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 __u16 num_raid_devs_created
; /* 0x4C - 0x4D Used for generating unique
236 * volume IDs for raid_dev created in this array
239 __u16 filler1
; /* 0x4E - 0x4F */
240 #define IMSM_FILLERS 34
241 __u32 filler
[IMSM_FILLERS
]; /* 0x50 - 0xD7 RAID_MPB_FILLERS */
242 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
243 /* here comes imsm_dev[num_raid_devs] */
244 /* here comes BBM logs */
245 } __attribute__ ((packed
));
247 #define BBM_LOG_MAX_ENTRIES 254
248 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
249 #define BBM_LOG_SIGNATURE 0xabadb10c
251 struct bbm_log_block_addr
{
254 } __attribute__ ((__packed__
));
256 struct bbm_log_entry
{
257 __u8 marked_count
; /* Number of blocks marked - 1 */
258 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
259 struct bbm_log_block_addr defective_block_start
;
260 } __attribute__ ((__packed__
));
263 __u32 signature
; /* 0xABADB10C */
265 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
266 } __attribute__ ((__packed__
));
268 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
270 #define BLOCKS_PER_KB (1024/512)
272 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
274 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
276 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
277 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
278 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
281 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
282 * be recovered using srcMap */
283 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
284 * already been migrated and must
285 * be recovered from checkpoint area */
287 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of the PPL, without the header */
290 __u32 rec_status
; /* Status used to determine how to restart
291 * migration in case it aborts
293 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
294 __u32 family_num
; /* Family number of MPB
295 * containing the RaidDev
296 * that is migrating */
297 __u32 ascending_migr
; /* True if migrating in increasing
299 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
300 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
302 * advances per unit-of-operation */
303 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
304 __u32 dest_1st_member_lba
; /* First member lba on first
305 * stripe of destination */
306 __u32 num_migr_units
; /* Total num migration units-of-op */
307 __u32 post_migr_vol_cap
; /* Size of volume after
308 * migration completes */
309 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
310 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
311 * migration ckpt record was read from
312 * (for recovered migrations) */
313 } __attribute__ ((__packed__
));
318 * 2: metadata does not match
326 struct md_list
*next
;
329 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
331 static __u8
migr_type(struct imsm_dev
*dev
)
333 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
334 dev
->status
& DEV_VERIFY_AND_FIX
)
337 return dev
->vol
.migr_type
;
340 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
342 /* for compatibility with older oroms convert MIGR_REPAIR, into
343 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
345 if (migr_type
== MIGR_REPAIR
) {
346 dev
->vol
.migr_type
= MIGR_VERIFY
;
347 dev
->status
|= DEV_VERIFY_AND_FIX
;
349 dev
->vol
.migr_type
= migr_type
;
350 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
354 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
356 return ROUND_UP(bytes
, sector_size
) / sector_size
;
359 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
360 unsigned int sector_size
)
362 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
366 struct imsm_dev
*dev
;
367 struct intel_dev
*next
;
372 enum sys_dev_type type
;
375 struct intel_hba
*next
;
382 /* internal representation of IMSM metadata */
385 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
386 struct imsm_super
*anchor
; /* immovable parameters */
389 void *migr_rec_buf
; /* buffer for I/O operations */
390 struct migr_record
*migr_rec
; /* migration record */
392 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
393 array, it indicates that mdmon is allowed to clean migration
395 size_t len
; /* size of the 'buf' allocation */
396 size_t extra_space
; /* extra space in 'buf' that is not used yet */
397 void *next_buf
; /* for realloc'ing buf from the manager */
399 int updates_pending
; /* count of pending updates for mdmon */
400 int current_vol
; /* index of raid device undergoing creation */
401 unsigned long long create_offset
; /* common start for 'current_vol' */
402 __u32 random
; /* random data for seeding new family numbers */
403 struct intel_dev
*devlist
;
404 unsigned int sector_size
; /* sector size of used member drives */
408 __u8 serial
[MAX_RAID_SERIAL_LEN
];
411 struct imsm_disk disk
;
414 struct extent
*e
; /* for determining freespace @ create */
415 int raiddisk
; /* slot to fill in autolayout */
417 } *disks
, *current_disk
;
418 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
420 struct dl
*missing
; /* disks removed while we weren't looking */
421 struct bbm_log
*bbm_log
;
422 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
423 const struct imsm_orom
*orom
; /* platform firmware support */
424 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
425 struct md_bb bb
; /* memory for get_bad_blocks call */
429 struct imsm_disk disk
;
430 #define IMSM_UNKNOWN_OWNER (-1)
432 struct intel_disk
*next
;
436 unsigned long long start
, size
;
439 /* definitions of reshape process types */
440 enum imsm_reshape_type
{
446 /* definition of messages passed to imsm_process_update */
447 enum imsm_update_type
{
448 update_activate_spare
,
452 update_add_remove_disk
,
453 update_reshape_container_disks
,
454 update_reshape_migration
,
456 update_general_migration_checkpoint
,
458 update_prealloc_badblocks_mem
,
462 struct imsm_update_activate_spare
{
463 enum imsm_update_type type
;
467 struct imsm_update_activate_spare
*next
;
473 unsigned long long size
;
480 enum takeover_direction
{
484 struct imsm_update_takeover
{
485 enum imsm_update_type type
;
487 enum takeover_direction direction
;
490 struct imsm_update_reshape
{
491 enum imsm_update_type type
;
495 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
498 struct imsm_update_reshape_migration
{
499 enum imsm_update_type type
;
502 /* fields for array migration changes
509 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
512 struct imsm_update_size_change
{
513 enum imsm_update_type type
;
518 struct imsm_update_general_migration_checkpoint
{
519 enum imsm_update_type type
;
520 __u32 curr_migr_unit
;
524 __u8 serial
[MAX_RAID_SERIAL_LEN
];
527 struct imsm_update_create_array
{
528 enum imsm_update_type type
;
533 struct imsm_update_kill_array
{
534 enum imsm_update_type type
;
538 struct imsm_update_rename_array
{
539 enum imsm_update_type type
;
540 __u8 name
[MAX_RAID_SERIAL_LEN
];
544 struct imsm_update_add_remove_disk
{
545 enum imsm_update_type type
;
548 struct imsm_update_prealloc_bb_mem
{
549 enum imsm_update_type type
;
552 struct imsm_update_rwh_policy
{
553 enum imsm_update_type type
;
558 static const char *_sys_dev_type
[] = {
559 [SYS_DEV_UNKNOWN
] = "Unknown",
560 [SYS_DEV_SAS
] = "SAS",
561 [SYS_DEV_SATA
] = "SATA",
562 [SYS_DEV_NVME
] = "NVMe",
563 [SYS_DEV_VMD
] = "VMD"
566 const char *get_sys_dev_type(enum sys_dev_type type
)
568 if (type
>= SYS_DEV_MAX
)
569 type
= SYS_DEV_UNKNOWN
;
571 return _sys_dev_type
[type
];
574 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
576 struct intel_hba
*result
= xmalloc(sizeof(*result
));
578 result
->type
= device
->type
;
579 result
->path
= xstrdup(device
->path
);
581 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
587 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
589 struct intel_hba
*result
;
591 for (result
= hba
; result
; result
= result
->next
) {
592 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
598 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
600 struct intel_hba
*hba
;
602 /* check if disk attached to Intel HBA */
603 hba
= find_intel_hba(super
->hba
, device
);
606 /* Check if HBA is already attached to super */
607 if (super
->hba
== NULL
) {
608 super
->hba
= alloc_intel_hba(device
);
613 /* Intel metadata allows for all disks attached to the same type HBA.
614 * Do not support HBA types mixing
616 if (device
->type
!= hba
->type
)
619 /* Multiple same type HBAs can be used if they share the same OROM */
620 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
622 if (device_orom
!= super
->orom
)
628 hba
->next
= alloc_intel_hba(device
);
632 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
634 struct sys_dev
*list
, *elem
;
637 if ((list
= find_intel_devices()) == NULL
)
641 disk_path
= (char *) devname
;
643 disk_path
= diskfd_to_devpath(fd
);
648 for (elem
= list
; elem
; elem
= elem
->next
)
649 if (path_attached_to_hba(disk_path
, elem
->path
))
652 if (disk_path
!= devname
)
658 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
661 static struct supertype
*match_metadata_desc_imsm(char *arg
)
663 struct supertype
*st
;
665 if (strcmp(arg
, "imsm") != 0 &&
666 strcmp(arg
, "default") != 0
670 st
= xcalloc(1, sizeof(*st
));
671 st
->ss
= &super_imsm
;
672 st
->max_devs
= IMSM_MAX_DEVICES
;
673 st
->minor_version
= 0;
678 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
680 return &mpb
->sig
[MPB_SIG_LEN
];
683 /* retrieve a disk directly from the anchor when the anchor is known to be
684 * up-to-date, currently only at load time
686 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
688 if (index
>= mpb
->num_disks
)
690 return &mpb
->disk
[index
];
693 /* retrieve the disk description based on a index of the disk
696 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
700 for (d
= super
->disks
; d
; d
= d
->next
)
701 if (d
->index
== index
)
706 /* retrieve a disk from the parsed metadata */
707 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
711 dl
= get_imsm_dl_disk(super
, index
);
718 /* generate a checksum directly from the anchor when the anchor is known to be
719 * up-to-date, currently only at load or write_super after coalescing
721 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
723 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
724 __u32
*p
= (__u32
*) mpb
;
728 sum
+= __le32_to_cpu(*p
);
732 return sum
- __le32_to_cpu(mpb
->check_sum
);
735 static size_t sizeof_imsm_map(struct imsm_map
*map
)
737 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
740 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
742 /* A device can have 2 maps if it is in the middle of a migration.
744 * MAP_0 - we return the first map
745 * MAP_1 - we return the second map if it exists, else NULL
746 * MAP_X - we return the second map if it exists, else the first
748 struct imsm_map
*map
= &dev
->vol
.map
[0];
749 struct imsm_map
*map2
= NULL
;
751 if (dev
->vol
.migr_state
)
752 map2
= (void *)map
+ sizeof_imsm_map(map
);
754 switch (second_map
) {
771 /* return the size of the device.
