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 | \
87 /* Define attributes that are unused but not harmful */
88 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
90 #define MPB_SECTOR_CNT 2210
91 #define IMSM_RESERVED_SECTORS 4096
92 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
93 #define SECT_PER_MB_SHIFT 11
94 #define MAX_SECTOR_SIZE 4096
96 /* Disk configuration info. */
97 #define IMSM_MAX_DEVICES 255
99 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
100 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
101 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
102 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
103 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
104 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
105 #define JOURNAL_DISK __cpu_to_le32(0x2000000) /* Device marked as Journaling Drive */
106 __u32 status
; /* 0xF0 - 0xF3 */
107 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
108 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
109 #define IMSM_DISK_FILLERS 3
110 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
113 /* map selector for map managment
119 /* RAID map configuration infos. */
121 __u32 pba_of_lba0_lo
; /* start address of partition */
122 __u32 blocks_per_member_lo
;/* blocks per member */
123 __u32 num_data_stripes_lo
; /* number of data stripes */
124 __u16 blocks_per_strip
;
125 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
126 #define IMSM_T_STATE_NORMAL 0
127 #define IMSM_T_STATE_UNINITIALIZED 1
128 #define IMSM_T_STATE_DEGRADED 2
129 #define IMSM_T_STATE_FAILED 3
131 #define IMSM_T_RAID0 0
132 #define IMSM_T_RAID1 1
133 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
134 __u8 num_members
; /* number of member disks */
135 __u8 num_domains
; /* number of parity domains */
136 __u8 failed_disk_num
; /* valid only when state is degraded */
138 __u32 pba_of_lba0_hi
;
139 __u32 blocks_per_member_hi
;
140 __u32 num_data_stripes_hi
;
141 __u32 filler
[4]; /* expansion area */
142 #define IMSM_ORD_REBUILD (1 << 24)
143 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
144 * top byte contains some flags
146 } __attribute__ ((packed
));
149 __u32 curr_migr_unit
;
150 __u32 checkpoint_id
; /* id to access curr_migr_unit */
151 __u8 migr_state
; /* Normal or Migrating */
153 #define MIGR_REBUILD 1
154 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
155 #define MIGR_GEN_MIGR 3
156 #define MIGR_STATE_CHANGE 4
157 #define MIGR_REPAIR 5
158 __u8 migr_type
; /* Initializing, Rebuilding, ... */
159 #define RAIDVOL_CLEAN 0
160 #define RAIDVOL_DIRTY 1
161 #define RAIDVOL_DSRECORD_VALID 2
163 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
164 __u16 verify_errors
; /* number of mismatches */
165 __u16 bad_blocks
; /* number of bad blocks during verify */
167 struct imsm_map map
[1];
168 /* here comes another one if migr_state */
169 } __attribute__ ((packed
));
172 __u8 volume
[MAX_RAID_SERIAL_LEN
];
175 #define DEV_BOOTABLE __cpu_to_le32(0x01)
176 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
177 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
178 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
179 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
180 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
181 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
182 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
183 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
184 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
185 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
186 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
187 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
188 __u32 status
; /* Persistent RaidDev status */
189 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
193 __u8 cng_master_disk
;
197 __u16 my_vol_raid_dev_num
; /* Used in Unique volume Id for this RaidDev */
203 /* Unique Volume Id of the NvCache Volume associated with this volume */
204 __u32 nvc_vol_orig_family_num
;
205 __u16 nvc_vol_raid_dev_num
;
208 #define RWH_DISTRIBUTED 1
209 #define RWH_JOURNALING_DRIVE 2
210 __u8 rwh_policy
; /* Raid Write Hole Policy */
211 __u8 jd_serial
[MAX_RAID_SERIAL_LEN
]; /* Journal Drive serial number */
214 #define IMSM_DEV_FILLERS 3
215 __u32 filler
[IMSM_DEV_FILLERS
];
217 } __attribute__ ((packed
));
220 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
221 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
222 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
223 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
224 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
225 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
226 __u32 attributes
; /* 0x34 - 0x37 */
227 __u8 num_disks
; /* 0x38 Number of configured disks */
228 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
229 __u8 error_log_pos
; /* 0x3A */
230 __u8 fill
[1]; /* 0x3B */
231 __u32 cache_size
; /* 0x3c - 0x40 in mb */
232 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
233 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
234 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
235 #define IMSM_FILLERS 35
236 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
237 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
238 /* here comes imsm_dev[num_raid_devs] */
239 /* here comes BBM logs */
240 } __attribute__ ((packed
));
242 #define BBM_LOG_MAX_ENTRIES 254
243 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
244 #define BBM_LOG_SIGNATURE 0xabadb10c
246 struct bbm_log_block_addr
{
249 } __attribute__ ((__packed__
));
251 struct bbm_log_entry
{
252 __u8 marked_count
; /* Number of blocks marked - 1 */
253 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
254 struct bbm_log_block_addr defective_block_start
;
255 } __attribute__ ((__packed__
));
258 __u32 signature
; /* 0xABADB10C */
260 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
261 } __attribute__ ((__packed__
));
264 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
267 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
269 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
271 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
272 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
273 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
276 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
277 * be recovered using srcMap */
278 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
279 * already been migrated and must
280 * be recovered from checkpoint area */
282 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of the PPL, without the header */
285 __u32 rec_status
; /* Status used to determine how to restart
286 * migration in case it aborts
288 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
289 __u32 family_num
; /* Family number of MPB
290 * containing the RaidDev
291 * that is migrating */
292 __u32 ascending_migr
; /* True if migrating in increasing
294 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
295 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
297 * advances per unit-of-operation */
298 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
299 __u32 dest_1st_member_lba
; /* First member lba on first
300 * stripe of destination */
301 __u32 num_migr_units
; /* Total num migration units-of-op */
302 __u32 post_migr_vol_cap
; /* Size of volume after
303 * migration completes */
304 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
305 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
306 * migration ckpt record was read from
307 * (for recovered migrations) */
308 } __attribute__ ((__packed__
));
313 * 2: metadata does not match
321 struct md_list
*next
;
324 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
326 static __u8
migr_type(struct imsm_dev
*dev
)
328 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
329 dev
->status
& DEV_VERIFY_AND_FIX
)
332 return dev
->vol
.migr_type
;
335 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
337 /* for compatibility with older oroms convert MIGR_REPAIR, into
338 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
340 if (migr_type
== MIGR_REPAIR
) {
341 dev
->vol
.migr_type
= MIGR_VERIFY
;
342 dev
->status
|= DEV_VERIFY_AND_FIX
;
344 dev
->vol
.migr_type
= migr_type
;
345 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
349 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
351 return ROUND_UP(bytes
, sector_size
) / sector_size
;
354 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
355 unsigned int sector_size
)
357 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
361 struct imsm_dev
*dev
;
362 struct intel_dev
*next
;
367 enum sys_dev_type type
;
370 struct intel_hba
*next
;
377 /* internal representation of IMSM metadata */
380 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
381 struct imsm_super
*anchor
; /* immovable parameters */
384 void *migr_rec_buf
; /* buffer for I/O operations */
385 struct migr_record
*migr_rec
; /* migration record */
387 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
388 array, it indicates that mdmon is allowed to clean migration
390 size_t len
; /* size of the 'buf' allocation */
391 size_t extra_space
; /* extra space in 'buf' that is not used yet */
392 void *next_buf
; /* for realloc'ing buf from the manager */
394 int updates_pending
; /* count of pending updates for mdmon */
395 int current_vol
; /* index of raid device undergoing creation */
396 unsigned long long create_offset
; /* common start for 'current_vol' */
397 __u32 random
; /* random data for seeding new family numbers */
398 struct intel_dev
*devlist
;
399 unsigned int sector_size
; /* sector size of used member drives */
403 __u8 serial
[MAX_RAID_SERIAL_LEN
];
406 struct imsm_disk disk
;
409 struct extent
*e
; /* for determining freespace @ create */
410 int raiddisk
; /* slot to fill in autolayout */
412 } *disks
, *current_disk
;
413 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
415 struct dl
*missing
; /* disks removed while we weren't looking */
416 struct bbm_log
*bbm_log
;
417 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
418 const struct imsm_orom
*orom
; /* platform firmware support */
419 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
420 struct md_bb bb
; /* memory for get_bad_blocks call */
424 struct imsm_disk disk
;
425 #define IMSM_UNKNOWN_OWNER (-1)
427 struct intel_disk
*next
;
431 unsigned long long start
, size
;
434 /* definitions of reshape process types */
435 enum imsm_reshape_type
{
441 /* definition of messages passed to imsm_process_update */
442 enum imsm_update_type
{
443 update_activate_spare
,
447 update_add_remove_disk
,
448 update_reshape_container_disks
,
449 update_reshape_migration
,
451 update_general_migration_checkpoint
,
453 update_prealloc_badblocks_mem
,
456 struct imsm_update_activate_spare
{
457 enum imsm_update_type type
;
461 struct imsm_update_activate_spare
*next
;
467 unsigned long long size
;
474 enum takeover_direction
{
478 struct imsm_update_takeover
{
479 enum imsm_update_type type
;
481 enum takeover_direction direction
;
484 struct imsm_update_reshape
{
485 enum imsm_update_type type
;
489 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
492 struct imsm_update_reshape_migration
{
493 enum imsm_update_type type
;
496 /* fields for array migration changes
503 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
506 struct imsm_update_size_change
{
507 enum imsm_update_type type
;
512 struct imsm_update_general_migration_checkpoint
{
513 enum imsm_update_type type
;
514 __u32 curr_migr_unit
;
518 __u8 serial
[MAX_RAID_SERIAL_LEN
];
521 struct imsm_update_create_array
{
522 enum imsm_update_type type
;
527 struct imsm_update_kill_array
{
528 enum imsm_update_type type
;
532 struct imsm_update_rename_array
{
533 enum imsm_update_type type
;
534 __u8 name
[MAX_RAID_SERIAL_LEN
];
538 struct imsm_update_add_remove_disk
{
539 enum imsm_update_type type
;
542 struct imsm_update_prealloc_bb_mem
{
543 enum imsm_update_type type
;
546 static const char *_sys_dev_type
[] = {
547 [SYS_DEV_UNKNOWN
] = "Unknown",
548 [SYS_DEV_SAS
] = "SAS",
549 [SYS_DEV_SATA
] = "SATA",
550 [SYS_DEV_NVME
] = "NVMe",
551 [SYS_DEV_VMD
] = "VMD"
554 const char *get_sys_dev_type(enum sys_dev_type type
)
556 if (type
>= SYS_DEV_MAX
)
557 type
= SYS_DEV_UNKNOWN
;
559 return _sys_dev_type
[type
];
562 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
564 struct intel_hba
*result
= xmalloc(sizeof(*result
));
566 result
->type
= device
->type
;
567 result
->path
= xstrdup(device
->path
);
569 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
575 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
577 struct intel_hba
*result
;
579 for (result
= hba
; result
; result
= result
->next
) {
580 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
586 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
588 struct intel_hba
*hba
;
590 /* check if disk attached to Intel HBA */
591 hba
= find_intel_hba(super
->hba
, device
);
594 /* Check if HBA is already attached to super */
595 if (super
->hba
== NULL
) {
596 super
->hba
= alloc_intel_hba(device
);
601 /* Intel metadata allows for all disks attached to the same type HBA.
602 * Do not support HBA types mixing
604 if (device
->type
!= hba
->type
)
607 /* Multiple same type HBAs can be used if they share the same OROM */
608 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
610 if (device_orom
!= super
->orom
)
616 hba
->next
= alloc_intel_hba(device
);
620 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
622 struct sys_dev
*list
, *elem
;
625 if ((list
= find_intel_devices()) == NULL
)
629 disk_path
= (char *) devname
;
631 disk_path
= diskfd_to_devpath(fd
);
636 for (elem
= list
; elem
; elem
= elem
->next
)
637 if (path_attached_to_hba(disk_path
, elem
->path
))
640 if (disk_path
!= devname
)
646 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
649 static struct supertype
*match_metadata_desc_imsm(char *arg
)
651 struct supertype
*st
;
653 if (strcmp(arg
, "imsm") != 0 &&
654 strcmp(arg
, "default") != 0
658 st
= xcalloc(1, sizeof(*st
));
659 st
->ss
= &super_imsm
;
660 st
->max_devs
= IMSM_MAX_DEVICES
;
661 st
->minor_version
= 0;
667 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
669 return &mpb
->sig
[MPB_SIG_LEN
];
673 /* retrieve a disk directly from the anchor when the anchor is known to be
674 * up-to-date, currently only at load time
676 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
678 if (index
>= mpb
->num_disks
)
680 return &mpb
->disk
[index
];
683 /* retrieve the disk description based on a index of the disk
686 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
690 for (d
= super
->disks
; d
; d
= d
->next
)
691 if (d
->index
== index
)
696 /* retrieve a disk from the parsed metadata */
697 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
701 dl
= get_imsm_dl_disk(super
, index
);
708 /* generate a checksum directly from the anchor when the anchor is known to be
709 * up-to-date, currently only at load or write_super after coalescing
711 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
713 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
714 __u32
*p
= (__u32
*) mpb
;
718 sum
+= __le32_to_cpu(*p
);
722 return sum
- __le32_to_cpu(mpb
->check_sum
);
725 static size_t sizeof_imsm_map(struct imsm_map
*map
)
727 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
730 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
732 /* A device can have 2 maps if it is in the middle of a migration.
734 * MAP_0 - we return the first map
735 * MAP_1 - we return the second map if it exists, else NULL
736 * MAP_X - we return the second map if it exists, else the first
738 struct imsm_map
*map
= &dev
->vol
.map
[0];
739 struct imsm_map
*map2
= NULL
;
741 if (dev
->vol
.migr_state
)
742 map2
= (void *)map
+ sizeof_imsm_map(map
);
744 switch (second_map
) {
761 /* return the size of the device.
762 * migr_state increases the returned size if map[0] were to be duplicated
764 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
766 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
767 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
769 /* migrating means an additional map */
770 if (dev
->vol
.migr_state
)
771 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
773 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
779 /* retrieve disk serial number list from a metadata update */
780 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
783 struct disk_info
*inf
;
785 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
786 sizeof_imsm_dev(&update
->dev
, 0);
792 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
798 if (index
>= mpb
->num_raid_devs
)
801 /* devices start after all disks */
802 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
804 for (i
= 0; i
<= index
; i
++)
806 return _mpb
+ offset
;
808 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
813 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
815 struct intel_dev
*dv
;
817 if (index
>= super
->anchor
->num_raid_devs
)
819 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
820 if (dv
->index
== index
)
825 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
828 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
829 __le16_to_cpu(addr
->w1
));
832 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
834 struct bbm_log_block_addr addr
;
836 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
837 addr
.dw1
= __cpu_to_le32((__u32
)(sec
>> 16) & 0xffffffff);
842 /* get size of the bbm log */
843 static __u32
get_imsm_bbm_log_size(struct bbm_log
*log
)
845 if (!log
|| log
->entry_count
== 0)
848 return sizeof(log
->signature
) +
849 sizeof(log
->entry_count
) +
850 log
->entry_count
* sizeof(struct bbm_log_entry
);
853 /* check if bad block is not partially stored in bbm log */
854 static int is_stored_in_bbm(struct bbm_log
*log
, const __u8 idx
, const unsigned
855 long long sector
, const int length
, __u32
*pos
)
859 for (i
= *pos
; i
< log
->entry_count
; i
++) {
860 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
861 unsigned long long bb_start
;
862 unsigned long long bb_end
;
864 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
865 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
867 if ((entry
->disk_ordinal
== idx
) && (bb_start
>= sector
) &&
868 (bb_end
<= sector
+ length
)) {
876 /* record new bad block in bbm log */
877 static int record_new_badblock(struct bbm_log
*log
, const __u8 idx
, unsigned
878 long long sector
, int length
)
882 struct bbm_log_entry
*entry
= NULL
;
884 while (is_stored_in_bbm(log
, idx
, sector
, length
, &pos
)) {
885 struct bbm_log_entry
*e
= &log
->marked_block_entries
[pos
];
887 if ((e
->marked_count
+ 1 == BBM_LOG_MAX_LBA_ENTRY_VAL
) &&
888 (__le48_to_cpu(&e
->defective_block_start
) == sector
)) {
889 sector
+= BBM_LOG_MAX_LBA_ENTRY_VAL
;
890 length
-= BBM_LOG_MAX_LBA_ENTRY_VAL
;
899 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
900 BBM_LOG_MAX_LBA_ENTRY_VAL
;
901 entry
->defective_block_start
= __cpu_to_le48(sector
);
902 entry
->marked_count
= cnt
- 1;
909 new_bb
= ROUND_UP(length
, BBM_LOG_MAX_LBA_ENTRY_VAL
) /
910 BBM_LOG_MAX_LBA_ENTRY_VAL
;
911 if (log
->entry_count
+ new_bb
> BBM_LOG_MAX_ENTRIES
)
915 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
916 BBM_LOG_MAX_LBA_ENTRY_VAL
;
917 struct bbm_log_entry
*entry
=
918 &log
->marked_block_entries
[log
->entry_count
];
920 entry
->defective_block_start
= __cpu_to_le48(sector
);
921 entry
->marked_count
= cnt
- 1;
922 entry
->disk_ordinal
= idx
;
933 /* clear all bad blocks for given disk */
934 static void clear_disk_badblocks(struct bbm_log
*log
, const __u8 idx
)
938 while (i
< log
->entry_count
) {
939 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
941 if (entries
[i
].disk_ordinal
== idx
) {
942 if (i
< log
->entry_count
- 1)
943 entries
[i
] = entries
[log
->entry_count
- 1];
951 /* clear given bad block */
952 static int clear_badblock(struct bbm_log
*log
, const __u8 idx
, const unsigned
953 long long sector
, const int length
) {
956 while (i
< log
->entry_count
) {
957 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
959 if ((entries
[i
].disk_ordinal
== idx
) &&
960 (__le48_to_cpu(&entries
[i
].defective_block_start
) ==
961 sector
) && (entries
[i
].marked_count
+ 1 == length
)) {
962 if (i
< log
->entry_count
- 1)
963 entries
[i
] = entries
[log
->entry_count
- 1];
972 #endif /* MDASSEMBLE */
974 /* allocate and load BBM log from metadata */
975 static int load_bbm_log(struct intel_super
*super
)
977 struct imsm_super
*mpb
= super
->anchor
;
978 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
980 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
985 struct bbm_log
*log
= (void *)mpb
+
986 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
990 if (bbm_log_size
< sizeof(log
->signature
) +
991 sizeof(log
->entry_count
))
994 entry_count
= __le32_to_cpu(log
->entry_count
);
995 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
996 (entry_count
> BBM_LOG_MAX_ENTRIES
))
1000 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
1001 entry_count
* sizeof(struct bbm_log_entry
))
1004 memcpy(super
->bbm_log
, log
, bbm_log_size
);
1006 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
1007 super
->bbm_log
->entry_count
= 0;
1013 /* checks if bad block is within volume boundaries */
1014 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
1015 const unsigned long long start_sector
,
1016 const unsigned long long size
)
1018 unsigned long long bb_start
;
1019 unsigned long long bb_end
;
1021 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1022 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1024 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1025 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1031 /* get list of bad blocks on a drive for a volume */
1032 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1033 const unsigned long long start_sector
,
1034 const unsigned long long size
,
1040 for (i
= 0; i
< log
->entry_count
; i
++) {
1041 const struct bbm_log_entry
*ent
=
1042 &log
->marked_block_entries
[i
];
1043 struct md_bb_entry
*bb
;
1045 if ((ent
->disk_ordinal
== idx
) &&
1046 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1048 if (!bbs
->entries
) {
1049 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1055 bb
= &bbs
->entries
[count
++];
1056 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1057 bb
->length
= ent
->marked_count
+ 1;
1065 * == MAP_0 get first map
1066 * == MAP_1 get second map
1067 * == MAP_X than get map according to the current migr_state
1069 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1073 struct imsm_map
*map
;
1075 map
= get_imsm_map(dev
, second_map
);
1077 /* top byte identifies disk under rebuild */
1078 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1081 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1082 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1084 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1086 return ord_to_idx(ord
);
1089 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1091 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1094 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
1099 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1100 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1101 if (ord_to_idx(ord
) == idx
)
1108 static int get_imsm_raid_level(struct imsm_map
*map
)
1110 if (map
->raid_level
== 1) {
1111 if (map
->num_members
== 2)
1117 return map
->raid_level
;
1120 static int cmp_extent(const void *av
, const void *bv
)
1122 const struct extent
*a
= av
;
1123 const struct extent
*b
= bv
;
1124 if (a
->start
< b
->start
)
1126 if (a
->start
> b
->start
)
1131 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1133 int memberships
= 0;
1136 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1137 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1138 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1140 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
1147 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1149 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
1151 if (lo
== 0 || hi
== 0)
1153 *lo
= __le32_to_cpu((unsigned)n
);
1154 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
1158 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1160 return (unsigned long long)__le32_to_cpu(lo
) |
1161 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1164 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1168 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1171 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1175 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1178 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1182 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1185 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1189 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1192 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1194 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1197 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1199 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1202 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1204 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1207 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1209 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1212 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
1214 /* find a list of used extents on the given physical device */
1215 struct extent
*rv
, *e
;
1217 int memberships
= count_memberships(dl
, super
);
1220 /* trim the reserved area for spares, so they can join any array
1221 * regardless of whether the OROM has assigned sectors from the
1222 * IMSM_RESERVED_SECTORS region
1224 if (dl
->index
== -1)
1225 reservation
= imsm_min_reserved_sectors(super
);
1227 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1229 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
1232 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1233 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1234 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1236 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1237 e
->start
= pba_of_lba0(map
);
1238 e
->size
= blocks_per_member(map
);
1242 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1244 /* determine the start of the metadata
1245 * when no raid devices are defined use the default
1246 * ...otherwise allow the metadata to truncate the value
1247 * as is the case with older versions of imsm
1250 struct extent
*last
= &rv
[memberships
- 1];
1251 unsigned long long remainder
;
1253 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1254 /* round down to 1k block to satisfy precision of the kernel
1258 /* make sure remainder is still sane */
1259 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1260 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1261 if (reservation
> remainder
)
1262 reservation
= remainder
;
1264 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1269 /* try to determine how much space is reserved for metadata from
1270 * the last get_extents() entry, otherwise fallback to the
1273 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1279 /* for spares just return a minimal reservation which will grow
1280 * once the spare is picked up by an array
1282 if (dl
->index
== -1)
1283 return MPB_SECTOR_CNT
;
1285 e
= get_extents(super
, dl
);
1287 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1289 /* scroll to last entry */
1290 for (i
= 0; e
[i
].size
; i
++)
1293 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1300 static int is_spare(struct imsm_disk
*disk
)
1302 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1305 static int is_configured(struct imsm_disk
*disk
)
1307 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1310 static int is_failed(struct imsm_disk
*disk
)
1312 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1315 static int is_journal(struct imsm_disk
*disk
)
1317 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1320 /* try to determine how much space is reserved for metadata from
1321 * the last get_extents() entry on the smallest active disk,
1322 * otherwise fallback to the default
1324 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1328 unsigned long long min_active
;
1330 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1331 struct dl
*dl
, *dl_min
= NULL
;
1337 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1340 unsigned long long blocks
= total_blocks(&dl
->disk
);
1341 if (blocks
< min_active
|| min_active
== 0) {
1343 min_active
= blocks
;
1349 /* find last lba used by subarrays on the smallest active disk */
1350 e
= get_extents(super
, dl_min
);
1353 for (i
= 0; e
[i
].size
; i
++)
1356 remainder
= min_active
- e
[i
].start
;
1359 /* to give priority to recovery we should not require full
1360 IMSM_RESERVED_SECTORS from the spare */
1361 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1363 /* if real reservation is smaller use that value */
1364 return (remainder
< rv
) ? remainder
: rv
;
1367 /* Return minimum size of a spare that can be used in this array*/
1368 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1370 struct intel_super
*super
= st
->sb
;
1374 unsigned long long rv
= 0;
1378 /* find first active disk in array */
1380 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1384 /* find last lba used by subarrays */
1385 e
= get_extents(super
, dl
);
1388 for (i
= 0; e
[i
].size
; i
++)
1391 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1394 /* add the amount of space needed for metadata */
1395 rv
= rv
+ imsm_min_reserved_sectors(super
);
1400 static int is_gen_migration(struct imsm_dev
*dev
);
1402 #define IMSM_4K_DIV 8
1405 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1406 struct imsm_dev
*dev
);
1408 static void print_imsm_dev(struct intel_super
*super
,
1409 struct imsm_dev
*dev
,
1415 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1416 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1420 printf("[%.16s]:\n", dev
->volume
);
1421 printf(" UUID : %s\n", uuid
);
1422 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1424 printf(" <-- %d", get_imsm_raid_level(map2
));
1426 printf(" Members : %d", map
->num_members
);
1428 printf(" <-- %d", map2
->num_members
);
1430 printf(" Slots : [");
1431 for (i
= 0; i
< map
->num_members
; i
++) {
1432 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1433 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1438 for (i
= 0; i
< map2
->num_members
; i
++) {
1439 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1440 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1445 printf(" Failed disk : ");
1446 if (map
->failed_disk_num
== 0xff)
1449 printf("%i", map
->failed_disk_num
);
1451 slot
= get_imsm_disk_slot(map
, disk_idx
);
1453 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1454 printf(" This Slot : %d%s\n", slot
,
1455 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1457 printf(" This Slot : ?\n");
1458 sz
= __le32_to_cpu(dev
->size_high
);
1460 sz
+= __le32_to_cpu(dev
->size_low
);
1461 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1462 human_size(sz
* 512));
1463 sz
= blocks_per_member(map
);
1464 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1465 human_size(sz
* 512));
1466 printf(" Sector Offset : %llu\n",
1468 printf(" Num Stripes : %llu\n",
1469 num_data_stripes(map
));
1470 printf(" Chunk Size : %u KiB",
1471 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1473 printf(" <-- %u KiB",
1474 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1476 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1477 printf(" Migrate State : ");
1478 if (dev
->vol
.migr_state
) {
1479 if (migr_type(dev
) == MIGR_INIT
)
1480 printf("initialize\n");
1481 else if (migr_type(dev
) == MIGR_REBUILD
)
1482 printf("rebuild\n");
1483 else if (migr_type(dev
) == MIGR_VERIFY
)
1485 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1486 printf("general migration\n");
1487 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1488 printf("state change\n");
1489 else if (migr_type(dev
) == MIGR_REPAIR
)
1492 printf("<unknown:%d>\n", migr_type(dev
));
1495 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1496 if (dev
->vol
.migr_state
) {
1497 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1499 printf(" <-- %s", map_state_str
[map
->map_state
]);
1500 printf("\n Checkpoint : %u ",
1501 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1502 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1505 printf("(%llu)", (unsigned long long)
1506 blocks_per_migr_unit(super
, dev
));
1509 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1511 printf(" RWH Policy : ");
1512 if (dev
->rwh_policy
== RWH_OFF
)
1514 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1515 printf("PPL distributed\n");
1516 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1517 printf("PPL journaling drive\n");
1519 printf("<unknown:%d>\n", dev
->rwh_policy
);
1522 static void print_imsm_disk(struct imsm_disk
*disk
,
1525 unsigned int sector_size
) {
1526 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1529 if (index
< -1 || !disk
)
1533 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1535 printf(" Disk%02d Serial : %s\n", index
, str
);
1537 printf(" Disk Serial : %s\n", str
);
1538 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1539 is_configured(disk
) ? " active" : "",
1540 is_failed(disk
) ? " failed" : "",
1541 is_journal(disk
) ? " journal" : "");
1542 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1543 sz
= total_blocks(disk
) - reserved
;
1544 printf(" Usable Size : %llu%s\n",
1545 (unsigned long long)sz
* 512 / sector_size
,
1546 human_size(sz
* 512));
1549 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1551 struct migr_record
*migr_rec
= super
->migr_rec
;
1553 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1554 migr_rec
->ckpt_area_pba
/= IMSM_4K_DIV
;
1555 migr_rec
->dest_1st_member_lba
/= IMSM_4K_DIV
;
1556 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1557 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1558 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1559 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1562 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1564 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1567 void convert_to_4k(struct intel_super
*super
)
1569 struct imsm_super
*mpb
= super
->anchor
;
1570 struct imsm_disk
*disk
;
1572 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1574 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1575 disk
= __get_imsm_disk(mpb
, i
);
1577 convert_to_4k_imsm_disk(disk
);
1579 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1580 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1581 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1583 split_ull((join_u32(dev
->size_low
, dev
->size_high
)/IMSM_4K_DIV
),
1584 &dev
->size_low
, &dev
->size_high
);
1585 dev
->vol
.curr_migr_unit
/= IMSM_4K_DIV
;
1588 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1589 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1590 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1592 if (dev
->vol
.migr_state
) {
1594 map
= get_imsm_map(dev
, MAP_1
);
1595 set_blocks_per_member(map
,
1596 blocks_per_member(map
)/IMSM_4K_DIV
);
1597 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1598 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1602 struct bbm_log
*log
= (void *)mpb
+
1603 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1606 for (i
= 0; i
< log
->entry_count
; i
++) {
1607 struct bbm_log_entry
*entry
=
1608 &log
->marked_block_entries
[i
];
1610 __u8 count
= entry
->marked_count
+ 1;
1611 unsigned long long sector
=
1612 __le48_to_cpu(&entry
->defective_block_start
);
1614 entry
->defective_block_start
=
1615 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1616 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1620 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1623 void examine_migr_rec_imsm(struct intel_super
*super
)
1625 struct migr_record
*migr_rec
= super
->migr_rec
;
1626 struct imsm_super
*mpb
= super
->anchor
;
1629 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1630 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1631 struct imsm_map
*map
;
1634 if (is_gen_migration(dev
) == 0)
1637 printf("\nMigration Record Information:");
1639 /* first map under migration */
1640 map
= get_imsm_map(dev
, MAP_0
);
1642 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1643 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1644 printf(" Empty\n ");
1645 printf("Examine one of first two disks in array\n");
1648 printf("\n Status : ");
1649 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1652 printf("Contains Data\n");
1653 printf(" Current Unit : %u\n",
1654 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1655 printf(" Family : %u\n",
1656 __le32_to_cpu(migr_rec
->family_num
));
1657 printf(" Ascending : %u\n",
1658 __le32_to_cpu(migr_rec
->ascending_migr
));
1659 printf(" Blocks Per Unit : %u\n",
1660 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1661 printf(" Dest. Depth Per Unit : %u\n",
1662 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1663 printf(" Checkpoint Area pba : %u\n",
1664 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1665 printf(" First member lba : %u\n",
1666 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1667 printf(" Total Number of Units : %u\n",
1668 __le32_to_cpu(migr_rec
->num_migr_units
));
1669 printf(" Size of volume : %u\n",
1670 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1671 printf(" Expansion space for LBA64 : %u\n",
1672 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1673 printf(" Record was read from : %u\n",
1674 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1679 #endif /* MDASSEMBLE */
1681 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1683 struct migr_record
*migr_rec
= super
->migr_rec
;
1685 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1686 migr_rec
->ckpt_area_pba
*= IMSM_4K_DIV
;
1687 migr_rec
->dest_1st_member_lba
*= IMSM_4K_DIV
;
1688 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1689 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1690 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1691 &migr_rec
->post_migr_vol_cap
,
1692 &migr_rec
->post_migr_vol_cap_hi
);
1695 void convert_from_4k(struct intel_super
*super
)
1697 struct imsm_super
*mpb
= super
->anchor
;
1698 struct imsm_disk
*disk
;
1700 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1702 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1703 disk
= __get_imsm_disk(mpb
, i
);
1705 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1708 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1709 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1710 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1712 split_ull((join_u32(dev
->size_low
, dev
->size_high
)*IMSM_4K_DIV
),
1713 &dev
->size_low
, &dev
->size_high
);
1714 dev
->vol
.curr_migr_unit
*= IMSM_4K_DIV
;
1717 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1718 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1719 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1721 if (dev
->vol
.migr_state
) {
1723 map
= get_imsm_map(dev
, MAP_1
);
1724 set_blocks_per_member(map
,
1725 blocks_per_member(map
)*IMSM_4K_DIV
);
1726 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1727 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1731 struct bbm_log
*log
= (void *)mpb
+
1732 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1735 for (i
= 0; i
< log
->entry_count
; i
++) {
1736 struct bbm_log_entry
*entry
=
1737 &log
->marked_block_entries
[i
];
1739 __u8 count
= entry
->marked_count
+ 1;
1740 unsigned long long sector
=
1741 __le48_to_cpu(&entry
->defective_block_start
);
1743 entry
->defective_block_start
=
1744 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1745 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1749 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1752 /*******************************************************************************
1753 * function: imsm_check_attributes
1754 * Description: Function checks if features represented by attributes flags
1755 * are supported by mdadm.
