_new
,
7568 struct mdinfo
*additional_test_list
)
7570 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7571 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7572 struct imsm_super
*mpb
= super
->anchor
;
7573 struct imsm_map
*map
;
7574 unsigned long long pos
;
7579 __u32 array_start
= 0;
7580 __u32 array_end
= 0;
7582 struct mdinfo
*test_list
;
7584 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7585 /* If in this array, skip */
7586 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7587 if (d
->state_fd
>= 0 &&
7588 d
->disk
.major
== dl
->major
&&
7589 d
->disk
.minor
== dl
->minor
) {
7590 dprintf("%x:%x already in array\n",
7591 dl
->major
, dl
->minor
);
7596 test_list
= additional_test_list
;
7598 if (test_list
->disk
.major
== dl
->major
&&
7599 test_list
->disk
.minor
== dl
->minor
) {
7600 dprintf("%x:%x already in additional test list\n",
7601 dl
->major
, dl
->minor
);
7604 test_list
= test_list
->next
;
7609 /* skip in use or failed drives */
7610 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7612 dprintf("%x:%x status (failed: %d index: %d)\n",
7613 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7617 /* skip pure spares when we are looking for partially
7618 * assimilated drives
7620 if (dl
->index
== -1 && !activate_new
)
7623 /* Does this unused device have the requisite free space?
7624 * It needs to be able to cover all member volumes
7626 ex
= get_extents(super
, dl
);
7628 dprintf("cannot get extents\n");
7631 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7632 dev
= get_imsm_dev(super
, i
);
7633 map
= get_imsm_map(dev
, MAP_0
);
7635 /* check if this disk is already a member of
7638 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7644 array_start
= pba_of_lba0(map
);
7645 array_end
= array_start
+
7646 blocks_per_member(map
) - 1;
7649 /* check that we can start at pba_of_lba0 with
7650 * blocks_per_member of space
7652 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7656 pos
= ex
[j
].start
+ ex
[j
].size
;
7658 } while (ex
[j
-1].size
);
7665 if (i
< mpb
->num_raid_devs
) {
7666 dprintf("%x:%x does not have %u to %u available\n",
7667 dl
->major
, dl
->minor
, array_start
, array_end
);
7677 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7679 struct imsm_dev
*dev2
;
7680 struct imsm_map
*map
;
7686 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7688 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7689 if (state
== IMSM_T_STATE_FAILED
) {
7690 map
= get_imsm_map(dev2
, MAP_0
);
7693 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7695 * Check if failed disks are deleted from intel
7696 * disk list or are marked to be deleted
7698 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7699 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7701 * Do not rebuild the array if failed disks
7702 * from failed sub-array are not removed from
7706 is_failed(&idisk
->disk
) &&
7707 (idisk
->action
!= DISK_REMOVE
))
7715 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7716 struct metadata_update
**updates
)
7719 * Find a device with unused free space and use it to replace a
7720 * failed/vacant region in an array. We replace failed regions one a
7721 * array at a time. The result is that a new spare disk will be added
7722 * to the first failed array and after the monitor has finished
7723 * propagating failures the remainder will be consumed.
7725 * FIXME add a capability for mdmon to request spares from another
7729 struct intel_super
*super
= a
->container
->sb
;
7730 int inst
= a
->info
.container_member
;
7731 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7732 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7733 int failed
= a
->info
.array
.raid_disks
;
7734 struct mdinfo
*rv
= NULL
;
7737 struct metadata_update
*mu
;
7739 struct imsm_update_activate_spare
*u
;
7744 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7745 if ((d
->curr_state
& DS_FAULTY
) &&
7747 /* wait for Removal to happen */
7749 if (d
->state_fd
>= 0)
7753 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7754 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7756 if (imsm_reshape_blocks_arrays_changes(super
))
7759 /* Cannot activate another spare if rebuild is in progress already
7761 if (is_rebuilding(dev
)) {
7762 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
7766 if (a
->info
.array
.level
== 4)
7767 /* No repair for takeovered array
7768 * imsm doesn't support raid4
7772 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7773 IMSM_T_STATE_DEGRADED
)
7777 * If there are any failed disks check state of the other volume.
7778 * Block rebuild if the another one is failed until failed disks
7779 * are removed from container.
7782 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
7783 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7784 /* check if states of the other volumes allow for rebuild */
7785 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7787 allowed
= imsm_rebuild_allowed(a
->container
,
7795 /* For each slot, if it is not working, find a spare */
7796 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7797 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7798 if (d
->disk
.raid_disk
== i
)
7800 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7801 if (d
&& (d
->state_fd
>= 0))
7805 * OK, this device needs recovery. Try to re-add the
7806 * previous occupant of this slot, if this fails see if
7807 * we can continue the assimilation of a spare that was
7808 * partially assimilated, finally try to activate a new
7811 dl
= imsm_readd(super
, i
, a
);
7813 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7815 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7819 /* found a usable disk with enough space */
7820 di
= xcalloc(1, sizeof(*di
));
7822 /* dl->index will be -1 in the case we are activating a
7823 * pristine spare. imsm_process_update() will create a
7824 * new index in this case. Once a disk is found to be
7825 * failed in all member arrays it is kicked from the
7828 di
->disk
.number
= dl
->index
;
7830 /* (ab)use di->devs to store a pointer to the device
7833 di
->devs
= (struct mdinfo
*) dl
;
7835 di
->disk
.raid_disk
= i
;
7836 di
->disk
.major
= dl
->major
;
7837 di
->disk
.minor
= dl
->minor
;
7839 di
->recovery_start
= 0;
7840 di
->data_offset
= pba_of_lba0(map
);
7841 di
->component_size
= a
->info
.component_size
;
7842 di
->container_member
= inst
;
7843 super
->random
= random32();
7847 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7848 i
, di
->data_offset
);
7852 /* No spares found */
7854 /* Now 'rv' has a list of devices to return.
7855 * Create a metadata_update record to update the
7856 * disk_ord_tbl for the array
7858 mu
= xmalloc(sizeof(*mu
));
7859 mu
->buf
= xcalloc(num_spares
,
7860 sizeof(struct imsm_update_activate_spare
));
7862 mu
->space_list
= NULL
;
7863 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7864 mu
->next
= *updates
;
7865 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7867 for (di
= rv
; di
; di
= di
->next
) {
7868 u
->type
= update_activate_spare
;
7869 u
->dl
= (struct dl
*) di
->devs
;
7871 u
->slot
= di
->disk
.raid_disk
;
7882 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7884 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7885 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7886 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7887 struct disk_info
*inf
= get_disk_info(u
);
7888 struct imsm_disk
*disk
;
7892 for (i
= 0; i
< map
->num_members
; i
++) {
7893 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7894 for (j
= 0; j
< new_map
->num_members
; j
++)
7895 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7902 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7904 struct dl
*dl
= NULL
;
7905 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7906 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7911 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7913 struct dl
*prev
= NULL
;
7917 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7918 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7921 prev
->next
= dl
->next
;
7923 super
->disks
= dl
->next
;
7925 __free_imsm_disk(dl
);
7926 dprintf("removed %x:%x\n", major
, minor
);
7934 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7936 static int add_remove_disk_update(struct intel_super
*super
)
7938 int check_degraded
= 0;
7939 struct dl
*disk
= NULL
;
7940 /* add/remove some spares to/from the metadata/contrainer */
7941 while (super
->disk_mgmt_list
) {
7942 struct dl
*disk_cfg
;
7944 disk_cfg
= super
->disk_mgmt_list
;
7945 super
->disk_mgmt_list
= disk_cfg
->next
;
7946 disk_cfg
->next
= NULL
;
7948 if (disk_cfg
->action
== DISK_ADD
) {
7949 disk_cfg
->next
= super
->disks
;
7950 super
->disks
= disk_cfg
;
7952 dprintf("added %x:%x\n",
7953 disk_cfg
->major
, disk_cfg
->minor
);
7954 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7955 dprintf("Disk remove action processed: %x.%x\n",
7956 disk_cfg
->major
, disk_cfg
->minor
);
7957 disk
= get_disk_super(super
,
7961 /* store action status */
7962 disk
->action
= DISK_REMOVE
;
7963 /* remove spare disks only */
7964 if (disk
->index
== -1) {
7965 remove_disk_super(super
,
7970 /* release allocate disk structure */
7971 __free_imsm_disk(disk_cfg
);
7974 return check_degraded
;
7977 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7978 struct intel_super
*super
,
7981 struct intel_dev
*id
;
7982 void **tofree
= NULL
;
7985 dprintf("(enter)\n");
7986 if ((u
->subdev
< 0) ||
7988 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7991 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7992 dprintf("imsm: Error: Memory is not allocated\n");
7996 for (id
= super
->devlist
; id
; id
= id
->next
) {
7997 if (id
->index
== (unsigned)u
->subdev
) {
7998 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7999 struct imsm_map
*map
;
8000 struct imsm_dev
*new_dev
=
8001 (struct imsm_dev
*)*space_list
;
8002 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8004 struct dl
*new_disk
;
8006 if (new_dev
== NULL
)
8008 *space_list
= **space_list
;
8009 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
8010 map
= get_imsm_map(new_dev
, MAP_0
);
8012 dprintf("imsm: Error: migration in progress");
8016 to_state
= map
->map_state
;
8017 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
8019 /* this should not happen */
8020 if (u
->new_disks
[0] < 0) {
8021 map
->failed_disk_num
=
8022 map
->num_members
- 1;
8023 to_state
= IMSM_T_STATE_DEGRADED
;
8025 to_state