772 * migr_state increases the returned size if map[0] were to be duplicated
774 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
776 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
777 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
779 /* migrating means an additional map */
780 if (dev
->vol
.migr_state
)
781 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
783 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
788 /* retrieve disk serial number list from a metadata update */
789 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
792 struct disk_info
*inf
;
794 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
795 sizeof_imsm_dev(&update
->dev
, 0);
800 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
806 if (index
>= mpb
->num_raid_devs
)
809 /* devices start after all disks */
810 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
812 for (i
= 0; i
<= index
; i
++)
814 return _mpb
+ offset
;
816 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
821 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
823 struct intel_dev
*dv
;
825 if (index
>= super
->anchor
->num_raid_devs
)
827 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
828 if (dv
->index
== index
)
833 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
836 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
837 __le16_to_cpu(addr
->w1
));
840 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
842 struct bbm_log_block_addr addr
;
844 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
845 addr
.dw1
= __cpu_to_le32((__u32
)(sec
>> 16) & 0xffffffff);
849 /* get size of the bbm log */
850 static __u32
get_imsm_bbm_log_size(struct bbm_log
*log
)
852 if (!log
|| log
->entry_count
== 0)
855 return sizeof(log
->signature
) +
856 sizeof(log
->entry_count
) +
857 log
->entry_count
* sizeof(struct bbm_log_entry
);
860 /* check if bad block is not partially stored in bbm log */
861 static int is_stored_in_bbm(struct bbm_log
*log
, const __u8 idx
, const unsigned
862 long long sector
, const int length
, __u32
*pos
)
866 for (i
= *pos
; i
< log
->entry_count
; i
++) {
867 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
868 unsigned long long bb_start
;
869 unsigned long long bb_end
;
871 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
872 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
874 if ((entry
->disk_ordinal
== idx
) && (bb_start
>= sector
) &&
875 (bb_end
<= sector
+ length
)) {
883 /* record new bad block in bbm log */
884 static int record_new_badblock(struct bbm_log
*log
, const __u8 idx
, unsigned
885 long long sector
, int length
)
889 struct bbm_log_entry
*entry
= NULL
;
891 while (is_stored_in_bbm(log
, idx
, sector
, length
, &pos
)) {
892 struct bbm_log_entry
*e
= &log
->marked_block_entries
[pos
];
894 if ((e
->marked_count
+ 1 == BBM_LOG_MAX_LBA_ENTRY_VAL
) &&
895 (__le48_to_cpu(&e
->defective_block_start
) == sector
)) {
896 sector
+= BBM_LOG_MAX_LBA_ENTRY_VAL
;
897 length
-= BBM_LOG_MAX_LBA_ENTRY_VAL
;
906 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
907 BBM_LOG_MAX_LBA_ENTRY_VAL
;
908 entry
->defective_block_start
= __cpu_to_le48(sector
);
909 entry
->marked_count
= cnt
- 1;
916 new_bb
= ROUND_UP(length
, BBM_LOG_MAX_LBA_ENTRY_VAL
) /
917 BBM_LOG_MAX_LBA_ENTRY_VAL
;
918 if (log
->entry_count
+ new_bb
> BBM_LOG_MAX_ENTRIES
)
922 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
923 BBM_LOG_MAX_LBA_ENTRY_VAL
;
924 struct bbm_log_entry
*entry
=
925 &log
->marked_block_entries
[log
->entry_count
];
927 entry
->defective_block_start
= __cpu_to_le48(sector
);
928 entry
->marked_count
= cnt
- 1;
929 entry
->disk_ordinal
= idx
;
940 /* clear all bad blocks for given disk */
941 static void clear_disk_badblocks(struct bbm_log
*log
, const __u8 idx
)
945 while (i
< log
->entry_count
) {
946 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
948 if (entries
[i
].disk_ordinal
== idx
) {
949 if (i
< log
->entry_count
- 1)
950 entries
[i
] = entries
[log
->entry_count
- 1];
958 /* clear given bad block */
959 static int clear_badblock(struct bbm_log
*log
, const __u8 idx
, const unsigned
960 long long sector
, const int length
) {
963 while (i
< log
->entry_count
) {
964 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
966 if ((entries
[i
].disk_ordinal
== idx
) &&
967 (__le48_to_cpu(&entries
[i
].defective_block_start
) ==
968 sector
) && (entries
[i
].marked_count
+ 1 == length
)) {
969 if (i
< log
->entry_count
- 1)
970 entries
[i
] = entries
[log
->entry_count
- 1];
980 /* allocate and load BBM log from metadata */
981 static int load_bbm_log(struct intel_super
*super
)
983 struct imsm_super
*mpb
= super
->anchor
;
984 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
986 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
991 struct bbm_log
*log
= (void *)mpb
+
992 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
996 if (bbm_log_size
< sizeof(log
->signature
) +
997 sizeof(log
->entry_count
))
1000 entry_count
= __le32_to_cpu(log
->entry_count
);
1001 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
1002 (entry_count
> BBM_LOG_MAX_ENTRIES
))
1006 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
1007 entry_count
* sizeof(struct bbm_log_entry
))
1010 memcpy(super
->bbm_log
, log
, bbm_log_size
);
1012 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
1013 super
->bbm_log
->entry_count
= 0;
1019 /* checks if bad block is within volume boundaries */
1020 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
1021 const unsigned long long start_sector
,
1022 const unsigned long long size
)
1024 unsigned long long bb_start
;
1025 unsigned long long bb_end
;
1027 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1028 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1030 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1031 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1037 /* get list of bad blocks on a drive for a volume */
1038 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1039 const unsigned long long start_sector
,
1040 const unsigned long long size
,
1046 for (i
= 0; i
< log
->entry_count
; i
++) {
1047 const struct bbm_log_entry
*ent
=
1048 &log
->marked_block_entries
[i
];
1049 struct md_bb_entry
*bb
;
1051 if ((ent
->disk_ordinal
== idx
) &&
1052 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1054 if (!bbs
->entries
) {
1055 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1061 bb
= &bbs
->entries
[count
++];
1062 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1063 bb
->length
= ent
->marked_count
+ 1;
1071 * == MAP_0 get first map
1072 * == MAP_1 get second map
1073 * == MAP_X than get map according to the current migr_state
1075 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1079 struct imsm_map
*map
;
1081 map
= get_imsm_map(dev
, second_map
);
1083 /* top byte identifies disk under rebuild */
1084 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1087 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1088 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1090 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1092 return ord_to_idx(ord
);
1095 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1097 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1100 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
1105 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1106 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1107 if (ord_to_idx(ord
) == idx
)
1114 static int get_imsm_raid_level(struct imsm_map
*map
)
1116 if (map
->raid_level
== 1) {
1117 if (map
->num_members
== 2)
1123 return map
->raid_level
;
1126 static int cmp_extent(const void *av
, const void *bv
)
1128 const struct extent
*a
= av
;
1129 const struct extent
*b
= bv
;
1130 if (a
->start
< b
->start
)
1132 if (a
->start
> b
->start
)
1137 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1139 int memberships
= 0;
1142 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1143 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1144 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1146 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
1153 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1155 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
1157 if (lo
== 0 || hi
== 0)
1159 *lo
= __le32_to_cpu((unsigned)n
);
1160 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
1164 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1166 return (unsigned long long)__le32_to_cpu(lo
) |
1167 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1170 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1174 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1177 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1181 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1184 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1188 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1191 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1195 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1198 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1200 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1203 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1205 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1208 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1210 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1213 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1215 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1218 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
1220 /* find a list of used extents on the given physical device */
1221 struct extent
*rv
, *e
;
1223 int memberships
= count_memberships(dl
, super
);
1226 /* trim the reserved area for spares, so they can join any array
1227 * regardless of whether the OROM has assigned sectors from the
1228 * IMSM_RESERVED_SECTORS region
1230 if (dl
->index
== -1)
1231 reservation
= imsm_min_reserved_sectors(super
);
1233 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1235 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
1238 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1239 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1240 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1242 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1243 e
->start
= pba_of_lba0(map
);
1244 e
->size
= blocks_per_member(map
);
1248 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1250 /* determine the start of the metadata
1251 * when no raid devices are defined use the default
1252 * ...otherwise allow the metadata to truncate the value
1253 * as is the case with older versions of imsm
1256 struct extent
*last
= &rv
[memberships
- 1];
1257 unsigned long long remainder
;
1259 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1260 /* round down to 1k block to satisfy precision of the kernel
1264 /* make sure remainder is still sane */
1265 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1266 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1267 if (reservation
> remainder
)
1268 reservation
= remainder
;
1270 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1275 /* try to determine how much space is reserved for metadata from
1276 * the last get_extents() entry, otherwise fallback to the
1279 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1285 /* for spares just return a minimal reservation which will grow
1286 * once the spare is picked up by an array
1288 if (dl
->index
== -1)
1289 return MPB_SECTOR_CNT
;
1291 e
= get_extents(super
, dl
);
1293 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1295 /* scroll to last entry */
1296 for (i
= 0; e
[i
].size
; i
++)
1299 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1306 static int is_spare(struct imsm_disk
*disk
)
1308 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1311 static int is_configured(struct imsm_disk
*disk
)
1313 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1316 static int is_failed(struct imsm_disk
*disk
)
1318 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1321 static int is_journal(struct imsm_disk
*disk
)
1323 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1326 /* round array size down to closest MB and ensure it splits evenly
1329 static unsigned long long round_size_to_mb(unsigned long long size
, unsigned int
1333 size
= (size
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
1339 /* try to determine how much space is reserved for metadata from
1340 * the last get_extents() entry on the smallest active disk,
1341 * otherwise fallback to the default
1343 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1347 unsigned long long min_active
;
1349 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1350 struct dl
*dl
, *dl_min
= NULL
;
1356 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1359 unsigned long long blocks
= total_blocks(&dl
->disk
);
1360 if (blocks
< min_active
|| min_active
== 0) {
1362 min_active
= blocks
;
1368 /* find last lba used by subarrays on the smallest active disk */
1369 e
= get_extents(super
, dl_min
);
1372 for (i
= 0; e
[i
].size
; i
++)
1375 remainder
= min_active
- e
[i
].start
;
1378 /* to give priority to recovery we should not require full
1379 IMSM_RESERVED_SECTORS from the spare */
1380 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1382 /* if real reservation is smaller use that value */
1383 return (remainder
< rv
) ? remainder
: rv
;
1387 * Return minimum size of a spare and sector size
1388 * that can be used in this array
1390 int get_spare_criteria_imsm(struct supertype
*st
, struct spare_criteria
*c
)
1392 struct intel_super
*super
= st
->sb
;
1396 unsigned long long size
= 0;
1403 /* find first active disk in array */
1405 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1409 /* find last lba used by subarrays */
1410 e
= get_extents(super
, dl
);
1413 for (i
= 0; e
[i
].size
; i
++)
1416 size
= e
[i
-1].start
+ e
[i
-1].size
;
1419 /* add the amount of space needed for metadata */
1420 size
+= imsm_min_reserved_sectors(super
);
1422 c
->min_size
= size
* 512;
1423 c
->sector_size
= super
->sector_size
;
1428 static int is_gen_migration(struct imsm_dev
*dev
);
1430 #define IMSM_4K_DIV 8
1432 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1433 struct imsm_dev
*dev
);
1435 static void print_imsm_dev(struct intel_super
*super
,
1436 struct imsm_dev
*dev
,
1442 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1443 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1447 printf("[%.16s]:\n", dev
->volume
);
1448 printf(" UUID : %s\n", uuid
);
1449 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1451 printf(" <-- %d", get_imsm_raid_level(map2
));
1453 printf(" Members : %d", map
->num_members
);
1455 printf(" <-- %d", map2
->num_members
);
1457 printf(" Slots : [");
1458 for (i
= 0; i
< map
->num_members
; i
++) {
1459 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1460 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1465 for (i
= 0; i
< map2
->num_members
; i
++) {
1466 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1467 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1472 printf(" Failed disk : ");
1473 if (map
->failed_disk_num
== 0xff)
1476 printf("%i", map
->failed_disk_num
);
1478 slot
= get_imsm_disk_slot(map
, disk_idx
);
1480 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1481 printf(" This Slot : %d%s\n", slot
,
1482 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1484 printf(" This Slot : ?\n");
1485 printf(" Sector Size : %u\n", super
->sector_size
);
1486 sz
= __le32_to_cpu(dev
->size_high
);
1488 sz
+= __le32_to_cpu(dev
->size_low
);
1489 printf(" Array Size : %llu%s\n",
1490 (unsigned long long)sz
* 512 / super
->sector_size
,
1491 human_size(sz
* 512));
1492 sz
= blocks_per_member(map
);
1493 printf(" Per Dev Size : %llu%s\n",
1494 (unsigned long long)sz
* 512 / super
->sector_size
,
1495 human_size(sz