1757 * attributes - Attributes read from metadata
1759 * 0 - passed attributes contains unsupported features flags
1760 * 1 - all features are supported
1761 ******************************************************************************/
1762 static int imsm_check_attributes(__u32 attributes
)
1765 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1767 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1769 not_supported
&= attributes
;
1770 if (not_supported
) {
1771 pr_err("(IMSM): Unsupported attributes : %x\n",
1772 (unsigned)__le32_to_cpu(not_supported
));
1773 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1774 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1775 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1777 if (not_supported
& MPB_ATTRIB_2TB
) {
1778 dprintf("\t\tMPB_ATTRIB_2TB\n");
1779 not_supported
^= MPB_ATTRIB_2TB
;
1781 if (not_supported
& MPB_ATTRIB_RAID0
) {
1782 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1783 not_supported
^= MPB_ATTRIB_RAID0
;
1785 if (not_supported
& MPB_ATTRIB_RAID1
) {
1786 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1787 not_supported
^= MPB_ATTRIB_RAID1
;
1789 if (not_supported
& MPB_ATTRIB_RAID10
) {
1790 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1791 not_supported
^= MPB_ATTRIB_RAID10
;
1793 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1794 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1795 not_supported
^= MPB_ATTRIB_RAID1E
;
1797 if (not_supported
& MPB_ATTRIB_RAID5
) {
1798 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1799 not_supported
^= MPB_ATTRIB_RAID5
;
1801 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1802 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1803 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1805 if (not_supported
& MPB_ATTRIB_BBM
) {
1806 dprintf("\t\tMPB_ATTRIB_BBM\n");
1807 not_supported
^= MPB_ATTRIB_BBM
;
1809 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1810 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1811 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1813 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1814 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1815 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1817 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1818 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1819 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1821 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1822 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1823 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1825 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1826 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1827 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1831 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1840 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1842 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1844 struct intel_super
*super
= st
->sb
;
1845 struct imsm_super
*mpb
= super
->anchor
;
1846 char str
[MAX_SIGNATURE_LENGTH
];
1851 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1854 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
1855 str
[MPB_SIG_LEN
-1] = '\0';
1856 printf(" Magic : %s\n", str
);
1857 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1858 printf(" Version : %s\n", get_imsm_version(mpb
));
1859 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1860 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1861 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1862 printf(" Attributes : ");
1863 if (imsm_check_attributes(mpb
->attributes
))
1864 printf("All supported\n");
1866 printf("not supported\n");
1867 getinfo_super_imsm(st
, &info
, NULL
);
1868 fname_from_uuid(st
, &info
, nbuf
, ':');
1869 printf(" UUID : %s\n", nbuf
+ 5);
1870 sum
= __le32_to_cpu(mpb
->check_sum
);
1871 printf(" Checksum : %08x %s\n", sum
,
1872 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1873 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
1874 printf(" Disks : %d\n", mpb
->num_disks
);
1875 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1876 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
1877 super
->disks
->index
, reserved
, super
->sector_size
);
1878 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
1879 struct bbm_log
*log
= super
->bbm_log
;
1882 printf("Bad Block Management Log:\n");
1883 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1884 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1885 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1887 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1889 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1891 super
->current_vol
= i
;
1892 getinfo_super_imsm(st
, &info
, NULL
);
1893 fname_from_uuid(st
, &info
, nbuf
, ':');
1894 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1896 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1897 if (i
== super
->disks
->index
)
1899 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
1900 super
->sector_size
);
1903 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1904 if (dl
->index
== -1)
1905 print_imsm_disk(&dl
->disk
, -1, reserved
,
1906 super
->sector_size
);
1908 examine_migr_rec_imsm(super
);
1911 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1913 /* We just write a generic IMSM ARRAY entry */
1916 struct intel_super
*super
= st
->sb
;
1918 if (!super
->anchor
->num_raid_devs
) {
1919 printf("ARRAY metadata=imsm\n");
1923 getinfo_super_imsm(st
, &info
, NULL
);
1924 fname_from_uuid(st
, &info
, nbuf
, ':');
1925 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1928 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1930 /* We just write a generic IMSM ARRAY entry */
1934 struct intel_super
*super
= st
->sb
;
1937 if (!super
->anchor
->num_raid_devs
)
1940 getinfo_super_imsm(st
, &info
, NULL
);
1941 fname_from_uuid(st
, &info
, nbuf
, ':');
1942 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1943 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1945 super
->current_vol
= i
;
1946 getinfo_super_imsm(st
, &info
, NULL
);
1947 fname_from_uuid(st
, &info
, nbuf1
, ':');
1948 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1949 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1953 static void export_examine_super_imsm(struct supertype
*st
)
1955 struct intel_super
*super
= st
->sb
;
1956 struct imsm_super
*mpb
= super
->anchor
;
1960 getinfo_super_imsm(st
, &info
, NULL
);
1961 fname_from_uuid(st
, &info
, nbuf
, ':');
1962 printf("MD_METADATA=imsm\n");
1963 printf("MD_LEVEL=container\n");
1964 printf("MD_UUID=%s\n", nbuf
+5);
1965 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1968 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
1970 /* The second last sector of the device contains
1971 * the "struct imsm_super" metadata.
1972 * This contains mpb_size which is the size in bytes of the
1973 * extended metadata. This is located immediately before
1975 * We want to read all that, plus the last sector which
1976 * may contain a migration record, and write it all
1980 unsigned long long dsize
, offset
;
1982 struct imsm_super
*sb
;
1983 struct intel_super
*super
= st
->sb
;
1984 unsigned int sector_size
= super
->sector_size
;
1985 unsigned int written
= 0;
1987 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
) != 0)
1990 if (!get_dev_size(from
, NULL
, &dsize
))
1993 if (lseek64(from
, dsize
-(2*sector_size
), 0) < 0)
1995 if ((unsigned int)read(from
, buf
, sector_size
) != sector_size
)
1998 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
2001 sectors
= mpb_sectors(sb
, sector_size
) + 2;
2002 offset
= dsize
- sectors
* sector_size
;
2003 if (lseek64(from
, offset
, 0) < 0 ||
2004 lseek64(to
, offset
, 0) < 0)
2006 while (written
< sectors
* sector_size
) {
2007 int n
= sectors
*sector_size
- written
;
2010 if (read(from
, buf
, n
) != n
)
2012 if (write(to
, buf
, n
) != n
)
2023 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
2028 getinfo_super_imsm(st
, &info
, NULL
);
2029 fname_from_uuid(st
, &info
, nbuf
, ':');
2030 printf("\n UUID : %s\n", nbuf
+ 5);
2033 static void brief_detail_super_imsm(struct supertype
*st
)
2037 getinfo_super_imsm(st
, &info
, NULL
);
2038 fname_from_uuid(st
, &info
, nbuf
, ':');
2039 printf(" UUID=%s", nbuf
+ 5);
2042 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
2043 static void fd2devname(int fd
, char *name
);
2045 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2047 /* dump an unsorted list of devices attached to AHCI Intel storage
2048 * controller, as well as non-connected ports
2050 int hba_len
= strlen(hba_path
) + 1;
2055 unsigned long port_mask
= (1 << port_count
) - 1;
2057 if (port_count
> (int)sizeof(port_mask
) * 8) {
2059 pr_err("port_count %d out of range\n", port_count
);
2063 /* scroll through /sys/dev/block looking for devices attached to
2066 dir
= opendir("/sys/dev/block");
2070 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2081 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2083 path
= devt_to_devpath(makedev(major
, minor
));
2086 if (!path_attached_to_hba(path
, hba_path
)) {
2092 /* retrieve the scsi device type */
2093 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2095 pr_err("failed to allocate 'device'\n");
2099 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2100 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2102 pr_err("failed to read device type for %s\n",
2108 type
= strtoul(buf
, NULL
, 10);
2110 /* if it's not a disk print the vendor and model */
2111 if (!(type
== 0 || type
== 7 || type
== 14)) {
2114 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2115 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2116 strncpy(vendor
, buf
, sizeof(vendor
));
2117 vendor
[sizeof(vendor
) - 1] = '\0';
2118 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2119 while (isspace(*c
) || *c
== '\0')
2123 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2124 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2125 strncpy(model
, buf
, sizeof(model
));
2126 model
[sizeof(model
) - 1] = '\0';
2127 c
= (char *) &model
[sizeof(model
) - 1];
2128 while (isspace(*c
) || *c
== '\0')
2132 if (vendor
[0] && model
[0])
2133 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2135 switch (type
) { /* numbers from hald/linux/device.c */
2136 case 1: sprintf(buf
, "tape"); break;
2137 case 2: sprintf(buf
, "printer"); break;
2138 case 3: sprintf(buf
, "processor"); break;
2140 case 5: sprintf(buf
, "cdrom"); break;
2141 case 6: sprintf(buf
, "scanner"); break;
2142 case 8: sprintf(buf
, "media_changer"); break;
2143 case 9: sprintf(buf
, "comm"); break;
2144 case 12: sprintf(buf
, "raid"); break;
2145 default: sprintf(buf
, "unknown");
2151 /* chop device path to 'host%d' and calculate the port number */
2152 c
= strchr(&path
[hba_len
], '/');
2155 pr_err("%s - invalid path name\n", path
+ hba_len
);
2160 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2161 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2165 *c
= '/'; /* repair the full string */
2166 pr_err("failed to determine port number for %s\n",
2173 /* mark this port as used */
2174 port_mask
&= ~(1 << port
);
2176 /* print out the device information */
2178 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2182 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2184 printf(" Port%d : - disk info unavailable -\n", port
);
2186 fd2devname(fd
, buf
);
2187 printf(" Port%d : %s", port
, buf
);
2188 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
2189 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
2204 for (i
= 0; i
< port_count
; i
++)
2205 if (port_mask
& (1 << i
))
2206 printf(" Port%d : - no device attached -\n", i
);
2212 static int print_vmd_attached_devs(struct sys_dev
*hba
)
2220 if (hba
->type
!= SYS_DEV_VMD
)
2223 /* scroll through /sys/dev/block looking for devices attached to
2226 dir
= opendir("/sys/bus/pci/drivers/nvme");
2230 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2233 /* is 'ent' a device? check that the 'subsystem' link exists and
2234 * that its target matches 'bus'
2236 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
2238 n
= readlink(path
, link
, sizeof(link
));
2239 if (n
< 0 || n
>= (int)sizeof(link
))
2242 c
= strrchr(link
, '/');
2245 if (strncmp("pci", c
+1, strlen("pci")) != 0)
2248 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
2250 rp
= realpath(path
, NULL
);
2254 if (path_attached_to_hba(rp
, hba
->path
)) {
2255 printf(" NVMe under VMD : %s\n", rp
);
2264 static void print_found_intel_controllers(struct sys_dev
*elem
)
2266 for (; elem
; elem
= elem
->next
) {
2267 pr_err("found Intel(R) ");
2268 if (elem
->type
== SYS_DEV_SATA
)
2269 fprintf(stderr
, "SATA ");
2270 else if (elem
->type
== SYS_DEV_SAS
)
2271 fprintf(stderr
, "SAS ");
2272 else if (elem
->type
== SYS_DEV_NVME
)
2273 fprintf(stderr
, "NVMe ");
2275 if (elem
->type
== SYS_DEV_VMD
)
2276 fprintf(stderr
, "VMD domain");
2278 fprintf(stderr
, "RAID controller");
2281 fprintf(stderr
, " at %s", elem
->pci_id
);
2282 fprintf(stderr
, ".\n");
2287 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2294 if ((dir
= opendir(hba_path
)) == NULL
)
2297 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2300 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2301 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2303 if (*port_count
== 0)
2305 else if (host
< host_base
)
2308 if (host
+ 1 > *port_count
+ host_base
)
2309 *port_count
= host
+ 1 - host_base
;
2315 static void print_imsm_capability(const struct imsm_orom
*orom
)
2317 printf(" Platform : Intel(R) ");
2318 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2319 printf("Matrix Storage Manager\n");
2321 printf("Rapid Storage Technology%s\n",
2322 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2323 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2324 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2325 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2326 printf(" RAID Levels :%s%s%s%s%s\n",
2327 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2328 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2329 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2330 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2331 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2332 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2333 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2334 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2335 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2336 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2337 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2338 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2339 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2340 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2341 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2342 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2343 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2344 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2345 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2346 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2347 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2348 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2349 printf(" 2TB volumes :%s supported\n",
2350 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2351 printf(" 2TB disks :%s supported\n",
2352 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2353 printf(" Max Disks : %d\n", orom
->tds
);
2354 printf(" Max Volumes : %d per array, %d per %s\n",
2355 orom
->vpa
, orom
->vphba
,
2356 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2360 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2362 printf("MD_FIRMWARE_TYPE=imsm\n");
2363 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2364 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2365 orom
->hotfix_ver
, orom
->build
);
2366 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2367 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2368 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2369 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2370 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2371 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2372 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2373 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2374 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2375 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2376 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2377 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2378 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2379 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2380 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2381 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2382 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2383 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2384 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2385 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2386 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2387 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2388 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2389 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2390 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2391 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2392 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2393 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2396 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2398 /* There are two components to imsm platform support, the ahci SATA
2399 * controller and the option-rom. To find the SATA controller we
2400 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2401 * controller with the Intel vendor id is present. This approach
2402 * allows mdadm to leverage the kernel's ahci detection logic, with the
2403 * caveat that if ahci.ko is not loaded mdadm will not be able to
2404 * detect platform raid capabilities. The option-rom resides in a
2405 * platform "Adapter ROM". We scan for its signature to retrieve the
2406 * platform capabilities. If raid support is disabled in the BIOS the
2407 * option-rom capability structure will not be available.
2409 struct sys_dev
*list
, *hba
;
2414 if (enumerate_only
) {
2415 if (check_env("IMSM_NO_PLATFORM"))
2417 list
= find_intel_devices();
2420 for (hba
= list
; hba
; hba
= hba
->next
) {
2421 if (find_imsm_capability(hba
)) {
2431 list
= find_intel_devices();
2434 pr_err("no active Intel(R) RAID controller found.\n");
2436 } else if (verbose
> 0)
2437 print_found_intel_controllers(list
);
2439 for (hba
= list
; hba
; hba
= hba
->next
) {
2440 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2442 if (!find_imsm_capability(hba
)) {
2444 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2445 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2446 get_sys_dev_type(hba
->type
));
2452 if (controller_path
&& result
== 1) {
2453 pr_err("no active Intel(R) RAID controller found under %s\n",
2458 const struct orom_entry
*entry
;
2460 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2461 if (entry
->type
== SYS_DEV_VMD
) {
2462 print_imsm_capability(&entry
->orom
);
2463 printf(" 3rd party NVMe :%s supported\n",
2464 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2465 for (hba
= list
; hba
; hba
= hba
->next
) {
2466 if (hba
->type
== SYS_DEV_VMD
) {
2468 printf(" I/O Controller : %s (%s)\n",
2469 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2470 if (print_vmd_attached_devs(hba
)) {
2472 pr_err("failed to get devices attached to VMD domain.\n");
2481 print_imsm_capability(&entry
->orom
);
2482 if (entry
->type
== SYS_DEV_NVME
) {
2483 for (hba
= list
; hba
; hba
= hba
->next
) {
2484 if (hba
->type
== SYS_DEV_NVME
)
2485 printf(" NVMe Device : %s\n", hba
->path
);
2491 struct devid_list
*devid
;
2492 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2493 hba
= device_by_id(devid
->devid
);
2497 printf(" I/O Controller : %s (%s)\n",
2498 hba
->path
, get_sys_dev_type(hba
->type
));
2499 if (hba
->type
== SYS_DEV_SATA
) {
2500 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2501 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2503 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2514 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2516 struct sys_dev
*list
, *hba
;
2519 list
= find_intel_devices();
2522 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2527 for (hba
= list
; hba
; hba
= hba
->next
) {
2528 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2530 if (!find_imsm_capability(hba
) && verbose
> 0) {
2532 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2533 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2539 const struct orom_entry
*entry
;
2541 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2542 if (entry
->type
== SYS_DEV_VMD
) {
2543 for (hba
= list
; hba
; hba
= hba
->next
)
2544 print_imsm_capability_export(&entry
->orom
);
2547 print_imsm_capability_export(&entry
->orom
);
2555 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2557 /* the imsm metadata format does not specify any host
2558 * identification information. We return -1 since we can never
2559 * confirm nor deny whether a given array is "meant" for this
2560 * host. We rely on compare_super and the 'family_num' fields to
2561 * exclude member disks that do not belong, and we rely on
2562 * mdadm.conf to specify the arrays that should be assembled.
2563 * Auto-assembly may still pick up "foreign" arrays.
2569 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2571 /* The uuid returned here is used for:
2572 * uuid to put into bitmap file (Create, Grow)
2573 * uuid for backup header when saving critical section (Grow)
2574 * comparing uuids when re-adding a device into an array
2575 * In these cases the uuid required is that of the data-array,
2576 * not the device-set.
2577 * uuid to recognise same set when adding a missing device back
2578 * to an array. This is a uuid for the device-set.
2580 * For each of these we can make do with a truncated
2581 * or hashed uuid rather than the original, as long as
2583 * In each case the uuid required is that of the data-array,
2584 * not the device-set.
2586 /* imsm does not track uuid's so we synthesis one using sha1 on
2587 * - The signature (Which is constant for all imsm array, but no matter)
2588 * - the orig_family_num of the container
2589 * - the index number of the volume
2590 * - the 'serial' number of the volume.
2591 * Hopefully these are all constant.
2593 struct intel_super
*super
= st
->sb
;
2596 struct sha1_ctx ctx
;
2597 struct imsm_dev
*dev
= NULL
;
2600 /* some mdadm versions failed to set ->orig_family_num, in which
2601 * case fall back to ->family_num. orig_family_num will be
2602 * fixed up with the first metadata update.
2604 family_num
= super
->anchor
->orig_family_num
;
2605 if (family_num
== 0)
2606 family_num
= super
->anchor
->family_num
;
2607 sha1_init_ctx(&ctx
);
2608 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2609 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2610 if (super
->current_vol
>= 0)
2611 dev
= get_imsm_dev(super
, super
->current_vol
);
2613 __u32 vol
= super
->current_vol
;
2614 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2615 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2617 sha1_finish_ctx(&ctx
, buf
);
2618 memcpy(uuid
, buf
, 4*4);
2623 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2625 __u8
*v
= get_imsm_version(mpb
);
2626 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2627 char major
[] = { 0, 0, 0 };
2628 char minor
[] = { 0 ,0, 0 };
2629 char patch
[] = { 0, 0, 0 };
2630 char *ver_parse
[] = { major
, minor
, patch
};
2634 while (*v
!= '\0' && v
< end
) {
2635 if (*v
!= '.' && j
< 2)
2636 ver_parse
[i
][j
++] = *v
;
2644 *m
= strtol(minor
, NULL
, 0);
2645 *p
= strtol(patch
, NULL
, 0);
2649 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2651 /* migr_strip_size when repairing or initializing parity */
2652 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2653 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2655 switch (get_imsm_raid_level(map
)) {
2660 return 128*1024 >> 9;
2664 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2666 /* migr_strip_size when rebuilding a degraded disk, no idea why
2667 * this is different than migr_strip_size_resync(), but it's good
2670 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2671 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2673 switch (get_imsm_raid_level(map
)) {
2676 if (map
->num_members
% map
->num_domains
== 0)
2677 return 128*1024 >> 9;
2681 return max((__u32
) 64*1024 >> 9, chunk
);
2683 return 128*1024 >> 9;
2687 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2689 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2690 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2691 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2692 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2694 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2697 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2699 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2700 int level
= get_imsm_raid_level(lo
);
2702 if (level
== 1 || level
== 10) {
2703 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2705 return hi
->num_domains
;
2707 return num_stripes_per_unit_resync(dev
);
2710 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2712 /* named 'imsm_' because raid0, raid1 and raid10
2713 * counter-intuitively have the same number of data disks
2715 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2717 switch (get_imsm_raid_level(map
)) {
2719 return map
->num_members
;
2723 return map
->num_members
/2;
2725 return map
->num_members
- 1;
2727 dprintf("unsupported raid level\n");
2732 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2734 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2735 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2737 switch(get_imsm_raid_level(map
)) {
2740 return chunk
* map
->num_domains
;
2742 return chunk
* map
->num_members
;
2748 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2750 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2751 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2752 __u32 strip
= block
/ chunk
;
2754 switch (get_imsm_raid_level(map
)) {
2757 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2758 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2760 return vol_stripe
* chunk
+ block
% chunk
;
2762 __u32 stripe
= strip
/ (map
->num_members
- 1);
2764 return stripe
* chunk
+ block
% chunk
;
2771 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2772 struct imsm_dev
*dev
)
2774 /* calculate the conversion factor between per member 'blocks'
2775 * (md/{resync,rebuild}_start) and imsm migration units, return
2776 * 0 for the 'not migrating' and 'unsupported migration' cases
2778 if (!dev
->vol
.migr_state
)
2781 switch (migr_type(dev
)) {
2782 case MIGR_GEN_MIGR
: {
2783 struct migr_record
*migr_rec
= super
->migr_rec
;
2784 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2789 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2790 __u32 stripes_per_unit
;
2791 __u32 blocks_per_unit
;
2800 /* yes, this is really the translation of migr_units to
2801 * per-member blocks in the 'resync' case
2803 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2804 migr_chunk
= migr_strip_blocks_resync(dev
);
2805 disks
= imsm_num_data_members(dev
, MAP_0
);
2806 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2807 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2808 segment
= blocks_per_unit
/ stripe
;
2809 block_rel
= blocks_per_unit
- segment
* stripe
;
2810 parity_depth
= parity_segment_depth(dev
);
2811 block_map
= map_migr_block(dev
, block_rel
);
2812 return block_map
+ parity_depth
* segment
;
2814 case MIGR_REBUILD
: {
2815 __u32 stripes_per_unit
;
2818 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2819 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2820 return migr_chunk
* stripes_per_unit
;
2822 case MIGR_STATE_CHANGE
:
2828 static int imsm_level_to_layout(int level
)
2836 return ALGORITHM_LEFT_ASYMMETRIC
;
2843 /*******************************************************************************
2844 * Function: read_imsm_migr_rec
2845 * Description: Function reads imsm migration record from last sector of disk
2847 * fd : disk descriptor
2848 * super : metadata info
2852 ******************************************************************************/
2853 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2856 unsigned int sector_size
= super
->sector_size
;
2857 unsigned long long dsize
;
2859 get_dev_size(fd
, NULL
, &dsize
);
2860 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
2862 pr_err("Cannot seek to anchor block: %s\n",
2866 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
2867 MIGR_REC_BUF_SECTORS
*sector_size
) !=
2868 MIGR_REC_BUF_SECTORS
*sector_size
) {
2869 pr_err("Cannot read migr record block: %s\n",
2874 if (sector_size
== 4096)
2875 convert_from_4k_imsm_migr_rec(super
);
2881 static struct imsm_dev
*imsm_get_device_during_migration(
2882 struct intel_super
*super
)
2885 struct intel_dev
*dv
;
2887 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2888 if (is_gen_migration(dv
->dev
))
2894 /*******************************************************************************
2895 * Function: load_imsm_migr_rec
2896 * Description: Function reads imsm migration record (it is stored at the last
2899 * super : imsm internal array info
2900 * info : general array info
2904 * -2 : no migration in progress
2905 ******************************************************************************/
2906 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2913 struct imsm_dev
*dev
;
2914 struct imsm_map
*map
;
2917 /* find map under migration */
2918 dev
= imsm_get_device_during_migration(super
);
2919 /* nothing to load,no migration in progress?
2925 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2926 /* read only from one of the first two slots */
2927 if ((sd
->disk
.raid_disk
< 0) ||
2928 (sd
->disk
.raid_disk
> 1))
2931 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2932 fd
= dev_open(nm
, O_RDONLY
);
2938 map
= get_imsm_map(dev
, MAP_0
);
2939 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2940 /* skip spare and failed disks
2944 /* read only from one of the first two slots */
2946 slot
= get_imsm_disk_slot(map
, dl
->index
);
2947 if (map
== NULL
|| slot
> 1 || slot
< 0)
2949 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2950 fd
= dev_open(nm
, O_RDONLY
);
2957 retval
= read_imsm_migr_rec(fd
, super
);
2966 /*******************************************************************************
2967 * function: imsm_create_metadata_checkpoint_update
2968 * Description: It creates update for checkpoint change.
2970 * super : imsm internal array info
2971 * u : pointer to prepared update
2974 * If length is equal to 0, input pointer u contains no update
2975 ******************************************************************************/
2976 static int imsm_create_metadata_checkpoint_update(
2977 struct intel_super
*super
,
2978 struct imsm_update_general_migration_checkpoint
**u
)
2981 int update_memory_size
= 0;
2983 dprintf("(enter)\n");
2989 /* size of all update data without anchor */
2990 update_memory_size
=
2991 sizeof(struct imsm_update_general_migration_checkpoint
);
2993 *u
= xcalloc(1, update_memory_size
);
2995 dprintf("error: cannot get memory\n");
2998 (*u
)->type
= update_general_migration_checkpoint
;
2999 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
3000 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3002 return update_memory_size
;
3005 static void imsm_update_metadata_locally(struct supertype
*st
,
3006 void *buf
, int len
);
3008 /*******************************************************************************
3009 * Function: write_imsm_migr_rec
3010 * Description: Function writes imsm migration record
3011 * (at the last sector of disk)
3013 * super : imsm internal array info
3017 ******************************************************************************/
3018 static int write_imsm_migr_rec(struct supertype
*st
)
3020 struct intel_super
*super
= st
->sb
;
3021 unsigned int sector_size
= super
->sector_size
;
3022 unsigned long long dsize
;
3028 struct imsm_update_general_migration_checkpoint
*u
;
3029 struct imsm_dev
*dev
;
3030 struct imsm_map
*map
;
3032 /* find map under migration */
3033 dev
= imsm_get_device_during_migration(super
);
3034 /* if no migration, write buffer anyway to clear migr_record
3035 * on disk based on first available device
3038 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3039 super
->current_vol
);
3041 map
= get_imsm_map(dev
, MAP_0
);
3043 if (sector_size
== 4096)
3044 convert_to_4k_imsm_migr_rec(super
);
3045 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3048 /* skip failed and spare devices */
3051 /* write to 2 first slots only */
3053 slot
= get_imsm_disk_slot(map
, sd
->index
);
3054 if (map
== NULL
|| slot
> 1 || slot
< 0)
3057 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3058 fd
= dev_open(nm
, O_RDWR
);
3061 get_dev_size(fd
, NULL
, &dsize
);
3062 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3064 pr_err("Cannot seek to anchor block: %s\n",
3068 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3069 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3070 MIGR_REC_BUF_SECTORS
*sector_size
) {
3071 pr_err("Cannot write migr record block: %s\n",
3078 if (sector_size
== 4096)
3079 convert_from_4k_imsm_migr_rec(super
);
3080 /* update checkpoint information in metadata */
3081 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3083 dprintf("imsm: Cannot prepare update\n");
3086 /* update metadata locally */
3087 imsm_update_metadata_locally(st
, u
, len
);
3088 /* and possibly remotely */
3089 if (st
->update_tail
) {
3090 append_metadata_update(st
, u
, len
);
3091 /* during reshape we do all work inside metadata handler
3092 * manage_reshape(), so metadata update has to be triggered
3095 flush_metadata_updates(st
);
3096 st
->update_tail
= &st
->updates
;
3106 #endif /* MDASSEMBLE */
3108 /* spare/missing disks activations are not allowe when
3109 * array/container performs reshape operation, because
3110 * all arrays in container works on the same disks set
3112 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3115 struct intel_dev
*i_dev
;
3116 struct imsm_dev
*dev
;
3118 /* check whole container
3120 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3122 if (is_gen_migration(dev
)) {
3123 /* No repair during any migration in container
3131 static unsigned long long imsm_component_size_aligment_check(int level
,
3133 unsigned int sector_size
,
3134 unsigned long long component_size
)
3136 unsigned int component_size_alligment
;
3138 /* check component size aligment
3140 component_size_alligment
= component_size
% (chunk_size
/sector_size
);
3142 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
3143 level
, chunk_size
, component_size
,
3144 component_size_alligment
);
3146 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
3147 dprintf("imsm: reported component size alligned from %llu ",
3149 component_size
-= component_size_alligment
;
3150 dprintf_cont("to %llu (%i).\n",
3151 component_size
, component_size_alligment
);
3154 return component_size
;
3157 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3159 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3160 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3162 return pba_of_lba0(map
) +
3163 (num_data_stripes(map
) * map
->blocks_per_strip
);
3166 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3168 struct intel_super
*super
= st
->sb
;
3169 struct migr_record
*migr_rec
= super
->migr_rec
;
3170 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3171 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3172 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3173 struct imsm_map
*map_to_analyse
= map
;
3175 int map_disks
= info
->array
.raid_disks
;
3177 memset(info
, 0, sizeof(*info
));
3179 map_to_analyse
= prev_map
;
3181 dl
= super
->current_disk
;
3183 info
->container_member
= super
->current_vol
;
3184 info
->array
.raid_disks
= map
->num_members
;
3185 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3186 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3187 info
->array
.md_minor
= -1;
3188 info
->array
.ctime
= 0;
3189 info
->array
.utime
= 0;
3190 info
->array
.chunk_size
=
3191 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3192 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3193 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
3194 info
->custom_array_size
<<= 32;
3195 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
3196 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3198 if (is_gen_migration(dev
)) {
3199 info
->reshape_active
= 1;
3200 info
->new_level
= get_imsm_raid_level(map
);
3201 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3202 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3203 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3204 if (info
->delta_disks
) {
3205 /* this needs to be applied to every array
3208 info
->reshape_active
= CONTAINER_RESHAPE
;
3210 /* We shape information that we give to md might have to be
3211 * modify to cope with md's requirement for reshaping arrays.