= IMSM_T_STATE_NORMAL
;
8027 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
8028 if (u
->new_level
> -1)
8029 map
->raid_level
= u
->new_level
;
8030 migr_map
= get_imsm_map(new_dev
, MAP_1
);
8031 if ((u
->new_level
== 5) &&
8032 (migr_map
->raid_level
== 0)) {
8033 int ord
= map
->num_members
- 1;
8034 migr_map
->num_members
--;
8035 if (u
->new_disks
[0] < 0)
8036 ord
|= IMSM_ORD_REBUILD
;
8037 set_imsm_ord_tbl_ent(map
,
8038 map
->num_members
- 1,
8042 tofree
= (void **)dev
;
8044 /* update chunk size
8046 if (u
->new_chunksize
> 0)
8047 map
->blocks_per_strip
=
8048 __cpu_to_le16(u
->new_chunksize
* 2);
8052 if ((u
->new_level
!= 5) ||
8053 (migr_map
->raid_level
!= 0) ||
8054 (migr_map
->raid_level
== map
->raid_level
))
8057 if (u
->new_disks
[0] >= 0) {
8060 new_disk
= get_disk_super(super
,
8061 major(u
->new_disks
[0]),
8062 minor(u
->new_disks
[0]));
8063 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8064 major(u
->new_disks
[0]),
8065 minor(u
->new_disks
[0]),
8066 new_disk
, new_disk
->index
);
8067 if (new_disk
== NULL
)
8068 goto error_disk_add
;
8070 new_disk
->index
= map
->num_members
- 1;
8071 /* slot to fill in autolayout
8073 new_disk
->raiddisk
= new_disk
->index
;
8074 new_disk
->disk
.status
|= CONFIGURED_DISK
;
8075 new_disk
->disk
.status
&= ~SPARE_DISK
;
8077 goto error_disk_add
;
8080 *tofree
= *space_list
;
8081 /* calculate new size
8083 imsm_set_array_size(new_dev
, -1);
8090 *space_list
= tofree
;
8094 dprintf("Error: imsm: Cannot find disk.\n");
8098 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8099 struct intel_super
*super
)
8101 struct intel_dev
*id
;
8104 dprintf("(enter)\n");
8105 if ((u
->subdev
< 0) ||
8107 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8111 for (id
= super
->devlist
; id
; id
= id
->next
) {
8112 if (id
->index
== (unsigned)u
->subdev
) {
8113 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8114 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8115 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8116 unsigned long long blocks_per_member
;
8118 /* calculate new size
8120 blocks_per_member
= u
->new_size
/ used_disks
;
8121 dprintf("(size: %llu, blocks per member: %llu)\n",
8122 u
->new_size
, blocks_per_member
);
8123 set_blocks_per_member(map
, blocks_per_member
);
8124 imsm_set_array_size(dev
, u
->new_size
);
8134 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8135 struct intel_super
*super
,
8136 struct active_array
*active_array
)
8138 struct imsm_super
*mpb
= super
->anchor
;
8139 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8140 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8141 struct imsm_map
*migr_map
;
8142 struct active_array
*a
;
8143 struct imsm_disk
*disk
;
8150 int second_map_created
= 0;
8152 for (; u
; u
= u
->next
) {
8153 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8158 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8163 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
8168 /* count failures (excluding rebuilds and the victim)
8169 * to determine map[0] state
8172 for (i
= 0; i
< map
->num_members
; i
++) {
8175 disk
= get_imsm_disk(super
,
8176 get_imsm_disk_idx(dev
, i
, MAP_X
));
8177 if (!disk
|| is_failed(disk
))
8181 /* adding a pristine spare, assign a new index */
8182 if (dl
->index
< 0) {
8183 dl
->index
= super
->anchor
->num_disks
;
8184 super
->anchor
->num_disks
++;
8187 disk
->status
|= CONFIGURED_DISK
;
8188 disk
->status
&= ~SPARE_DISK
;
8191 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8192 if (!second_map_created
) {
8193 second_map_created
= 1;
8194 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8195 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8197 map
->map_state
= to_state
;
8198 migr_map
= get_imsm_map(dev
, MAP_1
);
8199 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8200 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8201 dl
->index
| IMSM_ORD_REBUILD
);
8203 /* update the family_num to mark a new container
8204 * generation, being careful to record the existing
8205 * family_num in orig_family_num to clean up after
8206 * earlier mdadm versions that neglected to set it.
8208 if (mpb
->orig_family_num
== 0)
8209 mpb
->orig_family_num
= mpb
->family_num
;
8210 mpb
->family_num
+= super
->random
;
8212 /* count arrays using the victim in the metadata */
8214 for (a
= active_array
; a
; a
= a
->next
) {
8215 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8216 map
= get_imsm_map(dev
, MAP_0
);
8218 if (get_imsm_disk_slot(map
, victim
) >= 0)
8222 /* delete the victim if it is no longer being
8228 /* We know that 'manager' isn't touching anything,
8229 * so it is safe to delete
8231 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8232 if ((*dlp
)->index
== victim
)
8235 /* victim may be on the missing list */
8237 for (dlp
= &super
->missing
; *dlp
;
8238 dlp
= &(*dlp
)->next
)
8239 if ((*dlp
)->index
== victim
)
8241 imsm_delete(super
, dlp
, victim
);
8248 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8249 struct intel_super
*super
,
8252 struct dl
*new_disk
;
8253 struct intel_dev
*id
;
8255 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8256 int disk_count
= u
->old_raid_disks
;
8257 void **tofree
= NULL
;
8258 int devices_to_reshape
= 1;
8259 struct imsm_super
*mpb
= super
->anchor
;
8261 unsigned int dev_id
;
8263 dprintf("(enter)\n");
8265 /* enable spares to use in array */
8266 for (i
= 0; i
< delta_disks
; i
++) {
8267 new_disk
= get_disk_super(super
,
8268 major(u
->new_disks
[i
]),
8269 minor(u
->new_disks
[i
]));
8270 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8271 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8272 new_disk
, new_disk
->index
);
8273 if ((new_disk
== NULL
) ||
8274 ((new_disk
->index
>= 0) &&
8275 (new_disk
->index
< u
->old_raid_disks
)))
8276 goto update_reshape_exit
;
8277 new_disk
->index
= disk_count
++;
8278 /* slot to fill in autolayout
8280 new_disk
->raiddisk
= new_disk
->index
;
8281 new_disk
->disk
.status
|=
8283 new_disk
->disk
.status
&= ~SPARE_DISK
;
8286 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8287 mpb
->num_raid_devs
);
8288 /* manage changes in volume
8290 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8291 void **sp
= *space_list
;
8292 struct imsm_dev
*newdev
;
8293 struct imsm_map
*newmap
, *oldmap
;
8295 for (id
= super
->devlist
; id
; id
= id
->next
) {
8296 if (id
->index
== dev_id
)
8305 /* Copy the dev, but not (all of) the map */
8306 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8307 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8308 newmap
= get_imsm_map(newdev
, MAP_0
);
8309 /* Copy the current map */
8310 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8311 /* update one device only
8313 if (devices_to_reshape
) {
8314 dprintf("imsm: modifying subdev: %i\n",
8316 devices_to_reshape
--;
8317 newdev
->vol
.migr_state
= 1;
8318 newdev
->vol
.curr_migr_unit
= 0;
8319 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8320 newmap
->num_members
= u
->new_raid_disks
;
8321 for (i
= 0; i
< delta_disks
; i
++) {
8322 set_imsm_ord_tbl_ent(newmap
,
8323 u
->old_raid_disks
+ i
,
8324 u
->old_raid_disks
+ i
);
8326 /* New map is correct, now need to save old map
8328 newmap
= get_imsm_map(newdev
, MAP_1
);
8329 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8331 imsm_set_array_size(newdev
, -1);
8334 sp
= (void **)id
->dev
;
8339 /* Clear migration record */
8340 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8343 *space_list
= tofree
;
8346 update_reshape_exit
:
8351 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8352 struct intel_super
*super
,
8355 struct imsm_dev
*dev
= NULL
;
8356 struct intel_dev
*dv
;
8357 struct imsm_dev
*dev_new
;
8358 struct imsm_map
*map
;
8362 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8363 if (dv
->index
== (unsigned int)u
->subarray
) {
8371 map
= get_imsm_map(dev
, MAP_0
);
8373 if (u
->direction
== R10_TO_R0
) {
8374 /* Number of failed disks must be half of initial disk number */
8375 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8376 (map
->num_members
/ 2))
8379 /* iterate through devices to mark removed disks as spare */
8380 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8381 if (dm
->disk
.status
& FAILED_DISK
) {
8382 int idx
= dm
->index
;
8383 /* update indexes on the disk list */
8384 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8385 the index values will end up being correct.... NB */
8386 for (du
= super
->disks
; du
; du
= du
->next
)
8387 if (du
->index
> idx
)
8389 /* mark as spare disk */
8394 map
->num_members
= map
->num_members
/ 2;
8395 map
->map_state
= IMSM_T_STATE_NORMAL
;
8396 map
->num_domains
= 1;
8397 map
->raid_level
= 0;
8398 map
->failed_disk_num
= -1;
8401 if (u
->direction
== R0_TO_R10
) {
8403 /* update slots in current disk list */
8404 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8408 /* create new *missing* disks */
8409 for (i
= 0; i
< map
->num_members
; i
++) {
8410 space
= *space_list
;
8413 *space_list
= *space
;
8415 memcpy(du
, super
->disks
, sizeof(*du
));
8419 du
->index
= (i
* 2) + 1;
8420 sprintf((char *)du
->disk
.serial
,
8421 " MISSING_%d", du
->index
);
8422 sprintf((char *)du
->serial
,
8423 "MISSING_%d", du
->index
);
8424 du
->next
= super
->missing
;
8425 super
->missing
= du
;
8427 /* create new dev and map */
8428 space
= *space_list
;
8431 *space_list
= *space
;
8432 dev_new
= (void *)space
;
8433 memcpy(dev_new
, dev
, sizeof(*dev
));
8434 /* update new map */
8435 map
= get_imsm_map(dev_new
, MAP_0
);
8436 map
->num_members
= map
->num_members
* 2;
8437 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8438 map
->num_domains
= 2;
8439 map
->raid_level
= 1;
8440 /* replace dev<->dev_new */
8443 /* update disk order table */
8444 for (du
= super
->disks
; du
; du
= du
->next
)
8446 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8447 for (du
= super
->missing
; du
; du
= du
->next
)
8448 if (du
->index
>= 0) {
8449 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8450 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8456 static void imsm_process_update(struct supertype
*st
,
8457 struct metadata_update
*update
)
8460 * crack open the metadata_update envelope to find the update record
8461 * update can be one of:
8462 * update_reshape_container_disks - all the arrays in the container
8463 * are being reshaped to have more devices. We need to mark
8464 * the arrays for general migration and convert selected spares
8465 * into active devices.