* 512));
1496 printf(" Sector Offset : %llu\n",
1498 printf(" Num Stripes : %llu\n",
1499 num_data_stripes(map
));
1500 printf(" Chunk Size : %u KiB",
1501 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1503 printf(" <-- %u KiB",
1504 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1506 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1507 printf(" Migrate State : ");
1508 if (dev
->vol
.migr_state
) {
1509 if (migr_type(dev
) == MIGR_INIT
)
1510 printf("initialize\n");
1511 else if (migr_type(dev
) == MIGR_REBUILD
)
1512 printf("rebuild\n");
1513 else if (migr_type(dev
) == MIGR_VERIFY
)
1515 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1516 printf("general migration\n");
1517 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1518 printf("state change\n");
1519 else if (migr_type(dev
) == MIGR_REPAIR
)
1522 printf("<unknown:%d>\n", migr_type(dev
));
1525 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1526 if (dev
->vol
.migr_state
) {
1527 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1529 printf(" <-- %s", map_state_str
[map
->map_state
]);
1530 printf("\n Checkpoint : %u ",
1531 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1532 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1535 printf("(%llu)", (unsigned long long)
1536 blocks_per_migr_unit(super
, dev
));
1539 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1541 printf(" RWH Policy : ");
1542 if (dev
->rwh_policy
== RWH_OFF
)
1544 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1545 printf("PPL distributed\n");
1546 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1547 printf("PPL journaling drive\n");
1549 printf("<unknown:%d>\n", dev
->rwh_policy
);
1552 static void print_imsm_disk(struct imsm_disk
*disk
,
1555 unsigned int sector_size
) {
1556 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1559 if (index
< -1 || !disk
)
1563 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1565 printf(" Disk%02d Serial : %s\n", index
, str
);
1567 printf(" Disk Serial : %s\n", str
);
1568 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1569 is_configured(disk
) ? " active" : "",
1570 is_failed(disk
) ? " failed" : "",
1571 is_journal(disk
) ? " journal" : "");
1572 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1573 sz
= total_blocks(disk
) - reserved
;
1574 printf(" Usable Size : %llu%s\n",
1575 (unsigned long long)sz
* 512 / sector_size
,
1576 human_size(sz
* 512));
1579 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1581 struct migr_record
*migr_rec
= super
->migr_rec
;
1583 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1584 migr_rec
->ckpt_area_pba
/= IMSM_4K_DIV
;
1585 migr_rec
->dest_1st_member_lba
/= IMSM_4K_DIV
;
1586 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1587 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1588 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1589 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1592 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1594 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1597 void convert_to_4k(struct intel_super
*super
)
1599 struct imsm_super
*mpb
= super
->anchor
;
1600 struct imsm_disk
*disk
;
1602 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1604 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1605 disk
= __get_imsm_disk(mpb
, i
);
1607 convert_to_4k_imsm_disk(disk
);
1609 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1610 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1611 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1613 split_ull((join_u32(dev
->size_low
, dev
->size_high
)/IMSM_4K_DIV
),
1614 &dev
->size_low
, &dev
->size_high
);
1615 dev
->vol
.curr_migr_unit
/= IMSM_4K_DIV
;
1618 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1619 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1620 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1622 if (dev
->vol
.migr_state
) {
1624 map
= get_imsm_map(dev
, MAP_1
);
1625 set_blocks_per_member(map
,
1626 blocks_per_member(map
)/IMSM_4K_DIV
);
1627 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1628 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1632 struct bbm_log
*log
= (void *)mpb
+
1633 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1636 for (i
= 0; i
< log
->entry_count
; i
++) {
1637 struct bbm_log_entry
*entry
=
1638 &log
->marked_block_entries
[i
];
1640 __u8 count
= entry
->marked_count
+ 1;
1641 unsigned long long sector
=
1642 __le48_to_cpu(&entry
->defective_block_start
);
1644 entry
->defective_block_start
=
1645 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1646 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1650 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1653 void examine_migr_rec_imsm(struct intel_super
*super
)
1655 struct migr_record
*migr_rec
= super
->migr_rec
;
1656 struct imsm_super
*mpb
= super
->anchor
;
1659 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1660 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1661 struct imsm_map
*map
;
1664 if (is_gen_migration(dev
) == 0)
1667 printf("\nMigration Record Information:");
1669 /* first map under migration */
1670 map
= get_imsm_map(dev
, MAP_0
);
1672 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1673 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1674 printf(" Empty\n ");
1675 printf("Examine one of first two disks in array\n");
1678 printf("\n Status : ");
1679 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1682 printf("Contains Data\n");
1683 printf(" Current Unit : %u\n",
1684 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1685 printf(" Family : %u\n",
1686 __le32_to_cpu(migr_rec
->family_num
));
1687 printf(" Ascending : %u\n",
1688 __le32_to_cpu(migr_rec
->ascending_migr
));
1689 printf(" Blocks Per Unit : %u\n",
1690 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1691 printf(" Dest. Depth Per Unit : %u\n",
1692 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1693 printf(" Checkpoint Area pba : %u\n",
1694 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1695 printf(" First member lba : %u\n",
1696 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1697 printf(" Total Number of Units : %u\n",
1698 __le32_to_cpu(migr_rec
->num_migr_units
));
1699 printf(" Size of volume : %u\n",
1700 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1701 printf(" Expansion space for LBA64 : %u\n",
1702 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1703 printf(" Record was read from : %u\n",
1704 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1710 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1712 struct migr_record
*migr_rec
= super
->migr_rec
;
1714 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1715 migr_rec
->ckpt_area_pba
*= IMSM_4K_DIV
;
1716 migr_rec
->dest_1st_member_lba
*= IMSM_4K_DIV
;
1717 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1718 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1719 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1720 &migr_rec
->post_migr_vol_cap
,
1721 &migr_rec
->post_migr_vol_cap_hi
);
1724 void convert_from_4k(struct intel_super
*super
)
1726 struct imsm_super
*mpb
= super
->anchor
;
1727 struct imsm_disk
*disk
;
1729 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1731 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1732 disk
= __get_imsm_disk(mpb
, i
);
1734 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1737 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1738 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1739 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1741 split_ull((join_u32(dev
->size_low
, dev
->size_high
)*IMSM_4K_DIV
),
1742 &dev
->size_low
, &dev
->size_high
);
1743 dev
->vol
.curr_migr_unit
*= IMSM_4K_DIV
;
1746 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1747 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1748 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1750 if (dev
->vol
.migr_state
) {
1752 map
= get_imsm_map(dev
, MAP_1
);
1753 set_blocks_per_member(map
,
1754 blocks_per_member(map
)*IMSM_4K_DIV
);
1755 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1756 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1760 struct bbm_log
*log
= (void *)mpb
+
1761 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1764 for (i
= 0; i
< log
->entry_count
; i
++) {
1765 struct bbm_log_entry
*entry
=
1766 &log
->marked_block_entries
[i
];
1768 __u8 count
= entry
->marked_count
+ 1;
1769 unsigned long long sector
=
1770 __le48_to_cpu(&entry
->defective_block_start
);
1772 entry
->defective_block_start
=
1773 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1774 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1778 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1781 /*******************************************************************************
1782 * function: imsm_check_attributes
1783 * Description: Function checks if features represented by attributes flags
1784 * are supported by mdadm.
1786 * attributes - Attributes read from metadata
1788 * 0 - passed attributes contains unsupported features flags
1789 * 1 - all features are supported
1790 ******************************************************************************/
1791 static int imsm_check_attributes(__u32 attributes
)
1794 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1796 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1798 not_supported
&= attributes
;
1799 if (not_supported
) {
1800 pr_err("(IMSM): Unsupported attributes : %x\n",
1801 (unsigned)__le32_to_cpu(not_supported
));
1802 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1803 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1804 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1806 if (not_supported
& MPB_ATTRIB_2TB
) {
1807 dprintf("\t\tMPB_ATTRIB_2TB\n");
1808 not_supported
^= MPB_ATTRIB_2TB
;
1810 if (not_supported
& MPB_ATTRIB_RAID0
) {
1811 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1812 not_supported
^= MPB_ATTRIB_RAID0
;
1814 if (not_supported
& MPB_ATTRIB_RAID1
) {
1815 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1816 not_supported
^= MPB_ATTRIB_RAID1
;
1818 if (not_supported
& MPB_ATTRIB_RAID10
) {
1819 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1820 not_supported
^= MPB_ATTRIB_RAID10
;
1822 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1823 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1824 not_supported
^= MPB_ATTRIB_RAID1E
;
1826 if (not_supported
& MPB_ATTRIB_RAID5
) {
1827 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1828 not_supported
^= MPB_ATTRIB_RAID5
;
1830 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1831 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1832 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1834 if (not_supported
& MPB_ATTRIB_BBM
) {
1835 dprintf("\t\tMPB_ATTRIB_BBM\n");
1836 not_supported
^= MPB_ATTRIB_BBM
;
1838 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1839 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1840 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1842 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1843 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1844 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1846 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1847 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1848 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1850 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1851 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1852 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1854 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1855 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1856 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1860 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1868 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1870 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1872 struct intel_super
*super
= st
->sb
;
1873 struct imsm_super
*mpb
= super
->anchor
;
1874 char str
[MAX_SIGNATURE_LENGTH
];
1879 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1882 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
1883 str
[MPB_SIG_LEN
-1] = '\0';
1884 printf(" Magic : %s\n", str
);
1885 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1886 printf(" Version : %s\n", get_imsm_version(mpb
));
1887 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1888 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1889 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1890 printf(" Attributes : ");
1891 if (imsm_check_attributes(mpb
->attributes
))
1892 printf("All supported\n");
1894 printf("not supported\n");
1895 getinfo_super_imsm(st
, &info
, NULL
);
1896 fname_from_uuid(st
, &info
, nbuf
, ':');
1897 printf(" UUID : %s\n", nbuf
+ 5);
1898 sum
= __le32_to_cpu(mpb
->check_sum
);
1899 printf(" Checksum : %08x %s\n", sum
,
1900 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1901 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
1902 printf(" Disks : %d\n", mpb
->num_disks
);
1903 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1904 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
1905 super
->disks
->index
, reserved
, super
->sector_size
);
1906 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
1907 struct bbm_log
*log
= super
->bbm_log
;
1910 printf("Bad Block Management Log:\n");
1911 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1912 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1913 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1915 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1917 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1919 super
->current_vol
= i
;
1920 getinfo_super_imsm(st
, &info
, NULL
);
1921 fname_from_uuid(st
, &info
, nbuf
, ':');
1922 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1924 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1925 if (i
== super
->disks
->index
)
1927 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
1928 super
->sector_size
);
1931 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1932 if (dl
->index
== -1)
1933 print_imsm_disk(&dl
->disk
, -1, reserved
,
1934 super
->sector_size
);
1936 examine_migr_rec_imsm(super
);
1939 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1941 /* We just write a generic IMSM ARRAY entry */
1944 struct intel_super
*super
= st
->sb
;
1946 if (!super
->anchor
->num_raid_devs
) {
1947 printf("ARRAY metadata=imsm\n");
1951 getinfo_super_imsm(st
, &info
, NULL
);
1952 fname_from_uuid(st
, &info
, nbuf
, ':');
1953 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1956 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1958 /* We just write a generic IMSM ARRAY entry */
1962 struct intel_super
*super
= st
->sb
;
1965 if (!super
->anchor
->num_raid_devs
)
1968 getinfo_super_imsm(st
, &info
, NULL
);
1969 fname_from_uuid(st
, &info
, nbuf
, ':');
1970 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1971 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1973 super
->current_vol
= i
;
1974 getinfo_super_imsm(st
, &info
, NULL
);
1975 fname_from_uuid(st
, &info
, nbuf1
, ':');
1976 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1977 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1981 static void export_examine_super_imsm(struct supertype
*st
)
1983 struct intel_super
*super
= st
->sb
;
1984 struct imsm_super
*mpb
= super
->anchor
;
1988 getinfo_super_imsm(st
, &info
, NULL
);
1989 fname_from_uuid(st
, &info
, nbuf
, ':');
1990 printf("MD_METADATA=imsm\n");
1991 printf("MD_LEVEL=container\n");
1992 printf("MD_UUID=%s\n", nbuf
+5);
1993 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1996 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
1998 /* The second last sector of the device contains
1999 * the "struct imsm_super" metadata.