3212 * For example, when reshaping a RAID0, md requires it to be
3213 * presented as a degraded RAID4.
3214 * Also if a RAID0 is migrating to a RAID5 we need to specify
3215 * the array as already being RAID5, but the 'before' layout
3216 * is a RAID4-like layout.
3218 switch (info
->array
.level
) {
3220 switch(info
->new_level
) {
3222 /* conversion is happening as RAID4 */
3223 info
->array
.level
= 4;
3224 info
->array
.raid_disks
+= 1;
3227 /* conversion is happening as RAID5 */
3228 info
->array
.level
= 5;
3229 info
->array
.layout
= ALGORITHM_PARITY_N
;
3230 info
->delta_disks
-= 1;
3233 /* FIXME error message */
3234 info
->array
.level
= UnSet
;
3240 info
->new_level
= UnSet
;
3241 info
->new_layout
= UnSet
;
3242 info
->new_chunk
= info
->array
.chunk_size
;
3243 info
->delta_disks
= 0;
3247 info
->disk
.major
= dl
->major
;
3248 info
->disk
.minor
= dl
->minor
;
3249 info
->disk
.number
= dl
->index
;
3250 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3254 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3256 if (info
->array
.level
== 5) {
3257 info
->component_size
= num_data_stripes(map_to_analyse
) *
3258 map_to_analyse
->blocks_per_strip
;
3260 info
->component_size
= blocks_per_member(map_to_analyse
);
3263 info
->component_size
= imsm_component_size_aligment_check(
3265 info
->array
.chunk_size
,
3267 info
->component_size
);
3268 info
->bb
.supported
= 1;
3270 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3271 info
->recovery_start
= MaxSector
;
3273 if (info
->array
.level
== 5 && dev
->rwh_policy
== RWH_DISTRIBUTED
) {
3274 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3275 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3276 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
) >> 9;
3277 } else if (info
->array
.level
<= 0) {
3278 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3280 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3283 info
->reshape_progress
= 0;
3284 info
->resync_start
= MaxSector
;
3285 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3286 !(info
->array
.state
& 1)) &&
3287 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3288 info
->resync_start
= 0;
3290 if (dev
->vol
.migr_state
) {
3291 switch (migr_type(dev
)) {
3294 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3296 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3298 info
->resync_start
= blocks_per_unit
* units
;
3301 case MIGR_GEN_MIGR
: {
3302 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3304 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
3305 unsigned long long array_blocks
;
3308 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3310 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
3311 (super
->migr_rec
->rec_status
==
3312 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3315 info
->reshape_progress
= blocks_per_unit
* units
;
3317 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3318 (unsigned long long)units
,
3319 (unsigned long long)blocks_per_unit
,
3320 info
->reshape_progress
);
3322 used_disks
= imsm_num_data_members(dev
, MAP_1
);
3323 if (used_disks
> 0) {
3324 array_blocks
= blocks_per_member(map
) *
3326 /* round array size down to closest MB
3328 info
->custom_array_size
= (array_blocks
3329 >> SECT_PER_MB_SHIFT
)
3330 << SECT_PER_MB_SHIFT
;
3334 /* we could emulate the checkpointing of
3335 * 'sync_action=check' migrations, but for now
3336 * we just immediately complete them
3339 /* this is handled by container_content_imsm() */
3340 case MIGR_STATE_CHANGE
:
3341 /* FIXME handle other migrations */
3343 /* we are not dirty, so... */
3344 info
->resync_start
= MaxSector
;
3348 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3349 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3351 info
->array
.major_version
= -1;
3352 info
->array
.minor_version
= -2;
3353 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3354 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3355 uuid_from_super_imsm(st
, info
->uuid
);
3359 for (i
=0; i
<map_disks
; i
++) {
3361 if (i
< info
->array
.raid_disks
) {
3362 struct imsm_disk
*dsk
;
3363 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3364 dsk
= get_imsm_disk(super
, j
);
3365 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3372 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3373 int failed
, int look_in_map
);
3375 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3379 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3381 if (is_gen_migration(dev
)) {
3384 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3386 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3387 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3388 if (map2
->map_state
!= map_state
) {
3389 map2
->map_state
= map_state
;
3390 super
->updates_pending
++;
3396 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3400 for (d
= super
->missing
; d
; d
= d
->next
)
3401 if (d
->index
== index
)
3406 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3408 struct intel_super
*super
= st
->sb
;
3409 struct imsm_disk
*disk
;
3410 int map_disks
= info
->array
.raid_disks
;
3411 int max_enough
= -1;
3413 struct imsm_super
*mpb
;
3415 if (super
->current_vol
>= 0) {
3416 getinfo_super_imsm_volume(st
, info
, map
);
3419 memset(info
, 0, sizeof(*info
));
3421 /* Set raid_disks to zero so that Assemble will always pull in valid
3424 info
->array
.raid_disks
= 0;
3425 info
->array
.level
= LEVEL_CONTAINER
;
3426 info
->array
.layout
= 0;
3427 info
->array
.md_minor
= -1;
3428 info
->array
.ctime
= 0; /* N/A for imsm */
3429 info
->array
.utime
= 0;
3430 info
->array
.chunk_size
= 0;
3432 info
->disk
.major
= 0;
3433 info
->disk
.minor
= 0;
3434 info
->disk
.raid_disk
= -1;
3435 info
->reshape_active
= 0;
3436 info
->array
.major_version
= -1;
3437 info
->array
.minor_version
= -2;
3438 strcpy(info
->text_version
, "imsm");
3439 info
->safe_mode_delay
= 0;
3440 info
->disk
.number
= -1;
3441 info
->disk
.state
= 0;
3443 info
->recovery_start
= MaxSector
;
3444 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3445 info
->bb
.supported
= 1;
3447 /* do we have the all the insync disks that we expect? */
3448 mpb
= super
->anchor
;
3449 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3451 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3452 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3453 int failed
, enough
, j
, missing
= 0;
3454 struct imsm_map
*map
;
3457 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3458 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3459 map
= get_imsm_map(dev
, MAP_0
);
3461 /* any newly missing disks?
3462 * (catches single-degraded vs double-degraded)
3464 for (j
= 0; j
< map
->num_members
; j
++) {
3465 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3466 __u32 idx
= ord_to_idx(ord
);
3468 if (!(ord
& IMSM_ORD_REBUILD
) &&
3469 get_imsm_missing(super
, idx
)) {
3475 if (state
== IMSM_T_STATE_FAILED
)
3477 else if (state
== IMSM_T_STATE_DEGRADED
&&
3478 (state
!= map
->map_state
|| missing
))
3480 else /* we're normal, or already degraded */
3482 if (is_gen_migration(dev
) && missing
) {
3483 /* during general migration we need all disks
3484 * that process is running on.
3485 * No new missing disk is allowed.
3489 /* no more checks necessary
3493 /* in the missing/failed disk case check to see
3494 * if at least one array is runnable
3496 max_enough
= max(max_enough
, enough
);
3498 dprintf("enough: %d\n", max_enough
);
3499 info
->container_enough
= max_enough
;
3502 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3504 disk
= &super
->disks
->disk
;
3505 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3506 info
->component_size
= reserved
;
3507 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3508 /* we don't change info->disk.raid_disk here because
3509 * this state will be finalized in mdmon after we have
3510 * found the 'most fresh' version of the metadata
3512 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3513 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3514 0 : (1 << MD_DISK_SYNC
);
3517 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3518 * ->compare_super may have updated the 'num_raid_devs' field for spares
3520 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3521 uuid_from_super_imsm(st
, info
->uuid
);
3523 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3525 /* I don't know how to compute 'map' on imsm, so use safe default */
3528 for (i
= 0; i
< map_disks
; i
++)
3534 /* allocates memory and fills disk in mdinfo structure
3535 * for each disk in array */
3536 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3538 struct mdinfo
*mddev
;
3539 struct intel_super
*super
= st
->sb
;
3540 struct imsm_disk
*disk
;
3543 if (!super
|| !super
->disks
)
3546 mddev
= xcalloc(1, sizeof(*mddev
));
3550 tmp
= xcalloc(1, sizeof(*tmp
));
3552 tmp
->next
= mddev
->devs
;
3554 tmp
->disk
.number
= count
++;
3555 tmp
->disk
.major
= dl
->major
;
3556 tmp
->disk
.minor
= dl
->minor
;
3557 tmp
->disk
.state
= is_configured(disk
) ?
3558 (1 << MD_DISK_ACTIVE
) : 0;
3559 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3560 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3561 tmp
->disk
.raid_disk
= -1;
3567 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3568 char *update
, char *devname
, int verbose
,
3569 int uuid_set
, char *homehost
)
3571 /* For 'assemble' and 'force' we need to return non-zero if any
3572 * change was made. For others, the return value is ignored.
3573 * Update options are:
3574 * force-one : This device looks a bit old but needs to be included,
3575 * update age info appropriately.
3576 * assemble: clear any 'faulty' flag to allow this device to
3578 * force-array: Array is degraded but being forced, mark it clean
3579 * if that will be needed to assemble it.
3581 * newdev: not used ????
3582 * grow: Array has gained a new device - this is currently for
3584 * resync: mark as dirty so a resync will happen.
3585 * name: update the name - preserving the homehost
3586 * uuid: Change the uuid of the array to match watch is given
3588 * Following are not relevant for this imsm:
3589 * sparc2.2 : update from old dodgey metadata
3590 * super-minor: change the preferred_minor number
3591 * summaries: update redundant counters.
3592 * homehost: update the recorded homehost
3593 * _reshape_progress: record new reshape_progress position.
3596 struct intel_super
*super
= st
->sb
;
3597 struct imsm_super
*mpb
;
3599 /* we can only update container info */
3600 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3603 mpb
= super
->anchor
;
3605 if (strcmp(update
, "uuid") == 0) {
3606 /* We take this to mean that the family_num should be updated.
3607 * However that is much smaller than the uuid so we cannot really
3608 * allow an explicit uuid to be given. And it is hard to reliably
3610 * So if !uuid_set we know the current uuid is random and just used
3611 * the first 'int' and copy it to the other 3 positions.
3612 * Otherwise we require the 4 'int's to be the same as would be the
3613 * case if we are using a random uuid. So an explicit uuid will be
3614 * accepted as long as all for ints are the same... which shouldn't hurt
3617 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3620 if (info
->uuid
[0] != info
->uuid
[1] ||
3621 info
->uuid
[1] != info
->uuid
[2] ||
3622 info
->uuid
[2] != info
->uuid
[3])
3628 mpb
->orig_family_num
= info
->uuid
[0];
3629 } else if (strcmp(update
, "assemble") == 0)
3634 /* successful update? recompute checksum */
3636 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3641 static size_t disks_to_mpb_size(int disks
)
3645 size
= sizeof(struct imsm_super
);
3646 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3647 size
+= 2 * sizeof(struct imsm_dev
);
3648 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3649 size
+= (4 - 2) * sizeof(struct imsm_map
);
3650 /* 4 possible disk_ord_tbl's */
3651 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3652 /* maximum bbm log */
3653 size
+= sizeof(struct bbm_log
);
3658 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3659 unsigned long long data_offset
)
3661 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3664 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3667 static void free_devlist(struct intel_super
*super
)
3669 struct intel_dev
*dv
;
3671 while (super
->devlist
) {
3672 dv
= super
->devlist
->next
;
3673 free(super
->devlist
->dev
);
3674 free(super
->devlist
);
3675 super
->devlist
= dv
;
3679 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3681 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3684 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3688 * 0 same, or first was empty, and second was copied
3689 * 1 second had wrong number
3691 * 3 wrong other info
3693 struct intel_super
*first
= st
->sb
;
3694 struct intel_super
*sec
= tst
->sb
;
3701 /* in platform dependent environment test if the disks
3702 * use the same Intel hba
3703 * If not on Intel hba at all, allow anything.
3705 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3706 if (first
->hba
->type
!= sec
->hba
->type
) {
3708 "HBAs of devices do not match %s != %s\n",
3709 get_sys_dev_type(first
->hba
->type
),
3710 get_sys_dev_type(sec
->hba
->type
));
3713 if (first
->orom
!= sec
->orom
) {
3715 "HBAs of devices do not match %s != %s\n",
3716 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3721 /* if an anchor does not have num_raid_devs set then it is a free
3724 if (first
->anchor
->num_raid_devs
> 0 &&
3725 sec
->anchor
->num_raid_devs
> 0) {
3726 /* Determine if these disks might ever have been
3727 * related. Further disambiguation can only take place
3728 * in load_super_imsm_all
3730 __u32 first_family
= first
->anchor
->orig_family_num
;
3731 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3733 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3734 MAX_SIGNATURE_LENGTH
) != 0)
3737 if (first_family
== 0)
3738 first_family
= first
->anchor
->family_num
;
3739 if (sec_family
== 0)
3740 sec_family
= sec
->anchor
->family_num
;
3742 if (first_family
!= sec_family
)
3747 /* if 'first' is a spare promote it to a populated mpb with sec's
3750 if (first
->anchor
->num_raid_devs
== 0 &&
3751 sec
->anchor
->num_raid_devs
> 0) {
3753 struct intel_dev
*dv
;
3754 struct imsm_dev
*dev
;
3756 /* we need to copy raid device info from sec if an allocation
3757 * fails here we don't associate the spare
3759 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3760 dv
= xmalloc(sizeof(*dv
));
3761 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3764 dv
->next
= first
->devlist
;
3765 first
->devlist
= dv
;
3767 if (i
< sec
->anchor
->num_raid_devs
) {
3768 /* allocation failure */
3769 free_devlist(first
);
3770 pr_err("imsm: failed to associate spare\n");
3773 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3774 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3775 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3776 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3777 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3778 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3784 static void fd2devname(int fd
, char *name
)
3788 char dname
[PATH_MAX
];
3793 if (fstat(fd
, &st
) != 0)
3795 sprintf(path
, "/sys/dev/block/%d:%d",
3796 major(st
.st_rdev
), minor(st
.st_rdev
));
3798 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3803 nm
= strrchr(dname
, '/');
3806 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3810 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3813 char *name
= fd2kname(fd
);
3818 if (strncmp(name
, "nvme", 4) != 0)
3821 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3823 return load_sys(path
, buf
, buf_len
);
3826 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3828 static int imsm_read_serial(int fd
, char *devname
,
3829 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3838 memset(buf
, 0, sizeof(buf
));
3840 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3843 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3845 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3846 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3847 fd2devname(fd
, (char *) serial
);
3853 pr_err("Failed to retrieve serial for %s\n",
3858 /* trim all whitespace and non-printable characters and convert
3861 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
3864 /* ':' is reserved for use in placeholder serial
3865 * numbers for missing disks
3876 /* truncate leading characters */
3877 if (len
> MAX_RAID_SERIAL_LEN
) {
3878 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3879 len
= MAX_RAID_SERIAL_LEN
;
3882 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3883 memcpy(serial
, dest
, len
);
3888 static int serialcmp(__u8
*s1
, __u8
*s2
)
3890 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3893 static void serialcpy(__u8
*dest
, __u8
*src
)
3895 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3898 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3902 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3903 if (serialcmp(dl
->serial
, serial
) == 0)
3909 static struct imsm_disk
*
3910 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3914 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3915 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3917 if (serialcmp(disk
->serial
, serial
) == 0) {
3928 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3930 struct imsm_disk
*disk
;
3935 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3937 rv
= imsm_read_serial(fd
, devname
, serial
);
3942 dl
= xcalloc(1, sizeof(*dl
));
3945 dl
->major
= major(stb
.st_rdev
);
3946 dl
->minor
= minor(stb
.st_rdev
);
3947 dl
->next
= super
->disks
;
3948 dl
->fd
= keep_fd
? fd
: -1;
3949 assert(super
->disks
== NULL
);
3951 serialcpy(dl
->serial
, serial
);
3954 fd2devname(fd
, name
);
3956 dl
->devname
= xstrdup(devname
);
3958 dl
->devname
= xstrdup(name
);
3960 /* look up this disk's index in the current anchor */
3961 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3964 /* only set index on disks that are a member of a
3965 * populated contianer, i.e. one with raid_devs
3967 if (is_failed(&dl
->disk
))
3969 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
3977 /* When migrating map0 contains the 'destination' state while map1
3978 * contains the current state. When not migrating map0 contains the
3979 * current state. This routine assumes that map[0].map_state is set to
3980 * the current array state before being called.
3982 * Migration is indicated by one of the following states
3983 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3984 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3985 * map1state=unitialized)
3986 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3988 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3989 * map1state=degraded)
3990 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3993 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3994 __u8 to_state
, int migr_type
)
3996 struct imsm_map
*dest
;
3997 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3999 dev
->vol
.migr_state
= 1;
4000 set_migr_type(dev
, migr_type
);
4001 dev
->vol
.curr_migr_unit
= 0;
4002 dest
= get_imsm_map(dev
, MAP_1
);
4004 /* duplicate and then set the target end state in map[0] */
4005 memcpy(dest
, src
, sizeof_imsm_map(src
));
4006 if (migr_type
== MIGR_REBUILD
|| migr_type
== MIGR_GEN_MIGR
) {
4010 for (i
= 0; i
< src
->num_members
; i
++) {
4011 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4012 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4016 if (migr_type
== MIGR_GEN_MIGR
)
4017 /* Clear migration record */
4018 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4020 src
->map_state
= to_state
;
4023 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4026 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4027 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4031 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4032 * completed in the last migration.
4034 * FIXME add support for raid-level-migration
4036 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4037 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4038 /* when final map state is other than expected
4039 * merge maps (not for migration)
4043 for (i
= 0; i
< prev
->num_members
; i
++)
4044 for (j
= 0; j
< map
->num_members
; j
++)
4045 /* during online capacity expansion
4046 * disks position can be changed
4047 * if takeover is used
4049 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4050 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4051 map
->disk_ord_tbl
[j
] |=
4052 prev
->disk_ord_tbl
[i
];
4055 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4056 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4059 dev
->vol
.migr_state
= 0;
4060 set_migr_type(dev
, 0);
4061 dev
->vol
.curr_migr_unit
= 0;
4062 map
->map_state
= map_state
;
4066 static int parse_raid_devices(struct intel_super
*super
)
4069 struct imsm_dev
*dev_new
;
4070 size_t len
, len_migr
;
4072 size_t space_needed
= 0;
4073 struct imsm_super
*mpb
= super
->anchor
;
4075 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4076 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4077 struct intel_dev
*dv
;
4079 len
= sizeof_imsm_dev(dev_iter
, 0);
4080 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4082 space_needed
+= len_migr
- len
;
4084 dv
= xmalloc(sizeof(*dv
));
4085 if (max_len
< len_migr
)
4087 if (max_len
> len_migr
)
4088 space_needed
+= max_len
- len_migr
;
4089 dev_new
= xmalloc(max_len
);
4090 imsm_copy_dev(dev_new
, dev_iter
);
4093 dv
->next
= super
->devlist
;
4094 super
->devlist
= dv
;
4097 /* ensure that super->buf is large enough when all raid devices
4100 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4103 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4104 super
->sector_size
);
4105 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4108 memcpy(buf
, super
->buf
, super
->len
);
4109 memset(buf
+ super
->len
, 0, len
- super
->len
);
4115 super
->extra_space
+= space_needed
;
4120 /*******************************************************************************
4121 * Function: check_mpb_migr_compatibility
4122 * Description: Function checks for unsupported migration features:
4123 * - migration optimization area (pba_of_lba0)
4124 * - descending reshape (ascending_migr)
4126 * super : imsm metadata information
4128 * 0 : migration is compatible
4129 * -1 : migration is not compatible
4130 ******************************************************************************/
4131 int check_mpb_migr_compatibility(struct intel_super
*super
)
4133 struct imsm_map
*map0
, *map1
;
4134 struct migr_record
*migr_rec
= super
->migr_rec
;
4137 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4138 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4141 dev_iter
->vol
.migr_state
== 1 &&
4142 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4143 /* This device is migrating */
4144 map0
= get_imsm_map(dev_iter
, MAP_0
);
4145 map1
= get_imsm_map(dev_iter
, MAP_1
);
4146 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4147 /* migration optimization area was used */
4149 if (migr_rec
->ascending_migr
== 0
4150 && migr_rec
->dest_depth_per_unit
> 0)
4151 /* descending reshape not supported yet */
4158 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4160 /* load_imsm_mpb - read matrix metadata
4161 * allocates super->mpb to be freed by free_imsm
4163 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4165 unsigned long long dsize
;
4166 unsigned long long sectors
;
4167 unsigned int sector_size
= super
->sector_size
;
4169 struct imsm_super
*anchor
;
4172 get_dev_size(fd
, NULL
, &dsize
);
4173 if (dsize
< 2*sector_size
) {
4175 pr_err("%s: device to small for imsm\n",
4180 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4182 pr_err("Cannot seek to anchor block on %s: %s\n",
4183 devname
, strerror(errno
));
4187 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4189 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4192 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4194 pr_err("Cannot read anchor block on %s: %s\n",
4195 devname
, strerror(errno
));
4200 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4202 pr_err("no IMSM anchor on %s\n", devname
);
4207 __free_imsm(super
, 0);
4208 /* reload capability and hba */
4210 /* capability and hba must be updated with new super allocation */
4211 find_intel_hba_capability(fd
, super
, devname
);
4212 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4213 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4215 pr_err("unable to allocate %zu byte mpb buffer\n",
4220 memcpy(super
->buf
, anchor
, sector_size
);
4222 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4225 if (posix_memalign(&super
->migr_rec_buf
, sector_size
,
4226 MIGR_REC_BUF_SECTORS
*sector_size
) != 0) {
4227 pr_err("could not allocate migr_rec buffer\n");
4231 super
->clean_migration_record_by_mdmon
= 0;
4234 check_sum
= __gen_imsm_checksum(super
->anchor
);
4235 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4237 pr_err("IMSM checksum %x != %x on %s\n",
4239 __le32_to_cpu(super
->anchor
->check_sum
),
4247 /* read the extended mpb */
4248 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4250 pr_err("Cannot seek to extended mpb on %s: %s\n",
4251 devname
, strerror(errno
));
4255 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4256 super
->len
- sector_size
) != super
->len
- sector_size
) {
4258 pr_err("Cannot read extended mpb on %s: %s\n",
4259 devname
, strerror(errno
));
4263 check_sum
= __gen_imsm_checksum(super
->anchor
);
4264 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4266 pr_err("IMSM checksum %x != %x on %s\n",
4267 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4275 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4277 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4278 static void clear_hi(struct intel_super
*super
)
4280 struct imsm_super
*mpb
= super
->anchor
;
4282 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4284 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4285 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4286 disk
->total_blocks_hi
= 0;
4288 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4289 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4292 for (n
= 0; n
< 2; ++n
) {
4293 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4296 map
->pba_of_lba0_hi
= 0;
4297 map
->blocks_per_member_hi
= 0;
4298 map
->num_data_stripes_hi
= 0;
4304 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4308 err
= load_imsm_mpb(fd
, super
, devname
);
4311 if (super
->sector_size
== 4096)
4312 convert_from_4k(super
);
4313 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4316 err
= parse_raid_devices(super
);
4319 err
= load_bbm_log(super
);
4324 static void __free_imsm_disk(struct dl
*d
)
4336 static void free_imsm_disks(struct intel_super
*super
)
4340 while (super
->disks
) {
4342 super
->disks
= d
->next
;
4343 __free_imsm_disk(d
);
4345 while (super
->disk_mgmt_list
) {
4346 d
= super
->disk_mgmt_list
;
4347 super
->disk_mgmt_list
= d
->next
;
4348 __free_imsm_disk(d
);
4350 while (super
->missing
) {
4352 super
->missing
= d
->next
;
4353 __free_imsm_disk(d
);
4358 /* free all the pieces hanging off of a super pointer */
4359 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4361 struct intel_hba
*elem
, *next
;
4367 /* unlink capability description */
4369 if (super
->migr_rec_buf
) {
4370 free(super
->migr_rec_buf
);
4371 super
->migr_rec_buf
= NULL
;
4374 free_imsm_disks(super
);
4375 free_devlist(super
);
4379 free((void *)elem
->path
);
4385 free(super
->bbm_log
);
4389 static void free_imsm(struct intel_super
*super
)
4391 __free_imsm(super
, 1);
4392 free(super
->bb
.entries
);
4396 static void free_super_imsm(struct supertype
*st
)
4398 struct intel_super
*super
= st
->sb
;
4407 static struct intel_super
*alloc_super(void)
4409 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4411 super
->current_vol
= -1;
4412 super
->create_offset
= ~((unsigned long long) 0);
4414 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4415 sizeof(struct md_bb_entry
));
4416 if (!super
->bb
.entries
) {
4425 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4427 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4429 struct sys_dev
*hba_name
;
4432 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4437 hba_name
= find_disk_attached_hba(fd
, NULL
);
4440 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4444 rv
= attach_hba_to_super(super
, hba_name
);
4447 struct intel_hba
*hba
= super
->hba
;
4449 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4450 " but the container is assigned to Intel(R) %s %s (",
4452 get_sys_dev_type(hba_name
->type
),
4453 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4454 hba_name
->pci_id
? : "Err!",
4455 get_sys_dev_type(super
->hba
->type
),
4456 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4459 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4461 fprintf(stderr
, ", ");
4464 fprintf(stderr
, ").\n"
4465 " Mixing devices attached to different %s is not allowed.\n",
4466 hba_name
->type
== SYS_DEV_VMD
? "VMD domains" : "controllers");
4470 super
->orom
= find_imsm_capability(hba_name
);
4477 /* find_missing - helper routine for load_super_imsm_all that identifies
4478 * disks that have disappeared from the system. This routine relies on
4479 * the mpb being uptodate, which it is at load time.
4481 static int find_missing(struct intel_super
*super
)
4484 struct imsm_super
*mpb
= super
->anchor
;
4486 struct imsm_disk
*disk
;
4488 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4489 disk
= __get_imsm_disk(mpb
, i
);
4490 dl
= serial_to_dl(disk
->serial
, super
);
4494 dl
= xmalloc(sizeof(*dl
));
4498 dl
->devname
= xstrdup("missing");
4500 serialcpy(dl
->serial
, disk
->serial
);
4503 dl
->next
= super
->missing
;
4504 super
->missing
= dl
;
4511 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4513 struct intel_disk
*idisk
= disk_list
;
4516 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4518 idisk
= idisk
->next
;
4524 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4525 struct intel_super
*super
,
4526 struct intel_disk
**disk_list
)
4528 struct imsm_disk
*d
= &super
->disks
->disk
;
4529 struct imsm_super
*mpb
= super
->anchor
;
4532 for (i
= 0; i
< tbl_size
; i
++) {
4533 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4534 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4536 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4537 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4538 dprintf("mpb from %d:%d matches %d:%d\n",
4539 super
->disks
->major
,
4540 super
->disks
->minor
,
4541 table
[i
]->disks
->major
,
4542 table
[i
]->disks
->minor
);
4546 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4547 is_configured(d
) == is_configured(tbl_d
)) &&
4548 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4549 /* current version of the mpb is a
4550 * better candidate than the one in
4551 * super_table, but copy over "cross
4552 * generational" status
4554 struct intel_disk
*idisk
;
4556 dprintf("mpb from %d:%d replaces %d:%d\n",
4557 super
->disks
->major
,
4558 super
->disks
->minor
,
4559 table
[i
]->disks
->major
,
4560 table
[i
]->disks
->minor
);
4562 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4563 if (idisk
&& is_failed(&idisk
->disk
))
4564 tbl_d
->status
|= FAILED_DISK
;
4567 struct intel_disk
*idisk
;
4568 struct imsm_disk
*disk
;
4570 /* tbl_mpb is more up to date, but copy
4571 * over cross generational status before
4574 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4575 if (disk
&& is_failed(disk
))
4576 d
->status
|= FAILED_DISK
;
4578 idisk
= disk_list_get(d
->serial
, *disk_list
);
4581 if (disk
&& is_configured(disk
))
4582 idisk
->disk
.status
|= CONFIGURED_DISK
;
4585 dprintf("mpb from %d:%d prefer %d:%d\n",
4586 super
->disks
->major
,
4587 super
->disks
->minor
,
4588 table
[i
]->disks
->major
,
4589 table
[i
]->disks
->minor
);
4597 table
[tbl_size
++] = super
;
4601 /* update/extend the merged list of imsm_disk records */
4602 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4603 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4604 struct intel_disk
*idisk
;
4606 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4608 idisk
->disk
.status
|= disk
->status
;
4609 if (is_configured(&idisk
->disk
) ||
4610 is_failed(&idisk
->disk
))
4611 idisk
->disk
.status
&= ~(SPARE_DISK
);
4613 idisk
= xcalloc(1, sizeof(*idisk
));
4614 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4615 idisk
->disk
= *disk
;
4616 idisk
->next
= *disk_list
;
4620 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4627 static struct intel_super
*
4628 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4631 struct imsm_super
*mpb
= super
->anchor
;
4635 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4636 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4637 struct intel_disk
*idisk
;
4639 idisk
= disk_list_get(disk
->serial
, disk_list
);
4641 if (idisk
->owner
== owner
||
4642 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4645 dprintf("'%.16s' owner %d != %d\n",
4646 disk
->serial
, idisk
->owner
,
4649 dprintf("unknown disk %x [%d]: %.16s\n",
4650 __le32_to_cpu(mpb
->family_num
), i
,
4656 if (ok_count
== mpb
->num_disks
)
4661 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4663 struct intel_super
*s
;
4665 for (s
= super_list
; s
; s
= s
->next
) {
4666 if (family_num
!= s
->anchor
->family_num
)
4668 pr_err("Conflict, offlining family %#x on '%s'\n",
4669 __le32_to_cpu(family_num
), s
->disks
->devname
);
4673 static struct intel_super
*
4674 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4676 struct intel_super
*super_table
[len
];
4677 struct intel_disk
*disk_list
= NULL
;
4678 struct intel_super
*champion
, *spare
;
4679 struct intel_super
*s
, **del
;
4684 memset(super_table
, 0, sizeof(super_table
));
4685 for (s
= *super_list
; s
; s
= s
->next
)
4686 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4688 for (i
= 0; i
< tbl_size
; i
++) {
4689 struct imsm_disk
*d
;
4690 struct intel_disk
*idisk
;
4691 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4694 d
= &s
->disks
->disk
;
4696 /* 'd' must appear in merged disk list for its
4697 * configuration to be valid
4699 idisk
= disk_list_get(d
->serial
, disk_list
);
4700 if (idisk
&& idisk
->owner
== i
)
4701 s
= validate_members(s
, disk_list
, i
);
4706 dprintf("marking family: %#x from %d:%d offline\n",
4708 super_table
[i
]->disks
->major
,
4709 super_table
[i
]->disks
->minor
);
4713 /* This is where the mdadm implementation differs from the Windows
4714 * driver which has no strict concept of a container. We can only
4715 * assemble one family from a container, so when returning a prodigal
4716 * array member to this system the code will not be able to disambiguate
4717 * the container contents that should be assembled ("foreign" versus
4718 * "local"). It requires user intervention to set the orig_family_num
4719 * to a new value to establish a new container. The Windows driver in
4720 * this situation fixes up the volume name in place and manages the
4721 * foreign array as an independent entity.