8466 * update_activate_spare - a spare device has replaced a failed
8467 * device in an array, update the disk_ord_tbl. If this disk is
8468 * present in all member arrays then also clear the SPARE_DISK
8470 * update_create_array
8472 * update_rename_array
8473 * update_add_remove_disk
8475 struct intel_super
*super
= st
->sb
;
8476 struct imsm_super
*mpb
;
8477 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8479 /* update requires a larger buf but the allocation failed */
8480 if (super
->next_len
&& !super
->next_buf
) {
8481 super
->next_len
= 0;
8485 if (super
->next_buf
) {
8486 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8488 super
->len
= super
->next_len
;
8489 super
->buf
= super
->next_buf
;
8491 super
->next_len
= 0;
8492 super
->next_buf
= NULL
;
8495 mpb
= super
->anchor
;
8498 case update_general_migration_checkpoint
: {
8499 struct intel_dev
*id
;
8500 struct imsm_update_general_migration_checkpoint
*u
=
8501 (void *)update
->buf
;
8503 dprintf("called for update_general_migration_checkpoint\n");
8505 /* find device under general migration */
8506 for (id
= super
->devlist
; id
; id
= id
->next
) {
8507 if (is_gen_migration(id
->dev
)) {
8508 id
->dev
->vol
.curr_migr_unit
=
8509 __cpu_to_le32(u
->curr_migr_unit
);
8510 super
->updates_pending
++;
8515 case update_takeover
: {
8516 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8517 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8518 imsm_update_version_info(super
);
8519 super
->updates_pending
++;
8524 case update_reshape_container_disks
: {
8525 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8526 if (apply_reshape_container_disks_update(
8527 u
, super
, &update
->space_list
))
8528 super
->updates_pending
++;
8531 case update_reshape_migration
: {
8532 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8533 if (apply_reshape_migration_update(
8534 u
, super
, &update
->space_list
))
8535 super
->updates_pending
++;
8538 case update_size_change
: {
8539 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8540 if (apply_size_change_update(u
, super
))
8541 super
->updates_pending
++;
8544 case update_activate_spare
: {
8545 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8546 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8547 super
->updates_pending
++;
8550 case update_create_array
: {
8551 /* someone wants to create a new array, we need to be aware of
8552 * a few races/collisions:
8553 * 1/ 'Create' called by two separate instances of mdadm
8554 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8555 * devices that have since been assimilated via
8557 * In the event this update can not be carried out mdadm will
8558 * (FIX ME) notice that its update did not take hold.
8560 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8561 struct intel_dev
*dv
;
8562 struct imsm_dev
*dev
;
8563 struct imsm_map
*map
, *new_map
;
8564 unsigned long long start
, end
;
8565 unsigned long long new_start
, new_end
;
8567 struct disk_info
*inf
;
8570 /* handle racing creates: first come first serve */
8571 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8572 dprintf("subarray %d already defined\n", u
->dev_idx
);
8576 /* check update is next in sequence */
8577 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8578 dprintf("can not create array %d expected index %d\n",
8579 u
->dev_idx
, mpb
->num_raid_devs
);
8583 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8584 new_start
= pba_of_lba0(new_map
);
8585 new_end
= new_start
+ blocks_per_member(new_map
);
8586 inf
= get_disk_info(u
);
8588 /* handle activate_spare versus create race:
8589 * check to make sure that overlapping arrays do not include
8592 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8593 dev
= get_imsm_dev(super
, i
);
8594 map
= get_imsm_map(dev
, MAP_0
);
8595 start
= pba_of_lba0(map
);
8596 end
= start
+ blocks_per_member(map
);
8597 if ((new_start
>= start
&& new_start
<= end
) ||
8598 (start
>= new_start
&& start
<= new_end
))
8603 if (disks_overlap(super
, i
, u
)) {
8604 dprintf("arrays overlap\n");
8609 /* check that prepare update was successful */
8610 if (!update
->space
) {
8611 dprintf("prepare update failed\n");
8615 /* check that all disks are still active before committing
8616 * changes. FIXME: could we instead handle this by creating a
8617 * degraded array? That's probably not what the user expects,
8618 * so better to drop this update on the floor.
8620 for (i
= 0; i
< new_map
->num_members
; i
++) {
8621 dl
= serial_to_dl(inf
[i
].serial
, super
);
8623 dprintf("disk disappeared\n");
8628 super
->updates_pending
++;
8630 /* convert spares to members and fixup ord_tbl */
8631 for (i
= 0; i
< new_map
->num_members
; i
++) {
8632 dl
= serial_to_dl(inf
[i
].serial
, super
);
8633 if (dl
->index
== -1) {
8634 dl
->index
= mpb
->num_disks
;
8636 dl
->disk
.status
|= CONFIGURED_DISK
;
8637 dl
->disk
.status
&= ~SPARE_DISK
;
8639 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8644 update
->space
= NULL
;
8645 imsm_copy_dev(dev
, &u
->dev
);
8646 dv
->index
= u
->dev_idx
;
8647 dv
->next
= super
->devlist
;
8648 super
->devlist
= dv
;
8649 mpb
->num_raid_devs
++;
8651 imsm_update_version_info(super
);
8654 /* mdmon knows how to release update->space, but not
8655 * ((struct intel_dev *) update->space)->dev
8657 if (update
->space
) {
8663 case update_kill_array
: {
8664 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8665 int victim
= u
->dev_idx
;
8666 struct active_array
*a
;
8667 struct intel_dev
**dp
;
8668 struct imsm_dev
*dev
;
8670 /* sanity check that we are not affecting the uuid of
8671 * active arrays, or deleting an active array
8673 * FIXME when immutable ids are available, but note that
8674 * we'll also need to fixup the invalidated/active
8675 * subarray indexes in mdstat
8677 for (a
= st
->arrays
; a
; a
= a
->next
)
8678 if (a
->info
.container_member
>= victim
)
8680 /* by definition if mdmon is running at least one array
8681 * is active in the container, so checking
8682 * mpb->num_raid_devs is just extra paranoia
8684 dev
= get_imsm_dev(super
, victim
);
8685 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8686 dprintf("failed to delete subarray-%d\n", victim
);
8690 for (dp
= &super
->devlist
; *dp
;)
8691 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8694 if ((*dp
)->index
> (unsigned)victim
)
8698 mpb
->num_raid_devs
--;
8699 super
->updates_pending
++;
8702 case update_rename_array
: {
8703 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8704 char name
[MAX_RAID_SERIAL_LEN
+1];
8705 int target
= u
->dev_idx
;
8706 struct active_array
*a
;
8707 struct imsm_dev
*dev
;
8709 /* sanity check that we are not affecting the uuid of
8712 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8713 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8714 for (a
= st
->arrays
; a
; a
= a
->next
)
8715 if (a
->info
.container_member
== target
)
8717 dev
= get_imsm_dev(super
, u
->dev_idx
);
8718 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8719 dprintf("failed to rename subarray-%d\n", target
);
8723 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8724 super
->updates_pending
++;
8727 case update_add_remove_disk
: {
8728 /* we may be able to repair some arrays if disks are
8729 * being added, check the status of add_remove_disk
8730 * if discs has been added.
8732 if (add_remove_disk_update(super
)) {
8733 struct active_array
*a
;
8735 super
->updates_pending
++;
8736 for (a
= st
->arrays
; a
; a
= a
->next
)
8737 a
->check_degraded
= 1;
8742 pr_err("error: unsuported process update type:(type: %d)\n", type
);
8746 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8748 static int imsm_prepare_update(struct supertype
*st
,
8749 struct metadata_update
*update
)
8752 * Allocate space to hold new disk entries, raid-device entries or a new
8753 * mpb if necessary. The manager synchronously waits for updates to
8754 * complete in the monitor, so new mpb buffers allocated here can be
8755 * integrated by the monitor thread without worrying about live pointers
8756 * in the manager thread.