2000 * This contains mpb_size which is the size in bytes of the
2001 * extended metadata. This is located immediately before
2003 * We want to read all that, plus the last sector which
2004 * may contain a migration record, and write it all
2008 unsigned long long dsize
, offset
;
2010 struct imsm_super
*sb
;
2011 struct intel_super
*super
= st
->sb
;
2012 unsigned int sector_size
= super
->sector_size
;
2013 unsigned int written
= 0;
2015 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
) != 0)
2018 if (!get_dev_size(from
, NULL
, &dsize
))
2021 if (lseek64(from
, dsize
-(2*sector_size
), 0) < 0)
2023 if ((unsigned int)read(from
, buf
, sector_size
) != sector_size
)
2026 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
2029 sectors
= mpb_sectors(sb
, sector_size
) + 2;
2030 offset
= dsize
- sectors
* sector_size
;
2031 if (lseek64(from
, offset
, 0) < 0 ||
2032 lseek64(to
, offset
, 0) < 0)
2034 while (written
< sectors
* sector_size
) {
2035 int n
= sectors
*sector_size
- written
;
2038 if (read(from
, buf
, n
) != n
)
2040 if (write(to
, buf
, n
) != n
)
2051 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
2056 getinfo_super_imsm(st
, &info
, NULL
);
2057 fname_from_uuid(st
, &info
, nbuf
, ':');
2058 printf("\n UUID : %s\n", nbuf
+ 5);
2061 static void brief_detail_super_imsm(struct supertype
*st
)
2065 getinfo_super_imsm(st
, &info
, NULL
);
2066 fname_from_uuid(st
, &info
, nbuf
, ':');
2067 printf(" UUID=%s", nbuf
+ 5);
2070 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
2071 static void fd2devname(int fd
, char *name
);
2073 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2075 /* dump an unsorted list of devices attached to AHCI Intel storage
2076 * controller, as well as non-connected ports
2078 int hba_len
= strlen(hba_path
) + 1;
2083 unsigned long port_mask
= (1 << port_count
) - 1;
2085 if (port_count
> (int)sizeof(port_mask
) * 8) {
2087 pr_err("port_count %d out of range\n", port_count
);
2091 /* scroll through /sys/dev/block looking for devices attached to
2094 dir
= opendir("/sys/dev/block");
2098 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2109 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2111 path
= devt_to_devpath(makedev(major
, minor
));
2114 if (!path_attached_to_hba(path
, hba_path
)) {
2120 /* retrieve the scsi device type */
2121 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2123 pr_err("failed to allocate 'device'\n");
2127 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2128 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2130 pr_err("failed to read device type for %s\n",
2136 type
= strtoul(buf
, NULL
, 10);
2138 /* if it's not a disk print the vendor and model */
2139 if (!(type
== 0 || type
== 7 || type
== 14)) {
2142 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2143 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2144 strncpy(vendor
, buf
, sizeof(vendor
));
2145 vendor
[sizeof(vendor
) - 1] = '\0';
2146 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2147 while (isspace(*c
) || *c
== '\0')
2151 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2152 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2153 strncpy(model
, buf
, sizeof(model
));
2154 model
[sizeof(model
) - 1] = '\0';
2155 c
= (char *) &model
[sizeof(model
) - 1];
2156 while (isspace(*c
) || *c
== '\0')
2160 if (vendor
[0] && model
[0])
2161 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2163 switch (type
) { /* numbers from hald/linux/device.c */
2164 case 1: sprintf(buf
, "tape"); break;
2165 case 2: sprintf(buf
, "printer"); break;
2166 case 3: sprintf(buf
, "processor"); break;
2168 case 5: sprintf(buf
, "cdrom"); break;
2169 case 6: sprintf(buf
, "scanner"); break;
2170 case 8: sprintf(buf
, "media_changer"); break;
2171 case 9: sprintf(buf
, "comm"); break;
2172 case 12: sprintf(buf
, "raid"); break;
2173 default: sprintf(buf
, "unknown");
2179 /* chop device path to 'host%d' and calculate the port number */
2180 c
= strchr(&path
[hba_len
], '/');
2183 pr_err("%s - invalid path name\n", path
+ hba_len
);
2188 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2189 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2193 *c
= '/'; /* repair the full string */
2194 pr_err("failed to determine port number for %s\n",
2201 /* mark this port as used */
2202 port_mask
&= ~(1 << port
);
2204 /* print out the device information */
2206 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2210 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2212 printf(" Port%d : - disk info unavailable -\n", port
);
2214 fd2devname(fd
, buf
);
2215 printf(" Port%d : %s", port
, buf
);
2216 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
2217 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
2232 for (i
= 0; i
< port_count
; i
++)
2233 if (port_mask
& (1 << i
))
2234 printf(" Port%d : - no device attached -\n", i
);
2240 static int print_vmd_attached_devs(struct sys_dev
*hba
)
2248 if (hba
->type
!= SYS_DEV_VMD
)
2251 /* scroll through /sys/dev/block looking for devices attached to
2254 dir
= opendir("/sys/bus/pci/drivers/nvme");
2258 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2261 /* is 'ent' a device? check that the 'subsystem' link exists and
2262 * that its target matches 'bus'
2264 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
2266 n
= readlink(path
, link
, sizeof(link
));
2267 if (n
< 0 || n
>= (int)sizeof(link
))
2270 c
= strrchr(link
, '/');
2273 if (strncmp("pci", c
+1, strlen("pci")) != 0)
2276 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
2278 rp
= realpath(path
, NULL
);
2282 if (path_attached_to_hba(rp
, hba
->path
)) {
2283 printf(" NVMe under VMD : %s\n", rp
);
2292 static void print_found_intel_controllers(struct sys_dev
*elem
)
2294 for (; elem
; elem
= elem
->next
) {
2295 pr_err("found Intel(R) ");
2296 if (elem
->type
== SYS_DEV_SATA
)
2297 fprintf(stderr
, "SATA ");
2298 else if (elem
->type
== SYS_DEV_SAS
)
2299 fprintf(stderr
, "SAS ");
2300 else if (elem
->type
== SYS_DEV_NVME
)
2301 fprintf(stderr
, "NVMe ");
2303 if (elem
->type
== SYS_DEV_VMD
)
2304 fprintf(stderr
, "VMD domain");
2306 fprintf(stderr
, "RAID controller");
2309 fprintf(stderr
, " at %s", elem
->pci_id
);
2310 fprintf(stderr
, ".\n");
2315 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2322 if ((dir
= opendir(hba_path
)) == NULL
)
2325 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2328 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2329 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2331 if (*port_count
== 0)
2333 else if (host
< host_base
)
2336 if (host
+ 1 > *port_count
+ host_base
)
2337 *port_count
= host
+ 1 - host_base
;
2343 static void print_imsm_capability(const struct imsm_orom
*orom
)
2345 printf(" Platform : Intel(R) ");
2346 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2347 printf("Matrix Storage Manager\n");
2349 printf("Rapid Storage Technology%s\n",
2350 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2351 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2352 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2353 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2354 printf(" RAID Levels :%s%s%s%s%s\n",
2355 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2356 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2357 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2358 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2359 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2360 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2361 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2362 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2363 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2364 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2365 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2366 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2367 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2368 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2369 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2370 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2371 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2372 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2373 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2374 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2375 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2376 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2377 printf(" 2TB volumes :%s supported\n",
2378 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2379 printf(" 2TB disks :%s supported\n",
2380 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2381 printf(" Max Disks : %d\n", orom
->tds
);
2382 printf(" Max Volumes : %d per array, %d per %s\n",
2383 orom
->vpa
, orom
->vphba
,
2384 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2388 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2390 printf("MD_FIRMWARE_TYPE=imsm\n");
2391 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2392 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2393 orom
->hotfix_ver
, orom
->build
);
2394 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2395 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2396 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2397 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2398 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2399 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2400 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2401 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2402 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2403 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2404 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2405 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2406 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2407 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2408 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2409 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2410 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2411 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2412 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2413 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2414 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2415 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2416 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2417 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2418 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2419 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2420 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2421 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2424 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2426 /* There are two components to imsm platform support, the ahci SATA
2427 * controller and the option-rom. To find the SATA controller we
2428 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2429 * controller with the Intel vendor id is present. This approach
2430 * allows mdadm to leverage the kernel's ahci detection logic, with the
2431 * caveat that if ahci.ko is not loaded mdadm will not be able to
2432 * detect platform raid capabilities. The option-rom resides in a
2433 * platform "Adapter ROM". We scan for its signature to retrieve the
2434 * platform capabilities. If raid support is disabled in the BIOS the
2435 * option-rom capability structure will not be available.
2437 struct sys_dev
*list
, *hba
;
2442 if (enumerate_only
) {
2443 if (check_env("IMSM_NO_PLATFORM"))
2445 list
= find_intel_devices();
2448 for (hba
= list
; hba
; hba
= hba
->next
) {
2449 if (find_imsm_capability(hba
)) {
2459 list
= find_intel_devices();
2462 pr_err("no active Intel(R) RAID controller found.\n");
2464 } else if (verbose
> 0)
2465 print_found_intel_controllers(list
);
2467 for (hba
= list
; hba
; hba
= hba
->next
) {
2468 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2470 if (!find_imsm_capability(hba
)) {
2472 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2473 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2474 get_sys_dev_type(hba
->type
));
2480 if (controller_path
&& result
== 1) {
2481 pr_err("no active Intel(R) RAID controller found under %s\n",
2486 const struct orom_entry
*entry
;
2488 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2489 if (entry
->type
== SYS_DEV_VMD
) {
2490 print_imsm_capability(&entry
->orom
);
2491 printf(" 3rd party NVMe :%s supported\n",
2492 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2493 for (hba
= list
; hba
; hba
= hba
->next
) {
2494 if (hba
->type
== SYS_DEV_VMD
) {
2496 printf(" I/O Controller : %s (%s)\n",
2497 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2498 if (print_vmd_attached_devs(hba
)) {
2500 pr_err("failed to get devices attached to VMD domain.\n");
2509 print_imsm_capability(&entry
->orom
);
2510 if (entry
->type
== SYS_DEV_NVME
) {
2511 for (hba
= list
; hba
; hba
= hba
->next
) {
2512 if (hba
->type
== SYS_DEV_NVME
)
2513 printf(" NVMe Device : %s\n", hba
->path
);
2519 struct devid_list
*devid
;
2520 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2521 hba
= device_by_id(devid
->devid
);
2525 printf(" I/O Controller : %s (%s)\n",
2526 hba
->path
, get_sys_dev_type(hba
->type
));
2527 if (hba
->type
== SYS_DEV_SATA
) {
2528 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2529 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2531 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2542 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2544 struct sys_dev
*list
, *hba
;
2547 list
= find_intel_devices();
2550 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2555 for (hba
= list
; hba
; hba
= hba
->next
) {
2556 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2558 if (!find_imsm_capability(hba
) && verbose
> 0) {
2560 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2561 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2567 const struct orom_entry
*entry
;
2569 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2570 if (entry
->type
== SYS_DEV_VMD
) {
2571 for (hba
= list
; hba
; hba
= hba
->next
)
2572 print_imsm_capability_export(&entry
->orom
);
2575 print_imsm_capability_export(&entry
->orom
);
2581 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2583 /* the imsm metadata format does not specify any host
2584 * identification information. We return -1 since we can never
2585 * confirm nor deny whether a given array is "meant" for this
2586 * host. We rely on compare_super and the 'family_num' fields to
2587 * exclude member disks that do not belong, and we rely on
2588 * mdadm.conf to specify the arrays that should be assembled.
2589 * Auto-assembly may still pick up "foreign" arrays.
2595 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2597 /* The uuid returned here is used for:
2598 * uuid to put into bitmap file (Create, Grow)
2599 * uuid for backup header when saving critical section (Grow)
2600 * comparing uuids when re-adding a device into an array
2601 * In these cases the uuid required is that of the data-array,
2602 * not the device-set.
2603 * uuid to recognise same set when adding a missing device back
2604 * to an array. This is a uuid for the device-set.
2606 * For each of these we can make do with a truncated
2607 * or hashed uuid rather than the original, as long as
2609 * In each case the uuid required is that of the data-array,
2610 * not the device-set.
2612 /* imsm does not track uuid's so we synthesis one using sha1 on
2613 * - The signature (Which is constant for all imsm array, but no matter)
2614 * - the orig_family_num of the container
2615 * - the index number of the volume
2616 * - the 'serial' number of the volume.
2617 * Hopefully these are all constant.
2619 struct intel_super
*super
= st
->sb
;
2622 struct sha1_ctx ctx
;
2623 struct imsm_dev
*dev
= NULL
;
2626 /* some mdadm versions failed to set ->orig_family_num, in which
2627 * case fall back to ->family_num. orig_family_num will be
2628 * fixed up with the first metadata update.
2630 family_num
= super
->anchor
->orig_family_num
;
2631 if (family_num
== 0)
2632 family_num
= super
->anchor
->family_num
;
2633 sha1_init_ctx(&ctx
);
2634 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2635 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2636 if (super
->current_vol
>= 0)
2637 dev
= get_imsm_dev(super
, super
->current_vol
);
2639 __u32 vol
= super
->current_vol
;
2640 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2641 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2643 sha1_finish_ctx(&ctx
, buf
);
2644 memcpy(uuid
, buf
, 4*4);
2649 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2651 __u8
*v
= get_imsm_version(mpb
);
2652 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2653 char major
[] = { 0, 0, 0 };
2654 char minor
[] = { 0 ,0, 0 };
2655 char patch
[] = { 0, 0, 0 };
2656 char *ver_parse
[] = { major
, minor
, patch
};
2660 while (*v
!= '\0' && v
< end
) {
2661 if (*v
!= '.' && j
< 2)
2662 ver_parse
[i
][j
++] = *v
;
2670 *m
= strtol(minor
, NULL
, 0);
2671 *p
= strtol(patch
, NULL
, 0);
2675 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2677 /* migr_strip_size when repairing or initializing parity */
2678 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2679 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2681 switch (get_imsm_raid_level(map
)) {
2686 return 128*1024 >> 9;
2690 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2692 /* migr_strip_size when rebuilding a degraded disk, no idea why
2693 * this is different than migr_strip_size_resync(), but it's good
2696 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2697 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2699 switch (get_imsm_raid_level(map
)) {
2702 if (map
->num_members
% map
->num_domains
== 0)
2703 return 128*1024 >> 9;
2707 return max((__u32
) 64*1024 >> 9, chunk
);
2709 return 128*1024 >> 9;
2713 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2715 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2716 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2717 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2718 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2720 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2723 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2725 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2726 int level
= get_imsm_raid_level(lo
);
2728 if (level
== 1 || level
== 10) {
2729 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2731 return hi
->num_domains
;
2733 return num_stripes_per_unit_resync(dev
);
2736 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2738 /* named 'imsm_' because raid0, raid1 and raid10
2739 * counter-intuitively have the same number of data disks
2741 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2743 switch (get_imsm_raid_level(map
)) {
2745 return map
->num_members
;
2749 return map
->num_members
/2;
2751 return map
->num_members
- 1;
2753 dprintf("unsupported raid level\n");
2758 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2760 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2761 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2763 switch(get_imsm_raid_level(map
)) {
2766 return chunk
* map
->num_domains
;
2768 return chunk
* map
->num_members
;
2774 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2776 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2777 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2778 __u32 strip
= block
/ chunk
;
2780 switch (get_imsm_raid_level(map
)) {
2783 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2784 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2786 return vol_stripe
* chunk
+ block
% chunk
;
2788 __u32 stripe
= strip
/ (map
->num_members
- 1);
2790 return stripe
* chunk
+ block
% chunk
;
2797 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2798 struct imsm_dev
*dev
)
2800 /* calculate the conversion factor between per member 'blocks'
2801 * (md/{resync,rebuild}_start) and imsm migration units, return
2802 * 0 for the 'not migrating' and 'unsupported migration' cases
2804 if (!dev
->vol
.migr_state
)
2807 switch (migr_type(dev
)) {
2808 case MIGR_GEN_MIGR
: {
2809 struct migr_record
*migr_rec
= super
->migr_rec
;
2810 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2815 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2816 __u32 stripes_per_unit
;
2817 __u32 blocks_per_unit
;
2826 /* yes, this is really the translation of migr_units to
2827 * per-member blocks in the 'resync' case
2829 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2830 migr_chunk
= migr_strip_blocks_resync(dev
);
2831 disks
= imsm_num_data_members(dev
, MAP_0
);
2832 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2833 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2834 segment
= blocks_per_unit
/ stripe
;
2835 block_rel
= blocks_per_unit
- segment
* stripe
;
2836 parity_depth
= parity_segment_depth(dev
);
2837 block_map
= map_migr_block(dev
, block_rel
);
2838 return block_map
+ parity_depth
* segment
;
2840 case MIGR_REBUILD
: {
2841 __u32 stripes_per_unit
;
2844 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2845 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2846 return migr_chunk
* stripes_per_unit
;
2848 case MIGR_STATE_CHANGE
:
2854 static int imsm_level_to_layout(int level
)
2862 return ALGORITHM_LEFT_ASYMMETRIC
;
2869 /*******************************************************************************
2870 * Function: read_imsm_migr_rec
2871 * Description: Function reads imsm migration record from last sector of disk
2873 * fd : disk descriptor
2874 * super : metadata info
2878 ******************************************************************************/
2879 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2882 unsigned int sector_size
= super
->sector_size
;
2883 unsigned long long dsize
;
2885 get_dev_size(fd
, NULL
, &dsize
);
2886 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
2888 pr_err("Cannot seek to anchor block: %s\n",
2892 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
2893 MIGR_REC_BUF_SECTORS
*sector_size
) !=
2894 MIGR_REC_BUF_SECTORS
*sector_size
) {
2895 pr_err("Cannot read migr record block: %s\n",
2900 if (sector_size
== 4096)
2901 convert_from_4k_imsm_migr_rec(super
);
2907 static struct imsm_dev
*imsm_get_device_during_migration(
2908 struct intel_super
*super
)
2911 struct intel_dev
*dv
;
2913 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2914 if (is_gen_migration(dv
->dev
))
2920 /*******************************************************************************
2921 * Function: load_imsm_migr_rec
2922 * Description: Function reads imsm migration record (it is stored at the last
2925 * super : imsm internal array info
2926 * info : general array info
2930 * -2 : no migration in progress
2931 ******************************************************************************/
2932 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2939 struct imsm_dev
*dev
;
2940 struct imsm_map
*map
;
2943 /* find map under migration */
2944 dev
= imsm_get_device_during_migration(super
);
2945 /* nothing to load,no migration in progress?