4726 for (i
= 0; i
< tbl_size
; i
++) {
4727 struct intel_super
*tbl_ent
= super_table
[i
];
4733 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4738 if (s
&& !is_spare
) {
4739 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4741 } else if (!s
&& !is_spare
)
4754 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4755 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4757 /* collect all dl's onto 'champion', and update them to
4758 * champion's version of the status
4760 for (s
= *super_list
; s
; s
= s
->next
) {
4761 struct imsm_super
*mpb
= champion
->anchor
;
4762 struct dl
*dl
= s
->disks
;
4767 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4769 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4770 struct imsm_disk
*disk
;
4772 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4775 /* only set index on disks that are a member of
4776 * a populated contianer, i.e. one with
4779 if (is_failed(&dl
->disk
))
4781 else if (is_spare(&dl
->disk
))
4787 if (i
>= mpb
->num_disks
) {
4788 struct intel_disk
*idisk
;
4790 idisk
= disk_list_get(dl
->serial
, disk_list
);
4791 if (idisk
&& is_spare(&idisk
->disk
) &&
4792 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4800 dl
->next
= champion
->disks
;
4801 champion
->disks
= dl
;
4805 /* delete 'champion' from super_list */
4806 for (del
= super_list
; *del
; ) {
4807 if (*del
== champion
) {
4808 *del
= (*del
)->next
;
4811 del
= &(*del
)->next
;
4813 champion
->next
= NULL
;
4817 struct intel_disk
*idisk
= disk_list
;
4819 disk_list
= disk_list
->next
;
4827 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4828 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4829 int major
, int minor
, int keep_fd
);
4831 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4832 int *max
, int keep_fd
);
4834 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4835 char *devname
, struct md_list
*devlist
,
4838 struct intel_super
*super_list
= NULL
;
4839 struct intel_super
*super
= NULL
;
4844 /* 'fd' is an opened container */
4845 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4847 /* get super block from devlist devices */
4848 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4851 /* all mpbs enter, maybe one leaves */
4852 super
= imsm_thunderdome(&super_list
, i
);
4858 if (find_missing(super
) != 0) {
4864 /* load migration record */
4865 err
= load_imsm_migr_rec(super
, NULL
);
4867 /* migration is in progress,
4868 * but migr_rec cannot be loaded,
4874 /* Check migration compatibility */
4875 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
4876 pr_err("Unsupported migration detected");
4878 fprintf(stderr
, " on %s\n", devname
);
4880 fprintf(stderr
, " (IMSM).\n");
4889 while (super_list
) {
4890 struct intel_super
*s
= super_list
;
4892 super_list
= super_list
->next
;
4901 strcpy(st
->container_devnm
, fd2devnm(fd
));
4903 st
->container_devnm
[0] = 0;
4904 if (err
== 0 && st
->ss
== NULL
) {
4905 st
->ss
= &super_imsm
;
4906 st
->minor_version
= 0;
4907 st
->max_devs
= IMSM_MAX_DEVICES
;
4913 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4914 int *max
, int keep_fd
)
4916 struct md_list
*tmpdev
;
4920 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4921 if (tmpdev
->used
!= 1)
4923 if (tmpdev
->container
== 1) {
4925 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4927 pr_err("cannot open device %s: %s\n",
4928 tmpdev
->devname
, strerror(errno
));
4932 err
= get_sra_super_block(fd
, super_list
,
4933 tmpdev
->devname
, &lmax
,
4942 int major
= major(tmpdev
->st_rdev
);
4943 int minor
= minor(tmpdev
->st_rdev
);
4944 err
= get_super_block(super_list
,
4961 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4962 int major
, int minor
, int keep_fd
)
4964 struct intel_super
*s
;
4976 sprintf(nm
, "%d:%d", major
, minor
);
4977 dfd
= dev_open(nm
, O_RDWR
);
4983 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
4984 find_intel_hba_capability(dfd
, s
, devname
);
4985 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4987 /* retry the load if we might have raced against mdmon */
4988 if (err
== 3 && devnm
&& mdmon_running(devnm
))
4989 for (retry
= 0; retry
< 3; retry
++) {
4991 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4997 s
->next
= *super_list
;
5005 if (dfd
>= 0 && !keep_fd
)
5012 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5019 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5023 if (sra
->array
.major_version
!= -1 ||
5024 sra
->array
.minor_version
!= -2 ||
5025 strcmp(sra
->text_version
, "imsm") != 0) {
5030 devnm
= fd2devnm(fd
);
5031 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5032 if (get_super_block(super_list
, devnm
, devname
,
5033 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5044 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5046 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5050 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5052 struct intel_super
*super
;
5056 if (test_partition(fd
))
5057 /* IMSM not allowed on partitions */
5060 free_super_imsm(st
);
5062 super
= alloc_super();
5063 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5066 /* Load hba and capabilities if they exist.
5067 * But do not preclude loading metadata in case capabilities or hba are
5068 * non-compliant and ignore_hw_compat is set.
5070 rv
= find_intel_hba_capability(fd
, super
, devname
);
5071 /* no orom/efi or non-intel hba of the disk */
5072 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5074 pr_err("No OROM/EFI properties for %s\n", devname
);
5078 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5080 /* retry the load if we might have raced against mdmon */
5082 struct mdstat_ent
*mdstat
= NULL
;
5083 char *name
= fd2kname(fd
);
5086 mdstat
= mdstat_by_component(name
);
5088 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5089 for (retry
= 0; retry
< 3; retry
++) {
5091 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5097 free_mdstat(mdstat
);
5102 pr_err("Failed to load all information sections on %s\n", devname
);
5108 if (st
->ss
== NULL
) {
5109 st
->ss
= &super_imsm
;
5110 st
->minor_version
= 0;
5111 st
->max_devs
= IMSM_MAX_DEVICES
;
5114 /* load migration record */
5115 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5116 /* Check for unsupported migration features */
5117 if (check_mpb_migr_compatibility(super
) != 0) {
5118 pr_err("Unsupported migration detected");
5120 fprintf(stderr
, " on %s\n", devname
);
5122 fprintf(stderr
, " (IMSM).\n");
5130 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5132 if (info
->level
== 1)
5134 return info
->chunk_size
>> 9;
5137 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5138 unsigned long long size
)
5140 if (info
->level
== 1)
5143 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5146 static void imsm_update_version_info(struct intel_super
*super
)
5148 /* update the version and attributes */
5149 struct imsm_super
*mpb
= super
->anchor
;
5151 struct imsm_dev
*dev
;
5152 struct imsm_map
*map
;
5155 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5156 dev
= get_imsm_dev(super
, i
);
5157 map
= get_imsm_map(dev
, MAP_0
);
5158 if (__le32_to_cpu(dev
->size_high
) > 0)
5159 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5161 /* FIXME detect when an array spans a port multiplier */
5163 mpb
->attributes
|= MPB_ATTRIB_PM
;
5166 if (mpb
->num_raid_devs
> 1 ||
5167 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5168 version
= MPB_VERSION_ATTRIBS
;
5169 switch (get_imsm_raid_level(map
)) {
5170 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5171 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5172 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5173 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5176 if (map
->num_members
>= 5)
5177 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5178 else if (dev
->status
== DEV_CLONE_N_GO
)
5179 version
= MPB_VERSION_CNG
;
5180 else if (get_imsm_raid_level(map
) == 5)
5181 version
= MPB_VERSION_RAID5
;
5182 else if (map
->num_members
>= 3)
5183 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5184 else if (get_imsm_raid_level(map
) == 1)
5185 version
= MPB_VERSION_RAID1
;
5187 version
= MPB_VERSION_RAID0
;
5189 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5193 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5195 struct imsm_super
*mpb
= super
->anchor
;
5196 char *reason
= NULL
;
5199 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
5200 reason
= "must be 16 characters or less";
5202 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5203 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5205 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5206 reason
= "already exists";
5211 if (reason
&& !quiet
)
5212 pr_err("imsm volume name %s\n", reason
);
5217 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5218 struct shape
*s
, char *name
,
5219 char *homehost
, int *uuid
,
5220 long long data_offset
)
5222 /* We are creating a volume inside a pre-existing container.
5223 * so st->sb is already set.
5225 struct intel_super
*super
= st
->sb
;
5226 unsigned int sector_size
= super
->sector_size
;
5227 struct imsm_super
*mpb
= super
->anchor
;
5228 struct intel_dev
*dv
;
5229 struct imsm_dev
*dev
;
5230 struct imsm_vol
*vol
;
5231 struct imsm_map
*map
;
5232 int idx
= mpb
->num_raid_devs
;
5234 unsigned long long array_blocks
;
5235 size_t size_old
, size_new
;
5236 unsigned long long num_data_stripes
;
5238 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5239 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5243 /* ensure the mpb is large enough for the new data */
5244 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5245 size_new
= disks_to_mpb_size(info
->nr_disks
);
5246 if (size_new
> size_old
) {
5248 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5250 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5251 pr_err("could not allocate new mpb\n");
5254 if (posix_memalign(&super
->migr_rec_buf
, sector_size
,
5255 MIGR_REC_BUF_SECTORS
*sector_size
) != 0) {
5256 pr_err("could not allocate migr_rec buffer\n");
5262 memcpy(mpb_new
, mpb
, size_old
);
5265 super
->anchor
= mpb_new
;
5266 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5267 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5268 super
->len
= size_round
;
5270 super
->current_vol
= idx
;
5272 /* handle 'failed_disks' by either:
5273 * a) create dummy disk entries in the table if this the first
5274 * volume in the array. We add them here as this is the only
5275 * opportunity to add them. add_to_super_imsm_volume()
5276 * handles the non-failed disks and continues incrementing
5278 * b) validate that 'failed_disks' matches the current number
5279 * of missing disks if the container is populated
5281 if (super
->current_vol
== 0) {
5283 for (i
= 0; i
< info
->failed_disks
; i
++) {
5284 struct imsm_disk
*disk
;
5287 disk
= __get_imsm_disk(mpb
, i
);
5288 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5289 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5290 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5291 "missing:%d", (__u8
)i
);
5293 find_missing(super
);
5298 for (d
= super
->missing
; d
; d
= d
->next
)
5300 if (info
->failed_disks
> missing
) {
5301 pr_err("unable to add 'missing' disk to container\n");
5306 if (!check_name(super
, name
, 0))
5308 dv
= xmalloc(sizeof(*dv
));
5309 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5310 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
5311 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5312 info
->layout
, info
->chunk_size
,
5314 /* round array size down to closest MB */
5315 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
5317 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
5318 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
5319 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5321 vol
->migr_state
= 0;
5322 set_migr_type(dev
, MIGR_INIT
);
5323 vol
->dirty
= !info
->state
;
5324 vol
->curr_migr_unit
= 0;
5325 map
= get_imsm_map(dev
, MAP_0
);
5326 set_pba_of_lba0(map
, super
->create_offset
);
5327 set_blocks_per_member(map
, info_to_blocks_per_member(info
, s
->size
));
5328 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5329 map
->failed_disk_num
= ~0;
5330 if (info
->level
> 0)
5331 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5332 : IMSM_T_STATE_UNINITIALIZED
);
5334 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5335 IMSM_T_STATE_NORMAL
;
5338 if (info
->level
== 1 && info
->raid_disks
> 2) {
5341 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5345 map
->raid_level
= info
->level
;
5346 if (info
->level
== 10) {
5347 map
->raid_level
= 1;
5348 map
->num_domains
= info
->raid_disks
/ 2;
5349 } else if (info
->level
== 1)
5350 map
->num_domains
= info
->raid_disks
;
5352 map
->num_domains
= 1;
5354 /* info->size is only int so use the 'size' parameter instead */
5355 num_data_stripes
= (s
->size
* 2) / info_to_blocks_per_strip(info
);
5356 num_data_stripes
/= map
->num_domains
;
5357 set_num_data_stripes(map
, num_data_stripes
);
5359 map
->num_members
= info
->raid_disks
;
5360 for (i
= 0; i
< map
->num_members
; i
++) {
5361 /* initialized in add_to_super */
5362 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5364 mpb
->num_raid_devs
++;
5366 if (s
->consistency_policy
== UnSet
||
5367 s
->consistency_policy
== CONSISTENCY_POLICY_RESYNC
||
5368 s
->consistency_policy
== CONSISTENCY_POLICY_NONE
) {
5369 dev
->rwh_policy
= RWH_OFF
;
5370 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5371 dev
->rwh_policy
= RWH_DISTRIBUTED
;
5375 pr_err("imsm does not support consistency policy %s\n",
5376 map_num(consistency_policies
, s
->consistency_policy
));
5381 dv
->index
= super
->current_vol
;
5382 dv
->next
= super
->devlist
;
5383 super
->devlist
= dv
;
5385 imsm_update_version_info(super
);
5390 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5391 struct shape
*s
, char *name
,
5392 char *homehost
, int *uuid
,
5393 unsigned long long data_offset
)
5395 /* This is primarily called by Create when creating a new array.
5396 * We will then get add_to_super called for each component, and then
5397 * write_init_super called to write it out to each device.
5398 * For IMSM, Create can create on fresh devices or on a pre-existing
5400 * To create on a pre-existing array a different method will be called.
5401 * This one is just for fresh drives.
5403 struct intel_super
*super
;
5404 struct imsm_super
*mpb
;
5408 if (data_offset
!= INVALID_SECTORS
) {
5409 pr_err("data-offset not supported by imsm\n");
5414 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5418 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5420 mpb_size
= MAX_SECTOR_SIZE
;
5422 super
= alloc_super();
5424 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5429 pr_err("could not allocate superblock\n");
5432 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5433 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5434 pr_err("could not allocate migr_rec buffer\n");
5439 memset(super
->buf
, 0, mpb_size
);
5441 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5445 /* zeroing superblock */
5449 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5451 version
= (char *) mpb
->sig
;
5452 strcpy(version
, MPB_SIGNATURE
);
5453 version
+= strlen(MPB_SIGNATURE
);
5454 strcpy(version
, MPB_VERSION_RAID0
);
5460 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5461 int fd
, char *devname
)
5463 struct intel_super
*super
= st
->sb
;
5464 struct imsm_super
*mpb
= super
->anchor
;
5465 struct imsm_disk
*_disk
;
5466 struct imsm_dev
*dev
;
5467 struct imsm_map
*map
;
5471 dev
= get_imsm_dev(super
, super
->current_vol
);
5472 map
= get_imsm_map(dev
, MAP_0
);
5474 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5475 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5481 /* we're doing autolayout so grab the pre-marked (in
5482 * validate_geometry) raid_disk
5484 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5485 if (dl
->raiddisk
== dk
->raid_disk
)
5488 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5489 if (dl
->major
== dk
->major
&&
5490 dl
->minor
== dk
->minor
)
5495 pr_err("%s is not a member of the same container\n", devname
);
5499 /* add a pristine spare to the metadata */
5500 if (dl
->index
< 0) {
5501 dl
->index
= super
->anchor
->num_disks
;
5502 super
->anchor
->num_disks
++;
5504 /* Check the device has not already been added */
5505 slot
= get_imsm_disk_slot(map
, dl
->index
);
5507 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5508 pr_err("%s has been included in this array twice\n",
5512 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5513 dl
->disk
.status
= CONFIGURED_DISK
;
5515 /* update size of 'missing' disks to be at least as large as the
5516 * largest acitve member (we only have dummy missing disks when
5517 * creating the first volume)
5519 if (super
->current_vol
== 0) {
5520 for (df
= super
->missing
; df
; df
= df
->next
) {
5521 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5522 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5523 _disk
= __get_imsm_disk(mpb
, df
->index
);
5528 /* refresh unset/failed slots to point to valid 'missing' entries */
5529 for (df
= super
->missing
; df
; df
= df
->next
)
5530 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5531 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5533 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5535 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5536 if (is_gen_migration(dev
)) {
5537 struct imsm_map
*map2
= get_imsm_map(dev
,
5539 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5540 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5541 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5544 if ((unsigned)df
->index
==
5546 set_imsm_ord_tbl_ent(map2
,
5552 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5556 /* if we are creating the first raid device update the family number */
5557 if (super
->current_vol
== 0) {
5559 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5561 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5562 if (!_dev
|| !_disk
) {
5563 pr_err("BUG mpb setup error\n");
5569 sum
+= __gen_imsm_checksum(mpb
);
5570 mpb
->family_num
= __cpu_to_le32(sum
);
5571 mpb
->orig_family_num
= mpb
->family_num
;
5573 super
->current_disk
= dl
;
5578 * Function marks disk as spare and restores disk serial
5579 * in case it was previously marked as failed by takeover operation
5581 * -1 : critical error
5582 * 0 : disk is marked as spare but serial is not set
5585 int mark_spare(struct dl
*disk
)
5587 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5594 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5595 /* Restore disk serial number, because takeover marks disk
5596 * as failed and adds to serial ':0' before it becomes
5599 serialcpy(disk
->serial
, serial
);
5600 serialcpy(disk
->disk
.serial
, serial
);
5603 disk
->disk
.status
= SPARE_DISK
;
5609 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5610 int fd
, char *devname
,
5611 unsigned long long data_offset
)
5613 struct intel_super
*super
= st
->sb
;
5615 unsigned long long size
;
5616 unsigned int member_sector_size
;
5621 /* If we are on an RAID enabled platform check that the disk is
5622 * attached to the raid controller.
5623 * We do not need to test disks attachment for container based additions,
5624 * they shall be already tested when container was created/assembled.
5626 rv
= find_intel_hba_capability(fd
, super
, devname
);
5627 /* no orom/efi or non-intel hba of the disk */
5629 dprintf("capability: %p fd: %d ret: %d\n",
5630 super
->orom
, fd
, rv
);
5634 if (super
->current_vol
>= 0)
5635 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5638 dd
= xcalloc(sizeof(*dd
), 1);
5639 dd
->major
= major(stb
.st_rdev
);
5640 dd
->minor
= minor(stb
.st_rdev
);
5641 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5644 dd
->action
= DISK_ADD
;
5645 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5647 pr_err("failed to retrieve scsi serial, aborting\n");
5653 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5654 (super
->hba
->type
== SYS_DEV_VMD
))) {
5656 char *devpath
= diskfd_to_devpath(fd
);
5657 char controller_path
[PATH_MAX
];
5660 pr_err("failed to get devpath, aborting\n");
5667 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5670 if (devpath_to_vendor(controller_path
) == 0x8086) {
5672 * If Intel's NVMe drive has serial ended with
5673 * "-A","-B","-1" or "-2" it means that this is "x8"
5674 * device (double drive on single PCIe card).
5675 * User should be warned about potential data loss.
5677 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5678 /* Skip empty character at the end */
5679 if (dd
->serial
[i
] == 0)
5682 if (((dd
->serial
[i
] == 'A') ||
5683 (dd
->serial
[i
] == 'B') ||
5684 (dd
->serial
[i
] == '1') ||
5685 (dd
->serial
[i
] == '2')) &&
5686 (dd
->serial
[i
-1] == '-'))
5687 pr_err("\tThe action you are about to take may put your data at risk.\n"
5688 "\tPlease note that x8 devices may consist of two separate x4 devices "
5689 "located on a single PCIe port.\n"
5690 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5693 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5694 !imsm_orom_has_tpv_support(super
->orom
)) {
5695 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5696 "\tPlease refer to Intel(R) RSTe user guide.\n");
5703 get_dev_size(fd
, NULL
, &size
);
5704 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5706 if (super
->sector_size
== 0) {
5707 /* this a first device, so sector_size is not set yet */
5708 super
->sector_size
= member_sector_size
;
5709 } else if (member_sector_size
!= super
->sector_size
) {
5710 pr_err("Mixing between different sector size is forbidden, aborting...\n");
5717 /* clear migr_rec when adding disk to container */
5718 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*super
->sector_size
);
5719 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*super
->sector_size
,
5721 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5722 MIGR_REC_BUF_SECTORS
*super
->sector_size
) !=
5723 MIGR_REC_BUF_SECTORS
*super
->sector_size
)
5724 perror("Write migr_rec failed");
5728 serialcpy(dd
->disk
.serial
, dd
->serial
);
5729 set_total_blocks(&dd
->disk
, size
);
5730 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5731 struct imsm_super
*mpb
= super
->anchor
;
5732 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5735 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5736 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5738 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5740 if (st
->update_tail
) {
5741 dd
->next
= super
->disk_mgmt_list
;
5742 super
->disk_mgmt_list
= dd
;
5744 dd
->next
= super
->disks
;
5746 super
->updates_pending
++;
5752 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5754 struct intel_super
*super
= st
->sb
;
5757 /* remove from super works only in mdmon - for communication
5758 * manager - monitor. Check if communication memory buffer
5761 if (!st
->update_tail
) {
5762 pr_err("shall be used in mdmon context only\n");
5765 dd
= xcalloc(1, sizeof(*dd
));
5766 dd
->major
= dk
->major
;
5767 dd
->minor
= dk
->minor
;
5770 dd
->action
= DISK_REMOVE
;
5772 dd
->next
= super
->disk_mgmt_list
;
5773 super
->disk_mgmt_list
= dd
;
5778 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5781 char buf
[MAX_SECTOR_SIZE
];
5782 struct imsm_super anchor
;
5783 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5785 /* spare records have their own family number and do not have any defined raid
5788 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5790 struct imsm_super
*mpb
= super
->anchor
;
5791 struct imsm_super
*spare
= &spare_record
.anchor
;
5795 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5796 spare
->generation_num
= __cpu_to_le32(1UL);
5797 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5798 spare
->num_disks
= 1;
5799 spare
->num_raid_devs
= 0;
5800 spare
->cache_size
= mpb
->cache_size
;
5801 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5803 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5804 MPB_SIGNATURE MPB_VERSION_RAID0
);
5806 for (d
= super
->disks
; d
; d
= d
->next
) {
5810 spare
->disk
[0] = d
->disk
;
5811 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5812 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5814 if (super
->sector_size
== 4096)
5815 convert_to_4k_imsm_disk(&spare
->disk
[0]);
5817 sum
= __gen_imsm_checksum(spare
);
5818 spare
->family_num
= __cpu_to_le32(sum
);
5819 spare
->orig_family_num
= 0;
5820 sum
= __gen_imsm_checksum(spare
);
5821 spare
->check_sum
= __cpu_to_le32(sum
);
5823 if (store_imsm_mpb(d
->fd
, spare
)) {
5824 pr_err("failed for device %d:%d %s\n",
5825 d
->major
, d
->minor
, strerror(errno
));
5837 static int write_super_imsm(struct supertype
*st
, int doclose
)
5839 struct intel_super
*super
= st
->sb
;
5840 unsigned int sector_size
= super
->sector_size
;
5841 struct imsm_super
*mpb
= super
->anchor
;
5847 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5849 int clear_migration_record
= 1;
5852 /* 'generation' is incremented everytime the metadata is written */
5853 generation
= __le32_to_cpu(mpb
->generation_num
);
5855 mpb
->generation_num
= __cpu_to_le32(generation
);
5857 /* fix up cases where previous mdadm releases failed to set
5860 if (mpb
->orig_family_num
== 0)
5861 mpb
->orig_family_num
= mpb
->family_num
;
5863 for (d
= super
->disks
; d
; d
= d
->next
) {
5867 mpb
->disk
[d
->index
] = d
->disk
;
5871 for (d
= super
->missing
; d
; d
= d
->next
) {
5872 mpb
->disk
[d
->index
] = d
->disk
;
5875 mpb
->num_disks
= num_disks
;
5876 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5878 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5879 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5880 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5882 imsm_copy_dev(dev
, dev2
);
5883 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5885 if (is_gen_migration(dev2
))
5886 clear_migration_record
= 0;
5889 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
5892 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
5893 mpb
->attributes
|= MPB_ATTRIB_BBM
;
5895 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
5897 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
5898 mpb_size
+= bbm_log_size
;
5899 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5902 assert(super
->len
== 0 || mpb_size
<= super
->len
);
5905 /* recalculate checksum */
5906 sum
= __gen_imsm_checksum(mpb
);
5907 mpb
->check_sum
= __cpu_to_le32(sum
);
5909 if (super
->clean_migration_record_by_mdmon
) {
5910 clear_migration_record
= 1;
5911 super
->clean_migration_record_by_mdmon
= 0;
5913 if (clear_migration_record
)
5914 memset(super
->migr_rec_buf
, 0,
5915 MIGR_REC_BUF_SECTORS
*sector_size
);
5917 if (sector_size
== 4096)
5918 convert_to_4k(super
);
5920 /* write the mpb for disks that compose raid devices */
5921 for (d
= super
->disks
; d
; d
= d
->next
) {
5922 if (d
->index
< 0 || is_failed(&d
->disk
))
5925 if (clear_migration_record
) {
5926 unsigned long long dsize
;
5928 get_dev_size(d
->fd
, NULL
, &dsize
);
5929 if (lseek64(d
->fd
, dsize
- sector_size
,
5931 if ((unsigned int)write(d
->fd
,
5932 super
->migr_rec_buf
,
5933 MIGR_REC_BUF_SECTORS
*sector_size
) !=
5934 MIGR_REC_BUF_SECTORS
*sector_size
)
5935 perror("Write migr_rec failed");
5939 if (store_imsm_mpb(d
->fd
, mpb
))
5941 "failed for device %d:%d (fd: %d)%s\n",
5943 d
->fd
, strerror(errno
));
5952 return write_super_imsm_spares(super
, doclose
);
5957 static int create_array(struct supertype
*st
, int dev_idx
)
5960 struct imsm_update_create_array
*u
;
5961 struct intel_super
*super
= st
->sb
;
5962 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5963 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5964 struct disk_info
*inf
;
5965 struct imsm_disk
*disk
;
5968 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5969 sizeof(*inf
) * map
->num_members
;
5971 u
->type
= update_create_array
;
5972 u
->dev_idx
= dev_idx
;
5973 imsm_copy_dev(&u
->dev
, dev
);
5974 inf
= get_disk_info(u
);
5975 for (i
= 0; i
< map
->num_members
; i
++) {
5976 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5978 disk
= get_imsm_disk(super
, idx
);
5980 disk
= get_imsm_missing(super
, idx
);
5981 serialcpy(inf
[i
].serial
, disk
->serial
);
5983 append_metadata_update(st
, u
, len
);
5988 static int mgmt_disk(struct supertype
*st
)
5990 struct intel_super
*super
= st
->sb
;
5992 struct imsm_update_add_remove_disk
*u
;
5994 if (!super
->disk_mgmt_list
)
5999 u
->type
= update_add_remove_disk
;
6000 append_metadata_update(st
, u
, len
);
6006 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6008 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6010 struct intel_super
*super
= st
->sb
;
6012 struct ppl_header
*ppl_hdr
;
6015 ret
= posix_memalign(&buf
, 4096, PPL_HEADER_SIZE
);
6017 pr_err("Failed to allocate PPL header buffer\n");
6021 memset(buf
, 0, PPL_HEADER_SIZE
);
6023 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6024 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6025 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6027 if (lseek64(fd
, info
->ppl_sector
* 512, SEEK_SET
) < 0) {
6029 perror("Failed to seek to PPL header location");
6032 if (!ret
&& write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6034 perror("Write PPL header failed");
6044 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6045 struct mdinfo
*disk
)
6047 struct intel_super
*super
= st
->sb
;
6051 struct ppl_header
*ppl_hdr
;
6053 struct imsm_dev
*dev
;
6054 struct imsm_map
*map
;
6057 if (disk
->disk
.raid_disk
< 0)
6060 if (posix_memalign(&buf
, 4096, PPL_HEADER_SIZE
)) {
6061 pr_err("Failed to allocate PPL header buffer\n");
6065 dev
= get_imsm_dev(super
, info
->container_member
);
6066 map
= get_imsm_map(dev
, MAP_X
);
6067 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_X
);
6068 d
= get_imsm_dl_disk(super
, idx
);
6070 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6073 if (lseek64(d
->fd
, info
->ppl_sector
* 512, SEEK_SET
) < 0) {
6074 perror("Failed to seek to PPL header location");
6079 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6080 perror("Read PPL header failed");
6087 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6088 ppl_hdr
->checksum
= 0;
6090 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6091 dprintf("Wrong PPL header checksum on %s\n",
6096 if (!ret
&& (__le32_to_cpu(ppl_hdr
->signature
) !=
6097 super
->anchor
->orig_family_num
)) {
6098 dprintf("Wrong PPL header signature on %s\n",
6106 if (ret
== 1 && map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6107 return st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6114 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6116 struct intel_super
*super
= st
->sb
;
6120 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6121 info
->array
.level
!= 5)
6124 for (d
= super
->disks
; d
; d
= d
->next
) {
6125 if (d
->index
< 0 || is_failed(&d
->disk
))
6128 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6136 static int write_init_super_imsm(struct supertype
*st
)
6138 struct intel_super
*super
= st
->sb
;
6139 int current_vol
= super
->current_vol
;
6143 getinfo_super_imsm(st
, &info
, NULL
);
6145 /* we are done with current_vol reset it to point st at the container */
6146 super
->current_vol
= -1;
6148 if (st
->update_tail
) {
6149 /* queue the recently created array / added disk
6150 * as a metadata update */
6152 /* determine if we are creating a volume or adding a disk */
6153 if (current_vol
< 0) {
6154 /* in the mgmt (add/remove) disk case we are running
6155 * in mdmon context, so don't close fd's
6159 rv
= write_init_ppl_imsm_all(st
, &info
);
6161 rv
= create_array(st
, current_vol
);
6165 for (d
= super
->disks
; d
; d
= d
->next
)
6166 Kill(d
->devname
, NULL
, 0, -1, 1);
6167 if (current_vol
>= 0)
6168 rv
= write_init_ppl_imsm_all(st
, &info
);
6170 rv
= write_super_imsm(st
, 1);
6177 static int store_super_imsm(struct supertype
*st
, int fd
)
6179 struct intel_super
*super
= st
->sb
;
6180 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6186 if (super
->sector_size
== 4096)
6187 convert_to_4k(super
);
6188 return store_imsm_mpb(fd
, mpb
);
6195 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6196 int layout
, int raiddisks
, int chunk
,
6197 unsigned long long size
,
6198 unsigned long long data_offset
,
6200 unsigned long long *freesize
,
6204 unsigned long long ldsize
;
6205 struct intel_super
*super
;
6208 if (level
!= LEVEL_CONTAINER
)
6213 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6216 pr_err("imsm: Cannot open %s: %s\n",
6217 dev
, strerror(errno
));
6220 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6225 /* capabilities retrieve could be possible
6226 * note that there is no fd for the disks in array.