8758 enum imsm_update_type type
;
8759 struct intel_super
*super
= st
->sb
;
8760 struct imsm_super
*mpb
= super
->anchor
;
8764 if (update
->len
< (int)sizeof(type
))
8767 type
= *(enum imsm_update_type
*) update
->buf
;
8770 case update_general_migration_checkpoint
:
8771 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
8773 dprintf("called for update_general_migration_checkpoint\n");
8775 case update_takeover
: {
8776 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8777 if (update
->len
< (int)sizeof(*u
))
8779 if (u
->direction
== R0_TO_R10
) {
8780 void **tail
= (void **)&update
->space_list
;
8781 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8782 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8783 int num_members
= map
->num_members
;
8786 /* allocate memory for added disks */
8787 for (i
= 0; i
< num_members
; i
++) {
8788 size
= sizeof(struct dl
);
8789 space
= xmalloc(size
);
8794 /* allocate memory for new device */
8795 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8796 (num_members
* sizeof(__u32
));
8797 space
= xmalloc(size
);
8801 len
= disks_to_mpb_size(num_members
* 2);
8806 case update_reshape_container_disks
: {
8807 /* Every raid device in the container is about to
8808 * gain some more devices, and we will enter a
8810 * So each 'imsm_map' will be bigger, and the imsm_vol
8811 * will now hold 2 of them.
8812 * Thus we need new 'struct imsm_dev' allocations sized
8813 * as sizeof_imsm_dev but with more devices in both maps.
8815 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8816 struct intel_dev
*dl
;
8817 void **space_tail
= (void**)&update
->space_list
;
8819 if (update
->len
< (int)sizeof(*u
))
8822 dprintf("for update_reshape\n");
8824 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8825 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8827 if (u
->new_raid_disks
> u
->old_raid_disks
)
8828 size
+= sizeof(__u32
)*2*
8829 (u
->new_raid_disks
- u
->old_raid_disks
);
8836 len
= disks_to_mpb_size(u
->new_raid_disks
);
8837 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8840 case update_reshape_migration
: {
8841 /* for migration level 0->5 we need to add disks
8842 * so the same as for container operation we will copy
8843 * device to the bigger location.
8844 * in memory prepared device and new disk area are prepared
8845 * for usage in process update
8847 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8848 struct intel_dev
*id
;
8849 void **space_tail
= (void **)&update
->space_list
;
8852 int current_level
= -1;
8854 if (update
->len
< (int)sizeof(*u
))
8857 dprintf("for update_reshape\n");
8859 /* add space for bigger array in update
8861 for (id
= super
->devlist
; id
; id
= id
->next
) {
8862 if (id
->index
== (unsigned)u
->subdev
) {
8863 size
= sizeof_imsm_dev(id
->dev
, 1);
8864 if (u
->new_raid_disks
> u
->old_raid_disks
)
8865 size
+= sizeof(__u32
)*2*
8866 (u
->new_raid_disks
- u
->old_raid_disks
);
8874 if (update
->space_list
== NULL
)
8877 /* add space for disk in update
8879 size
= sizeof(struct dl
);
8885 /* add spare device to update
8887 for (id
= super
->devlist
; id
; id
= id
->next
)
8888 if (id
->index
== (unsigned)u
->subdev
) {
8889 struct imsm_dev
*dev
;
8890 struct imsm_map
*map
;
8892 dev
= get_imsm_dev(super
, u
->subdev
);
8893 map
= get_imsm_map(dev
, MAP_0
);
8894 current_level
= map
->raid_level
;
8897 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8898 struct mdinfo
*spares
;
8900 spares
= get_spares_for_grow(st
);
8908 makedev(dev
->disk
.major
,
8910 dl
= get_disk_super(super
,
8913 dl
->index
= u
->old_raid_disks
;
8919 len
= disks_to_mpb_size(u
->new_raid_disks
);
8920 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8923 case update_size_change
: {
8924 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
8928 case update_activate_spare
: {
8929 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
8933 case update_create_array
: {
8934 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8935 struct intel_dev
*dv
;
8936 struct imsm_dev
*dev
= &u
->dev
;
8937 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8939 struct disk_info
*inf
;
8943 if (update
->len
< (int)sizeof(*u
))
8946 inf
= get_disk_info(u
);
8947 len
= sizeof_imsm_dev(dev
, 1);
8948 /* allocate a new super->devlist entry */
8949 dv
= xmalloc(sizeof(*dv
));
8950 dv
->dev
= xmalloc(len
);
8953 /* count how many spares will be converted to members */
8954 for (i
= 0; i
< map
->num_members
; i
++) {
8955 dl
= serial_to_dl(inf
[i
].serial
, super
);
8957 /* hmm maybe it failed?, nothing we can do about
8962 if (count_memberships(dl
, super
) == 0)
8965 len
+= activate
* sizeof(struct imsm_disk
);
8968 case update_kill_array
: {
8969 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
8973 case update_rename_array
: {
8974 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
8978 case update_add_remove_disk
:
8979 /* no update->len needed */
8985 /* check if we need a larger metadata buffer */
8986 if (super
->next_buf
)
8987 buf_len
= super
->next_len
;
8989 buf_len
= super
->len
;
8991 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8992 /* ok we need a larger buf than what is currently allocated
8993 * if this allocation fails process_update will notice that
8994 * ->next_len is set and ->next_buf is NULL
8996 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8997 if (super
->next_buf
)
8998 free(super
->next_buf
);
9000 super
->next_len
= buf_len
;
9001 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
9002 memset(super
->next_buf
, 0, buf_len
);
9004 super
->next_buf
= NULL
;
9009 /* must be called while manager is quiesced */
9010 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
9012 struct imsm_super
*mpb
= super
->anchor
;
9014 struct imsm_dev
*dev
;
9015 struct imsm_map
*map
;
9016 int i
, j
, num_members
;
9019 dprintf("deleting device[%d] from imsm_super\n", index
);
9021 /* shift all indexes down one */
9022 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
9023 if (iter
->index
> (int)index
)
9025 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
9026 if (iter
->index
> (int)index
)
9029 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9030 dev
= get_imsm_dev(super
, i
);
9031 map
= get_imsm_map(dev
, MAP_0
);
9032 num_members
= map
->num_members
;
9033 for (j
= 0; j
< num_members
; j
++) {
9034 /* update ord entries being careful not to propagate
9035 * ord-flags to the first map
9037 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
9039 if (ord_to_idx(ord
) <= index
)
9042 map
= get_imsm_map(dev
, MAP_0
);
9043 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
9044 map
= get_imsm_map(dev
, MAP_1
);
9046 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
9051 super
->updates_pending
++;
9053 struct dl
*dl
= *dlp
;
9055 *dlp
= (*dlp
)->next
;
9056 __free_imsm_disk(dl
);
9059 #endif /* MDASSEMBLE */
9061 static void close_targets(int *targets
, int new_disks
)
9068 for (i
= 0; i
< new_disks
; i
++) {
9069 if (targets
[i
] >= 0) {
9076 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
9077 struct intel_super
*super
,
9078 struct imsm_dev
*dev
)
9084 struct imsm_map
*map
;
9087 ret_val
= raid_disks
/2;
9088 /* check map if all disks pairs not failed
9091 map
= get_imsm_map(dev
, MAP_0
);
9092 for (i
= 0; i
< ret_val
; i
++) {
9093 int degradation
= 0;
9094 if (get_imsm_disk(super
, i
) == NULL
)
9096 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9098 if (degradation
== 2)
9101 map
= get_imsm_map(dev
, MAP_1
);
9102 /* if there is no second map
9103 * result can be returned
9107 /* check degradation in second map
9109 for (i
= 0; i
< ret_val
; i
++) {
9110 int degradation
= 0;
9111 if (get_imsm_disk(super
, i
) == NULL
)
9113 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9115 if (degradation
== 2)
9129 /*******************************************************************************
9130 * Function: open_backup_targets
9131 * Description: Function opens file descriptors for all devices given in
9134 * info : general array info
9135 * raid_disks : number of disks
9136 * raid_fds : table of device's file descriptors
9137 * super : intel super for raid10 degradation check
9138 * dev : intel device for raid10 degradation check
9142 ******************************************************************************/
9143 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9144 struct intel_super
*super
, struct imsm_dev
*dev
)
9150 for (i
= 0; i
< raid_disks
; i
++)
9153 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9156 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9157 dprintf("disk is faulty!!\n");
9161 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
9162 (sd
->disk
.raid_disk
< 0))
9165 dn
= map_dev(sd
->disk
.major
,
9167 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9168 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9169 pr_err("cannot open component\n");
9174 /* check if maximum array degradation level is not exceeded
9176 if ((raid_disks
- opened
) >
9177 imsm_get_allowed_degradation(info
->new_level
,
9180 pr_err("Not enough disks can be opened.\n");
9181 close_targets(raid_fds
, raid_disks
);
9187 /*******************************************************************************
9188 * Function: validate_container_imsm
9189 * Description: This routine validates container after assemble,
9190 * eg. if devices in container are under the same controller.