2951 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2952 /* read only from one of the first two slots */
2953 if ((sd
->disk
.raid_disk
< 0) ||
2954 (sd
->disk
.raid_disk
> 1))
2957 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2958 fd
= dev_open(nm
, O_RDONLY
);
2964 map
= get_imsm_map(dev
, MAP_0
);
2965 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2966 /* skip spare and failed disks
2970 /* read only from one of the first two slots */
2972 slot
= get_imsm_disk_slot(map
, dl
->index
);
2973 if (map
== NULL
|| slot
> 1 || slot
< 0)
2975 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2976 fd
= dev_open(nm
, O_RDONLY
);
2983 retval
= read_imsm_migr_rec(fd
, super
);
2991 /*******************************************************************************
2992 * function: imsm_create_metadata_checkpoint_update
2993 * Description: It creates update for checkpoint change.
2995 * super : imsm internal array info
2996 * u : pointer to prepared update
2999 * If length is equal to 0, input pointer u contains no update
3000 ******************************************************************************/
3001 static int imsm_create_metadata_checkpoint_update(
3002 struct intel_super
*super
,
3003 struct imsm_update_general_migration_checkpoint
**u
)
3006 int update_memory_size
= 0;
3008 dprintf("(enter)\n");
3014 /* size of all update data without anchor */
3015 update_memory_size
=
3016 sizeof(struct imsm_update_general_migration_checkpoint
);
3018 *u
= xcalloc(1, update_memory_size
);
3020 dprintf("error: cannot get memory\n");
3023 (*u
)->type
= update_general_migration_checkpoint
;
3024 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
3025 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3027 return update_memory_size
;
3030 static void imsm_update_metadata_locally(struct supertype
*st
,
3031 void *buf
, int len
);
3033 /*******************************************************************************
3034 * Function: write_imsm_migr_rec
3035 * Description: Function writes imsm migration record
3036 * (at the last sector of disk)
3038 * super : imsm internal array info
3042 ******************************************************************************/
3043 static int write_imsm_migr_rec(struct supertype
*st
)
3045 struct intel_super
*super
= st
->sb
;
3046 unsigned int sector_size
= super
->sector_size
;
3047 unsigned long long dsize
;
3053 struct imsm_update_general_migration_checkpoint
*u
;
3054 struct imsm_dev
*dev
;
3055 struct imsm_map
*map
;
3057 /* find map under migration */
3058 dev
= imsm_get_device_during_migration(super
);
3059 /* if no migration, write buffer anyway to clear migr_record
3060 * on disk based on first available device
3063 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3064 super
->current_vol
);
3066 map
= get_imsm_map(dev
, MAP_0
);
3068 if (sector_size
== 4096)
3069 convert_to_4k_imsm_migr_rec(super
);
3070 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3073 /* skip failed and spare devices */
3076 /* write to 2 first slots only */
3078 slot
= get_imsm_disk_slot(map
, sd
->index
);
3079 if (map
== NULL
|| slot
> 1 || slot
< 0)
3082 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3083 fd
= dev_open(nm
, O_RDWR
);
3086 get_dev_size(fd
, NULL
, &dsize
);
3087 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3089 pr_err("Cannot seek to anchor block: %s\n",
3093 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3094 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3095 MIGR_REC_BUF_SECTORS
*sector_size
) {
3096 pr_err("Cannot write migr record block: %s\n",
3103 if (sector_size
== 4096)
3104 convert_from_4k_imsm_migr_rec(super
);
3105 /* update checkpoint information in metadata */
3106 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3108 dprintf("imsm: Cannot prepare update\n");
3111 /* update metadata locally */
3112 imsm_update_metadata_locally(st
, u
, len
);
3113 /* and possibly remotely */
3114 if (st
->update_tail
) {
3115 append_metadata_update(st
, u
, len
);
3116 /* during reshape we do all work inside metadata handler
3117 * manage_reshape(), so metadata update has to be triggered
3120 flush_metadata_updates(st
);
3121 st
->update_tail
= &st
->updates
;
3132 /* spare/missing disks activations are not allowe when
3133 * array/container performs reshape operation, because
3134 * all arrays in container works on the same disks set
3136 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3139 struct intel_dev
*i_dev
;
3140 struct imsm_dev
*dev
;
3142 /* check whole container
3144 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3146 if (is_gen_migration(dev
)) {
3147 /* No repair during any migration in container
3155 static unsigned long long imsm_component_size_aligment_check(int level
,
3157 unsigned int sector_size
,
3158 unsigned long long component_size
)
3160 unsigned int component_size_alligment
;
3162 /* check component size aligment
3164 component_size_alligment
= component_size
% (chunk_size
/sector_size
);
3166 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
3167 level
, chunk_size
, component_size
,
3168 component_size_alligment
);
3170 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
3171 dprintf("imsm: reported component size alligned from %llu ",
3173 component_size
-= component_size_alligment
;
3174 dprintf_cont("to %llu (%i).\n",
3175 component_size
, component_size_alligment
);
3178 return component_size
;
3181 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3183 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3184 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3186 return pba_of_lba0(map
) +
3187 (num_data_stripes(map
) * map
->blocks_per_strip
);
3190 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3192 struct intel_super
*super
= st
->sb
;
3193 struct migr_record
*migr_rec
= super
->migr_rec
;
3194 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3195 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3196 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3197 struct imsm_map
*map_to_analyse
= map
;
3199 int map_disks
= info
->array
.raid_disks
;
3201 memset(info
, 0, sizeof(*info
));
3203 map_to_analyse
= prev_map
;
3205 dl
= super
->current_disk
;
3207 info
->container_member
= super
->current_vol
;
3208 info
->array
.raid_disks
= map
->num_members
;
3209 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3210 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3211 info
->array
.md_minor
= -1;
3212 info
->array
.ctime
= 0;
3213 info
->array
.utime
= 0;
3214 info
->array
.chunk_size
=
3215 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3216 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3217 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
3218 info
->custom_array_size
<<= 32;
3219 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
3220 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3222 if (is_gen_migration(dev
)) {
3223 info
->reshape_active
= 1;
3224 info
->new_level
= get_imsm_raid_level(map
);
3225 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3226 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3227 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3228 if (info
->delta_disks
) {
3229 /* this needs to be applied to every array
3232 info
->reshape_active
= CONTAINER_RESHAPE
;
3234 /* We shape information that we give to md might have to be
3235 * modify to cope with md's requirement for reshaping arrays.
3236 * For example, when reshaping a RAID0, md requires it to be
3237 * presented as a degraded RAID4.
3238 * Also if a RAID0 is migrating to a RAID5 we need to specify
3239 * the array as already being RAID5, but the 'before' layout
3240 * is a RAID4-like layout.
3242 switch (info
->array
.level
) {
3244 switch(info
->new_level
) {
3246 /* conversion is happening as RAID4 */
3247 info
->array
.level
= 4;
3248 info
->array
.raid_disks
+= 1;
3251 /* conversion is happening as RAID5 */
3252 info
->array
.level
= 5;
3253 info
->array
.layout
= ALGORITHM_PARITY_N
;
3254 info
->delta_disks
-= 1;
3257 /* FIXME error message */
3258 info
->array
.level
= UnSet
;
3264 info
->new_level
= UnSet
;
3265 info
->new_layout
= UnSet
;
3266 info
->new_chunk
= info
->array
.chunk_size
;
3267 info
->delta_disks
= 0;
3271 info
->disk
.major
= dl
->major
;
3272 info
->disk
.minor
= dl
->minor
;
3273 info
->disk
.number
= dl
->index
;
3274 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3278 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3280 if (info
->array
.level
== 5) {
3281 info
->component_size
= num_data_stripes(map_to_analyse
) *
3282 map_to_analyse
->blocks_per_strip
;
3284 info
->component_size
= blocks_per_member(map_to_analyse
);
3287 info
->component_size
= imsm_component_size_aligment_check(
3289 info
->array
.chunk_size
,
3291 info
->component_size
);
3292 info
->bb
.supported
= 1;
3294 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3295 info
->recovery_start
= MaxSector
;
3297 if (info
->array
.level
== 5 && dev
->rwh_policy
== RWH_DISTRIBUTED
) {
3298 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3299 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3300 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
) >> 9;
3301 } else if (info
->array
.level
<= 0) {
3302 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3304 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3307 info
->reshape_progress
= 0;
3308 info
->resync_start
= MaxSector
;
3309 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3310 !(info
->array
.state
& 1)) &&
3311 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3312 info
->resync_start
= 0;
3314 if (dev
->vol
.migr_state
) {
3315 switch (migr_type(dev
)) {
3318 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3320 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3322 info
->resync_start
= blocks_per_unit
* units
;
3325 case MIGR_GEN_MIGR
: {
3326 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3328 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
3329 unsigned long long array_blocks
;
3332 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3334 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
3335 (super
->migr_rec
->rec_status
==
3336 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3339 info
->reshape_progress
= blocks_per_unit
* units
;
3341 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3342 (unsigned long long)units
,
3343 (unsigned long long)blocks_per_unit
,
3344 info
->reshape_progress
);
3346 used_disks
= imsm_num_data_members(dev
, MAP_1
);
3347 if (used_disks
> 0) {
3348 array_blocks
= blocks_per_member(map
) *
3350 info
->custom_array_size
=
3351 round_size_to_mb(array_blocks
,
3357 /* we could emulate the checkpointing of
3358 * 'sync_action=check' migrations, but for now
3359 * we just immediately complete them
3362 /* this is handled by container_content_imsm() */
3363 case MIGR_STATE_CHANGE
:
3364 /* FIXME handle other migrations */
3366 /* we are not dirty, so... */
3367 info
->resync_start
= MaxSector
;
3371 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3372 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3374 info
->array
.major_version
= -1;
3375 info
->array
.minor_version
= -2;
3376 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3377 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3378 uuid_from_super_imsm(st
, info
->uuid
);
3382 for (i
=0; i
<map_disks
; i
++) {
3384 if (i
< info
->array
.raid_disks
) {
3385 struct imsm_disk
*dsk
;
3386 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3387 dsk
= get_imsm_disk(super
, j
);
3388 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3395 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3396 int failed
, int look_in_map
);
3398 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3401 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3403 if (is_gen_migration(dev
)) {
3406 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3408 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3409 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3410 if (map2
->map_state
!= map_state
) {
3411 map2
->map_state
= map_state
;
3412 super
->updates_pending
++;
3417 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3421 for (d
= super
->missing
; d
; d
= d
->next
)
3422 if (d
->index
== index
)
3427 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3429 struct intel_super
*super
= st
->sb
;
3430 struct imsm_disk
*disk
;
3431 int map_disks
= info
->array
.raid_disks
;
3432 int max_enough
= -1;
3434 struct imsm_super
*mpb
;
3436 if (super
->current_vol
>= 0) {
3437 getinfo_super_imsm_volume(st
, info
, map
);
3440 memset(info
, 0, sizeof(*info
));
3442 /* Set raid_disks to zero so that Assemble will always pull in valid
3445 info
->array
.raid_disks
= 0;
3446 info
->array
.level
= LEVEL_CONTAINER
;
3447 info
->array
.layout
= 0;
3448 info
->array
.md_minor
= -1;
3449 info
->array
.ctime
= 0; /* N/A for imsm */
3450 info
->array
.utime
= 0;
3451 info
->array
.chunk_size
= 0;
3453 info
->disk
.major
= 0;
3454 info
->disk
.minor
= 0;
3455 info
->disk
.raid_disk
= -1;
3456 info
->reshape_active
= 0;
3457 info
->array
.major_version
= -1;
3458 info
->array
.minor_version
= -2;
3459 strcpy(info
->text_version
, "imsm");
3460 info
->safe_mode_delay
= 0;
3461 info
->disk
.number
= -1;
3462 info
->disk
.state
= 0;
3464 info
->recovery_start
= MaxSector
;
3465 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3466 info
->bb
.supported
= 1;
3468 /* do we have the all the insync disks that we expect? */
3469 mpb
= super
->anchor
;
3470 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3472 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3473 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3474 int failed
, enough
, j
, missing
= 0;
3475 struct imsm_map
*map
;
3478 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3479 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3480 map
= get_imsm_map(dev
, MAP_0
);
3482 /* any newly missing disks?