6228 super
= alloc_super();
6233 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6239 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6243 fd2devname(fd
, str
);
6244 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6245 fd
, str
, super
->orom
, rv
, raiddisks
);
6247 /* no orom/efi or non-intel hba of the disk */
6254 if (raiddisks
> super
->orom
->tds
) {
6256 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6257 raiddisks
, super
->orom
->tds
);
6261 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6262 (ldsize
>> 9) >> 32 > 0) {
6264 pr_err("%s exceeds maximum platform supported size\n", dev
);
6270 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6276 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6278 const unsigned long long base_start
= e
[*idx
].start
;
6279 unsigned long long end
= base_start
+ e
[*idx
].size
;
6282 if (base_start
== end
)
6286 for (i
= *idx
; i
< num_extents
; i
++) {
6287 /* extend overlapping extents */
6288 if (e
[i
].start
>= base_start
&&
6289 e
[i
].start
<= end
) {
6292 if (e
[i
].start
+ e
[i
].size
> end
)
6293 end
= e
[i
].start
+ e
[i
].size
;
6294 } else if (e
[i
].start
> end
) {
6300 return end
- base_start
;
6303 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6305 /* build a composite disk with all known extents and generate a new
6306 * 'maxsize' given the "all disks in an array must share a common start
6307 * offset" constraint
6309 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6313 unsigned long long pos
;
6314 unsigned long long start
= 0;
6315 unsigned long long maxsize
;
6316 unsigned long reserve
;
6318 /* coalesce and sort all extents. also, check to see if we need to
6319 * reserve space between member arrays
6322 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6325 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6328 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6333 while (i
< sum_extents
) {
6334 e
[j
].start
= e
[i
].start
;
6335 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6337 if (e
[j
-1].size
== 0)
6346 unsigned long long esize
;
6348 esize
= e
[i
].start
- pos
;
6349 if (esize
>= maxsize
) {
6354 pos
= e
[i
].start
+ e
[i
].size
;
6356 } while (e
[i
-1].size
);
6362 /* FIXME assumes volume at offset 0 is the first volume in a
6365 if (start_extent
> 0)
6366 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6370 if (maxsize
< reserve
)
6373 super
->create_offset
= ~((unsigned long long) 0);
6374 if (start
+ reserve
> super
->create_offset
)
6375 return 0; /* start overflows create_offset */
6376 super
->create_offset
= start
+ reserve
;
6378 return maxsize
- reserve
;
6381 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6383 if (level
< 0 || level
== 6 || level
== 4)
6386 /* if we have an orom prevent invalid raid levels */
6389 case 0: return imsm_orom_has_raid0(orom
);
6392 return imsm_orom_has_raid1e(orom
);
6393 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6394 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6395 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6398 return 1; /* not on an Intel RAID platform so anything goes */
6404 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6405 int dpa
, int verbose
)
6407 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6408 struct mdstat_ent
*memb
;
6414 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6415 if (memb
->metadata_version
&&
6416 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6417 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6418 !is_subarray(memb
->metadata_version
+9) &&
6420 struct dev_member
*dev
= memb
->members
;
6422 while(dev
&& (fd
< 0)) {
6423 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6424 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6426 fd
= open(path
, O_RDONLY
, 0);
6427 if (num
<= 0 || fd
< 0) {
6428 pr_vrb("Cannot open %s: %s\n",
6429 dev
->name
, strerror(errno
));
6435 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6436 struct mdstat_ent
*vol
;
6437 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6438 if (vol
->active
> 0 &&
6439 vol
->metadata_version
&&
6440 is_container_member(vol
, memb
->devnm
)) {
6445 if (*devlist
&& (found
< dpa
)) {
6446 dv
= xcalloc(1, sizeof(*dv
));
6447 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6448 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6451 dv
->next
= *devlist
;
6459 free_mdstat(mdstat
);
6464 static struct md_list
*
6465 get_loop_devices(void)
6468 struct md_list
*devlist
= NULL
;
6471 for(i
= 0; i
< 12; i
++) {
6472 dv
= xcalloc(1, sizeof(*dv
));
6473 dv
->devname
= xmalloc(40);
6474 sprintf(dv
->devname
, "/dev/loop%d", i
);
6482 static struct md_list
*
6483 get_devices(const char *hba_path
)
6485 struct md_list
*devlist
= NULL
;
6492 devlist
= get_loop_devices();
6495 /* scroll through /sys/dev/block looking for devices attached to
6498 dir
= opendir("/sys/dev/block");
6499 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6504 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6506 path
= devt_to_devpath(makedev(major
, minor
));
6509 if (!path_attached_to_hba(path
, hba_path
)) {
6516 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6518 fd2devname(fd
, buf
);
6521 pr_err("cannot open device: %s\n",
6526 dv
= xcalloc(1, sizeof(*dv
));
6527 dv
->devname
= xstrdup(buf
);
6534 devlist
= devlist
->next
;
6544 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6545 int verbose
, int *found
)
6547 struct md_list
*tmpdev
;
6549 struct supertype
*st
;
6551 /* first walk the list of devices to find a consistent set
6552 * that match the criterea, if that is possible.
6553 * We flag the ones we like with 'used'.
6556 st
= match_metadata_desc_imsm("imsm");
6558 pr_vrb("cannot allocate memory for imsm supertype\n");
6562 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6563 char *devname
= tmpdev
->devname
;
6565 struct supertype
*tst
;
6567 if (tmpdev
->used
> 1)
6569 tst
= dup_super(st
);
6571 pr_vrb("cannot allocate memory for imsm supertype\n");
6574 tmpdev
->container
= 0;
6575 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6577 dprintf("cannot open device %s: %s\n",
6578 devname
, strerror(errno
));
6580 } else if (fstat(dfd
, &stb
)< 0) {
6582 dprintf("fstat failed for %s: %s\n",
6583 devname
, strerror(errno
));
6585 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
6586 dprintf("%s is not a block device.\n",
6589 } else if (must_be_container(dfd
)) {
6590 struct supertype
*cst
;
6591 cst
= super_by_fd(dfd
, NULL
);
6593 dprintf("cannot recognize container type %s\n",
6596 } else if (tst
->ss
!= st
->ss
) {
6597 dprintf("non-imsm container - ignore it: %s\n",
6600 } else if (!tst
->ss
->load_container
||
6601 tst
->ss
->load_container(tst
, dfd
, NULL
))
6604 tmpdev
->container
= 1;
6607 cst
->ss
->free_super(cst
);
6609 tmpdev
->st_rdev
= stb
.st_rdev
;
6610 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6611 dprintf("no RAID superblock on %s\n",
6614 } else if (tst
->ss
->compare_super
== NULL
) {
6615 dprintf("Cannot assemble %s metadata on %s\n",
6616 tst
->ss
->name
, devname
);
6622 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6623 /* Ignore unrecognised devices during auto-assembly */
6628 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6630 if (st
->minor_version
== -1)
6631 st
->minor_version
= tst
->minor_version
;
6633 if (memcmp(info
.uuid
, uuid_zero
,
6634 sizeof(int[4])) == 0) {
6635 /* this is a floating spare. It cannot define
6636 * an array unless there are no more arrays of
6637 * this type to be found. It can be included
6638 * in an array of this type though.
6644 if (st
->ss
!= tst
->ss
||
6645 st
->minor_version
!= tst
->minor_version
||
6646 st
->ss
->compare_super(st
, tst
) != 0) {
6647 /* Some mismatch. If exactly one array matches this host,
6648 * we can resolve on that one.
6649 * Or, if we are auto assembling, we just ignore the second
6652 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6658 dprintf("found: devname: %s\n", devname
);
6662 tst
->ss
->free_super(tst
);
6666 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6667 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6668 for (iter
= head
; iter
; iter
= iter
->next
) {
6669 dprintf("content->text_version: %s vol\n",
6670 iter
->text_version
);
6671 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6672 /* do not assemble arrays with unsupported
6674 dprintf("Cannot activate member %s.\n",
6675 iter
->text_version
);
6682 dprintf("No valid super block on device list: err: %d %p\n",
6686 dprintf("no more devices to examine\n");
6689 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6690 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6692 if (count
< tmpdev
->found
)
6695 count
-= tmpdev
->found
;
6698 if (tmpdev
->used
== 1)
6703 st
->ss
->free_super(st
);
6707 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
6710 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6712 const struct orom_entry
*entry
;
6713 struct devid_list
*dv
, *devid_list
;
6718 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
6719 if (strstr(idev
->path
, hba_path
))
6723 if (!idev
|| !idev
->dev_id
)
6726 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
6728 if (!entry
|| !entry
->devid_list
)
6731 devid_list
= entry
->devid_list
;
6732 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
6733 struct md_list
*devlist
;
6734 struct sys_dev
*device
= NULL
;
6739 device
= device_by_id_and_path(dv
->devid
, hba_path
);
6741 device
= device_by_id(dv
->devid
);
6744 hpath
= device
->path
;
6748 devlist
= get_devices(hpath
);
6749 /* if no intel devices return zero volumes */
6750 if (devlist
== NULL
)
6753 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
6755 dprintf("path: %s active arrays: %d\n", hpath
, count
);
6756 if (devlist
== NULL
)
6760 count
+= count_volumes_list(devlist
,
6764 dprintf("found %d count: %d\n", found
, count
);
6767 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
6770 struct md_list
*dv
= devlist
;
6771 devlist
= devlist
->next
;
6779 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
6783 if (hba
->type
== SYS_DEV_VMD
) {
6784 struct sys_dev
*dev
;
6787 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
6788 if (dev
->type
== SYS_DEV_VMD
)
6789 count
+= __count_volumes(dev
->path
, dpa
,
6794 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
6797 static int imsm_default_chunk(const struct imsm_orom
*orom
)
6799 /* up to 512 if the plaform supports it, otherwise the platform max.
6800 * 128 if no platform detected
6802 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
6804 return min(512, (1 << fs
));
6808 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
6809 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
6811 /* check/set platform and metadata limits/defaults */
6812 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
6813 pr_vrb("platform supports a maximum of %d disks per array\n",
6818 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6819 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
6820 pr_vrb("platform does not support raid%d with %d disk%s\n",
6821 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
6825 if (*chunk
== 0 || *chunk
== UnSet
)
6826 *chunk
= imsm_default_chunk(super
->orom
);
6828 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
6829 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
6833 if (layout
!= imsm_level_to_layout(level
)) {
6835 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
6836 else if (level
== 10)
6837 pr_vrb("imsm raid 10 only supports the n2 layout\n");
6839 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
6844 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
6845 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
6846 pr_vrb("platform does not support a volume size over 2TB\n");
6853 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
6854 * FIX ME add ahci details
6856 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
6857 int layout
, int raiddisks
, int *chunk
,
6858 unsigned long long size
,
6859 unsigned long long data_offset
,
6861 unsigned long long *freesize
,
6865 struct intel_super
*super
= st
->sb
;
6866 struct imsm_super
*mpb
;
6868 unsigned long long pos
= 0;
6869 unsigned long long maxsize
;
6873 /* We must have the container info already read in. */
6877 mpb
= super
->anchor
;
6879 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
6880 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6884 /* General test: make sure there is space for
6885 * 'raiddisks' device extents of size 'size' at a given
6888 unsigned long long minsize
= size
;
6889 unsigned long long start_offset
= MaxSector
;
6892 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
6893 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6898 e
= get_extents(super
, dl
);
6901 unsigned long long esize
;
6902 esize
= e
[i
].start
- pos
;
6903 if (esize
>= minsize
)
6905 if (found
&& start_offset
== MaxSector
) {
6908 } else if (found
&& pos
!= start_offset
) {
6912 pos
= e
[i
].start
+ e
[i
].size
;
6914 } while (e
[i
-1].size
);
6919 if (dcnt
< raiddisks
) {
6921 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6928 /* This device must be a member of the set */
6929 if (stat(dev
, &stb
) < 0)
6931 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6933 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6934 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6935 dl
->minor
== (int)minor(stb
.st_rdev
))
6940 pr_err("%s is not in the same imsm set\n", dev
);
6942 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6943 /* If a volume is present then the current creation attempt
6944 * cannot incorporate new spares because the orom may not
6945 * understand this configuration (all member disks must be
6946 * members of each array in the container).
6948 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6949 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6951 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6952 mpb
->num_disks
!= raiddisks
) {
6953 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6957 /* retrieve the largest free space block */
6958 e
= get_extents(super
, dl
);
6963 unsigned long long esize
;
6965 esize
= e
[i
].start
- pos
;
6966 if (esize
>= maxsize
)
6968 pos
= e
[i
].start
+ e
[i
].size
;
6970 } while (e
[i
-1].size
);
6975 pr_err("unable to determine free space for: %s\n",
6979 if (maxsize
< size
) {
6981 pr_err("%s not enough space (%llu < %llu)\n",
6982 dev
, maxsize
, size
);
6986 /* count total number of extents for merge */
6988 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6990 i
+= dl
->extent_cnt
;
6992 maxsize
= merge_extents(super
, i
);
6994 if (!check_env("IMSM_NO_PLATFORM") &&
6995 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6996 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7000 if (maxsize
< size
|| maxsize
== 0) {
7003 pr_err("no free space left on device. Aborting...\n");
7005 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7011 *freesize
= maxsize
;
7014 int count
= count_volumes(super
->hba
,
7015 super
->orom
->dpa
, verbose
);
7016 if (super
->orom
->vphba
<= count
) {
7017 pr_vrb("platform does not support more than %d raid volumes.\n",
7018 super
->orom
->vphba
);
7025 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7026 unsigned long long size
, int chunk
,
7027 unsigned long long *freesize
)
7029 struct intel_super
*super
= st
->sb
;
7030 struct imsm_super
*mpb
= super
->anchor
;
7035 unsigned long long maxsize
;
7036 unsigned long long minsize
;
7040 /* find the largest common start free region of the possible disks */
7044 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7050 /* don't activate new spares if we are orom constrained
7051 * and there is already a volume active in the container
7053 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7056 e
= get_extents(super
, dl
);
7059 for (i
= 1; e
[i
-1].size
; i
++)
7067 maxsize
= merge_extents(super
, extent_cnt
);
7071 minsize
= chunk
* 2;
7073 if (cnt
< raiddisks
||
7074 (super
->orom
&& used
&& used
!= raiddisks
) ||
7075 maxsize
< minsize
||
7077 pr_err("not enough devices with space to create array.\n");
7078 return 0; /* No enough free spaces large enough */
7089 if (!check_env("IMSM_NO_PLATFORM") &&
7090 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7091 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7095 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7097 dl
->raiddisk
= cnt
++;
7101 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7106 static int reserve_space(struct supertype
*st
, int raiddisks
,
7107 unsigned long long size
, int chunk
,
7108 unsigned long long *freesize
)
7110 struct intel_super
*super
= st
->sb
;
7115 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7118 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7120 dl
->raiddisk
= cnt
++;
7127 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7128 int raiddisks
, int *chunk
, unsigned long long size
,
7129 unsigned long long data_offset
,
7130 char *dev
, unsigned long long *freesize
,
7131 int consistency_policy
, int verbose
)
7138 * if given unused devices create a container
7139 * if given given devices in a container create a member volume
7141 if (level
== LEVEL_CONTAINER
) {
7142 /* Must be a fresh device to add to a container */
7143 return validate_geometry_imsm_container(st
, level
, layout
,
7153 struct intel_super
*super
= st
->sb
;
7154 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7155 raiddisks
, chunk
, size
,
7158 /* we are being asked to automatically layout a
7159 * new volume based on the current contents of
7160 * the container. If the the parameters can be
7161 * satisfied reserve_space will record the disks,
7162 * start offset, and size of the volume to be
7163 * created. add_to_super and getinfo_super
7164 * detect when autolayout is in progress.
7166 /* assuming that freesize is always given when array is
7168 if (super
->orom
&& freesize
) {
7170 count
= count_volumes(super
->hba
,
7171 super
->orom
->dpa
, verbose
);
7172 if (super
->orom
->vphba
<= count
) {
7173 pr_vrb("platform does not support more than %d raid volumes.\n",
7174 super
->orom
->vphba
);
7179 return reserve_space(st
, raiddisks
, size
,
7185 /* creating in a given container */
7186 return validate_geometry_imsm_volume(st
, level
, layout
,
7187 raiddisks
, chunk
, size
,
7189 dev
, freesize
, verbose
);
7192 /* This device needs to be a device in an 'imsm' container */
7193 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7196 pr_err("Cannot create this array on device %s\n",
7201 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7203 pr_err("Cannot open %s: %s\n",
7204 dev
, strerror(errno
));
7207 /* Well, it is in use by someone, maybe an 'imsm' container. */
7208 cfd
= open_container(fd
);
7212 pr_err("Cannot use %s: It is busy\n",
7216 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7217 if (sra
&& sra
->array
.major_version
== -1 &&
7218 strcmp(sra
->text_version
, "imsm") == 0)
7222 /* This is a member of a imsm container. Load the container
7223 * and try to create a volume
7225 struct intel_super
*super
;
7227 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7229 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7231 return validate_geometry_imsm_volume(st
, level
, layout
,
7233 size
, data_offset
, dev
,
7240 pr_err("failed container membership check\n");
7246 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7248 struct intel_super
*super
= st
->sb
;
7250 if (level
&& *level
== UnSet
)
7251 *level
= LEVEL_CONTAINER
;
7253 if (level
&& layout
&& *layout
== UnSet
)
7254 *layout
= imsm_level_to_layout(*level
);
7256 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7257 *chunk
= imsm_default_chunk(super
->orom
);
7260 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7262 static int kill_subarray_imsm(struct supertype
*st
)
7264 /* remove the subarray currently referenced by ->current_vol */
7266 struct intel_dev
**dp
;
7267 struct intel_super
*super
= st
->sb
;
7268 __u8 current_vol
= super
->current_vol
;
7269 struct imsm_super
*mpb
= super
->anchor
;
7271 if (super
->current_vol
< 0)
7273 super
->current_vol
= -1; /* invalidate subarray cursor */
7275 /* block deletions that would change the uuid of active subarrays
7277 * FIXME when immutable ids are available, but note that we'll
7278 * also need to fixup the invalidated/active subarray indexes in
7281 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7284 if (i
< current_vol
)
7286 sprintf(subarray
, "%u", i
);
7287 if (is_subarray_active(subarray
, st
->devnm
)) {
7288 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7295 if (st
->update_tail
) {
7296 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7298 u
->type
= update_kill_array
;
7299 u
->dev_idx
= current_vol
;
7300 append_metadata_update(st
, u
, sizeof(*u
));
7305 for (dp
= &super
->devlist
; *dp
;)
7306 if ((*dp
)->index
== current_vol
) {
7309 handle_missing(super
, (*dp
)->dev
);
7310 if ((*dp
)->index
> current_vol
)
7315 /* no more raid devices, all active components are now spares,
7316 * but of course failed are still failed
7318 if (--mpb
->num_raid_devs
== 0) {
7321 for (d
= super
->disks
; d
; d
= d
->next
)
7326 super
->updates_pending
++;
7331 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7332 char *update
, struct mddev_ident
*ident
)
7334 /* update the subarray currently referenced by ->current_vol */
7335 struct intel_super
*super
= st
->sb
;
7336 struct imsm_super
*mpb
= super
->anchor
;
7338 if (strcmp(update
, "name") == 0) {
7339 char *name
= ident
->name
;
7343 if (is_subarray_active(subarray
, st
->devnm
)) {
7344 pr_err("Unable to update name of active subarray\n");
7348 if (!check_name(super
, name
, 0))
7351 vol
= strtoul(subarray
, &ep
, 10);
7352 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7355 if (st
->update_tail
) {
7356 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7358 u
->type
= update_rename_array
;
7360 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7361 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7362 append_metadata_update(st
, u
, sizeof(*u
));
7364 struct imsm_dev
*dev
;
7367 dev
= get_imsm_dev(super
, vol
);
7368 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
7369 dev
->volume
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7370 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7371 dev
= get_imsm_dev(super
, i
);
7372 handle_missing(super
, dev
);
7374 super
->updates_pending
++;
7381 #endif /* MDASSEMBLE */
7383 static int is_gen_migration(struct imsm_dev
*dev
)
7388 if (!dev
->vol
.migr_state
)
7391 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7397 static int is_rebuilding(struct imsm_dev
*dev
)
7399 struct imsm_map
*migr_map
;
7401 if (!dev
->vol
.migr_state
)
7404 if (migr_type(dev
) != MIGR_REBUILD
)
7407 migr_map
= get_imsm_map(dev
, MAP_1
);
7409 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7416 static int is_initializing(struct imsm_dev
*dev
)
7418 struct imsm_map
*migr_map
;
7420 if (!dev
->vol
.migr_state
)
7423 if (migr_type(dev
) != MIGR_INIT
)
7426 migr_map
= get_imsm_map(dev
, MAP_1
);
7428 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7435 static void update_recovery_start(struct intel_super
*super
,
7436 struct imsm_dev
*dev
,
7437 struct mdinfo
*array
)
7439 struct mdinfo
*rebuild
= NULL
;
7443 if (!is_rebuilding(dev
))
7446 /* Find the rebuild target, but punt on the dual rebuild case */
7447 for (d
= array
->devs
; d
; d
= d
->next
)
7448 if (d
->recovery_start
== 0) {
7455 /* (?) none of the disks are marked with
7456 * IMSM_ORD_REBUILD, so assume they are missing and the
7457 * disk_ord_tbl was not correctly updated
7459 dprintf("failed to locate out-of-sync disk\n");
7463 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7464 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7468 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7471 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7473 /* Given a container loaded by load_super_imsm_all,
7474 * extract information about all the arrays into
7476 * If 'subarray' is given, just extract info about that array.
7478 * For each imsm_dev create an mdinfo, fill it in,
7479 * then look for matching devices in super->disks
7480 * and create appropriate device mdinfo.
7482 struct intel_super
*super
= st
->sb
;
7483 struct imsm_super
*mpb
= super
->anchor
;
7484 struct mdinfo
*rest
= NULL
;
7488 int spare_disks
= 0;
7490 /* do not assemble arrays when not all attributes are supported */
7491 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7493 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7496 /* count spare devices, not used in maps
7498 for (d
= super
->disks
; d
; d
= d
->next
)
7502 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7503 struct imsm_dev
*dev
;
7504 struct imsm_map
*map
;
7505 struct imsm_map
*map2
;
7506 struct mdinfo
*this;
7514 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7517 dev
= get_imsm_dev(super
, i
);
7518 map
= get_imsm_map(dev
, MAP_0
);
7519 map2
= get_imsm_map(dev
, MAP_1
);
7521 /* do not publish arrays that are in the middle of an
7522 * unsupported migration
7524 if (dev
->vol
.migr_state
&&
7525 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7526 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7530 /* do not publish arrays that are not support by controller's
7534 this = xmalloc(sizeof(*this));
7536 super
->current_vol
= i
;
7537 getinfo_super_imsm_volume(st
, this, NULL
);
7540 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7541 /* mdadm does not support all metadata features- set the bit in all arrays state */
7542 if (!validate_geometry_imsm_orom(super
,
7543 get_imsm_raid_level(map
), /* RAID level */
7544 imsm_level_to_layout(get_imsm_raid_level(map
)),
7545 map
->num_members
, /* raid disks */
7546 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
7548 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7550 this->array
.state
|=
7551 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7552 (1<<MD_SB_BLOCK_VOLUME
);
7556 /* if array has bad blocks, set suitable bit in all arrays state */
7558 this->array
.state
|=
7559 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7560 (1<<MD_SB_BLOCK_VOLUME
);
7562 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7563 unsigned long long recovery_start
;
7564 struct mdinfo
*info_d
;
7571 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7572 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7573 for (d
= super
->disks
; d
; d
= d
->next
)
7574 if (d
->index
== idx
)
7577 recovery_start
= MaxSector
;
7580 if (d
&& is_failed(&d
->disk
))
7582 if (ord
& IMSM_ORD_REBUILD
)
7586 * if we skip some disks the array will be assmebled degraded;
7587 * reset resync start to avoid a dirty-degraded
7588 * situation when performing the intial sync
7590 * FIXME handle dirty degraded
7592 if ((skip
|| recovery_start
== 0) &&
7593 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
))
7594 this->resync_start
= MaxSector
;
7598 info_d
= xcalloc(1, sizeof(*info_d
));
7599 info_d
->next
= this->devs
;
7600 this->devs
= info_d
;
7602 info_d
->disk
.number
= d
->index
;
7603 info_d
->disk
.major
= d
->major
;
7604 info_d
->disk
.minor
= d
->minor
;
7605 info_d
->disk
.raid_disk
= slot
;
7606 info_d
->recovery_start
= recovery_start
;
7608 if (slot
< map2
->num_members
)
7609 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7611 this->array
.spare_disks
++;
7613 if (slot
< map
->num_members
)
7614 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7616 this->array
.spare_disks
++;
7618 if (info_d
->recovery_start
== MaxSector
)
7619 this->array
.working_disks
++;
7621 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7622 info_d
->data_offset
= pba_of_lba0(map
);
7624 if (map
->raid_level
== 5) {
7625 info_d
->component_size
=
7626 num_data_stripes(map
) *
7627 map
->blocks_per_strip
;
7628 info_d
->ppl_sector
= this->ppl_sector
;
7629 info_d
->ppl_size
= this->ppl_size
;
7631 info_d
->component_size
= blocks_per_member(map
);
7633 info_d
->consistency_policy
= this->consistency_policy
;
7635 info_d
->bb
.supported
= 1;
7636 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7637 info_d
->data_offset
,
7638 info_d
->component_size
,
7641 /* now that the disk list is up-to-date fixup recovery_start */
7642 update_recovery_start(super
, dev
, this);
7643 this->array
.spare_disks
+= spare_disks
;
7646 /* check for reshape */
7647 if (this->reshape_active
== 1)
7648 recover_backup_imsm(st
, this);
7656 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7657 int failed
, int look_in_map
)
7659 struct imsm_map
*map
;
7661 map
= get_imsm_map(dev
, look_in_map
);
7664 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
7665 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
7667 switch (get_imsm_raid_level(map
)) {
7669 return IMSM_T_STATE_FAILED
;
7672 if (failed
< map
->num_members
)
7673 return IMSM_T_STATE_DEGRADED
;
7675 return IMSM_T_STATE_FAILED
;
7680 * check to see if any mirrors have failed, otherwise we
7681 * are degraded. Even numbered slots are mirrored on
7685 /* gcc -Os complains that this is unused */
7686 int insync
= insync
;
7688 for (i
= 0; i
< map
->num_members
; i
++) {
7689 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
7690 int idx
= ord_to_idx(ord
);
7691 struct imsm_disk
*disk
;
7693 /* reset the potential in-sync count on even-numbered
7694 * slots. num_copies is always 2 for imsm raid10
7699 disk
= get_imsm_disk(super
, idx
);
7700 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7703 /* no in-sync disks left in this mirror the
7707 return IMSM_T_STATE_FAILED
;
7710 return IMSM_T_STATE_DEGRADED
;
7714 return IMSM_T_STATE_DEGRADED
;
7716 return IMSM_T_STATE_FAILED
;
7722 return map
->map_state
;
7725 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
7730 struct imsm_disk
*disk
;
7731 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7732 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
7733 struct imsm_map
*map_for_loop
;
7738 /* at the beginning of migration we set IMSM_ORD_REBUILD on
7739 * disks that are being rebuilt. New failures are recorded to
7740 * map[0]. So we look through all the disks we started with and
7741 * see if any failures are still present, or if any new ones
7745 if (prev
&& (map
->num_members
< prev
->num_members
))
7746 map_for_loop
= prev
;
7748 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
7750 /* when MAP_X is passed both maps failures are counted
7753 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
7754 i
< prev
->num_members
) {
7755 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
7756 idx_1
= ord_to_idx(ord
);
7758 disk
= get_imsm_disk(super
, idx_1
);
7759 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7762 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
7763 i
< map
->num_members
) {
7764 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
7765 idx
= ord_to_idx(ord
);
7768 disk
= get_imsm_disk(super
, idx
);
7769 if (!disk
|| is_failed(disk
) ||
7770 ord
& IMSM_ORD_REBUILD
)
7780 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
7783 struct intel_super
*super
= c
->sb
;
7784 struct imsm_super
*mpb
= super
->anchor
;
7785 struct imsm_update_prealloc_bb_mem u
;
7787 if (atoi(inst
) >= mpb
->num_raid_devs
) {
7788 pr_err("subarry index %d, out of range\n", atoi(inst
));
7792 dprintf("imsm: open_new %s\n", inst
);
7793 a
->info
.container_member
= atoi(inst
);
7795 u
.type
= update_prealloc_badblocks_mem
;
7796 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
7801 static int is_resyncing(struct imsm_dev
*dev
)
7803 struct imsm_map
*migr_map
;
7805 if (!dev
->vol
.migr_state
)
7808 if (migr_type(dev
) == MIGR_INIT
||
7809 migr_type(dev
) == MIGR_REPAIR
)
7812 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7815 migr_map
= get_imsm_map(dev
, MAP_1
);
7817 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
7818 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
7824 /* return true if we recorded new information */
7825 static int mark_failure(struct intel_super
*super
,
7826 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7830 struct imsm_map
*map
;
7831 char buf
[MAX_RAID_SERIAL_LEN
+3];
7832 unsigned int len
, shift
= 0;
7834 /* new failures are always set in map[0] */
7835 map
= get_imsm_map(dev
, MAP_0
);
7837 slot
= get_imsm_disk_slot(map
, idx
);
7841 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
7842 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
7845 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
7846 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
7848 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
7849 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
7850 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
7852 disk
->status
|= FAILED_DISK
;
7853 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
7854 /* mark failures in second map if second map exists and this disk
7856 * This is valid for migration, initialization and rebuild
7858 if (dev
->vol
.migr_state
) {
7859 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
7860 int slot2
= get_imsm_disk_slot(map2
, idx
);
7862 if (slot2
< map2
->num_members
&& slot2
>= 0)
7863 set_imsm_ord_tbl_ent(map2
, slot2
,
7864 idx
| IMSM_ORD_REBUILD
);
7866 if (map
->failed_disk_num
== 0xff)
7867 map
->failed_disk_num
= slot
;
7869 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
7874 static void mark_missing(struct intel_super
*super
,
7875 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7877 mark_failure(super
, dev
, disk
, idx
);
7879 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
7882 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
7883 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
7886 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7890 if (!super
->missing
)
7893 /* When orom adds replacement for missing disk it does
7894 * not remove entry of missing disk, but just updates map with
7895 * new added disk. So it is not enough just to test if there is
7896 * any missing disk, we have to look if there are any failed disks
7897 * in map to stop migration */
7899 dprintf("imsm: mark missing\n");
7900 /* end process for initialization and rebuild only
7902 if (is_gen_migration(dev
) == 0) {
7906 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7907 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7910 end_migration(dev
, super
, map_state
);
7912 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7913 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
7914 super
->updates_pending
++;
7917 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7920 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7921 unsigned long long array_blocks
;
7922 struct imsm_map
*map
;
7924 if (used_disks
== 0) {
7925 /* when problems occures
7926 * return current array_blocks value
7928 array_blocks
= __le32_to_cpu(dev
->size_high
);
7929 array_blocks
= array_blocks
<< 32;
7930 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7932 return array_blocks
;
7935 /* set array size in metadata
7937 if (new_size
<= 0) {
7938 /* OLCE size change is caused by added disks
7940 map
= get_imsm_map(dev
, MAP_0
);
7941 array_blocks
= blocks_per_member(map
) * used_disks
;
7943 /* Online Volume Size Change
7944 * Using available free space
7946 array_blocks
= new_size
;
7949 /* round array size down to closest MB
7951 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7952 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7953 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7955 return array_blocks
;
7958 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7960 static void imsm_progress_container_reshape(struct intel_super
*super
)
7962 /* if no device has a migr_state, but some device has a
7963 * different number of members than the previous device, start
7964 * changing the number of devices in this device to match
7967 struct imsm_super
*mpb
= super
->anchor
;
7968 int prev_disks
= -1;
7972 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7973 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7974 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7975 struct imsm_map
*map2
;
7976 int prev_num_members
;
7978 if (dev
->vol
.migr_state
)
7981 if (prev_disks
== -1)
7982 prev_disks
= map
->num_members
;
7983 if (prev_disks
== map
->num_members
)
7986 /* OK, this array needs to enter reshape mode.