9193 * info : linked list with info about devices used in array
9197 ******************************************************************************/
9198 int validate_container_imsm(struct mdinfo
*info
)
9200 if (check_env("IMSM_NO_PLATFORM"))
9203 struct sys_dev
*idev
;
9204 struct sys_dev
*hba
= NULL
;
9205 struct sys_dev
*intel_devices
= find_intel_devices();
9206 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
9209 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9210 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9219 pr_err("WARNING - Cannot detect HBA for device %s!\n",
9220 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
9224 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
9227 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
9228 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
9230 struct sys_dev
*hba2
= NULL
;
9231 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9232 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9240 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
9241 get_orom_by_device_id(hba2
->dev_id
);
9243 if (hba2
&& hba
->type
!= hba2
->type
) {
9244 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
9245 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
9249 if ((orom
!= orom2
) || ((hba
->type
== SYS_DEV_VMD
) && (hba
!= hba2
))) {
9250 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
9251 " This operation is not supported and can lead to data loss.\n");
9256 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
9257 " This operation is not supported and can lead to data loss.\n");
9265 /*******************************************************************************
9266 * Function: init_migr_record_imsm
9267 * Description: Function inits imsm migration record
9269 * super : imsm internal array info
9270 * dev : device under migration
9271 * info : general array info to find the smallest device
9274 ******************************************************************************/
9275 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9276 struct mdinfo
*info
)
9278 struct intel_super
*super
= st
->sb
;
9279 struct migr_record
*migr_rec
= super
->migr_rec
;
9281 unsigned long long dsize
, dev_sectors
;
9282 long long unsigned min_dev_sectors
= -1LLU;
9286 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9287 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9288 unsigned long long num_migr_units
;
9289 unsigned long long array_blocks
;
9291 memset(migr_rec
, 0, sizeof(struct migr_record
));
9292 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9294 /* only ascending reshape supported now */
9295 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9297 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9298 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9299 migr_rec
->dest_depth_per_unit
*=
9300 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9301 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9302 migr_rec
->blocks_per_unit
=
9303 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9304 migr_rec
->dest_depth_per_unit
=
9305 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9306 array_blocks
= info
->component_size
* new_data_disks
;
9308 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9310 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9312 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9314 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9315 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9317 /* Find the smallest dev */
9318 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9319 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9320 fd
= dev_open(nm
, O_RDONLY
);
9323 get_dev_size(fd
, NULL
, &dsize
);
9324 dev_sectors
= dsize
/ 512;
9325 if (dev_sectors
< min_dev_sectors
)
9326 min_dev_sectors
= dev_sectors
;
9329 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9330 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9332 write_imsm_migr_rec(st
);
9337 /*******************************************************************************
9338 * Function: save_backup_imsm
9339 * Description: Function saves critical data stripes to Migration Copy Area
9340 * and updates the current migration unit status.
9341 * Use restore_stripes() to form a destination stripe,
9342 * and to write it to the Copy Area.
9344 * st : supertype information
9345 * dev : imsm device that backup is saved for
9346 * info : general array info
9347 * buf : input buffer
9348 * length : length of data to backup (blocks_per_unit)
9352 ******************************************************************************/
9353 int save_backup_imsm(struct supertype
*st
,
9354 struct imsm_dev
*dev
,
9355 struct mdinfo
*info
,
9360 struct intel_super
*super
= st
->sb
;
9361 unsigned long long *target_offsets
= NULL
;
9362 int *targets
= NULL
;
9364 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9365 int new_disks
= map_dest
->num_members
;
9366 int dest_layout
= 0;
9368 unsigned long long start
;
9369 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9371 targets
= xmalloc(new_disks
* sizeof(int));
9373 for (i
= 0; i
< new_disks
; i
++)
9376 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
9378 start
= info
->reshape_progress
* 512;
9379 for (i
= 0; i
< new_disks
; i
++) {
9380 target_offsets
[i
] = (unsigned long long)
9381 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9382 /* move back copy area adderss, it will be moved forward
9383 * in restore_stripes() using start input variable
9385 target_offsets
[i
] -= start
/data_disks
;
9388 if (open_backup_targets(info
, new_disks
, targets
,
9392 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9393 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9395 if (restore_stripes(targets
, /* list of dest devices */
9396 target_offsets
, /* migration record offsets */
9399 map_dest
->raid_level
,
9401 -1, /* source backup file descriptor */
9402 0, /* input buf offset
9403 * always 0 buf is already offseted */
9407 pr_err("Error restoring stripes\n");
9415 close_targets(targets
, new_disks
);
9418 free(target_offsets
);
9423 /*******************************************************************************
9424 * Function: save_checkpoint_imsm
9425 * Description: Function called for current unit status update
9426 * in the migration record. It writes it to disk.
9428 * super : imsm internal array info
9429 * info : general array info
9433 * 2: failure, means no valid migration record
9434 * / no general migration in progress /
9435 ******************************************************************************/
9436 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9438 struct intel_super
*super
= st
->sb
;
9439 unsigned long long blocks_per_unit
;
9440 unsigned long long curr_migr_unit
;
9442 if (load_imsm_migr_rec(super
, info
) != 0) {
9443 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
9447 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9448 if (blocks_per_unit
== 0) {
9449 dprintf("imsm: no migration in progress.\n");
9452 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9453 /* check if array is alligned to copy area
9454 * if it is not alligned, add one to current migration unit value
9455 * this can happend on array reshape finish only
9457 if (info
->reshape_progress
% blocks_per_unit
)
9460 super
->migr_rec
->curr_migr_unit
=
9461 __cpu_to_le32(curr_migr_unit
);
9462 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9463 super
->migr_rec
->dest_1st_member_lba
=
9464 __cpu_to_le32(curr_migr_unit
*
9465 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9466 if (write_imsm_migr_rec(st
) < 0) {
9467 dprintf("imsm: Cannot write migration record outside backup area\n");
9474 /*******************************************************************************
9475 * Function: recover_backup_imsm
9476 * Description: Function recovers critical data from the Migration Copy Area
9477 * while assembling an array.
9479 * super : imsm internal array info
9480 * info : general array info
9482 * 0 : success (or there is no data to recover)
9484 ******************************************************************************/
9485 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9487 struct intel_super
*super
= st
->sb
;
9488 struct migr_record
*migr_rec
= super
->migr_rec
;
9489 struct imsm_map
*map_dest
= NULL
;
9490 struct intel_dev
*id
= NULL
;
9491 unsigned long long read_offset
;
9492 unsigned long long write_offset
;
9494 int *targets
= NULL
;
9495 int new_disks
, i
, err
;
9498 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9499 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9501 int skipped_disks
= 0;
9503 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9507 /* recover data only during assemblation */
9508 if (strncmp(buffer
, "inactive", 8) != 0)
9510 /* no data to recover */
9511 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9513 if (curr_migr_unit
>= num_migr_units
)
9516 /* find device during reshape */
9517 for (id
= super
->devlist
; id
; id
= id
->next
)
9518 if (is_gen_migration(id
->dev
))
9523 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9524 new_disks
= map_dest
->num_members
;
9526 read_offset
= (unsigned long long)
9527 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9529 write_offset
= ((unsigned long long)
9530 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9531 pba_of_lba0(map_dest
)) * 512;
9533 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9534 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9536 targets
= xcalloc(new_disks
, sizeof(int));
9538 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9539 pr_err("Cannot open some devices belonging to array.\n");
9543 for (i
= 0; i
< new_disks
; i
++) {
9544 if (targets
[i
] < 0) {
9548 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9549 pr_err("Cannot seek to block: %s\n",
9554 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9555 pr_err("Cannot read copy area block: %s\n",
9560 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9561 pr_err("Cannot seek to block: %s\n",
9566 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9567 pr_err("Cannot restore block: %s\n",
9574 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9578 pr_err("Cannot restore data from backup. Too many failed disks\n");
9582 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9583 /* ignore error == 2, this can mean end of reshape here
9585 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
9591 for (i
= 0; i
< new_disks
; i
++)
9600 static char disk_by_path
[] = "/dev/disk/by-path/";
9602 static const char *imsm_get_disk_controller_domain(const char *path
)
9604 char disk_path
[PATH_MAX
];
9608 strcpy(disk_path
, disk_by_path
);
9609 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9610 if (stat(disk_path
, &st
) == 0) {
9611 struct sys_dev
* hba
;
9614 path
= devt_to_devpath(st
.st_rdev
);
9617 hba
= find_disk_attached_hba(-1, path
);
9618 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9620 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9624 dprintf("path: %s hba: %s attached: %s\n",
9625 path
, (hba
) ? hba
->path
: "NULL", drv
);
9631 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
9633 static char devnm
[32];
9634 char subdev_name
[20];
9635 struct mdstat_ent
*mdstat
;
9637 sprintf(subdev_name
, "%d", subdev
);
9638 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9642 strcpy(devnm
, mdstat
->devnm
);
9643 free_mdstat(mdstat
);
9647 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9648 struct geo_params
*geo
,
9649 int *old_raid_disks
,
9652 /* currently we only support increasing the number of devices
9653 * for a container. This increases the number of device for each
9654 * member array. They must all be RAID0 or RAID5.
9657 struct mdinfo
*info
, *member
;
9658 int devices_that_can_grow
= 0;
9660 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
9662 if (geo
->size
> 0 ||
9663 geo
->level
!= UnSet
||
9664 geo
->layout
!= UnSet
||
9665 geo
->chunksize
!= 0 ||
9666 geo
->raid_disks
== UnSet
) {
9667 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
9671 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9672 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
9676 info
= container_content_imsm(st
, NULL
);
9677 for (member
= info
; member
; member
= member
->next
) {
9680 dprintf("imsm: checking device_num: %i\n",
9681 member
->container_member
);
9683 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9684 /* we work on container for Online Capacity Expansion
9685 * only so raid_disks has to grow
9687 dprintf("imsm: for container operation raid disks increase is required\n");
9691 if ((info
->array
.level
!= 0) &&
9692 (info
->array
.level
!= 5)) {
9693 /* we cannot use this container with other raid level
9695 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
9699 /* check for platform support
9700 * for this raid level configuration
9702 struct intel_super
*super
= st
->sb
;
9703 if (!is_raid_level_supported(super
->orom
,
9704 member
->array
.level
,
9706 dprintf("platform does not support raid%d with %d disk%s\n",
9709 geo
->raid_disks
> 1 ? "s" : "");
9712 /* check if component size is aligned to chunk size
9714 if (info
->component_size
%
9715 (info
->array
.chunk_size
/512)) {
9716 dprintf("Component size is not aligned to chunk size\n");
9721 if (*old_raid_disks
&&
9722 info
->array
.raid_disks
!= *old_raid_disks
)
9724 *old_raid_disks
= info
->array
.raid_disks
;
9726 /* All raid5 and raid0 volumes in container
9727 * have to be ready for Online Capacity Expansion
9728 * so they need to be assembled. We have already
9729 * checked that no recovery etc is happening.