3483 * (catches single-degraded vs double-degraded)
3485 for (j
= 0; j
< map
->num_members
; j
++) {
3486 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3487 __u32 idx
= ord_to_idx(ord
);
3489 if (!(ord
& IMSM_ORD_REBUILD
) &&
3490 get_imsm_missing(super
, idx
)) {
3496 if (state
== IMSM_T_STATE_FAILED
)
3498 else if (state
== IMSM_T_STATE_DEGRADED
&&
3499 (state
!= map
->map_state
|| missing
))
3501 else /* we're normal, or already degraded */
3503 if (is_gen_migration(dev
) && missing
) {
3504 /* during general migration we need all disks
3505 * that process is running on.
3506 * No new missing disk is allowed.
3510 /* no more checks necessary
3514 /* in the missing/failed disk case check to see
3515 * if at least one array is runnable
3517 max_enough
= max(max_enough
, enough
);
3519 dprintf("enough: %d\n", max_enough
);
3520 info
->container_enough
= max_enough
;
3523 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3525 disk
= &super
->disks
->disk
;
3526 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3527 info
->component_size
= reserved
;
3528 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3529 /* we don't change info->disk.raid_disk here because
3530 * this state will be finalized in mdmon after we have
3531 * found the 'most fresh' version of the metadata
3533 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3534 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3535 0 : (1 << MD_DISK_SYNC
);
3538 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3539 * ->compare_super may have updated the 'num_raid_devs' field for spares
3541 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3542 uuid_from_super_imsm(st
, info
->uuid
);
3544 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3546 /* I don't know how to compute 'map' on imsm, so use safe default */
3549 for (i
= 0; i
< map_disks
; i
++)
3555 /* allocates memory and fills disk in mdinfo structure
3556 * for each disk in array */
3557 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3559 struct mdinfo
*mddev
;
3560 struct intel_super
*super
= st
->sb
;
3561 struct imsm_disk
*disk
;
3564 if (!super
|| !super
->disks
)
3567 mddev
= xcalloc(1, sizeof(*mddev
));
3571 tmp
= xcalloc(1, sizeof(*tmp
));
3573 tmp
->next
= mddev
->devs
;
3575 tmp
->disk
.number
= count
++;
3576 tmp
->disk
.major
= dl
->major
;
3577 tmp
->disk
.minor
= dl
->minor
;
3578 tmp
->disk
.state
= is_configured(disk
) ?
3579 (1 << MD_DISK_ACTIVE
) : 0;
3580 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3581 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3582 tmp
->disk
.raid_disk
= -1;
3588 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3589 char *update
, char *devname
, int verbose
,
3590 int uuid_set
, char *homehost
)
3592 /* For 'assemble' and 'force' we need to return non-zero if any
3593 * change was made. For others, the return value is ignored.
3594 * Update options are:
3595 * force-one : This device looks a bit old but needs to be included,
3596 * update age info appropriately.
3597 * assemble: clear any 'faulty' flag to allow this device to
3599 * force-array: Array is degraded but being forced, mark it clean
3600 * if that will be needed to assemble it.
3602 * newdev: not used ????
3603 * grow: Array has gained a new device - this is currently for
3605 * resync: mark as dirty so a resync will happen.
3606 * name: update the name - preserving the homehost
3607 * uuid: Change the uuid of the array to match watch is given
3609 * Following are not relevant for this imsm:
3610 * sparc2.2 : update from old dodgey metadata
3611 * super-minor: change the preferred_minor number
3612 * summaries: update redundant counters.
3613 * homehost: update the recorded homehost
3614 * _reshape_progress: record new reshape_progress position.
3617 struct intel_super
*super
= st
->sb
;
3618 struct imsm_super
*mpb
;
3620 /* we can only update container info */
3621 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3624 mpb
= super
->anchor
;
3626 if (strcmp(update
, "uuid") == 0) {
3627 /* We take this to mean that the family_num should be updated.
3628 * However that is much smaller than the uuid so we cannot really
3629 * allow an explicit uuid to be given. And it is hard to reliably
3631 * So if !uuid_set we know the current uuid is random and just used
3632 * the first 'int' and copy it to the other 3 positions.
3633 * Otherwise we require the 4 'int's to be the same as would be the
3634 * case if we are using a random uuid. So an explicit uuid will be
3635 * accepted as long as all for ints are the same... which shouldn't hurt
3638 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3641 if (info
->uuid
[0] != info
->uuid
[1] ||
3642 info
->uuid
[1] != info
->uuid
[2] ||
3643 info
->uuid
[2] != info
->uuid
[3])
3649 mpb
->orig_family_num
= info
->uuid
[0];
3650 } else if (strcmp(update
, "assemble") == 0)
3655 /* successful update? recompute checksum */
3657 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3662 static size_t disks_to_mpb_size(int disks
)
3666 size
= sizeof(struct imsm_super
);
3667 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3668 size
+= 2 * sizeof(struct imsm_dev
);
3669 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3670 size
+= (4 - 2) * sizeof(struct imsm_map
);
3671 /* 4 possible disk_ord_tbl's */
3672 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3673 /* maximum bbm log */
3674 size
+= sizeof(struct bbm_log
);
3679 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3680 unsigned long long data_offset
)
3682 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3685 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3688 static void free_devlist(struct intel_super
*super
)
3690 struct intel_dev
*dv
;
3692 while (super
->devlist
) {
3693 dv
= super
->devlist
->next
;
3694 free(super
->devlist
->dev
);
3695 free(super
->devlist
);
3696 super
->devlist
= dv
;
3700 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3702 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3705 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3709 * 0 same, or first was empty, and second was copied
3710 * 1 second had wrong number
3712 * 3 wrong other info
3714 struct intel_super
*first
= st
->sb
;
3715 struct intel_super
*sec
= tst
->sb
;
3722 /* in platform dependent environment test if the disks
3723 * use the same Intel hba
3724 * If not on Intel hba at all, allow anything.
3726 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3727 if (first
->hba
->type
!= sec
->hba
->type
) {
3729 "HBAs of devices do not match %s != %s\n",
3730 get_sys_dev_type(first
->hba
->type
),
3731 get_sys_dev_type(sec
->hba
->type
));
3734 if (first
->orom
!= sec
->orom
) {
3736 "HBAs of devices do not match %s != %s\n",
3737 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3742 /* if an anchor does not have num_raid_devs set then it is a free
3745 if (first
->anchor
->num_raid_devs
> 0 &&
3746 sec
->anchor
->num_raid_devs
> 0) {
3747 /* Determine if these disks might ever have been
3748 * related. Further disambiguation can only take place
3749 * in load_super_imsm_all
3751 __u32 first_family
= first
->anchor
->orig_family_num
;
3752 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3754 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3755 MAX_SIGNATURE_LENGTH
) != 0)
3758 if (first_family
== 0)
3759 first_family
= first
->anchor
->family_num
;
3760 if (sec_family
== 0)
3761 sec_family
= sec
->anchor
->family_num
;
3763 if (first_family
!= sec_family
)
3768 /* if 'first' is a spare promote it to a populated mpb with sec's
3771 if (first
->anchor
->num_raid_devs
== 0 &&
3772 sec
->anchor
->num_raid_devs
> 0) {
3774 struct intel_dev
*dv
;
3775 struct imsm_dev
*dev
;
3777 /* we need to copy raid device info from sec if an allocation
3778 * fails here we don't associate the spare
3780 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3781 dv
= xmalloc(sizeof(*dv
));
3782 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3785 dv
->next
= first
->devlist
;
3786 first
->devlist
= dv
;
3788 if (i
< sec
->anchor
->num_raid_devs
) {
3789 /* allocation failure */
3790 free_devlist(first
);
3791 pr_err("imsm: failed to associate spare\n");
3794 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3795 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3796 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3797 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3798 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3799 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3805 static void fd2devname(int fd
, char *name
)
3809 char dname
[PATH_MAX
];
3814 if (fstat(fd
, &st
) != 0)
3816 sprintf(path
, "/sys/dev/block/%d:%d",
3817 major(st
.st_rdev
), minor(st
.st_rdev
));
3819 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3824 nm
= strrchr(dname
, '/');
3827 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3831 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3834 char *name
= fd2kname(fd
);
3839 if (strncmp(name
, "nvme", 4) != 0)
3842 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3844 return load_sys(path
, buf
, buf_len
);
3847 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3849 static int imsm_read_serial(int fd
, char *devname
,
3850 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3859 memset(buf
, 0, sizeof(buf
));
3861 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3864 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3866 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3867 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3868 fd2devname(fd
, (char *) serial
);
3874 pr_err("Failed to retrieve serial for %s\n",
3879 /* trim all whitespace and non-printable characters and convert
3882 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
3885 /* ':' is reserved for use in placeholder serial
3886 * numbers for missing disks
3897 /* truncate leading characters */
3898 if (len
> MAX_RAID_SERIAL_LEN
) {
3899 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3900 len
= MAX_RAID_SERIAL_LEN
;
3903 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3904 memcpy(serial
, dest
, len
);
3909 static int serialcmp(__u8
*s1
, __u8
*s2
)
3911 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3914 static void serialcpy(__u8
*dest
, __u8
*src
)
3916 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3919 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3923 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3924 if (serialcmp(dl
->serial
, serial
) == 0)
3930 static struct imsm_disk
*
3931 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3935 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3936 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3938 if (serialcmp(disk
->serial
, serial
) == 0) {
3949 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3951 struct imsm_disk
*disk
;
3956 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3958 rv
= imsm_read_serial(fd
, devname
, serial
);
3963 dl
= xcalloc(1, sizeof(*dl
));
3966 dl
->major
= major(stb
.st_rdev
);
3967 dl
->minor
= minor(stb
.st_rdev
);
3968 dl
->next
= super
->disks
;
3969 dl
->fd
= keep_fd
? fd
: -1;
3970 assert(super
->disks
== NULL
);
3972 serialcpy(dl
->serial
, serial
);
3975 fd2devname(fd
, name
);
3977 dl
->devname
= xstrdup(devname
);
3979 dl
->devname
= xstrdup(name
);
3981 /* look up this disk's index in the current anchor */
3982 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3985 /* only set index on disks that are a member of a
3986 * populated contianer, i.e. one with raid_devs
3988 if (is_failed(&dl
->disk
))
3990 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
3997 /* When migrating map0 contains the 'destination' state while map1
3998 * contains the current state. When not migrating map0 contains the
3999 * current state. This routine assumes that map[0].map_state is set to
4000 * the current array state before being called.