7987 * i.e it needs a migr_state
7990 copy_map_size
= sizeof_imsm_map(map
);
7991 prev_num_members
= map
->num_members
;
7992 map
->num_members
= prev_disks
;
7993 dev
->vol
.migr_state
= 1;
7994 dev
->vol
.curr_migr_unit
= 0;
7995 set_migr_type(dev
, MIGR_GEN_MIGR
);
7996 for (i
= prev_num_members
;
7997 i
< map
->num_members
; i
++)
7998 set_imsm_ord_tbl_ent(map
, i
, i
);
7999 map2
= get_imsm_map(dev
, MAP_1
);
8000 /* Copy the current map */
8001 memcpy(map2
, map
, copy_map_size
);
8002 map2
->num_members
= prev_num_members
;
8004 imsm_set_array_size(dev
, -1);
8005 super
->clean_migration_record_by_mdmon
= 1;
8006 super
->updates_pending
++;
8010 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8011 * states are handled in imsm_set_disk() with one exception, when a
8012 * resync is stopped due to a new failure this routine will set the
8013 * 'degraded' state for the array.
8015 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8017 int inst
= a
->info
.container_member
;
8018 struct intel_super
*super
= a
->container
->sb
;
8019 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8020 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8021 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8022 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8023 __u32 blocks_per_unit
;
8025 if (dev
->vol
.migr_state
&&
8026 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8027 /* array state change is blocked due to reshape action
8029 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8030 * - finish the reshape (if last_checkpoint is big and action != reshape)
8031 * - update curr_migr_unit
8033 if (a
->curr_action
== reshape
) {
8034 /* still reshaping, maybe update curr_migr_unit */
8035 goto mark_checkpoint
;
8037 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8038 /* for some reason we aborted the reshape.
8040 * disable automatic metadata rollback
8041 * user action is required to recover process
8044 struct imsm_map
*map2
=
8045 get_imsm_map(dev
, MAP_1
);
8046 dev
->vol
.migr_state
= 0;
8047 set_migr_type(dev
, 0);
8048 dev
->vol
.curr_migr_unit
= 0;
8050 sizeof_imsm_map(map2
));
8051 super
->updates_pending
++;
8054 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8055 unsigned long long array_blocks
;
8059 used_disks
= imsm_num_data_members(dev
, MAP_0
);
8060 if (used_disks
> 0) {
8062 blocks_per_member(map
) *
8064 /* round array size down to closest MB
8066 array_blocks
= (array_blocks
8067 >> SECT_PER_MB_SHIFT
)
8068 << SECT_PER_MB_SHIFT
;
8069 a
->info
.custom_array_size
= array_blocks
;
8070 /* encourage manager to update array
8074 a
->check_reshape
= 1;
8076 /* finalize online capacity expansion/reshape */
8077 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8079 mdi
->disk
.raid_disk
,
8082 imsm_progress_container_reshape(super
);
8087 /* before we activate this array handle any missing disks */
8088 if (consistent
== 2)
8089 handle_missing(super
, dev
);
8091 if (consistent
== 2 &&
8092 (!is_resync_complete(&a
->info
) ||
8093 map_state
!= IMSM_T_STATE_NORMAL
||
8094 dev
->vol
.migr_state
))
8097 if (is_resync_complete(&a
->info
)) {
8098 /* complete intialization / resync,
8099 * recovery and interrupted recovery is completed in
8102 if (is_resyncing(dev
)) {
8103 dprintf("imsm: mark resync done\n");
8104 end_migration(dev
, super
, map_state
);
8105 super
->updates_pending
++;
8106 a
->last_checkpoint
= 0;
8108 } else if ((!is_resyncing(dev
) && !failed
) &&
8109 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8110 /* mark the start of the init process if nothing is failed */
8111 dprintf("imsm: mark resync start\n");
8112 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8113 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8115 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8116 super
->updates_pending
++;
8120 /* skip checkpointing for general migration,
8121 * it is controlled in mdadm
8123 if (is_gen_migration(dev
))
8124 goto skip_mark_checkpoint
;
8126 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8127 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8128 if (blocks_per_unit
) {
8132 units
= a
->last_checkpoint
/ blocks_per_unit
;
8135 /* check that we did not overflow 32-bits, and that
8136 * curr_migr_unit needs updating
8138 if (units32
== units
&&
8140 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8141 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8142 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8143 super
->updates_pending
++;
8147 skip_mark_checkpoint
:
8148 /* mark dirty / clean */
8149 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8150 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8151 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8153 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8155 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8156 if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
8157 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8159 super
->updates_pending
++;
8165 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8167 int inst
= a
->info
.container_member
;
8168 struct intel_super
*super
= a
->container
->sb
;
8169 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8170 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8172 if (slot
> map
->num_members
) {
8173 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8174 slot
, map
->num_members
- 1);
8181 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8184 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8186 int inst
= a
->info
.container_member
;
8187 struct intel_super
*super
= a
->container
->sb
;
8188 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8189 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8190 struct imsm_disk
*disk
;
8192 int recovery_not_finished
= 0;
8197 ord
= imsm_disk_slot_to_ord(a
, n
);
8201 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8202 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8204 /* check for new failures */
8205 if (state
& DS_FAULTY
) {
8206 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8207 super
->updates_pending
++;
8210 /* check if in_sync */
8211 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8212 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8214 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8215 super
->updates_pending
++;
8218 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8219 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8221 /* check if recovery complete, newly degraded, or failed */
8222 dprintf("imsm: Detected transition to state ");
8223 switch (map_state
) {
8224 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8225 dprintf("normal: ");
8226 if (is_rebuilding(dev
)) {
8227 dprintf_cont("while rebuilding");
8228 /* check if recovery is really finished */
8229 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8230 if (mdi
->recovery_start
!= MaxSector
) {
8231 recovery_not_finished
= 1;
8234 if (recovery_not_finished
) {
8236 dprintf("Rebuild has not finished yet, state not changed");
8237 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8238 a
->last_checkpoint
= mdi
->recovery_start
;
8239 super
->updates_pending
++;
8243 end_migration(dev
, super
, map_state
);
8244 map
= get_imsm_map(dev
, MAP_0
);
8245 map
->failed_disk_num
= ~0;
8246 super
->updates_pending
++;
8247 a
->last_checkpoint
= 0;
8250 if (is_gen_migration(dev
)) {
8251 dprintf_cont("while general migration");
8252 if (a
->last_checkpoint
>= a
->info
.component_size
)
8253 end_migration(dev
, super
, map_state
);
8255 map
->map_state
= map_state
;
8256 map
= get_imsm_map(dev
, MAP_0
);
8257 map
->failed_disk_num
= ~0;
8258 super
->updates_pending
++;
8262 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8263 dprintf_cont("degraded: ");
8264 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8265 dprintf_cont("mark degraded");
8266 map
->map_state
= map_state
;
8267 super
->updates_pending
++;
8268 a
->last_checkpoint
= 0;
8271 if (is_rebuilding(dev
)) {
8272 dprintf_cont("while rebuilding.");
8273 if (map
->map_state
!= map_state
) {
8274 dprintf_cont(" Map state change");
8275 end_migration(dev
, super
, map_state
);
8276 super
->updates_pending
++;
8280 if (is_gen_migration(dev
)) {
8281 dprintf_cont("while general migration");
8282 if (a
->last_checkpoint
>= a
->info
.component_size
)
8283 end_migration(dev
, super
, map_state
);
8285 map
->map_state
= map_state
;
8286 manage_second_map(super
, dev
);
8288 super
->updates_pending
++;
8291 if (is_initializing(dev
)) {
8292 dprintf_cont("while initialization.");
8293 map
->map_state
= map_state
;
8294 super
->updates_pending
++;
8298 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8299 dprintf_cont("failed: ");
8300 if (is_gen_migration(dev
)) {
8301 dprintf_cont("while general migration");
8302 map
->map_state
= map_state
;
8303 super
->updates_pending
++;
8306 if (map
->map_state
!= map_state
) {
8307 dprintf_cont("mark failed");
8308 end_migration(dev
, super
, map_state
);
8309 super
->updates_pending
++;
8310 a
->last_checkpoint
= 0;
8315 dprintf_cont("state %i\n", map_state
);
8320 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8323 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8324 unsigned long long dsize
;
8325 unsigned long long sectors
;
8326 unsigned int sector_size
;
8328 get_dev_sector_size(fd
, NULL
, §or_size
);
8329 get_dev_size(fd
, NULL
, &dsize
);
8331 if (mpb_size
> sector_size
) {
8332 /* -1 to account for anchor */
8333 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8335 /* write the extended mpb to the sectors preceeding the anchor */
8336 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8340 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8341 sector_size
* sectors
) != sector_size
* sectors
)
8345 /* first block is stored on second to last sector of the disk */
8346 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8349 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8355 static void imsm_sync_metadata(struct supertype
*container
)
8357 struct intel_super
*super
= container
->sb
;
8359 dprintf("sync metadata: %d\n", super
->updates_pending
);
8360 if (!super
->updates_pending
)
8363 write_super_imsm(container
, 0);
8365 super
->updates_pending
= 0;
8368 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8370 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8371 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8374 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8378 if (dl
&& is_failed(&dl
->disk
))
8382 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8387 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8388 struct active_array
*a
, int activate_new
,
8389 struct mdinfo
*additional_test_list
)
8391 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8392 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8393 struct imsm_super
*mpb
= super
->anchor
;
8394 struct imsm_map
*map
;
8395 unsigned long long pos
;
8400 __u32 array_start
= 0;
8401 __u32 array_end
= 0;
8403 struct mdinfo
*test_list
;
8405 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8406 /* If in this array, skip */
8407 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8408 if (d
->state_fd
>= 0 &&
8409 d
->disk
.major
== dl
->major
&&
8410 d
->disk
.minor
== dl
->minor
) {
8411 dprintf("%x:%x already in array\n",
8412 dl
->major
, dl
->minor
);
8417 test_list
= additional_test_list
;
8419 if (test_list
->disk
.major
== dl
->major
&&
8420 test_list
->disk
.minor
== dl
->minor
) {
8421 dprintf("%x:%x already in additional test list\n",
8422 dl
->major
, dl
->minor
);
8425 test_list
= test_list
->next
;
8430 /* skip in use or failed drives */
8431 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8433 dprintf("%x:%x status (failed: %d index: %d)\n",
8434 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8438 /* skip pure spares when we are looking for partially
8439 * assimilated drives
8441 if (dl
->index
== -1 && !activate_new
)
8444 /* Does this unused device have the requisite free space?
8445 * It needs to be able to cover all member volumes
8447 ex
= get_extents(super
, dl
);
8449 dprintf("cannot get extents\n");
8452 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8453 dev
= get_imsm_dev(super
, i
);
8454 map
= get_imsm_map(dev
, MAP_0
);
8456 /* check if this disk is already a member of
8459 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8465 array_start
= pba_of_lba0(map
);
8466 array_end
= array_start
+
8467 blocks_per_member(map
) - 1;
8470 /* check that we can start at pba_of_lba0 with
8471 * blocks_per_member of space
8473 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8477 pos
= ex
[j
].start
+ ex
[j
].size
;
8479 } while (ex
[j
-1].size
);
8486 if (i
< mpb
->num_raid_devs
) {
8487 dprintf("%x:%x does not have %u to %u available\n",
8488 dl
->major
, dl
->minor
, array_start
, array_end
);
8498 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8500 struct imsm_dev
*dev2
;
8501 struct imsm_map
*map
;
8507 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8509 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8510 if (state
== IMSM_T_STATE_FAILED
) {
8511 map
= get_imsm_map(dev2
, MAP_0
);
8514 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8516 * Check if failed disks are deleted from intel
8517 * disk list or are marked to be deleted
8519 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8520 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8522 * Do not rebuild the array if failed disks
8523 * from failed sub-array are not removed from
8527 is_failed(&idisk
->disk
) &&
8528 (idisk
->action
!= DISK_REMOVE
))
8536 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8537 struct metadata_update
**updates
)
8540 * Find a device with unused free space and use it to replace a
8541 * failed/vacant region in an array. We replace failed regions one a
8542 * array at a time. The result is that a new spare disk will be added
8543 * to the first failed array and after the monitor has finished
8544 * propagating failures the remainder will be consumed.
8546 * FIXME add a capability for mdmon to request spares from another
8550 struct intel_super
*super
= a
->container
->sb
;
8551 int inst
= a
->info
.container_member
;
8552 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8553 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8554 int failed
= a
->info
.array
.raid_disks
;
8555 struct mdinfo
*rv
= NULL
;
8558 struct metadata_update
*mu
;
8560 struct imsm_update_activate_spare
*u
;
8565 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8566 if ((d
->curr_state
& DS_FAULTY
) &&
8568 /* wait for Removal to happen */
8570 if (d
->state_fd
>= 0)
8574 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8575 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8577 if (imsm_reshape_blocks_arrays_changes(super
))
8580 /* Cannot activate another spare if rebuild is in progress already
8582 if (is_rebuilding(dev
)) {
8583 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8587 if (a
->info
.array
.level
== 4)
8588 /* No repair for takeovered array
8589 * imsm doesn't support raid4
8593 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8594 IMSM_T_STATE_DEGRADED
)
8597 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8598 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8603 * If there are any failed disks check state of the other volume.
8604 * Block rebuild if the another one is failed until failed disks
8605 * are removed from container.
8608 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8609 MAX_RAID_SERIAL_LEN
, dev
->volume
);
8610 /* check if states of the other volumes allow for rebuild */
8611 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
8613 allowed
= imsm_rebuild_allowed(a
->container
,
8621 /* For each slot, if it is not working, find a spare */
8622 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
8623 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8624 if (d
->disk
.raid_disk
== i
)
8626 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
8627 if (d
&& (d
->state_fd
>= 0))
8631 * OK, this device needs recovery. Try to re-add the
8632 * previous occupant of this slot, if this fails see if
8633 * we can continue the assimilation of a spare that was
8634 * partially assimilated, finally try to activate a new
8637 dl
= imsm_readd(super
, i
, a
);
8639 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
8641 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
8645 /* found a usable disk with enough space */
8646 di
= xcalloc(1, sizeof(*di
));
8648 /* dl->index will be -1 in the case we are activating a
8649 * pristine spare. imsm_process_update() will create a
8650 * new index in this case. Once a disk is found to be
8651 * failed in all member arrays it is kicked from the
8654 di
->disk
.number
= dl
->index
;
8656 /* (ab)use di->devs to store a pointer to the device
8659 di
->devs
= (struct mdinfo
*) dl
;
8661 di
->disk
.raid_disk
= i
;
8662 di
->disk
.major
= dl
->major
;
8663 di
->disk
.minor
= dl
->minor
;
8665 di
->recovery_start
= 0;
8666 di
->data_offset
= pba_of_lba0(map
);
8667 di
->component_size
= a
->info
.component_size
;
8668 di
->container_member
= inst
;
8669 di
->bb
.supported
= 1;
8670 if (dev
->rwh_policy
== RWH_DISTRIBUTED
) {
8671 di
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
8672 di
->ppl_sector
= get_ppl_sector(super
, inst
);
8673 di
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
) >> 9;
8675 super
->random
= random32();
8679 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
8680 i
, di
->data_offset
);
8684 /* No spares found */
8686 /* Now 'rv' has a list of devices to return.
8687 * Create a metadata_update record to update the
8688 * disk_ord_tbl for the array
8690 mu
= xmalloc(sizeof(*mu
));
8691 mu
->buf
= xcalloc(num_spares
,
8692 sizeof(struct imsm_update_activate_spare
));
8694 mu
->space_list
= NULL
;
8695 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
8696 mu
->next
= *updates
;
8697 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
8699 for (di
= rv
; di
; di
= di
->next
) {
8700 u
->type
= update_activate_spare
;
8701 u
->dl
= (struct dl
*) di
->devs
;
8703 u
->slot
= di
->disk
.raid_disk
;
8714 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
8716 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
8717 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8718 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8719 struct disk_info
*inf
= get_disk_info(u
);
8720 struct imsm_disk
*disk
;
8724 for (i
= 0; i
< map
->num_members
; i
++) {
8725 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
8726 for (j
= 0; j
< new_map
->num_members
; j
++)
8727 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
8734 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
8738 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8739 if (dl
->major
== major
&& dl
->minor
== minor
)
8744 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
8750 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8751 if (dl
->major
== major
&& dl
->minor
== minor
) {
8754 prev
->next
= dl
->next
;
8756 super
->disks
= dl
->next
;
8758 __free_imsm_disk(dl
);
8759 dprintf("removed %x:%x\n", major
, minor
);
8767 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
8769 static int add_remove_disk_update(struct intel_super
*super
)
8771 int check_degraded
= 0;
8774 /* add/remove some spares to/from the metadata/contrainer */
8775 while (super
->disk_mgmt_list
) {
8776 struct dl
*disk_cfg
;
8778 disk_cfg
= super
->disk_mgmt_list
;
8779 super
->disk_mgmt_list
= disk_cfg
->next
;
8780 disk_cfg
->next
= NULL
;
8782 if (disk_cfg
->action
== DISK_ADD
) {
8783 disk_cfg
->next
= super
->disks
;
8784 super
->disks
= disk_cfg
;
8786 dprintf("added %x:%x\n",
8787 disk_cfg
->major
, disk_cfg
->minor
);
8788 } else if (disk_cfg
->action
== DISK_REMOVE
) {
8789 dprintf("Disk remove action processed: %x.%x\n",
8790 disk_cfg
->major
, disk_cfg
->minor
);
8791 disk
= get_disk_super(super
,
8795 /* store action status */
8796 disk
->action
= DISK_REMOVE
;
8797 /* remove spare disks only */
8798 if (disk
->index
== -1) {
8799 remove_disk_super(super
,
8804 /* release allocate disk structure */
8805 __free_imsm_disk(disk_cfg
);
8808 return check_degraded
;
8811 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
8812 struct intel_super
*super
,
8815 struct intel_dev
*id
;
8816 void **tofree
= NULL
;
8819 dprintf("(enter)\n");
8820 if (u
->subdev
< 0 || u
->subdev
> 1) {
8821 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8824 if (space_list
== NULL
|| *space_list
== NULL
) {
8825 dprintf("imsm: Error: Memory is not allocated\n");
8829 for (id
= super
->devlist
; id
; id
= id
->next
) {
8830 if (id
->index
== (unsigned)u
->subdev
) {
8831 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8832 struct imsm_map
*map
;
8833 struct imsm_dev
*new_dev
=
8834 (struct imsm_dev
*)*space_list
;
8835 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8837 struct dl
*new_disk
;
8839 if (new_dev
== NULL
)
8841 *space_list
= **space_list
;
8842 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
8843 map
= get_imsm_map(new_dev
, MAP_0
);
8845 dprintf("imsm: Error: migration in progress");
8849 to_state
= map
->map_state
;
8850 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
8852 /* this should not happen */
8853 if (u
->new_disks
[0] < 0) {
8854 map
->failed_disk_num
=
8855 map
->num_members
- 1;
8856 to_state
= IMSM_T_STATE_DEGRADED
;
8858 to_state
= IMSM_T_STATE_NORMAL
;
8860 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
8861 if (u
->new_level
> -1)
8862 map
->raid_level
= u
->new_level
;
8863 migr_map
= get_imsm_map(new_dev
, MAP_1
);
8864 if ((u
->new_level
== 5) &&
8865 (migr_map
->raid_level
== 0)) {
8866 int ord
= map
->num_members
- 1;
8867 migr_map
->num_members
--;
8868 if (u
->new_disks
[0] < 0)
8869 ord
|= IMSM_ORD_REBUILD
;
8870 set_imsm_ord_tbl_ent(map
,
8871 map
->num_members
- 1,
8875 tofree
= (void **)dev
;
8877 /* update chunk size
8879 if (u
->new_chunksize
> 0) {
8880 unsigned long long num_data_stripes
;
8882 imsm_num_data_members(dev
, MAP_0
);
8884 if (used_disks
== 0)
8887 map
->blocks_per_strip
=
8888 __cpu_to_le16(u
->new_chunksize
* 2);
8890 (join_u32(dev
->size_low
, dev
->size_high
)
8892 num_data_stripes
/= map
->blocks_per_strip
;
8893 num_data_stripes
/= map
->num_domains
;
8894 set_num_data_stripes(map
, num_data_stripes
);
8899 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
8900 migr_map
->raid_level
== map
->raid_level
)
8903 if (u
->new_disks
[0] >= 0) {
8906 new_disk
= get_disk_super(super
,
8907 major(u
->new_disks
[0]),
8908 minor(u
->new_disks
[0]));
8909 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8910 major(u
->new_disks
[0]),
8911 minor(u
->new_disks
[0]),
8912 new_disk
, new_disk
->index
);
8913 if (new_disk
== NULL
)
8914 goto error_disk_add
;
8916 new_disk
->index
= map
->num_members
- 1;
8917 /* slot to fill in autolayout
8919 new_disk
->raiddisk
= new_disk
->index
;
8920 new_disk
->disk
.status
|= CONFIGURED_DISK
;
8921 new_disk
->disk
.status
&= ~SPARE_DISK
;
8923 goto error_disk_add
;
8926 *tofree
= *space_list
;
8927 /* calculate new size
8929 imsm_set_array_size(new_dev
, -1);
8936 *space_list
= tofree
;
8940 dprintf("Error: imsm: Cannot find disk.\n");
8944 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8945 struct intel_super
*super
)
8947 struct intel_dev
*id
;
8950 dprintf("(enter)\n");
8951 if (u
->subdev
< 0 || u
->subdev
> 1) {
8952 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8956 for (id
= super
->devlist
; id
; id
= id
->next
) {
8957 if (id
->index
== (unsigned)u
->subdev
) {
8958 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8959 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8960 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8961 unsigned long long blocks_per_member
;
8962 unsigned long long num_data_stripes
;
8964 /* calculate new size
8966 blocks_per_member
= u
->new_size
/ used_disks
;
8967 num_data_stripes
= blocks_per_member
/
8968 map
->blocks_per_strip
;
8969 num_data_stripes
/= map
->num_domains
;
8970 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
8971 u
->new_size
, blocks_per_member
,
8973 set_blocks_per_member(map
, blocks_per_member
);
8974 set_num_data_stripes(map
, num_data_stripes
);
8975 imsm_set_array_size(dev
, u
->new_size
);
8985 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8986 struct intel_super
*super
,
8987 struct active_array
*active_array
)
8989 struct imsm_super
*mpb
= super
->anchor
;
8990 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8991 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8992 struct imsm_map
*migr_map
;
8993 struct active_array
*a
;
8994 struct imsm_disk
*disk
;
9001 int second_map_created
= 0;
9003 for (; u
; u
= u
->next
) {
9004 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9009 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9014 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9019 /* count failures (excluding rebuilds and the victim)
9020 * to determine map[0] state
9023 for (i
= 0; i
< map
->num_members
; i
++) {
9026 disk
= get_imsm_disk(super
,
9027 get_imsm_disk_idx(dev
, i
, MAP_X
));
9028 if (!disk
|| is_failed(disk
))
9032 /* adding a pristine spare, assign a new index */
9033 if (dl
->index
< 0) {
9034 dl
->index
= super
->anchor
->num_disks
;
9035 super
->anchor
->num_disks
++;
9038 disk
->status
|= CONFIGURED_DISK
;
9039 disk
->status
&= ~SPARE_DISK
;
9042 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9043 if (!second_map_created
) {
9044 second_map_created
= 1;
9045 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9046 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9048 map
->map_state
= to_state
;
9049 migr_map
= get_imsm_map(dev
, MAP_1
);
9050 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9051 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9052 dl
->index
| IMSM_ORD_REBUILD
);
9054 /* update the family_num to mark a new container
9055 * generation, being careful to record the existing
9056 * family_num in orig_family_num to clean up after
9057 * earlier mdadm versions that neglected to set it.
9059 if (mpb
->orig_family_num
== 0)
9060 mpb
->orig_family_num
= mpb
->family_num
;
9061 mpb
->family_num
+= super
->random
;
9063 /* count arrays using the victim in the metadata */
9065 for (a
= active_array
; a
; a
= a
->next
) {
9066 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9067 map
= get_imsm_map(dev
, MAP_0
);
9069 if (get_imsm_disk_slot(map
, victim
) >= 0)
9073 /* delete the victim if it is no longer being
9079 /* We know that 'manager' isn't touching anything,
9080 * so it is safe to delete
9082 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9083 if ((*dlp
)->index
== victim
)
9086 /* victim may be on the missing list */
9088 for (dlp
= &super
->missing
; *dlp
;
9089 dlp
= &(*dlp
)->next
)
9090 if ((*dlp
)->index
== victim
)
9092 imsm_delete(super
, dlp
, victim
);
9099 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9100 struct intel_super
*super
,
9103 struct dl
*new_disk
;
9104 struct intel_dev
*id
;
9106 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9107 int disk_count
= u
->old_raid_disks
;
9108 void **tofree
= NULL
;
9109 int devices_to_reshape
= 1;
9110 struct imsm_super
*mpb
= super
->anchor
;
9112 unsigned int dev_id
;
9114 dprintf("(enter)\n");
9116 /* enable spares to use in array */
9117 for (i
= 0; i
< delta_disks
; i
++) {
9118 new_disk
= get_disk_super(super
,
9119 major(u
->new_disks
[i
]),
9120 minor(u
->new_disks
[i
]));
9121 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9122 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9123 new_disk
, new_disk
->index
);
9124 if (new_disk
== NULL
||
9125 (new_disk
->index
>= 0 &&
9126 new_disk
->index
< u
->old_raid_disks
))
9127 goto update_reshape_exit
;
9128 new_disk
->index
= disk_count
++;
9129 /* slot to fill in autolayout
9131 new_disk
->raiddisk
= new_disk
->index
;
9132 new_disk
->disk
.status
|=
9134 new_disk
->disk
.status
&= ~SPARE_DISK
;
9137 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9138 mpb
->num_raid_devs
);
9139 /* manage changes in volume
9141 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9142 void **sp
= *space_list
;
9143 struct imsm_dev
*newdev
;
9144 struct imsm_map
*newmap
, *oldmap
;
9146 for (id
= super
->devlist
; id
; id
= id
->next
) {
9147 if (id
->index
== dev_id
)
9156 /* Copy the dev, but not (all of) the map */
9157 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9158 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9159 newmap
= get_imsm_map(newdev
, MAP_0
);
9160 /* Copy the current map */
9161 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9162 /* update one device only
9164 if (devices_to_reshape
) {
9165 dprintf("imsm: modifying subdev: %i\n",
9167 devices_to_reshape
--;
9168 newdev
->vol
.migr_state
= 1;
9169 newdev
->vol
.curr_migr_unit
= 0;
9170 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9171 newmap
->num_members
= u
->new_raid_disks
;
9172 for (i
= 0; i
< delta_disks
; i
++) {
9173 set_imsm_ord_tbl_ent(newmap
,
9174 u
->old_raid_disks
+ i
,
9175 u
->old_raid_disks
+ i
);
9177 /* New map is correct, now need to save old map
9179 newmap
= get_imsm_map(newdev
, MAP_1
);
9180 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9182 imsm_set_array_size(newdev
, -1);
9185 sp
= (void **)id
->dev
;
9190 /* Clear migration record */
9191 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9194 *space_list
= tofree
;
9197 update_reshape_exit
:
9202 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9203 struct intel_super
*super
,
9206 struct imsm_dev
*dev
= NULL
;
9207 struct intel_dev
*dv
;
9208 struct imsm_dev
*dev_new
;
9209 struct imsm_map
*map
;
9213 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9214 if (dv
->index
== (unsigned int)u
->subarray
) {
9222 map
= get_imsm_map(dev
, MAP_0
);
9224 if (u
->direction
== R10_TO_R0
) {
9225 unsigned long long num_data_stripes
;
9227 map
->num_domains
= 1;
9228 num_data_stripes
= blocks_per_member(map
);
9229 num_data_stripes
/= map
->blocks_per_strip
;
9230 num_data_stripes
/= map
->num_domains
;
9231 set_num_data_stripes(map
, num_data_stripes
);
9233 /* Number of failed disks must be half of initial disk number */
9234 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9235 (map
->num_members
/ 2))
9238 /* iterate through devices to mark removed disks as spare */
9239 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9240 if (dm
->disk
.status
& FAILED_DISK
) {
9241 int idx
= dm
->index
;
9242 /* update indexes on the disk list */
9243 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9244 the index values will end up being correct.... NB */
9245 for (du
= super
->disks
; du
; du
= du
->next
)
9246 if (du
->index
> idx
)
9248 /* mark as spare disk */
9253 map
->num_members
= map
->num_members
/ 2;
9254 map
->map_state
= IMSM_T_STATE_NORMAL
;
9255 map
->num_domains
= 1;
9256 map
->raid_level
= 0;
9257 map
->failed_disk_num
= -1;
9260 if (u
->direction
== R0_TO_R10
) {
9262 /* update slots in current disk list */
9263 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9267 /* create new *missing* disks */
9268 for (i
= 0; i
< map
->num_members
; i
++) {
9269 space
= *space_list
;
9272 *space_list
= *space
;
9274 memcpy(du
, super
->disks
, sizeof(*du
));
9278 du
->index
= (i
* 2) + 1;
9279 sprintf((char *)du
->disk
.serial
,
9280 " MISSING_%d", du
->index
);
9281 sprintf((char *)du
->serial
,
9282 "MISSING_%d", du
->index
);
9283 du
->next
= super
->missing
;
9284 super
->missing
= du
;
9286 /* create new dev and map */
9287 space
= *space_list
;
9290 *space_list
= *space
;
9291 dev_new
= (void *)space
;
9292 memcpy(dev_new
, dev
, sizeof(*dev
));
9293 /* update new map */
9294 map
= get_imsm_map(dev_new
, MAP_0
);
9295 map
->num_members
= map
->num_members
* 2;
9296 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9297 map
->num_domains
= 2;
9298 map
->raid_level
= 1;
9299 /* replace dev<->dev_new */
9302 /* update disk order table */
9303 for (du
= super
->disks
; du
; du
= du
->next
)
9305 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9306 for (du
= super
->missing
; du
; du
= du
->next
)
9307 if (du
->index
>= 0) {
9308 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9309 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9315 static void imsm_process_update(struct supertype
*st
,
9316 struct metadata_update
*update
)
9319 * crack open the metadata_update envelope to find the update record
9320 * update can be one of:
9321 * update_reshape_container_disks - all the arrays in the container
9322 * are being reshaped to have more devices. We need to mark
9323 * the arrays for general migration and convert selected spares
9324 * into active devices.