9731 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
9732 st
->container_devnm
);
9733 if (result
== NULL
) {
9734 dprintf("imsm: cannot find array\n");
9737 devices_that_can_grow
++;
9740 if (!member
&& devices_that_can_grow
)
9744 dprintf("Container operation allowed\n");
9746 dprintf("Error: %i\n", ret_val
);
9751 /* Function: get_spares_for_grow
9752 * Description: Allocates memory and creates list of spare devices
9753 * avaliable in container. Checks if spare drive size is acceptable.
9754 * Parameters: Pointer to the supertype structure
9755 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9758 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9760 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9761 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9764 /******************************************************************************
9765 * function: imsm_create_metadata_update_for_reshape
9766 * Function creates update for whole IMSM container.
9768 ******************************************************************************/
9769 static int imsm_create_metadata_update_for_reshape(
9770 struct supertype
*st
,
9771 struct geo_params
*geo
,
9773 struct imsm_update_reshape
**updatep
)
9775 struct intel_super
*super
= st
->sb
;
9776 struct imsm_super
*mpb
= super
->anchor
;
9777 int update_memory_size
= 0;
9778 struct imsm_update_reshape
*u
= NULL
;
9779 struct mdinfo
*spares
= NULL
;
9781 int delta_disks
= 0;
9784 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
9786 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9788 /* size of all update data without anchor */
9789 update_memory_size
= sizeof(struct imsm_update_reshape
);
9791 /* now add space for spare disks that we need to add. */
9792 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9794 u
= xcalloc(1, update_memory_size
);
9795 u
->type
= update_reshape_container_disks
;
9796 u
->old_raid_disks
= old_raid_disks
;
9797 u
->new_raid_disks
= geo
->raid_disks
;
9799 /* now get spare disks list
9801 spares
= get_spares_for_grow(st
);
9804 || delta_disks
> spares
->array
.spare_disks
) {
9805 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
9810 /* we have got spares
9811 * update disk list in imsm_disk list table in anchor
9813 dprintf("imsm: %i spares are available.\n\n",
9814 spares
->array
.spare_disks
);
9817 for (i
= 0; i
< delta_disks
; i
++) {
9822 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9824 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9825 dl
->index
= mpb
->num_disks
;
9835 dprintf("imsm: reshape update preparation :");
9836 if (i
== delta_disks
) {
9837 dprintf_cont(" OK\n");
9839 return update_memory_size
;
9842 dprintf_cont(" Error\n");
9847 /******************************************************************************
9848 * function: imsm_create_metadata_update_for_size_change()
9849 * Creates update for IMSM array for array size change.
9851 ******************************************************************************/
9852 static int imsm_create_metadata_update_for_size_change(
9853 struct supertype
*st
,
9854 struct geo_params
*geo
,
9855 struct imsm_update_size_change
**updatep
)
9857 struct intel_super
*super
= st
->sb
;
9858 int update_memory_size
= 0;
9859 struct imsm_update_size_change
*u
= NULL
;
9861 dprintf("(enter) New size = %llu\n", geo
->size
);
9863 /* size of all update data without anchor */
9864 update_memory_size
= sizeof(struct imsm_update_size_change
);
9866 u
= xcalloc(1, update_memory_size
);
9867 u
->type
= update_size_change
;
9868 u
->subdev
= super
->current_vol
;
9869 u
->new_size
= geo
->size
;
9871 dprintf("imsm: reshape update preparation : OK\n");
9874 return update_memory_size
;
9877 /******************************************************************************
9878 * function: imsm_create_metadata_update_for_migration()
9879 * Creates update for IMSM array.
9881 ******************************************************************************/
9882 static int imsm_create_metadata_update_for_migration(
9883 struct supertype
*st
,
9884 struct geo_params
*geo
,
9885 struct imsm_update_reshape_migration
**updatep
)
9887 struct intel_super
*super
= st
->sb
;
9888 int update_memory_size
= 0;
9889 struct imsm_update_reshape_migration
*u
= NULL
;
9890 struct imsm_dev
*dev
;
9891 int previous_level
= -1;
9893 dprintf("(enter) New Level = %i\n", geo
->level
);
9895 /* size of all update data without anchor */
9896 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9898 u
= xcalloc(1, update_memory_size
);
9899 u
->type
= update_reshape_migration
;
9900 u
->subdev
= super
->current_vol
;
9901 u
->new_level
= geo
->level
;
9902 u
->new_layout
= geo
->layout
;
9903 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9904 u
->new_disks
[0] = -1;
9905 u
->new_chunksize
= -1;
9907 dev
= get_imsm_dev(super
, u
->subdev
);
9909 struct imsm_map
*map
;
9911 map
= get_imsm_map(dev
, MAP_0
);
9913 int current_chunk_size
=
9914 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9916 if (geo
->chunksize
!= current_chunk_size
) {
9917 u
->new_chunksize
= geo
->chunksize
/ 1024;
9918 dprintf("imsm: chunk size change from %i to %i\n",
9919 current_chunk_size
, u
->new_chunksize
);
9921 previous_level
= map
->raid_level
;
9924 if ((geo
->level
== 5) && (previous_level
== 0)) {
9925 struct mdinfo
*spares
= NULL
;
9927 u
->new_raid_disks
++;
9928 spares
= get_spares_for_grow(st
);
9929 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9932 update_memory_size
= 0;
9933 dprintf("error: cannot get spare device for requested migration");
9938 dprintf("imsm: reshape update preparation : OK\n");
9941 return update_memory_size
;
9944 static void imsm_update_metadata_locally(struct supertype
*st
,
9947 struct metadata_update mu
;
9952 mu
.space_list
= NULL
;
9954 if (imsm_prepare_update(st
, &mu
))
9955 imsm_process_update(st
, &mu
);
9957 while (mu
.space_list
) {
9958 void **space
= mu
.space_list
;
9959 mu
.space_list
= *space
;
9964 /***************************************************************************
9965 * Function: imsm_analyze_change
9966 * Description: Function analyze change for single volume
9967 * and validate if transition is supported
9968 * Parameters: Geometry parameters, supertype structure,
9969 * metadata change direction (apply/rollback)
9970 * Returns: Operation type code on success, -1 if fail
9971 ****************************************************************************/
9972 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9973 struct geo_params
*geo
,
9980 /* number of added/removed disks in operation result */
9981 int devNumChange
= 0;
9982 /* imsm compatible layout value for array geometry verification */
9983 int imsm_layout
= -1;
9985 struct imsm_dev
*dev
;
9986 struct intel_super
*super
;
9987 unsigned long long current_size
;
9988 unsigned long long free_size
;
9989 unsigned long long max_size
;
9992 getinfo_super_imsm_volume(st
, &info
, NULL
);
9993 if ((geo
->level
!= info
.array
.level
) &&
9994 (geo
->level
>= 0) &&
9995 (geo
->level
!= UnSet
)) {
9996 switch (info
.array
.level
) {
9998 if (geo
->level
== 5) {
9999 change
= CH_MIGRATION
;
10000 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
10001 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
10003 goto analyse_change_exit
;
10005 imsm_layout
= geo
->layout
;
10007 devNumChange
= 1; /* parity disk added */
10008 } else if (geo
->level
== 10) {
10009 change
= CH_TAKEOVER
;
10011 devNumChange
= 2; /* two mirrors added */
10012 imsm_layout
= 0x102; /* imsm supported layout */
10017 if (geo
->level
== 0) {
10018 change
= CH_TAKEOVER
;
10020 devNumChange
= -(geo
->raid_disks
/2);
10021 imsm_layout
= 0; /* imsm raid0 layout */
10025 if (change
== -1) {
10026 pr_err("Error. Level Migration from %d to %d not supported!\n",
10027 info
.array
.level
, geo
->level
);
10028 goto analyse_change_exit
;
10031 geo
->level
= info
.array
.level
;
10033 if ((geo
->layout
!= info
.array
.layout
)
10034 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
10035 change
= CH_MIGRATION
;
10036 if ((info
.array
.layout
== 0)
10037 && (info
.array
.level
== 5)
10038 && (geo
->layout
== 5)) {
10039 /* reshape 5 -> 4 */
10040 } else if ((info
.array
.layout
== 5)
10041 && (info
.array
.level
== 5)
10042 && (geo
->layout
== 0)) {
10043 /* reshape 4 -> 5 */
10047 pr_err("Error. Layout Migration from %d to %d not supported!\n",
10048 info
.array
.layout
, geo
->layout
);
10050 goto analyse_change_exit
;
10053 geo
->layout
= info
.array
.layout
;
10054 if (imsm_layout
== -1)
10055 imsm_layout
= info
.array
.layout
;
10058 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