4002 * Migration is indicated by one of the following states
4003 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4004 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4005 * map1state=unitialized)
4006 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4008 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4009 * map1state=degraded)
4010 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4013 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4014 __u8 to_state
, int migr_type
)
4016 struct imsm_map
*dest
;
4017 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4019 dev
->vol
.migr_state
= 1;
4020 set_migr_type(dev
, migr_type
);
4021 dev
->vol
.curr_migr_unit
= 0;
4022 dest
= get_imsm_map(dev
, MAP_1
);
4024 /* duplicate and then set the target end state in map[0] */
4025 memcpy(dest
, src
, sizeof_imsm_map(src
));
4026 if (migr_type
== MIGR_GEN_MIGR
) {
4030 for (i
= 0; i
< src
->num_members
; i
++) {
4031 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4032 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4036 if (migr_type
== MIGR_GEN_MIGR
)
4037 /* Clear migration record */
4038 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4040 src
->map_state
= to_state
;
4043 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4046 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4047 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4051 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4052 * completed in the last migration.
4054 * FIXME add support for raid-level-migration
4056 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4057 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4058 /* when final map state is other than expected
4059 * merge maps (not for migration)
4063 for (i
= 0; i
< prev
->num_members
; i
++)
4064 for (j
= 0; j
< map
->num_members
; j
++)
4065 /* during online capacity expansion
4066 * disks position can be changed
4067 * if takeover is used
4069 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4070 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4071 map
->disk_ord_tbl
[j
] |=
4072 prev
->disk_ord_tbl
[i
];
4075 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4076 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4079 dev
->vol
.migr_state
= 0;
4080 set_migr_type(dev
, 0);
4081 dev
->vol
.curr_migr_unit
= 0;
4082 map
->map_state
= map_state
;
4085 static int parse_raid_devices(struct intel_super
*super
)
4088 struct imsm_dev
*dev_new
;
4089 size_t len
, len_migr
;
4091 size_t space_needed
= 0;
4092 struct imsm_super
*mpb
= super
->anchor
;
4094 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4095 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4096 struct intel_dev
*dv
;
4098 len
= sizeof_imsm_dev(dev_iter
, 0);
4099 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4101 space_needed
+= len_migr
- len
;
4103 dv
= xmalloc(sizeof(*dv
));
4104 if (max_len
< len_migr
)
4106 if (max_len
> len_migr
)
4107 space_needed
+= max_len
- len_migr
;
4108 dev_new
= xmalloc(max_len
);
4109 imsm_copy_dev(dev_new
, dev_iter
);
4112 dv
->next
= super
->devlist
;
4113 super
->devlist
= dv
;
4116 /* ensure that super->buf is large enough when all raid devices
4119 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4122 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4123 super
->sector_size
);
4124 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4127 memcpy(buf
, super
->buf
, super
->len
);
4128 memset(buf
+ super
->len
, 0, len
- super
->len
);
4134 super
->extra_space
+= space_needed
;
4139 /*******************************************************************************
4140 * Function: check_mpb_migr_compatibility
4141 * Description: Function checks for unsupported migration features:
4142 * - migration optimization area (pba_of_lba0)
4143 * - descending reshape (ascending_migr)
4145 * super : imsm metadata information
4147 * 0 : migration is compatible
4148 * -1 : migration is not compatible
4149 ******************************************************************************/
4150 int check_mpb_migr_compatibility(struct intel_super
*super
)
4152 struct imsm_map
*map0
, *map1
;
4153 struct migr_record
*migr_rec
= super
->migr_rec
;
4156 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4157 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4160 dev_iter
->vol
.migr_state
== 1 &&
4161 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4162 /* This device is migrating */
4163 map0
= get_imsm_map(dev_iter
, MAP_0
);
4164 map1
= get_imsm_map(dev_iter
, MAP_1
);
4165 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4166 /* migration optimization area was used */
4168 if (migr_rec
->ascending_migr
== 0 &&
4169 migr_rec
->dest_depth_per_unit
> 0)
4170 /* descending reshape not supported yet */
4177 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4179 /* load_imsm_mpb - read matrix metadata
4180 * allocates super->mpb to be freed by free_imsm
4182 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4184 unsigned long long dsize
;
4185 unsigned long long sectors
;
4186 unsigned int sector_size
= super
->sector_size
;
4188 struct imsm_super
*anchor
;
4191 get_dev_size(fd
, NULL
, &dsize
);
4192 if (dsize
< 2*sector_size
) {
4194 pr_err("%s: device to small for imsm\n",
4199 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4201 pr_err("Cannot seek to anchor block on %s: %s\n",
4202 devname
, strerror(errno
));
4206 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4208 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4211 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4213 pr_err("Cannot read anchor block on %s: %s\n",
4214 devname
, strerror(errno
));
4219 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4221 pr_err("no IMSM anchor on %s\n", devname
);
4226 __free_imsm(super
, 0);
4227 /* reload capability and hba */
4229 /* capability and hba must be updated with new super allocation */
4230 find_intel_hba_capability(fd
, super
, devname
);
4231 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4232 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4234 pr_err("unable to allocate %zu byte mpb buffer\n",
4239 memcpy(super
->buf
, anchor
, sector_size
);
4241 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4244 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4245 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4246 pr_err("could not allocate migr_rec buffer\n");
4250 super
->clean_migration_record_by_mdmon
= 0;
4253 check_sum
= __gen_imsm_checksum(super
->anchor
);
4254 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4256 pr_err("IMSM checksum %x != %x on %s\n",
4258 __le32_to_cpu(super
->anchor
->check_sum
),
4266 /* read the extended mpb */
4267 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4269 pr_err("Cannot seek to extended mpb on %s: %s\n",
4270 devname
, strerror(errno
));
4274 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4275 super
->len
- sector_size
) != super
->len
- sector_size
) {
4277 pr_err("Cannot read extended mpb on %s: %s\n",
4278 devname
, strerror(errno
));
4282 check_sum
= __gen_imsm_checksum(super
->anchor
);
4283 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4285 pr_err("IMSM checksum %x != %x on %s\n",
4286 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4294 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4296 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4297 static void clear_hi(struct intel_super
*super
)
4299 struct imsm_super
*mpb
= super
->anchor
;
4301 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4303 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4304 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4305 disk
->total_blocks_hi
= 0;
4307 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4308 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4311 for (n
= 0; n
< 2; ++n
) {
4312 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4315 map
->pba_of_lba0_hi
= 0;
4316 map
->blocks_per_member_hi
= 0;
4317 map
->num_data_stripes_hi
= 0;
4323 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4327 err
= load_imsm_mpb(fd
, super
, devname
);
4330 if (super
->sector_size
== 4096)
4331 convert_from_4k(super
);
4332 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4335 err
= parse_raid_devices(super
);
4338 err
= load_bbm_log(super
);
4343 static void __free_imsm_disk(struct dl
*d
)
4355 static void free_imsm_disks(struct intel_super
*super
)
4359 while (super
->disks
) {
4361 super
->disks
= d
->next
;
4362 __free_imsm_disk(d
);
4364 while (super
->disk_mgmt_list
) {
4365 d
= super
->disk_mgmt_list
;
4366 super
->disk_mgmt_list
= d
->next
;
4367 __free_imsm_disk(d
);
4369 while (super
->missing
) {
4371 super
->missing
= d
->next
;
4372 __free_imsm_disk(d
);
4377 /* free all the pieces hanging off of a super pointer */
4378 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4380 struct intel_hba
*elem
, *next
;
4386 /* unlink capability description */
4388 if (super
->migr_rec_buf
) {
4389 free(super
->migr_rec_buf
);
4390 super
->migr_rec_buf
= NULL
;
4393 free_imsm_disks(super
);
4394 free_devlist(super
);
4398 free((void *)elem
->path
);
4404 free(super
->bbm_log
);
4408 static void free_imsm(struct intel_super
*super
)
4410 __free_imsm(super
, 1);
4411 free(super
->bb
.entries
);
4415 static void free_super_imsm(struct supertype
*st
)
4417 struct intel_super
*super
= st
->sb
;
4426 static struct intel_super
*alloc_super(void)
4428 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4430 super
->current_vol
= -1;
4431 super
->create_offset
= ~((unsigned long long) 0);
4433 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4434 sizeof(struct md_bb_entry
));
4435 if (!super
->bb
.entries
) {
4444 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4446 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4448 struct sys_dev
*hba_name
;
4451 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4456 hba_name
= find_disk_attached_hba(fd
, NULL
);
4459 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4463 rv
= attach_hba_to_super(super
, hba_name
);
4466 struct intel_hba
*hba
= super
->hba
;
4468 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4469 " but the container is assigned to Intel(R) %s %s (",
4471 get_sys_dev_type(hba_name
->type
),
4472 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4473 hba_name
->pci_id
? : "Err!",
4474 get_sys_dev_type(super
->hba
->type
),
4475 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4478 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4480 fprintf(stderr
, ", ");
4483 fprintf(stderr
, ").\n"
4484 " Mixing devices attached to different %s is not allowed.\n",
4485 hba_name
->type
== SYS_DEV_VMD
? "VMD domains" : "controllers");
4489 super
->orom
= find_imsm_capability(hba_name
);
4496 /* find_missing - helper routine for load_super_imsm_all that identifies
4497 * disks that have disappeared from the system. This routine relies on
4498 * the mpb being uptodate, which it is at load time.
4500 static int find_missing(struct intel_super
*super
)
4503 struct imsm_super
*mpb
= super
->anchor
;
4505 struct imsm_disk
*disk
;
4507 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4508 disk
= __get_imsm_disk(mpb
, i
);
4509 dl
= serial_to_dl(disk
->serial
, super
);
4513 dl
= xmalloc(sizeof(*dl
));
4517 dl
->devname
= xstrdup("missing");
4519 serialcpy(dl
->serial
, disk
->serial
);
4522 dl
->next
= super
->missing
;
4523 super
->missing
= dl
;
4529 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4531 struct intel_disk
*idisk
= disk_list
;
4534 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4536 idisk
= idisk
->next
;
4542 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4543 struct intel_super
*super
,
4544 struct intel_disk
**disk_list
)
4546 struct imsm_disk
*d
= &super
->disks
->disk
;
4547 struct imsm_super
*mpb
= super
->anchor
;
4550 for (i
= 0; i
< tbl_size
; i
++) {
4551 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4552 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4554 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4555 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4556 dprintf("mpb from %d:%d matches %d:%d\n",
4557 super
->disks
->major
,
4558 super
->disks
->minor
,
4559 table
[i
]->disks
->major
,
4560 table
[i
]->disks
->minor
);
4564 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4565 is_configured(d
) == is_configured(tbl_d
)) &&
4566 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4567 /* current version of the mpb is a
4568 * better candidate than the one in
4569 * super_table, but copy over "cross
4570 * generational" status
4572 struct intel_disk
*idisk
;
4574 dprintf("mpb from %d:%d replaces %d:%d\n",
4575 super
->disks
->major
,
4576 super
->disks
->minor
,
4577 table
[i
]->disks
->major
,
4578 table
[i
]->disks
->minor
);
4580 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4581 if (idisk
&& is_failed(&idisk
->disk
))
4582 tbl_d
->status
|= FAILED_DISK
;
4585 struct intel_disk
*idisk
;
4586 struct imsm_disk
*disk
;
4588 /* tbl_mpb is more up to date, but copy
4589 * over cross generational status before
4592 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4593 if (disk
&& is_failed(disk
))
4594 d
->status
|= FAILED_DISK
;
4596 idisk
= disk_list_get(d
->serial
, *disk_list
);
4599 if (disk
&& is_configured(disk
))
4600 idisk
->disk
.status
|= CONFIGURED_DISK
;
4603 dprintf("mpb from %d:%d prefer %d:%d\n",
4604 super
->disks
->major
,
4605 super
->disks
->minor
,
4606 table
[i
]->disks
->major
,
4607 table
[i
]->disks
->minor
);
4615 table
[tbl_size
++] = super
;
4619 /* update/extend the merged list of imsm_disk records */
4620 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4621 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4622 struct intel_disk
*idisk
;
4624 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4626 idisk
->disk
.status
|= disk
->status
;
4627 if (is_configured(&idisk
->disk
) ||
4628 is_failed(&idisk
->disk
))
4629 idisk
->disk
.status
&= ~(SPARE_DISK
);
4631 idisk
= xcalloc(1, sizeof(*idisk
));
4632 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4633 idisk
->disk
= *disk
;
4634 idisk
->next
= *disk_list
;
4638 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4645 static struct intel_super
*
4646 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4649 struct imsm_super
*mpb
= super
->anchor
;
4653 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4654 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4655 struct intel_disk
*idisk
;
4657 idisk
= disk_list_get(disk
->serial
, disk_list
);
4659 if (idisk
->owner
== owner
||
4660 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4663 dprintf("'%.16s' owner %d != %d\n",
4664 disk
->serial
, idisk
->owner
,
4667 dprintf("unknown disk %x [%d]: %.16s\n",
4668 __le32_to_cpu(mpb
->family_num
), i
,
4674 if (ok_count
== mpb
->num_disks
)
4679 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4681 struct intel_super
*s
;
4683 for (s
= super_list
; s
; s
= s
->next
) {
4684 if (family_num
!= s
->anchor
->family_num
)
4686 pr_err("Conflict, offlining family %#x on '%s'\n",
4687 __le32_to_cpu(family_num
), s
->disks
->devname
);
4691 static struct intel_super
*
4692 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4694 struct intel_super
*super_table
[len
];
4695 struct intel_disk
*disk_list
= NULL
;
4696 struct intel_super
*champion
, *spare
;
4697 struct intel_super
*s
, **del
;
4702 memset(super_table
, 0, sizeof(super_table
));
4703 for (s
= *super_list
; s
; s
= s
->next
)
4704 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4706 for (i
= 0; i
< tbl_size
; i
++) {
4707 struct imsm_disk
*d
;
4708 struct intel_disk
*idisk
;
4709 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4712 d
= &s
->disks
->disk
;
4714 /* 'd' must appear in merged disk list for its
4715 * configuration to be valid
4717 idisk
= disk_list_get(d
->serial
, disk_list
);
4718 if (idisk
&& idisk
->owner
== i
)
4719 s
= validate_members(s
, disk_list
, i
);
4724 dprintf("marking family: %#x from %d:%d offline\n",
4726 super_table
[i
]->disks
->major
,
4727 super_table
[i
]->disks
->minor
);
4731 /* This is where the mdadm implementation differs from the Windows
4732 * driver which has no strict concept of a container. We can only
4733 * assemble one family from a container, so when returning a prodigal
4734 * array member to this system the code will not be able to disambiguate
4735 * the container contents that should be assembled ("foreign" versus
4736 * "local"). It requires user intervention to set the orig_family_num
4737 * to a new value to establish a new container. The Windows driver in
4738 * this situation fixes up the volume name in place and manages the
4739 * foreign array as an independent entity.