9325 * update_activate_spare - a spare device has replaced a failed
9326 * device in an array, update the disk_ord_tbl. If this disk is
9327 * present in all member arrays then also clear the SPARE_DISK
9329 * update_create_array
9331 * update_rename_array
9332 * update_add_remove_disk
9334 struct intel_super
*super
= st
->sb
;
9335 struct imsm_super
*mpb
;
9336 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9338 /* update requires a larger buf but the allocation failed */
9339 if (super
->next_len
&& !super
->next_buf
) {
9340 super
->next_len
= 0;
9344 if (super
->next_buf
) {
9345 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9347 super
->len
= super
->next_len
;
9348 super
->buf
= super
->next_buf
;
9350 super
->next_len
= 0;
9351 super
->next_buf
= NULL
;
9354 mpb
= super
->anchor
;
9357 case update_general_migration_checkpoint
: {
9358 struct intel_dev
*id
;
9359 struct imsm_update_general_migration_checkpoint
*u
=
9360 (void *)update
->buf
;
9362 dprintf("called for update_general_migration_checkpoint\n");
9364 /* find device under general migration */
9365 for (id
= super
->devlist
; id
; id
= id
->next
) {
9366 if (is_gen_migration(id
->dev
)) {
9367 id
->dev
->vol
.curr_migr_unit
=
9368 __cpu_to_le32(u
->curr_migr_unit
);
9369 super
->updates_pending
++;
9374 case update_takeover
: {
9375 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9376 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9377 imsm_update_version_info(super
);
9378 super
->updates_pending
++;
9383 case update_reshape_container_disks
: {
9384 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9385 if (apply_reshape_container_disks_update(
9386 u
, super
, &update
->space_list
))
9387 super
->updates_pending
++;
9390 case update_reshape_migration
: {
9391 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9392 if (apply_reshape_migration_update(
9393 u
, super
, &update
->space_list
))
9394 super
->updates_pending
++;
9397 case update_size_change
: {
9398 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9399 if (apply_size_change_update(u
, super
))
9400 super
->updates_pending
++;
9403 case update_activate_spare
: {
9404 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9405 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9406 super
->updates_pending
++;
9409 case update_create_array
: {
9410 /* someone wants to create a new array, we need to be aware of
9411 * a few races/collisions:
9412 * 1/ 'Create' called by two separate instances of mdadm
9413 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9414 * devices that have since been assimilated via
9416 * In the event this update can not be carried out mdadm will
9417 * (FIX ME) notice that its update did not take hold.
9419 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9420 struct intel_dev
*dv
;
9421 struct imsm_dev
*dev
;
9422 struct imsm_map
*map
, *new_map
;
9423 unsigned long long start
, end
;
9424 unsigned long long new_start
, new_end
;
9426 struct disk_info
*inf
;
9429 /* handle racing creates: first come first serve */
9430 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9431 dprintf("subarray %d already defined\n", u
->dev_idx
);
9435 /* check update is next in sequence */
9436 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9437 dprintf("can not create array %d expected index %d\n",
9438 u
->dev_idx
, mpb
->num_raid_devs
);
9442 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9443 new_start
= pba_of_lba0(new_map
);
9444 new_end
= new_start
+ blocks_per_member(new_map
);
9445 inf
= get_disk_info(u
);
9447 /* handle activate_spare versus create race:
9448 * check to make sure that overlapping arrays do not include
9451 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9452 dev
= get_imsm_dev(super
, i
);
9453 map
= get_imsm_map(dev
, MAP_0
);
9454 start
= pba_of_lba0(map
);
9455 end
= start
+ blocks_per_member(map
);
9456 if ((new_start
>= start
&& new_start
<= end
) ||
9457 (start
>= new_start
&& start
<= new_end
))
9462 if (disks_overlap(super
, i
, u
)) {
9463 dprintf("arrays overlap\n");
9468 /* check that prepare update was successful */
9469 if (!update
->space
) {
9470 dprintf("prepare update failed\n");
9474 /* check that all disks are still active before committing
9475 * changes. FIXME: could we instead handle this by creating a
9476 * degraded array? That's probably not what the user expects,
9477 * so better to drop this update on the floor.
9479 for (i
= 0; i
< new_map
->num_members
; i
++) {
9480 dl
= serial_to_dl(inf
[i
].serial
, super
);
9482 dprintf("disk disappeared\n");
9487 super
->updates_pending
++;
9489 /* convert spares to members and fixup ord_tbl */
9490 for (i
= 0; i
< new_map
->num_members
; i
++) {
9491 dl
= serial_to_dl(inf
[i
].serial
, super
);
9492 if (dl
->index
== -1) {
9493 dl
->index
= mpb
->num_disks
;
9495 dl
->disk
.status
|= CONFIGURED_DISK
;
9496 dl
->disk
.status
&= ~SPARE_DISK
;
9498 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9503 update
->space
= NULL
;
9504 imsm_copy_dev(dev
, &u
->dev
);
9505 dv
->index
= u
->dev_idx
;
9506 dv
->next
= super
->devlist
;
9507 super
->devlist
= dv
;
9508 mpb
->num_raid_devs
++;
9510 imsm_update_version_info(super
);
9513 /* mdmon knows how to release update->space, but not
9514 * ((struct intel_dev *) update->space)->dev
9516 if (update
->space
) {
9522 case update_kill_array
: {
9523 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9524 int victim
= u
->dev_idx
;
9525 struct active_array
*a
;
9526 struct intel_dev
**dp
;
9527 struct imsm_dev
*dev
;
9529 /* sanity check that we are not affecting the uuid of
9530 * active arrays, or deleting an active array
9532 * FIXME when immutable ids are available, but note that
9533 * we'll also need to fixup the invalidated/active
9534 * subarray indexes in mdstat
9536 for (a
= st
->arrays
; a
; a
= a
->next
)
9537 if (a
->info
.container_member
>= victim
)
9539 /* by definition if mdmon is running at least one array
9540 * is active in the container, so checking
9541 * mpb->num_raid_devs is just extra paranoia
9543 dev
= get_imsm_dev(super
, victim
);
9544 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9545 dprintf("failed to delete subarray-%d\n", victim
);
9549 for (dp
= &super
->devlist
; *dp
;)
9550 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9553 if ((*dp
)->index
> (unsigned)victim
)
9557 mpb
->num_raid_devs
--;
9558 super
->updates_pending
++;
9561 case update_rename_array
: {
9562 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9563 char name
[MAX_RAID_SERIAL_LEN
+1];
9564 int target
= u
->dev_idx
;
9565 struct active_array
*a
;
9566 struct imsm_dev
*dev
;
9568 /* sanity check that we are not affecting the uuid of
9571 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9572 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9573 for (a
= st
->arrays
; a
; a
= a
->next
)
9574 if (a
->info
.container_member
== target
)
9576 dev
= get_imsm_dev(super
, u
->dev_idx
);
9577 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9578 dprintf("failed to rename subarray-%d\n", target
);
9582 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
9583 super
->updates_pending
++;
9586 case update_add_remove_disk
: {
9587 /* we may be able to repair some arrays if disks are
9588 * being added, check the status of add_remove_disk
9589 * if discs has been added.
9591 if (add_remove_disk_update(super
)) {
9592 struct active_array
*a
;
9594 super
->updates_pending
++;
9595 for (a
= st
->arrays
; a
; a
= a
->next
)
9596 a
->check_degraded
= 1;
9600 case update_prealloc_badblocks_mem
:
9603 pr_err("error: unsuported process update type:(type: %d)\n", type
);
9607 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
9609 static int imsm_prepare_update(struct supertype
*st
,
9610 struct metadata_update
*update
)
9613 * Allocate space to hold new disk entries, raid-device entries or a new
9614 * mpb if necessary. The manager synchronously waits for updates to
9615 * complete in the monitor, so new mpb buffers allocated here can be
9616 * integrated by the monitor thread without worrying about live pointers
9617 * in the manager thread.
9619 enum imsm_update_type type
;
9620 struct intel_super
*super
= st
->sb
;
9621 unsigned int sector_size
= super
->sector_size
;
9622 struct imsm_super
*mpb
= super
->anchor
;
9626 if (update
->len
< (int)sizeof(type
))
9629 type
= *(enum imsm_update_type
*) update
->buf
;
9632 case update_general_migration_checkpoint
:
9633 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
9635 dprintf("called for update_general_migration_checkpoint\n");
9637 case update_takeover
: {
9638 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9639 if (update
->len
< (int)sizeof(*u
))
9641 if (u
->direction
== R0_TO_R10
) {
9642 void **tail
= (void **)&update
->space_list
;
9643 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
9644 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9645 int num_members
= map
->num_members
;
9648 /* allocate memory for added disks */
9649 for (i
= 0; i
< num_members
; i
++) {
9650 size
= sizeof(struct dl
);
9651 space
= xmalloc(size
);
9656 /* allocate memory for new device */
9657 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
9658 (num_members
* sizeof(__u32
));
9659 space
= xmalloc(size
);
9663 len
= disks_to_mpb_size(num_members
* 2);
9668 case update_reshape_container_disks
: {
9669 /* Every raid device in the container is about to
9670 * gain some more devices, and we will enter a
9672 * So each 'imsm_map' will be bigger, and the imsm_vol
9673 * will now hold 2 of them.
9674 * Thus we need new 'struct imsm_dev' allocations sized
9675 * as sizeof_imsm_dev but with more devices in both maps.
9677 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9678 struct intel_dev
*dl
;
9679 void **space_tail
= (void**)&update
->space_list
;
9681 if (update
->len
< (int)sizeof(*u
))
9684 dprintf("for update_reshape\n");
9686 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
9687 int size
= sizeof_imsm_dev(dl
->dev
, 1);
9689 if (u
->new_raid_disks
> u
->old_raid_disks
)
9690 size
+= sizeof(__u32
)*2*
9691 (u
->new_raid_disks
- u
->old_raid_disks
);
9698 len
= disks_to_mpb_size(u
->new_raid_disks
);
9699 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9702 case update_reshape_migration
: {
9703 /* for migration level 0->5 we need to add disks
9704 * so the same as for container operation we will copy
9705 * device to the bigger location.
9706 * in memory prepared device and new disk area are prepared
9707 * for usage in process update
9709 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9710 struct intel_dev
*id
;
9711 void **space_tail
= (void **)&update
->space_list
;
9714 int current_level
= -1;
9716 if (update
->len
< (int)sizeof(*u
))
9719 dprintf("for update_reshape\n");
9721 /* add space for bigger array in update
9723 for (id
= super
->devlist
; id
; id
= id
->next
) {
9724 if (id
->index
== (unsigned)u
->subdev
) {
9725 size
= sizeof_imsm_dev(id
->dev
, 1);
9726 if (u
->new_raid_disks
> u
->old_raid_disks
)
9727 size
+= sizeof(__u32
)*2*
9728 (u
->new_raid_disks
- u
->old_raid_disks
);
9736 if (update
->space_list
== NULL
)
9739 /* add space for disk in update
9741 size
= sizeof(struct dl
);
9747 /* add spare device to update
9749 for (id
= super
->devlist
; id
; id
= id
->next
)
9750 if (id
->index
== (unsigned)u
->subdev
) {
9751 struct imsm_dev
*dev
;
9752 struct imsm_map
*map
;
9754 dev
= get_imsm_dev(super
, u
->subdev
);
9755 map
= get_imsm_map(dev
, MAP_0
);
9756 current_level
= map
->raid_level
;
9759 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
9760 struct mdinfo
*spares
;
9762 spares
= get_spares_for_grow(st
);
9770 makedev(dev
->disk
.major
,
9772 dl
= get_disk_super(super
,
9775 dl
->index
= u
->old_raid_disks
;
9781 len
= disks_to_mpb_size(u
->new_raid_disks
);
9782 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9785 case update_size_change
: {
9786 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
9790 case update_activate_spare
: {
9791 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
9795 case update_create_array
: {
9796 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9797 struct intel_dev
*dv
;
9798 struct imsm_dev
*dev
= &u
->dev
;
9799 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9801 struct disk_info
*inf
;
9805 if (update
->len
< (int)sizeof(*u
))
9808 inf
= get_disk_info(u
);
9809 len
= sizeof_imsm_dev(dev
, 1);
9810 /* allocate a new super->devlist entry */
9811 dv
= xmalloc(sizeof(*dv
));
9812 dv
->dev
= xmalloc(len
);
9815 /* count how many spares will be converted to members */
9816 for (i
= 0; i
< map
->num_members
; i
++) {
9817 dl
= serial_to_dl(inf
[i
].serial
, super
);
9819 /* hmm maybe it failed?, nothing we can do about
9824 if (count_memberships(dl
, super
) == 0)
9827 len
+= activate
* sizeof(struct imsm_disk
);
9830 case update_kill_array
: {
9831 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
9835 case update_rename_array
: {
9836 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
9840 case update_add_remove_disk
:
9841 /* no update->len needed */
9843 case update_prealloc_badblocks_mem
:
9844 super
->extra_space
+= sizeof(struct bbm_log
) -
9845 get_imsm_bbm_log_size(super
->bbm_log
);
9851 /* check if we need a larger metadata buffer */
9852 if (super
->next_buf
)
9853 buf_len
= super
->next_len
;
9855 buf_len
= super
->len
;
9857 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
9858 /* ok we need a larger buf than what is currently allocated
9859 * if this allocation fails process_update will notice that
9860 * ->next_len is set and ->next_buf is NULL
9862 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
9863 super
->extra_space
+ len
, sector_size
);
9864 if (super
->next_buf
)
9865 free(super
->next_buf
);
9867 super
->next_len
= buf_len
;
9868 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
9869 memset(super
->next_buf
, 0, buf_len
);
9871 super
->next_buf
= NULL
;
9876 /* must be called while manager is quiesced */
9877 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
9879 struct imsm_super
*mpb
= super
->anchor
;
9881 struct imsm_dev
*dev
;
9882 struct imsm_map
*map
;
9883 unsigned int i
, j
, num_members
;
9885 struct bbm_log
*log
= super
->bbm_log
;
9887 dprintf("deleting device[%d] from imsm_super\n", index
);
9889 /* shift all indexes down one */
9890 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
9891 if (iter
->index
> (int)index
)
9893 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
9894 if (iter
->index
> (int)index
)
9897 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9898 dev
= get_imsm_dev(super
, i
);
9899 map
= get_imsm_map(dev
, MAP_0
);
9900 num_members
= map
->num_members
;
9901 for (j
= 0; j
< num_members
; j
++) {
9902 /* update ord entries being careful not to propagate
9903 * ord-flags to the first map
9905 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
9907 if (ord_to_idx(ord
) <= index
)
9910 map
= get_imsm_map(dev
, MAP_0
);
9911 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
9912 map
= get_imsm_map(dev
, MAP_1
);
9914 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
9918 for (i
= 0; i
< log
->entry_count
; i
++) {
9919 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
9921 if (entry
->disk_ordinal
<= index
)
9923 entry
->disk_ordinal
--;
9927 super
->updates_pending
++;
9929 struct dl
*dl
= *dlp
;
9931 *dlp
= (*dlp
)->next
;
9932 __free_imsm_disk(dl
);
9935 #endif /* MDASSEMBLE */
9937 static void close_targets(int *targets
, int new_disks
)
9944 for (i
= 0; i
< new_disks
; i
++) {
9945 if (targets
[i
] >= 0) {
9952 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
9953 struct intel_super
*super
,
9954 struct imsm_dev
*dev
)
9960 struct imsm_map
*map
;
9963 ret_val
= raid_disks
/2;
9964 /* check map if all disks pairs not failed
9967 map
= get_imsm_map(dev
, MAP_0
);
9968 for (i
= 0; i
< ret_val
; i
++) {
9969 int degradation
= 0;
9970 if (get_imsm_disk(super
, i
) == NULL
)
9972 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9974 if (degradation
== 2)
9977 map
= get_imsm_map(dev
, MAP_1
);
9978 /* if there is no second map
9979 * result can be returned
9983 /* check degradation in second map
9985 for (i
= 0; i
< ret_val
; i
++) {
9986 int degradation
= 0;
9987 if (get_imsm_disk(super
, i
) == NULL
)
9989 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9991 if (degradation
== 2)
10005 /*******************************************************************************
10006 * Function: open_backup_targets
10007 * Description: Function opens file descriptors for all devices given in
10010 * info : general array info
10011 * raid_disks : number of disks
10012 * raid_fds : table of device's file descriptors
10013 * super : intel super for raid10 degradation check
10014 * dev : intel device for raid10 degradation check
10018 ******************************************************************************/
10019 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10020 struct intel_super
*super
, struct imsm_dev
*dev
)
10026 for (i
= 0; i
< raid_disks
; i
++)
10029 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10032 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10033 dprintf("disk is faulty!!\n");
10037 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10040 dn
= map_dev(sd
->disk
.major
,
10041 sd
->disk
.minor
, 1);
10042 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10043 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10044 pr_err("cannot open component\n");
10049 /* check if maximum array degradation level is not exceeded
10051 if ((raid_disks
- opened
) >
10052 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10054 pr_err("Not enough disks can be opened.\n");
10055 close_targets(raid_fds
, raid_disks
);
10061 /*******************************************************************************
10062 * Function: validate_container_imsm
10063 * Description: This routine validates container after assemble,
10064 * eg. if devices in container are under the same controller.
10067 * info : linked list with info about devices used in array
10071 ******************************************************************************/
10072 int validate_container_imsm(struct mdinfo
*info
)
10074 if (check_env("IMSM_NO_PLATFORM"))
10077 struct sys_dev
*idev
;
10078 struct sys_dev
*hba
= NULL
;
10079 struct sys_dev
*intel_devices
= find_intel_devices();
10080 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10081 info
->disk
.minor
));
10083 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10084 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10093 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10094 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10098 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10099 struct mdinfo
*dev
;
10101 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10102 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10104 struct sys_dev
*hba2
= NULL
;
10105 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10106 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10114 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10115 get_orom_by_device_id(hba2
->dev_id
);
10117 if (hba2
&& hba
->type
!= hba2
->type
) {
10118 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10119 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10123 if (orom
!= orom2
) {
10124 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10125 " This operation is not supported and can lead to data loss.\n");
10130 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10131 " This operation is not supported and can lead to data loss.\n");
10139 /*******************************************************************************
10140 * Function: imsm_record_badblock
10141 * Description: This routine stores new bad block record in BBM log
10144 * a : array containing a bad block
10145 * slot : disk number containing a bad block
10146 * sector : bad block sector
10147 * length : bad block sectors range
10151 ******************************************************************************/
10152 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10153 unsigned long long sector
, int length
)
10155 struct intel_super
*super
= a
->container
->sb
;
10159 ord
= imsm_disk_slot_to_ord(a
, slot
);
10163 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10166 super
->updates_pending
++;
10170 /*******************************************************************************
10171 * Function: imsm_clear_badblock
10172 * Description: This routine clears bad block record from BBM log
10175 * a : array containing a bad block
10176 * slot : disk number containing a bad block
10177 * sector : bad block sector
10178 * length : bad block sectors range
10182 ******************************************************************************/
10183 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10184 unsigned long long sector
, int length
)
10186 struct intel_super
*super
= a
->container
->sb
;
10190 ord
= imsm_disk_slot_to_ord(a
, slot
);
10194 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10196 super
->updates_pending
++;
10200 /*******************************************************************************
10201 * Function: imsm_get_badblocks
10202 * Description: This routine get list of bad blocks for an array
10206 * slot : disk number
10208 * bb : structure containing bad blocks
10210 ******************************************************************************/
10211 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10213 int inst
= a
->info
.container_member
;
10214 struct intel_super
*super
= a
->container
->sb
;
10215 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10216 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10219 ord
= imsm_disk_slot_to_ord(a
, slot
);
10223 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10224 blocks_per_member(map
), &super
->bb
);
10228 /*******************************************************************************
10229 * Function: examine_badblocks_imsm
10230 * Description: Prints list of bad blocks on a disk to the standard output
10233 * st : metadata handler
10234 * fd : open file descriptor for device
10235 * devname : device name
10239 ******************************************************************************/
10240 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10242 struct intel_super
*super
= st
->sb
;
10243 struct bbm_log
*log
= super
->bbm_log
;
10244 struct dl
*d
= NULL
;
10247 for (d
= super
->disks
; d
; d
= d
->next
) {
10248 if (strcmp(d
->devname
, devname
) == 0)
10252 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10253 pr_err("%s doesn't appear to be part of a raid array\n",
10260 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10262 for (i
= 0; i
< log
->entry_count
; i
++) {
10263 if (entry
[i
].disk_ordinal
== d
->index
) {
10264 unsigned long long sector
= __le48_to_cpu(
10265 &entry
[i
].defective_block_start
);
10266 int cnt
= entry
[i
].marked_count
+ 1;
10269 printf("Bad-blocks on %s:\n", devname
);
10273 printf("%20llu for %d sectors\n", sector
, cnt
);
10279 printf("No bad-blocks list configured on %s\n", devname
);
10283 /*******************************************************************************
10284 * Function: init_migr_record_imsm
10285 * Description: Function inits imsm migration record
10287 * super : imsm internal array info
10288 * dev : device under migration
10289 * info : general array info to find the smallest device
10292 ******************************************************************************/
10293 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10294 struct mdinfo
*info
)
10296 struct intel_super
*super
= st
->sb
;
10297 struct migr_record
*migr_rec
= super
->migr_rec
;
10298 int new_data_disks
;
10299 unsigned long long dsize
, dev_sectors
;
10300 long long unsigned min_dev_sectors
= -1LLU;
10304 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10305 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10306 unsigned long long num_migr_units
;
10307 unsigned long long array_blocks
;
10309 memset(migr_rec
, 0, sizeof(struct migr_record
));
10310 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10312 /* only ascending reshape supported now */
10313 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10315 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10316 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10317 migr_rec
->dest_depth_per_unit
*=
10318 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10319 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
10320 migr_rec
->blocks_per_unit
=
10321 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10322 migr_rec
->dest_depth_per_unit
=
10323 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10324 array_blocks
= info
->component_size
* new_data_disks
;
10326 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10328 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10330 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
10332 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10333 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10335 /* Find the smallest dev */
10336 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10337 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10338 fd
= dev_open(nm
, O_RDONLY
);
10341 get_dev_size(fd
, NULL
, &dsize
);
10342 dev_sectors
= dsize
/ 512;
10343 if (dev_sectors
< min_dev_sectors
)
10344 min_dev_sectors
= dev_sectors
;
10347 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
10348 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10350 write_imsm_migr_rec(st
);
10355 /*******************************************************************************
10356 * Function: save_backup_imsm
10357 * Description: Function saves critical data stripes to Migration Copy Area
10358 * and updates the current migration unit status.
10359 * Use restore_stripes() to form a destination stripe,
10360 * and to write it to the Copy Area.
10362 * st : supertype information
10363 * dev : imsm device that backup is saved for
10364 * info : general array info
10365 * buf : input buffer
10366 * length : length of data to backup (blocks_per_unit)
10370 ******************************************************************************/
10371 int save_backup_imsm(struct supertype
*st
,
10372 struct imsm_dev
*dev
,
10373 struct mdinfo
*info
,
10378 struct intel_super
*super
= st
->sb
;
10379 unsigned long long *target_offsets
;
10382 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10383 int new_disks
= map_dest
->num_members
;
10384 int dest_layout
= 0;
10386 unsigned long long start
;
10387 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
10389 targets
= xmalloc(new_disks
* sizeof(int));
10391 for (i
= 0; i
< new_disks
; i
++)
10394 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10396 start
= info
->reshape_progress
* 512;
10397 for (i
= 0; i
< new_disks
; i
++) {
10398 target_offsets
[i
] = (unsigned long long)
10399 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
10400 /* move back copy area adderss, it will be moved forward
10401 * in restore_stripes() using start input variable
10403 target_offsets
[i
] -= start
/data_disks
;
10406 if (open_backup_targets(info
, new_disks
, targets
,
10410 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10411 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10413 if (restore_stripes(targets
, /* list of dest devices */
10414 target_offsets
, /* migration record offsets */
10417 map_dest
->raid_level
,
10419 -1, /* source backup file descriptor */
10420 0, /* input buf offset
10421 * always 0 buf is already offseted */
10425 pr_err("Error restoring stripes\n");
10433 close_targets(targets
, new_disks
);
10436 free(target_offsets
);
10441 /*******************************************************************************
10442 * Function: save_checkpoint_imsm
10443 * Description: Function called for current unit status update
10444 * in the migration record. It writes it to disk.
10446 * super : imsm internal array info
10447 * info : general array info
10451 * 2: failure, means no valid migration record
10452 * / no general migration in progress /
10453 ******************************************************************************/
10454 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10456 struct intel_super
*super
= st
->sb
;
10457 unsigned long long blocks_per_unit
;
10458 unsigned long long curr_migr_unit
;
10460 if (load_imsm_migr_rec(super
, info
) != 0) {
10461 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10465 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10466 if (blocks_per_unit
== 0) {
10467 dprintf("imsm: no migration in progress.\n");
10470 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10471 /* check if array is alligned to copy area
10472 * if it is not alligned, add one to current migration unit value
10473 * this can happend on array reshape finish only
10475 if (info
->reshape_progress
% blocks_per_unit
)
10478 super
->migr_rec
->curr_migr_unit
=
10479 __cpu_to_le32(curr_migr_unit
);
10480 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10481 super
->migr_rec
->dest_1st_member_lba
=
10482 __cpu_to_le32(curr_migr_unit
*
10483 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
10484 if (write_imsm_migr_rec(st
) < 0) {
10485 dprintf("imsm: Cannot write migration record outside backup area\n");
10492 /*******************************************************************************
10493 * Function: recover_backup_imsm
10494 * Description: Function recovers critical data from the Migration Copy Area
10495 * while assembling an array.
10497 * super : imsm internal array info
10498 * info : general array info
10500 * 0 : success (or there is no data to recover)
10502 ******************************************************************************/
10503 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10505 struct intel_super
*super
= st
->sb
;
10506 struct migr_record
*migr_rec
= super
->migr_rec
;
10507 struct imsm_map
*map_dest
;
10508 struct intel_dev
*id
= NULL
;
10509 unsigned long long read_offset
;
10510 unsigned long long write_offset
;
10512 int *targets
= NULL
;
10513 int new_disks
, i
, err
;
10516 unsigned int sector_size
= super
->sector_size
;
10517 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
10518 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
10520 int skipped_disks
= 0;
10522 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10526 /* recover data only during assemblation */
10527 if (strncmp(buffer
, "inactive", 8) != 0)
10529 /* no data to recover */
10530 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10532 if (curr_migr_unit
>= num_migr_units
)
10535 /* find device during reshape */
10536 for (id
= super
->devlist
; id
; id
= id
->next
)
10537 if (is_gen_migration(id
->dev
))
10542 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10543 new_disks
= map_dest
->num_members
;
10545 read_offset
= (unsigned long long)
10546 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
10548 write_offset
= ((unsigned long long)
10549 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
10550 pba_of_lba0(map_dest
)) * 512;
10552 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10553 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10555 targets
= xcalloc(new_disks
, sizeof(int));
10557 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10558 pr_err("Cannot open some devices belonging to array.\n");
10562 for (i
= 0; i
< new_disks
; i
++) {
10563 if (targets
[i
] < 0) {
10567 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
10568 pr_err("Cannot seek to block: %s\n",
10573 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
10574 pr_err("Cannot read copy area block: %s\n",
10579 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
10580 pr_err("Cannot seek to block: %s\n",
10585 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
10586 pr_err("Cannot restore block: %s\n",
10593 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
10597 pr_err("Cannot restore data from backup. Too many failed disks\n");
10601 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
10602 /* ignore error == 2, this can mean end of reshape here
10604 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
10610 for (i
= 0; i
< new_disks
; i
++)
10619 static char disk_by_path
[] = "/dev/disk/by-path/";
10621 static const char *imsm_get_disk_controller_domain(const char *path
)
10623 char disk_path
[PATH_MAX
];
10627 strcpy(disk_path
, disk_by_path
);
10628 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
10629 if (stat(disk_path
, &st
) == 0) {
10630 struct sys_dev
* hba
;
10633 path
= devt_to_devpath(st
.st_rdev
);
10636 hba
= find_disk_attached_hba(-1, path
);
10637 if (hba
&& hba
->type
== SYS_DEV_SAS
)
10639 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
10643 dprintf("path: %s hba: %s attached: %s\n",
10644 path
, (hba
) ? hba
->path
: "NULL", drv
);
10650 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
10652 static char devnm
[32];
10653 char subdev_name
[20];
10654 struct mdstat_ent
*mdstat
;
10656 sprintf(subdev_name
, "%d", subdev
);
10657 mdstat
= mdstat_by_subdev(subdev_name
, container
);
10661 strcpy(devnm
, mdstat
->devnm
);
10662 free_mdstat(mdstat
);
10666 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
10667 struct geo_params
*geo
,
10668 int *old_raid_disks
,
10671 /* currently we only support increasing the number of devices
10672 * for a container. This increases the number of device for each
10673 * member array. They must all be RAID0 or RAID5.