10059 && (geo
->chunksize
!= info
.array
.chunk_size
))
10060 change
= CH_MIGRATION
;
10062 geo
->chunksize
= info
.array
.chunk_size
;
10064 chunk
= geo
->chunksize
/ 1024;
10067 dev
= get_imsm_dev(super
, super
->current_vol
);
10068 data_disks
= imsm_num_data_members(dev
, MAP_0
);
10069 /* compute current size per disk member
10071 current_size
= info
.custom_array_size
/ data_disks
;
10073 if ((geo
->size
> 0) && (geo
->size
!= MAX_SIZE
)) {
10074 /* align component size
10076 geo
->size
= imsm_component_size_aligment_check(
10077 get_imsm_raid_level(dev
->vol
.map
),
10080 if (geo
->size
== 0) {
10081 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
10083 goto analyse_change_exit
;
10087 if ((current_size
!= geo
->size
) && (geo
->size
> 0)) {
10088 if (change
!= -1) {
10089 pr_err("Error. Size change should be the only one at a time.\n");
10091 goto analyse_change_exit
;
10093 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
10094 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
10095 super
->current_vol
, st
->devnm
);
10096 goto analyse_change_exit
;
10098 /* check the maximum available size
10100 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
10101 0, chunk
, &free_size
);
10103 /* Cannot find maximum available space
10107 max_size
= free_size
+ current_size
;
10108 /* align component size
10110 max_size
= imsm_component_size_aligment_check(
10111 get_imsm_raid_level(dev
->vol
.map
),
10115 if (geo
->size
== MAX_SIZE
) {
10116 /* requested size change to the maximum available size
10118 if (max_size
== 0) {
10119 pr_err("Error. Cannot find maximum available space.\n");
10121 goto analyse_change_exit
;
10123 geo
->size
= max_size
;
10126 if ((direction
== ROLLBACK_METADATA_CHANGES
)) {
10127 /* accept size for rollback only
10130 /* round size due to metadata compatibility
10132 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
10133 << SECT_PER_MB_SHIFT
;
10134 dprintf("Prepare update for size change to %llu\n",
10136 if (current_size
>= geo
->size
) {
10137 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
10138 current_size
, geo
->size
);
10139 goto analyse_change_exit
;
10141 if (max_size
&& geo
->size
> max_size
) {
10142 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
10143 max_size
, geo
->size
);
10144 goto analyse_change_exit
;
10147 geo
->size
*= data_disks
;
10148 geo
->raid_disks
= dev
->vol
.map
->num_members
;
10149 change
= CH_ARRAY_SIZE
;
10151 if (!validate_geometry_imsm(st
,
10154 geo
->raid_disks
+ devNumChange
,
10156 geo
->size
, INVALID_SECTORS
,
10161 struct intel_super
*super
= st
->sb
;
10162 struct imsm_super
*mpb
= super
->anchor
;
10164 if (mpb
->num_raid_devs
> 1) {
10165 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
10171 analyse_change_exit
:
10172 if ((direction
== ROLLBACK_METADATA_CHANGES
) &&
10173 ((change
== CH_MIGRATION
) || (change
== CH_TAKEOVER
))) {
10174 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
10180 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10182 struct intel_super
*super
= st
->sb
;
10183 struct imsm_update_takeover
*u
;
10185 u
= xmalloc(sizeof(struct imsm_update_takeover
));
10187 u
->type
= update_takeover
;
10188 u
->subarray
= super
->current_vol
;
10190 /* 10->0 transition */
10191 if (geo
->level
== 0)
10192 u
->direction
= R10_TO_R0
;
10194 /* 0->10 transition */
10195 if (geo
->level
== 10)
10196 u
->direction
= R0_TO_R10
;
10198 /* update metadata locally */
10199 imsm_update_metadata_locally(st
, u
,
10200 sizeof(struct imsm_update_takeover
));
10201 /* and possibly remotely */
10202 if (st
->update_tail
)
10203 append_metadata_update(st
, u
,
10204 sizeof(struct imsm_update_takeover
));
10211 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
10213 int layout
, int chunksize
, int raid_disks
,
10214 int delta_disks
, char *backup
, char *dev
,
10215 int direction
, int verbose
)
10218 struct geo_params geo
;
10220 dprintf("(enter)\n");
10222 memset(&geo
, 0, sizeof(struct geo_params
));
10224 geo
.dev_name
= dev
;
10225 strcpy(geo
.devnm
, st
->devnm
);
10228 geo
.layout
= layout
;
10229 geo
.chunksize
= chunksize
;
10230 geo
.raid_disks
= raid_disks
;
10231 if (delta_disks
!= UnSet
)
10232 geo
.raid_disks
+= delta_disks
;
10234 dprintf("for level : %i\n", geo
.level
);
10235 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
10237 if (experimental() == 0)
10240 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
10241 /* On container level we can only increase number of devices. */
10242 dprintf("imsm: info: Container operation\n");
10243 int old_raid_disks
= 0;
10245 if (imsm_reshape_is_allowed_on_container(
10246 st
, &geo
, &old_raid_disks
, direction
)) {
10247 struct imsm_update_reshape
*u
= NULL
;
10250 len
= imsm_create_metadata_update_for_reshape(
10251 st
, &geo
, old_raid_disks
, &u
);
10254 dprintf("imsm: Cannot prepare update\n");
10255 goto exit_imsm_reshape_super
;
10259 /* update metadata locally */
10260 imsm_update_metadata_locally(st
, u
, len
);
10261 /* and possibly remotely */
10262 if (st
->update_tail
)
10263 append_metadata_update(st
, u
, len
);
10268 pr_err("(imsm) Operation is not allowed on this container\n");
10271 /* On volume level we support following operations
10272 * - takeover: raid10 -> raid0; raid0 -> raid10
10273 * - chunk size migration
10274 * - migration: raid5 -> raid0; raid0 -> raid5
10276 struct intel_super
*super
= st
->sb
;
10277 struct intel_dev
*dev
= super
->devlist
;
10279 dprintf("imsm: info: Volume operation\n");
10280 /* find requested device */
10283 imsm_find_array_devnm_by_subdev(
10284 dev
->index
, st
->container_devnm
);
10285 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
10290 pr_err("Cannot find %s (%s) subarray\n",
10291 geo
.dev_name
, geo
.devnm
);
10292 goto exit_imsm_reshape_super
;
10294 super
->current_vol
= dev
->index
;
10295 change
= imsm_analyze_change(st
, &geo
, direction
);
10298 ret_val
= imsm_takeover(st
, &geo
);
10300 case CH_MIGRATION
: {
10301 struct imsm_update_reshape_migration
*u
= NULL
;
10303 imsm_create_metadata_update_for_migration(
10306 dprintf("imsm: Cannot prepare update\n");
10310 /* update metadata locally */
10311 imsm_update_metadata_locally(st
, u
, len
);
10312 /* and possibly remotely */
10313 if (st
->update_tail
)
10314 append_metadata_update(st
, u
, len
);
10319 case CH_ARRAY_SIZE
: {
10320 struct imsm_update_size_change
*u
= NULL
;
10322 imsm_create_metadata_update_for_size_change(
10325 dprintf("imsm: Cannot prepare update\n");
10329 /* update metadata locally */
10330 imsm_update_metadata_locally(st
, u
, len
);
10331 /* and possibly remotely */
10332 if (st
->update_tail
)
10333 append_metadata_update(st
, u
, len
);
10343 exit_imsm_reshape_super
:
10344 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10348 /*******************************************************************************
10349 * Function: wait_for_reshape_imsm
10350 * Description: Function writes new sync_max value and waits until
10351 * reshape process reach new position
10353 * sra : general array info
10354 * ndata : number of disks in new array's layout
10357 * 1 : there is no reshape in progress,
10359 ******************************************************************************/
10360 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10362 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
10363 unsigned long long completed
;
10364 /* to_complete : new sync_max position */
10365 unsigned long long to_complete
= sra
->reshape_progress
;
10366 unsigned long long position_to_set
= to_complete
/ ndata
;
10369 dprintf("cannot open reshape_position\n");
10373 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10374 dprintf("cannot read reshape_position (no reshape in progres)\n");
10379 if (completed
> position_to_set
) {
10380 dprintf("wrong next position to set %llu (%llu)\n",
10381 to_complete
, position_to_set
);
10385 dprintf("Position set: %llu\n", position_to_set
);
10386 if (sysfs_set_num(sra
, NULL
, "sync_max",
10387 position_to_set
) != 0) {
10388 dprintf("cannot set reshape position to %llu\n",
10396 int timeout
= 3000;
10397 sysfs_wait(fd
, &timeout
);
10398 if (sysfs_get_str(sra
, NULL
, "sync_action",
10400 strncmp(action
, "reshape", 7) != 0) {
10404 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10405 dprintf("cannot read reshape_position (in loop)\n");
10409 } while (completed
< position_to_set
);
10415 /*******************************************************************************
10416 * Function: check_degradation_change
10417 * Description: Check that array hasn't become failed.