10676 struct mdinfo
*info
, *member
;
10677 int devices_that_can_grow
= 0;
10679 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
10681 if (geo
->size
> 0 ||
10682 geo
->level
!= UnSet
||
10683 geo
->layout
!= UnSet
||
10684 geo
->chunksize
!= 0 ||
10685 geo
->raid_disks
== UnSet
) {
10686 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
10690 if (direction
== ROLLBACK_METADATA_CHANGES
) {
10691 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
10695 info
= container_content_imsm(st
, NULL
);
10696 for (member
= info
; member
; member
= member
->next
) {
10699 dprintf("imsm: checking device_num: %i\n",
10700 member
->container_member
);
10702 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
10703 /* we work on container for Online Capacity Expansion
10704 * only so raid_disks has to grow
10706 dprintf("imsm: for container operation raid disks increase is required\n");
10710 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
10711 /* we cannot use this container with other raid level
10713 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
10714 info
->array
.level
);
10717 /* check for platform support
10718 * for this raid level configuration
10720 struct intel_super
*super
= st
->sb
;
10721 if (!is_raid_level_supported(super
->orom
,
10722 member
->array
.level
,
10723 geo
->raid_disks
)) {
10724 dprintf("platform does not support raid%d with %d disk%s\n",
10727 geo
->raid_disks
> 1 ? "s" : "");
10730 /* check if component size is aligned to chunk size
10732 if (info
->component_size
%
10733 (info
->array
.chunk_size
/512)) {
10734 dprintf("Component size is not aligned to chunk size\n");
10739 if (*old_raid_disks
&&
10740 info
->array
.raid_disks
!= *old_raid_disks
)
10742 *old_raid_disks
= info
->array
.raid_disks
;
10744 /* All raid5 and raid0 volumes in container
10745 * have to be ready for Online Capacity Expansion
10746 * so they need to be assembled. We have already
10747 * checked that no recovery etc is happening.
10749 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
10750 st
->container_devnm
);
10751 if (result
== NULL
) {
10752 dprintf("imsm: cannot find array\n");
10755 devices_that_can_grow
++;
10758 if (!member
&& devices_that_can_grow
)
10762 dprintf("Container operation allowed\n");
10764 dprintf("Error: %i\n", ret_val
);
10769 /* Function: get_spares_for_grow
10770 * Description: Allocates memory and creates list of spare devices
10771 * avaliable in container. Checks if spare drive size is acceptable.
10772 * Parameters: Pointer to the supertype structure
10773 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
10776 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
10778 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
10779 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
10782 /******************************************************************************
10783 * function: imsm_create_metadata_update_for_reshape
10784 * Function creates update for whole IMSM container.
10786 ******************************************************************************/
10787 static int imsm_create_metadata_update_for_reshape(
10788 struct supertype
*st
,
10789 struct geo_params
*geo
,
10790 int old_raid_disks
,
10791 struct imsm_update_reshape
**updatep
)
10793 struct intel_super
*super
= st
->sb
;
10794 struct imsm_super
*mpb
= super
->anchor
;
10795 int update_memory_size
;
10796 struct imsm_update_reshape
*u
;
10797 struct mdinfo
*spares
;
10800 struct mdinfo
*dev
;
10802 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
10804 delta_disks
= geo
->raid_disks
- old_raid_disks
;
10806 /* size of all update data without anchor */
10807 update_memory_size
= sizeof(struct imsm_update_reshape
);
10809 /* now add space for spare disks that we need to add. */
10810 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
10812 u
= xcalloc(1, update_memory_size
);
10813 u
->type
= update_reshape_container_disks
;
10814 u
->old_raid_disks
= old_raid_disks
;
10815 u
->new_raid_disks
= geo
->raid_disks
;
10817 /* now get spare disks list
10819 spares
= get_spares_for_grow(st
);
10822 || delta_disks
> spares
->array
.spare_disks
) {
10823 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
10828 /* we have got spares
10829 * update disk list in imsm_disk list table in anchor
10831 dprintf("imsm: %i spares are available.\n\n",
10832 spares
->array
.spare_disks
);
10834 dev
= spares
->devs
;
10835 for (i
= 0; i
< delta_disks
; i
++) {
10840 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
10842 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
10843 dl
->index
= mpb
->num_disks
;
10851 sysfs_free(spares
);
10853 dprintf("imsm: reshape update preparation :");
10854 if (i
== delta_disks
) {
10855 dprintf_cont(" OK\n");
10857 return update_memory_size
;
10860 dprintf_cont(" Error\n");
10865 /******************************************************************************
10866 * function: imsm_create_metadata_update_for_size_change()
10867 * Creates update for IMSM array for array size change.
10869 ******************************************************************************/
10870 static int imsm_create_metadata_update_for_size_change(
10871 struct supertype
*st
,
10872 struct geo_params
*geo
,
10873 struct imsm_update_size_change
**updatep
)
10875 struct intel_super
*super
= st
->sb
;
10876 int update_memory_size
;
10877 struct imsm_update_size_change
*u
;
10879 dprintf("(enter) New size = %llu\n", geo
->size
);
10881 /* size of all update data without anchor */
10882 update_memory_size
= sizeof(struct imsm_update_size_change
);
10884 u
= xcalloc(1, update_memory_size
);
10885 u
->type
= update_size_change
;
10886 u
->subdev
= super
->current_vol
;
10887 u
->new_size
= geo
->size
;
10889 dprintf("imsm: reshape update preparation : OK\n");
10892 return update_memory_size
;
10895 /******************************************************************************
10896 * function: imsm_create_metadata_update_for_migration()
10897 * Creates update for IMSM array.
10899 ******************************************************************************/
10900 static int imsm_create_metadata_update_for_migration(
10901 struct supertype
*st
,
10902 struct geo_params
*geo
,
10903 struct imsm_update_reshape_migration
**updatep
)
10905 struct intel_super
*super
= st
->sb
;
10906 int update_memory_size
;
10907 struct imsm_update_reshape_migration
*u
;
10908 struct imsm_dev
*dev
;
10909 int previous_level
= -1;
10911 dprintf("(enter) New Level = %i\n", geo
->level
);
10913 /* size of all update data without anchor */
10914 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
10916 u
= xcalloc(1, update_memory_size
);
10917 u
->type
= update_reshape_migration
;
10918 u
->subdev
= super
->current_vol
;
10919 u
->new_level
= geo
->level
;
10920 u
->new_layout
= geo
->layout
;
10921 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
10922 u
->new_disks
[0] = -1;
10923 u
->new_chunksize
= -1;
10925 dev
= get_imsm_dev(super
, u
->subdev
);
10927 struct imsm_map
*map
;
10929 map
= get_imsm_map(dev
, MAP_0
);
10931 int current_chunk_size
=
10932 __le16_to_cpu(map
->blocks_per_strip
) / 2;
10934 if (geo
->chunksize
!= current_chunk_size
) {
10935 u
->new_chunksize
= geo
->chunksize
/ 1024;
10936 dprintf("imsm: chunk size change from %i to %i\n",
10937 current_chunk_size
, u
->new_chunksize
);
10939 previous_level
= map
->raid_level
;
10942 if (geo
->level
== 5 && previous_level
== 0) {
10943 struct mdinfo
*spares
= NULL
;
10945 u
->new_raid_disks
++;
10946 spares
= get_spares_for_grow(st
);
10947 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
10949 sysfs_free(spares
);
10950 update_memory_size
= 0;
10951 pr_err("cannot get spare device for requested migration\n");
10954 sysfs_free(spares
);
10956 dprintf("imsm: reshape update preparation : OK\n");
10959 return update_memory_size
;
10962 static void imsm_update_metadata_locally(struct supertype
*st
,
10963 void *buf
, int len
)
10965 struct metadata_update mu
;
10970 mu
.space_list
= NULL
;
10972 if (imsm_prepare_update(st
, &mu
))
10973 imsm_process_update(st
, &mu
);
10975 while (mu
.space_list
) {
10976 void **space
= mu
.space_list
;
10977 mu
.space_list
= *space
;
10982 /***************************************************************************
10983 * Function: imsm_analyze_change
10984 * Description: Function analyze change for single volume
10985 * and validate if transition is supported
10986 * Parameters: Geometry parameters, supertype structure,
10987 * metadata change direction (apply/rollback)
10988 * Returns: Operation type code on success, -1 if fail
10989 ****************************************************************************/
10990 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
10991 struct geo_params
*geo
,
10994 struct mdinfo info
;
10996 int check_devs
= 0;
10998 /* number of added/removed disks in operation result */
10999 int devNumChange
= 0;
11000 /* imsm compatible layout value for array geometry verification */
11001 int imsm_layout
= -1;
11003 struct imsm_dev
*dev
;
11004 struct intel_super
*super
;
11005 unsigned long long current_size
;
11006 unsigned long long free_size
;
11007 unsigned long long max_size
;
11010 getinfo_super_imsm_volume(st
, &info
, NULL
);
11011 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11012 geo
->level
!= UnSet
) {
11013 switch (info
.array
.level
) {
11015 if (geo
->level
== 5) {
11016 change
= CH_MIGRATION
;
11017 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11018 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11020 goto analyse_change_exit
;
11022 imsm_layout
= geo
->layout
;
11024 devNumChange
= 1; /* parity disk added */
11025 } else if (geo
->level
== 10) {
11026 change
= CH_TAKEOVER
;
11028 devNumChange
= 2; /* two mirrors added */
11029 imsm_layout
= 0x102; /* imsm supported layout */
11034 if (geo
->level
== 0) {
11035 change
= CH_TAKEOVER
;
11037 devNumChange
= -(geo
->raid_disks
/2);
11038 imsm_layout
= 0; /* imsm raid0 layout */
11042 if (change
== -1) {
11043 pr_err("Error. Level Migration from %d to %d not supported!\n",
11044 info
.array
.level
, geo
->level
);
11045 goto analyse_change_exit
;
11048 geo
->level
= info
.array
.level
;
11050 if (geo
->layout
!= info
.array
.layout
&&
11051 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11052 change
= CH_MIGRATION
;
11053 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11054 geo
->layout
== 5) {
11055 /* reshape 5 -> 4 */
11056 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11057 geo
->layout
== 0) {
11058 /* reshape 4 -> 5 */
11062 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11063 info
.array
.layout
, geo
->layout
);
11065 goto analyse_change_exit
;
11068 geo
->layout
= info
.array
.layout
;
11069 if (imsm_layout
== -1)
11070 imsm_layout
= info
.array
.layout
;
11073 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11074 geo
->chunksize
!= info
.array
.chunk_size
) {
11075 if (info
.array
.level
== 10) {
11076 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11078 goto analyse_change_exit
;
11079 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11080 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11081 geo
->chunksize
/1024, info
.component_size
/2);
11083 goto analyse_change_exit
;
11085 change
= CH_MIGRATION
;
11087 geo
->chunksize
= info
.array
.chunk_size
;
11090 chunk
= geo
->chunksize
/ 1024;
11093 dev
= get_imsm_dev(super
, super
->current_vol
);
11094 data_disks
= imsm_num_data_members(dev
, MAP_0
);
11095 /* compute current size per disk member
11097 current_size
= info
.custom_array_size
/ data_disks
;
11099 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11100 /* align component size
11102 geo
->size
= imsm_component_size_aligment_check(
11103 get_imsm_raid_level(dev
->vol
.map
),
11104 chunk
* 1024, super
->sector_size
,
11106 if (geo
->size
== 0) {
11107 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11109 goto analyse_change_exit
;
11113 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11114 if (change
!= -1) {
11115 pr_err("Error. Size change should be the only one at a time.\n");
11117 goto analyse_change_exit
;
11119 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11120 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11121 super
->current_vol
, st
->devnm
);
11122 goto analyse_change_exit
;
11124 /* check the maximum available size
11126 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11127 0, chunk
, &free_size
);
11129 /* Cannot find maximum available space
11133 max_size
= free_size
+ current_size
;
11134 /* align component size
11136 max_size
= imsm_component_size_aligment_check(
11137 get_imsm_raid_level(dev
->vol
.map
),
11138 chunk
* 1024, super
->sector_size
,
11141 if (geo
->size
== MAX_SIZE
) {
11142 /* requested size change to the maximum available size
11144 if (max_size
== 0) {
11145 pr_err("Error. Cannot find maximum available space.\n");
11147 goto analyse_change_exit
;
11149 geo
->size
= max_size
;
11152 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11153 /* accept size for rollback only
11156 /* round size due to metadata compatibility
11158 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11159 << SECT_PER_MB_SHIFT
;
11160 dprintf("Prepare update for size change to %llu\n",
11162 if (current_size
>= geo
->size
) {
11163 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11164 current_size
, geo
->size
);
11165 goto analyse_change_exit
;
11167 if (max_size
&& geo
->size
> max_size
) {
11168 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11169 max_size
, geo
->size
);
11170 goto analyse_change_exit
;
11173 geo
->size
*= data_disks
;
11174 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11175 change
= CH_ARRAY_SIZE
;
11177 if (!validate_geometry_imsm(st
,
11180 geo
->raid_disks
+ devNumChange
,
11182 geo
->size
, INVALID_SECTORS
,
11183 0, 0, info
.consistency_policy
, 1))
11187 struct intel_super
*super
= st
->sb
;
11188 struct imsm_super
*mpb
= super
->anchor
;
11190 if (mpb
->num_raid_devs
> 1) {
11191 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11197 analyse_change_exit
:
11198 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11199 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11200 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11206 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11208 struct intel_super
*super
= st
->sb
;
11209 struct imsm_update_takeover
*u
;
11211 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11213 u
->type
= update_takeover
;
11214 u
->subarray
= super
->current_vol
;
11216 /* 10->0 transition */
11217 if (geo
->level
== 0)
11218 u
->direction
= R10_TO_R0
;
11220 /* 0->10 transition */
11221 if (geo
->level
== 10)
11222 u
->direction
= R0_TO_R10
;
11224 /* update metadata locally */
11225 imsm_update_metadata_locally(st
, u
,
11226 sizeof(struct imsm_update_takeover
));
11227 /* and possibly remotely */
11228 if (st
->update_tail
)
11229 append_metadata_update(st
, u
,
11230 sizeof(struct imsm_update_takeover
));
11237 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11239 int layout
, int chunksize
, int raid_disks
,
11240 int delta_disks
, char *backup
, char *dev
,
11241 int direction
, int verbose
)
11244 struct geo_params geo
;
11246 dprintf("(enter)\n");
11248 memset(&geo
, 0, sizeof(struct geo_params
));
11250 geo
.dev_name
= dev
;
11251 strcpy(geo
.devnm
, st
->devnm
);
11254 geo
.layout
= layout
;
11255 geo
.chunksize
= chunksize
;
11256 geo
.raid_disks
= raid_disks
;
11257 if (delta_disks
!= UnSet
)
11258 geo
.raid_disks
+= delta_disks
;
11260 dprintf("for level : %i\n", geo
.level
);
11261 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11263 if (experimental() == 0)
11266 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11267 /* On container level we can only increase number of devices. */
11268 dprintf("imsm: info: Container operation\n");
11269 int old_raid_disks
= 0;
11271 if (imsm_reshape_is_allowed_on_container(
11272 st
, &geo
, &old_raid_disks
, direction
)) {
11273 struct imsm_update_reshape
*u
= NULL
;
11276 len
= imsm_create_metadata_update_for_reshape(
11277 st
, &geo
, old_raid_disks
, &u
);
11280 dprintf("imsm: Cannot prepare update\n");
11281 goto exit_imsm_reshape_super
;
11285 /* update metadata locally */
11286 imsm_update_metadata_locally(st
, u
, len
);
11287 /* and possibly remotely */
11288 if (st
->update_tail
)
11289 append_metadata_update(st
, u
, len
);
11294 pr_err("(imsm) Operation is not allowed on this container\n");
11297 /* On volume level we support following operations
11298 * - takeover: raid10 -> raid0; raid0 -> raid10
11299 * - chunk size migration
11300 * - migration: raid5 -> raid0; raid0 -> raid5
11302 struct intel_super
*super
= st
->sb
;
11303 struct intel_dev
*dev
= super
->devlist
;
11305 dprintf("imsm: info: Volume operation\n");
11306 /* find requested device */
11309 imsm_find_array_devnm_by_subdev(
11310 dev
->index
, st
->container_devnm
);
11311 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11316 pr_err("Cannot find %s (%s) subarray\n",
11317 geo
.dev_name
, geo
.devnm
);
11318 goto exit_imsm_reshape_super
;
11320 super
->current_vol
= dev
->index
;
11321 change
= imsm_analyze_change(st
, &geo
, direction
);
11324 ret_val
= imsm_takeover(st
, &geo
);
11326 case CH_MIGRATION
: {
11327 struct imsm_update_reshape_migration
*u
= NULL
;
11329 imsm_create_metadata_update_for_migration(
11332 dprintf("imsm: Cannot prepare update\n");
11336 /* update metadata locally */
11337 imsm_update_metadata_locally(st
, u
, len
);
11338 /* and possibly remotely */
11339 if (st
->update_tail
)
11340 append_metadata_update(st
, u
, len
);
11345 case CH_ARRAY_SIZE
: {
11346 struct imsm_update_size_change
*u
= NULL
;
11348 imsm_create_metadata_update_for_size_change(
11351 dprintf("imsm: Cannot prepare update\n");
11355 /* update metadata locally */
11356 imsm_update_metadata_locally(st
, u
, len
);
11357 /* and possibly remotely */
11358 if (st
->update_tail
)
11359 append_metadata_update(st
, u
, len
);
11369 exit_imsm_reshape_super
:
11370 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11374 #define COMPLETED_OK 0
11375 #define COMPLETED_NONE 1
11376 #define COMPLETED_DELAYED 2
11378 static int read_completed(int fd
, unsigned long long *val
)
11383 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11387 ret
= COMPLETED_OK
;
11388 if (strncmp(buf
, "none", 4) == 0) {
11389 ret
= COMPLETED_NONE
;
11390 } else if (strncmp(buf
, "delayed", 7) == 0) {
11391 ret
= COMPLETED_DELAYED
;
11394 *val
= strtoull(buf
, &ep
, 0);
11395 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11401 /*******************************************************************************
11402 * Function: wait_for_reshape_imsm
11403 * Description: Function writes new sync_max value and waits until
11404 * reshape process reach new position
11406 * sra : general array info
11407 * ndata : number of disks in new array's layout
11410 * 1 : there is no reshape in progress,
11412 ******************************************************************************/
11413 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11415 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11417 unsigned long long completed
;
11418 /* to_complete : new sync_max position */
11419 unsigned long long to_complete
= sra
->reshape_progress
;
11420 unsigned long long position_to_set
= to_complete
/ ndata
;
11423 dprintf("cannot open reshape_position\n");
11428 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11430 dprintf("cannot read reshape_position (no reshape in progres)\n");
11439 if (completed
> position_to_set
) {
11440 dprintf("wrong next position to set %llu (%llu)\n",
11441 to_complete
, position_to_set
);
11445 dprintf("Position set: %llu\n", position_to_set
);
11446 if (sysfs_set_num(sra
, NULL
, "sync_max",
11447 position_to_set
) != 0) {
11448 dprintf("cannot set reshape position to %llu\n",
11457 int timeout
= 3000;
11459 sysfs_wait(fd
, &timeout
);
11460 if (sysfs_get_str(sra
, NULL
, "sync_action",
11462 strncmp(action
, "reshape", 7) != 0) {
11463 if (strncmp(action
, "idle", 4) == 0)
11469 rc
= read_completed(fd
, &completed
);
11471 dprintf("cannot read reshape_position (in loop)\n");
11474 } else if (rc
== COMPLETED_NONE
)
11476 } while (completed
< position_to_set
);
11482 /*******************************************************************************
11483 * Function: check_degradation_change
11484 * Description: Check that array hasn't become failed.
11486 * info : for sysfs access
11487 * sources : source disks descriptors
11488 * degraded: previous degradation level
11490 * degradation level
11491 ******************************************************************************/
11492 int check_degradation_change(struct mdinfo
*info
,
11496 unsigned long long new_degraded
;
11499 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11500 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11501 /* check each device to ensure it is still working */
11504 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11505 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11507 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11510 if (sysfs_get_str(info
,
11511 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11512 strstr(sbuf
, "faulty") ||
11513 strstr(sbuf
, "in_sync") == NULL
) {
11514 /* this device is dead */
11515 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11516 if (sd
->disk
.raid_disk
>= 0 &&
11517 sources
[sd
->disk
.raid_disk
] >= 0) {
11519 sd
->disk
.raid_disk
]);
11520 sources
[sd
->disk
.raid_disk
] =
11529 return new_degraded
;
11532 /*******************************************************************************
11533 * Function: imsm_manage_reshape
11534 * Description: Function finds array under reshape and it manages reshape
11535 * process. It creates stripes backups (if required) and sets
11538 * afd : Backup handle (nattive) - not used
11539 * sra : general array info
11540 * reshape : reshape parameters - not used
11541 * st : supertype structure
11542 * blocks : size of critical section [blocks]
11543 * fds : table of source device descriptor
11544 * offsets : start of array (offest per devices)
11546 * destfd : table of destination device descriptor
11547 * destoffsets : table of destination offsets (per device)
11549 * 1 : success, reshape is done
11551 ******************************************************************************/
11552 static int imsm_manage_reshape(
11553 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
11554 struct supertype
*st
, unsigned long backup_blocks
,
11555 int *fds
, unsigned long long *offsets
,
11556 int dests
, int *destfd
, unsigned long long *destoffsets
)
11559 struct intel_super
*super
= st
->sb
;
11560 struct intel_dev
*dv
;
11561 unsigned int sector_size
= super
->sector_size
;
11562 struct imsm_dev
*dev
= NULL
;
11563 struct imsm_map
*map_src
;
11564 int migr_vol_qan
= 0;
11565 int ndata
, odata
; /* [bytes] */
11566 int chunk
; /* [bytes] */
11567 struct migr_record
*migr_rec
;
11569 unsigned int buf_size
; /* [bytes] */
11570 unsigned long long max_position
; /* array size [bytes] */
11571 unsigned long long next_step
; /* [blocks]/[bytes] */
11572 unsigned long long old_data_stripe_length
;
11573 unsigned long long start_src
; /* [bytes] */
11574 unsigned long long start
; /* [bytes] */
11575 unsigned long long start_buf_shift
; /* [bytes] */
11577 int source_layout
= 0;
11582 if (!fds
|| !offsets
)
11585 /* Find volume during the reshape */
11586 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
11587 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
11588 && dv
->dev
->vol
.migr_state
== 1) {
11593 /* Only one volume can migrate at the same time */
11594 if (migr_vol_qan
!= 1) {
11595 pr_err("%s", migr_vol_qan
?
11596 "Number of migrating volumes greater than 1\n" :
11597 "There is no volume during migrationg\n");
11601 map_src
= get_imsm_map(dev
, MAP_1
);
11602 if (map_src
== NULL
)
11605 ndata
= imsm_num_data_members(dev
, MAP_0
);
11606 odata
= imsm_num_data_members(dev
, MAP_1
);
11608 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
11609 old_data_stripe_length
= odata
* chunk
;
11611 migr_rec
= super
->migr_rec
;
11613 /* initialize migration record for start condition */
11614 if (sra
->reshape_progress
== 0)
11615 init_migr_record_imsm(st
, dev
, sra
);
11617 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
11618 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
11621 /* Save checkpoint to update migration record for current
11622 * reshape position (in md). It can be farther than current
11623 * reshape position in metadata.
11625 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11626 /* ignore error == 2, this can mean end of reshape here
11628 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
11633 /* size for data */
11634 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
11635 /* extend buffer size for parity disk */
11636 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11637 /* add space for stripe aligment */
11638 buf_size
+= old_data_stripe_length
;
11639 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
11640 dprintf("imsm: Cannot allocate checkpoint buffer\n");
11644 max_position
= sra
->component_size
* ndata
;
11645 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
11647 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
11648 __le32_to_cpu(migr_rec
->num_migr_units
)) {
11649 /* current reshape position [blocks] */
11650 unsigned long long current_position
=
11651 __le32_to_cpu(migr_rec
->blocks_per_unit
)
11652 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
11653 unsigned long long border
;
11655 /* Check that array hasn't become failed.
11657 degraded
= check_degradation_change(sra
, fds
, degraded
);
11658 if (degraded
> 1) {
11659 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
11663 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
11665 if ((current_position
+ next_step
) > max_position
)
11666 next_step
= max_position
- current_position
;
11668 start
= current_position
* 512;
11670 /* align reading start to old geometry */
11671 start_buf_shift
= start
% old_data_stripe_length
;
11672 start_src
= start
- start_buf_shift
;
11674 border
= (start_src
/ odata
) - (start
/ ndata
);
11676 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
11677 /* save critical stripes to buf
11678 * start - start address of current unit
11679 * to backup [bytes]
11680 * start_src - start address of current unit
11681 * to backup alligned to source array
11684 unsigned long long next_step_filler
;
11685 unsigned long long copy_length
= next_step
* 512;
11687 /* allign copy area length to stripe in old geometry */
11688 next_step_filler
= ((copy_length
+ start_buf_shift
)
11689 % old_data_stripe_length
);
11690 if (next_step_filler
)
11691 next_step_filler
= (old_data_stripe_length
11692 - next_step_filler
);
11693 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
11694 start
, start_src
, copy_length
,
11695 start_buf_shift
, next_step_filler
);
11697 if (save_stripes(fds
, offsets
, map_src
->num_members
,
11698 chunk
, map_src
->raid_level
,
11699 source_layout
, 0, NULL
, start_src
,
11701 next_step_filler
+ start_buf_shift
,
11703 dprintf("imsm: Cannot save stripes to buffer\n");
11706 /* Convert data to destination format and store it
11707 * in backup general migration area
11709 if (save_backup_imsm(st
, dev
, sra
,
11710 buf
+ start_buf_shift
, copy_length
)) {
11711 dprintf("imsm: Cannot save stripes to target devices\n");
11714 if (save_checkpoint_imsm(st
, sra
,
11715 UNIT_SRC_IN_CP_AREA
)) {
11716 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
11720 /* set next step to use whole border area */
11721 border
/= next_step
;
11723 next_step
*= border
;
11725 /* When data backed up, checkpoint stored,
11726 * kick the kernel to reshape unit of data
11728 next_step
= next_step
+ sra
->reshape_progress
;
11729 /* limit next step to array max position */
11730 if (next_step
> max_position
)
11731 next_step
= max_position
;
11732 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
11733 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
11734 sra
->reshape_progress
= next_step
;
11736 /* wait until reshape finish */
11737 if (wait_for_reshape_imsm(sra
, ndata
)) {
11738 dprintf("wait_for_reshape_imsm returned error!\n");
11744 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11745 /* ignore error == 2, this can mean end of reshape here
11747 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
11753 /* clear migr_rec on disks after successful migration */
11756 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*sector_size
);
11757 for (d
= super
->disks
; d
; d
= d
->next
) {
11758 if (d
->index
< 0 || is_failed(&d
->disk
))
11760 unsigned long long dsize
;
11762 get_dev_size(d
->fd
, NULL
, &dsize
);
11763 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
11765 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
11766 MIGR_REC_BUF_SECTORS
*sector_size
) !=
11767 MIGR_REC_BUF_SECTORS
*sector_size
)
11768 perror("Write migr_rec failed");
11772 /* return '1' if done */
11776 /* See Grow.c: abort_reshape() for further explanation */
11777 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
11778 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
11779 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
11784 #endif /* MDASSEMBLE */
11786 struct superswitch super_imsm
= {
11788 .examine_super
= examine_super_imsm
,
11789 .brief_examine_super
= brief_examine_super_imsm
,
11790 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
11791 .export_examine_super
= export_examine_super_imsm
,
11792 .detail_super
= detail_super_imsm
,
11793 .brief_detail_super
= brief_detail_super_imsm
,
11794 .write_init_super
= write_init_super_imsm
,
11795 .validate_geometry
= validate_geometry_imsm
,
11796 .add_to_super
= add_to_super_imsm
,
11797 .remove_from_super
= remove_from_super_imsm
,
11798 .detail_platform
= detail_platform_imsm
,
11799 .export_detail_platform
= export_detail_platform_imsm
,
11800 .kill_subarray
= kill_subarray_imsm
,
11801 .update_subarray
= update_subarray_imsm
,
11802 .load_container
= load_container_imsm
,
11803 .default_geometry
= default_geometry_imsm
,
11804 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
11805 .reshape_super
= imsm_reshape_super
,
11806 .manage_reshape
= imsm_manage_reshape
,
11807 .recover_backup
= recover_backup_imsm
,
11808 .copy_metadata
= copy_metadata_imsm
,
11809 .examine_badblocks
= examine_badblocks_imsm
,
11811 .match_home
= match_home_imsm
,
11812 .uuid_from_super
= uuid_from_super_imsm
,
11813 .getinfo_super
= getinfo_super_imsm
,
11814 .getinfo_super_disks
= getinfo_super_disks_imsm
,
11815 .update_super
= update_super_imsm
,
11817 .avail_size
= avail_size_imsm
,
11818 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
11820 .compare_super
= compare_super_imsm
,
11822 .load_super
= load_super_imsm
,
11823 .init_super
= init_super_imsm
,
11824 .store_super
= store_super_imsm
,
11825 .free_super
= free_super_imsm
,
11826 .match_metadata_desc
= match_metadata_desc_imsm
,
11827 .container_content
= container_content_imsm
,
11828 .validate_container
= validate_container_imsm
,
11830 .write_init_ppl
= write_init_ppl_imsm
,
11831 .validate_ppl
= validate_ppl_imsm
,
11838 .open_new
= imsm_open_new
,
11839 .set_array_state
= imsm_set_array_state
,
11840 .set_disk
= imsm_set_disk
,
11841 .sync_metadata
= imsm_sync_metadata
,
11842 .activate_spare
= imsm_activate_spare
,
11843 .process_update
= imsm_process_update
,
11844 .prepare_update
= imsm_prepare_update
,
11845 .record_bad_block
= imsm_record_badblock
,
11846 .clear_bad_block
= imsm_clear_badblock
,
11847 .get_bad_blocks
= imsm_get_badblocks
,
11848 #endif /* MDASSEMBLE */