10419 * info : for sysfs access
10420 * sources : source disks descriptors
10421 * degraded: previous degradation level
10423 * degradation level
10424 ******************************************************************************/
10425 int check_degradation_change(struct mdinfo
*info
,
10429 unsigned long long new_degraded
;
10432 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10433 if ((rv
== -1) || (new_degraded
!= (unsigned long long)degraded
)) {
10434 /* check each device to ensure it is still working */
10437 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10438 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10440 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10442 if (sysfs_get_str(info
,
10443 sd
, "state", sbuf
, 20) < 0 ||
10444 strstr(sbuf
, "faulty") ||
10445 strstr(sbuf
, "in_sync") == NULL
) {
10446 /* this device is dead */
10447 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10448 if (sd
->disk
.raid_disk
>= 0 &&
10449 sources
[sd
->disk
.raid_disk
] >= 0) {
10451 sd
->disk
.raid_disk
]);
10452 sources
[sd
->disk
.raid_disk
] =
10461 return new_degraded
;
10464 /*******************************************************************************
10465 * Function: imsm_manage_reshape
10466 * Description: Function finds array under reshape and it manages reshape
10467 * process. It creates stripes backups (if required) and sets
10470 * afd : Backup handle (nattive) - not used
10471 * sra : general array info
10472 * reshape : reshape parameters - not used
10473 * st : supertype structure
10474 * blocks : size of critical section [blocks]
10475 * fds : table of source device descriptor
10476 * offsets : start of array (offest per devices)
10478 * destfd : table of destination device descriptor
10479 * destoffsets : table of destination offsets (per device)
10481 * 1 : success, reshape is done
10483 ******************************************************************************/
10484 static int imsm_manage_reshape(
10485 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10486 struct supertype
*st
, unsigned long backup_blocks
,
10487 int *fds
, unsigned long long *offsets
,
10488 int dests
, int *destfd
, unsigned long long *destoffsets
)
10491 struct intel_super
*super
= st
->sb
;
10492 struct intel_dev
*dv
= NULL
;
10493 struct imsm_dev
*dev
= NULL
;
10494 struct imsm_map
*map_src
;
10495 int migr_vol_qan
= 0;
10496 int ndata
, odata
; /* [bytes] */
10497 int chunk
; /* [bytes] */
10498 struct migr_record
*migr_rec
;
10500 unsigned int buf_size
; /* [bytes] */
10501 unsigned long long max_position
; /* array size [bytes] */
10502 unsigned long long next_step
; /* [blocks]/[bytes] */
10503 unsigned long long old_data_stripe_length
;
10504 unsigned long long start_src
; /* [bytes] */
10505 unsigned long long start
; /* [bytes] */
10506 unsigned long long start_buf_shift
; /* [bytes] */
10508 int source_layout
= 0;
10510 if (!fds
|| !offsets
|| !sra
)
10513 /* Find volume during the reshape */
10514 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10515 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10516 && dv
->dev
->vol
.migr_state
== 1) {
10521 /* Only one volume can migrate at the same time */
10522 if (migr_vol_qan
!= 1) {
10523 pr_err(": %s", migr_vol_qan
?
10524 "Number of migrating volumes greater than 1\n" :
10525 "There is no volume during migrationg\n");
10529 map_src
= get_imsm_map(dev
, MAP_1
);
10530 if (map_src
== NULL
)
10533 ndata
= imsm_num_data_members(dev
, MAP_0
);
10534 odata
= imsm_num_data_members(dev
, MAP_1
);
10536 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10537 old_data_stripe_length
= odata
* chunk
;
10539 migr_rec
= super
->migr_rec
;
10541 /* initialize migration record for start condition */
10542 if (sra
->reshape_progress
== 0)
10543 init_migr_record_imsm(st
, dev
, sra
);
10545 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10546 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
10549 /* Save checkpoint to update migration record for current
10550 * reshape position (in md). It can be farther than current
10551 * reshape position in metadata.
10553 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10554 /* ignore error == 2, this can mean end of reshape here
10556 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
10561 /* size for data */
10562 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10563 /* extend buffer size for parity disk */
10564 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10565 /* add space for stripe aligment */
10566 buf_size
+= old_data_stripe_length
;
10567 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10568 dprintf("imsm: Cannot allocate checpoint buffer\n");
10572 max_position
= sra
->component_size
* ndata
;
10573 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10575 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10576 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10577 /* current reshape position [blocks] */
10578 unsigned long long current_position
=
10579 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10580 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10581 unsigned long long border
;
10583 /* Check that array hasn't become failed.
10585 degraded
= check_degradation_change(sra
, fds
, degraded
);
10586 if (degraded
> 1) {
10587 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
10591 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10593 if ((current_position
+ next_step
) > max_position
)
10594 next_step
= max_position
- current_position
;
10596 start
= current_position
* 512;
10598 /* allign reading start to old geometry */
10599 start_buf_shift
= start
% old_data_stripe_length
;
10600 start_src
= start
- start_buf_shift
;
10602 border
= (start_src
/ odata
) - (start
/ ndata
);
10604 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10605 /* save critical stripes to buf
10606 * start - start address of current unit
10607 * to backup [bytes]
10608 * start_src - start address of current unit
10609 * to backup alligned to source array
10612 unsigned long long next_step_filler
= 0;
10613 unsigned long long copy_length
= next_step
* 512;
10615 /* allign copy area length to stripe in old geometry */
10616 next_step_filler
= ((copy_length
+ start_buf_shift
)
10617 % old_data_stripe_length
);
10618 if (next_step_filler
)
10619 next_step_filler
= (old_data_stripe_length
10620 - next_step_filler
);
10621 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
10622 start
, start_src
, copy_length
,
10623 start_buf_shift
, next_step_filler
);
10625 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10626 chunk
, map_src
->raid_level
,
10627 source_layout
, 0, NULL
, start_src
,
10629 next_step_filler
+ start_buf_shift
,
10631 dprintf("imsm: Cannot save stripes to buffer\n");
10634 /* Convert data to destination format and store it
10635 * in backup general migration area
10637 if (save_backup_imsm(st
, dev
, sra
,
10638 buf
+ start_buf_shift
, copy_length
)) {
10639 dprintf("imsm: Cannot save stripes to target devices\n");
10642 if (save_checkpoint_imsm(st
, sra
,
10643 UNIT_SRC_IN_CP_AREA
)) {
10644 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
10648 /* set next step to use whole border area */
10649 border
/= next_step
;
10651 next_step
*= border
;
10653 /* When data backed up, checkpoint stored,
10654 * kick the kernel to reshape unit of data
10656 next_step
= next_step
+ sra
->reshape_progress
;
10657 /* limit next step to array max position */
10658 if (next_step
> max_position
)
10659 next_step
= max_position
;
10660 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10661 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10662 sra
->reshape_progress
= next_step
;
10664 /* wait until reshape finish */
10665 if (wait_for_reshape_imsm(sra
, ndata
)) {
10666 dprintf("wait_for_reshape_imsm returned error!\n");
10672 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10673 /* ignore error == 2, this can mean end of reshape here
10675 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
10681 /* clear migr_rec on disks after successful migration */
10684 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
10685 for (d
= super
->disks
; d
; d
= d
->next
) {
10686 if (d
->index
< 0 || is_failed(&d
->disk
))
10688 unsigned long long dsize
;
10690 get_dev_size(d
->fd
, NULL
, &dsize
);
10691 if (lseek64(d
->fd
, dsize
- MIGR_REC_POSITION
,
10693 if (write(d
->fd
, super
->migr_rec_buf
,
10694 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
10695 perror("Write migr_rec failed");
10699 /* return '1' if done */
10707 #endif /* MDASSEMBLE */
10709 struct superswitch super_imsm
= {
10711 .examine_super
= examine_super_imsm
,
10712 .brief_examine_super
= brief_examine_super_imsm
,
10713 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10714 .export_examine_super
= export_examine_super_imsm
,
10715 .detail_super
= detail_super_imsm
,
10716 .brief_detail_super
= brief_detail_super_imsm
,
10717 .write_init_super
= write_init_super_imsm
,
10718 .validate_geometry
= validate_geometry_imsm
,
10719 .add_to_super
= add_to_super_imsm
,
10720 .remove_from_super
= remove_from_super_imsm
,
10721 .detail_platform
= detail_platform_imsm
,
10722 .export_detail_platform
= export_detail_platform_imsm
,
10723 .kill_subarray
= kill_subarray_imsm
,
10724 .update_subarray
= update_subarray_imsm
,
10725 .load_container
= load_container_imsm
,
10726 .default_geometry
= default_geometry_imsm
,
10727 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10728 .reshape_super
= imsm_reshape_super
,
10729 .manage_reshape
= imsm_manage_reshape
,
10730 .recover_backup
= recover_backup_imsm
,
10731 .copy_metadata
= copy_metadata_imsm
,
10733 .match_home
= match_home_imsm
,
10734 .uuid_from_super
= uuid_from_super_imsm
,
10735 .getinfo_super
= getinfo_super_imsm
,
10736 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10737 .update_super
= update_super_imsm
,
10739 .avail_size
= avail_size_imsm
,
10740 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10742 .compare_super
= compare_super_imsm
,
10744 .load_super
= load_super_imsm
,
10745 .init_super
= init_super_imsm
,
10746 .store_super
= store_super_imsm
,
10747 .free_super
= free_super_imsm
,
10748 .match_metadata_desc
= match_metadata_desc_imsm
,
10749 .container_content
= container_content_imsm
,
10750 .validate_container
= validate_container_imsm
,
10757 .open_new
= imsm_open_new
,
10758 .set_array_state
= imsm_set_array_state
,
10759 .set_disk
= imsm_set_disk
,
10760 .sync_metadata
= imsm_sync_metadata
,
10761 .activate_spare
= imsm_activate_spare
,
10762 .process_update
= imsm_process_update
,
10763 .prepare_update
= imsm_prepare_update
,
10764 #endif /* MDASSEMBLE */