4744 for (i
= 0; i
< tbl_size
; i
++) {
4745 struct intel_super
*tbl_ent
= super_table
[i
];
4751 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4756 if (s
&& !is_spare
) {
4757 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4759 } else if (!s
&& !is_spare
)
4772 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4773 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4775 /* collect all dl's onto 'champion', and update them to
4776 * champion's version of the status
4778 for (s
= *super_list
; s
; s
= s
->next
) {
4779 struct imsm_super
*mpb
= champion
->anchor
;
4780 struct dl
*dl
= s
->disks
;
4785 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4787 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4788 struct imsm_disk
*disk
;
4790 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4793 /* only set index on disks that are a member of
4794 * a populated contianer, i.e. one with
4797 if (is_failed(&dl
->disk
))
4799 else if (is_spare(&dl
->disk
))
4805 if (i
>= mpb
->num_disks
) {
4806 struct intel_disk
*idisk
;
4808 idisk
= disk_list_get(dl
->serial
, disk_list
);
4809 if (idisk
&& is_spare(&idisk
->disk
) &&
4810 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4818 dl
->next
= champion
->disks
;
4819 champion
->disks
= dl
;
4823 /* delete 'champion' from super_list */
4824 for (del
= super_list
; *del
; ) {
4825 if (*del
== champion
) {
4826 *del
= (*del
)->next
;
4829 del
= &(*del
)->next
;
4831 champion
->next
= NULL
;
4835 struct intel_disk
*idisk
= disk_list
;
4837 disk_list
= disk_list
->next
;
4845 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4846 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4847 int major
, int minor
, int keep_fd
);
4849 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4850 int *max
, int keep_fd
);
4852 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4853 char *devname
, struct md_list
*devlist
,
4856 struct intel_super
*super_list
= NULL
;
4857 struct intel_super
*super
= NULL
;
4862 /* 'fd' is an opened container */
4863 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4865 /* get super block from devlist devices */
4866 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4869 /* all mpbs enter, maybe one leaves */
4870 super
= imsm_thunderdome(&super_list
, i
);
4876 if (find_missing(super
) != 0) {
4882 /* load migration record */
4883 err
= load_imsm_migr_rec(super
, NULL
);
4885 /* migration is in progress,
4886 * but migr_rec cannot be loaded,
4892 /* Check migration compatibility */
4893 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
4894 pr_err("Unsupported migration detected");
4896 fprintf(stderr
, " on %s\n", devname
);
4898 fprintf(stderr
, " (IMSM).\n");
4907 while (super_list
) {
4908 struct intel_super
*s
= super_list
;
4910 super_list
= super_list
->next
;
4919 strcpy(st
->container_devnm
, fd2devnm(fd
));
4921 st
->container_devnm
[0] = 0;
4922 if (err
== 0 && st
->ss
== NULL
) {
4923 st
->ss
= &super_imsm
;
4924 st
->minor_version
= 0;
4925 st
->max_devs
= IMSM_MAX_DEVICES
;
4931 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4932 int *max
, int keep_fd
)
4934 struct md_list
*tmpdev
;
4938 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4939 if (tmpdev
->used
!= 1)
4941 if (tmpdev
->container
== 1) {
4943 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4945 pr_err("cannot open device %s: %s\n",
4946 tmpdev
->devname
, strerror(errno
));
4950 err
= get_sra_super_block(fd
, super_list
,
4951 tmpdev
->devname
, &lmax
,
4960 int major
= major(tmpdev
->st_rdev
);
4961 int minor
= minor(tmpdev
->st_rdev
);
4962 err
= get_super_block(super_list
,
4979 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4980 int major
, int minor
, int keep_fd
)
4982 struct intel_super
*s
;
4994 sprintf(nm
, "%d:%d", major
, minor
);
4995 dfd
= dev_open(nm
, O_RDWR
);
5001 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
5002 find_intel_hba_capability(dfd
, s
, devname
);
5003 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5005 /* retry the load if we might have raced against mdmon */
5006 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5007 for (retry
= 0; retry
< 3; retry
++) {
5009 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5015 s
->next
= *super_list
;
5023 if (dfd
>= 0 && !keep_fd
)
5030 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5037 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5041 if (sra
->array
.major_version
!= -1 ||
5042 sra
->array
.minor_version
!= -2 ||
5043 strcmp(sra
->text_version
, "imsm") != 0) {
5048 devnm
= fd2devnm(fd
);
5049 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5050 if (get_super_block(super_list
, devnm
, devname
,
5051 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5062 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5064 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5067 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5069 struct intel_super
*super
;
5073 if (test_partition(fd
))
5074 /* IMSM not allowed on partitions */
5077 free_super_imsm(st
);
5079 super
= alloc_super();
5080 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5083 /* Load hba and capabilities if they exist.
5084 * But do not preclude loading metadata in case capabilities or hba are
5085 * non-compliant and ignore_hw_compat is set.
5087 rv
= find_intel_hba_capability(fd
, super
, devname
);
5088 /* no orom/efi or non-intel hba of the disk */
5089 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5091 pr_err("No OROM/EFI properties for %s\n", devname
);
5095 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5097 /* retry the load if we might have raced against mdmon */
5099 struct mdstat_ent
*mdstat
= NULL
;
5100 char *name
= fd2kname(fd
);
5103 mdstat
= mdstat_by_component(name
);
5105 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5106 for (retry
= 0; retry
< 3; retry
++) {
5108 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5114 free_mdstat(mdstat
);
5119 pr_err("Failed to load all information sections on %s\n", devname
);
5125 if (st
->ss
== NULL
) {
5126 st
->ss
= &super_imsm
;
5127 st
->minor_version
= 0;
5128 st
->max_devs
= IMSM_MAX_DEVICES
;
5131 /* load migration record */
5132 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5133 /* Check for unsupported migration features */
5134 if (check_mpb_migr_compatibility(super
) != 0) {
5135 pr_err("Unsupported migration detected");
5137 fprintf(stderr
, " on %s\n", devname
);
5139 fprintf(stderr
, " (IMSM).\n");
5147 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5149 if (info
->level
== 1)
5151 return info
->chunk_size
>> 9;
5154 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5155 unsigned long long size
)
5157 if (info
->level
== 1)
5160 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5163 static void imsm_update_version_info(struct intel_super
*super
)
5165 /* update the version and attributes */
5166 struct imsm_super
*mpb
= super
->anchor
;
5168 struct imsm_dev
*dev
;
5169 struct imsm_map
*map
;
5172 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5173 dev
= get_imsm_dev(super
, i
);
5174 map
= get_imsm_map(dev
, MAP_0
);
5175 if (__le32_to_cpu(dev
->size_high
) > 0)
5176 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5178 /* FIXME detect when an array spans a port multiplier */
5180 mpb
->attributes
|= MPB_ATTRIB_PM
;
5183 if (mpb
->num_raid_devs
> 1 ||
5184 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5185 version
= MPB_VERSION_ATTRIBS
;
5186 switch (get_imsm_raid_level(map
)) {
5187 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5188 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5189 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5190 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5193 if (map
->num_members
>= 5)
5194 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5195 else if (dev
->status
== DEV_CLONE_N_GO
)
5196 version
= MPB_VERSION_CNG
;
5197 else if (get_imsm_raid_level(map
) == 5)
5198 version
= MPB_VERSION_RAID5
;
5199 else if (map
->num_members
>= 3)
5200 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5201 else if (get_imsm_raid_level(map
) == 1)
5202 version
= MPB_VERSION_RAID1
;
5204 version
= MPB_VERSION_RAID0
;
5206 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5210 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5212 struct imsm_super
*mpb
= super
->anchor
;
5213 char *reason
= NULL
;
5216 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
5217 reason
= "must be 16 characters or less";
5219 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5220 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5222 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5223 reason
= "already exists";
5228 if (reason
&& !quiet
)
5229 pr_err("imsm volume name %s\n", reason
);
5234 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5235 struct shape
*s
, char *name
,
5236 char *homehost
, int *uuid
,
5237 long long data_offset
)
5239 /* We are creating a volume inside a pre-existing container.
5240 * so st->sb is already set.
5242 struct intel_super
*super
= st
->sb
;
5243 unsigned int sector_size
= super
->sector_size
;
5244 struct imsm_super
*mpb
= super
->anchor
;
5245 struct intel_dev
*dv
;
5246 struct imsm_dev
*dev
;
5247 struct imsm_vol
*vol
;
5248 struct imsm_map
*map
;
5249 int idx
= mpb
->num_raid_devs
;
5251 unsigned long long array_blocks
;
5252 size_t size_old
, size_new
;
5253 unsigned long long num_data_stripes
;
5254 unsigned int data_disks
;
5255 unsigned long long size_per_member
;
5257 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5258 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5262 /* ensure the mpb is large enough for the new data */
5263 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5264 size_new
= disks_to_mpb_size(info
->nr_disks
);
5265 if (size_new
> size_old
) {
5267 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5269 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5270 pr_err("could not allocate new mpb\n");
5273 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5274 MIGR_REC_BUF_SECTORS
*
5275 MAX_SECTOR_SIZE
) != 0) {
5276 pr_err("could not allocate migr_rec buffer\n");
5282 memcpy(mpb_new
, mpb
, size_old
);
5285 super
->anchor
= mpb_new
;
5286 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5287 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5288 super
->len
= size_round
;
5290 super
->current_vol
= idx
;
5292 /* handle 'failed_disks' by either:
5293 * a) create dummy disk entries in the table if this the first
5294 * volume in the array. We add them here as this is the only
5295 * opportunity to add them. add_to_super_imsm_volume()
5296 * handles the non-failed disks and continues incrementing
5298 * b) validate that 'failed_disks' matches the current number
5299 * of missing disks if the container is populated
5301 if (super
->current_vol
== 0) {
5303 for (i
= 0; i
< info
->failed_disks
; i
++) {
5304 struct imsm_disk
*disk
;
5307 disk
= __get_imsm_disk(mpb
, i
);
5308 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5309 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5310 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5311 "missing:%d", (__u8
)i
);
5313 find_missing(super
);
5318 for (d
= super
->missing
; d
; d
= d
->next
)
5320 if (info
->failed_disks
> missing
) {
5321 pr_err("unable to add 'missing' disk to container\n");
5326 if (!check_name(super
, name
, 0))
5328 dv
= xmalloc(sizeof(*dv
));
5329 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5330 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
5331 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5332 info
->layout
, info
->chunk_size
,
5333 s
->size
* BLOCKS_PER_KB
);
5334 data_disks
= get_data_disks(info
->level
, info
->layout
,
5336 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5337 size_per_member
= array_blocks
/ data_disks
;
5339 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
5340 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
5341 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5343 vol
->migr_state
= 0;
5344 set_migr_type(dev
, MIGR_INIT
);
5345 vol
->dirty
= !info
->state
;
5346 vol
->curr_migr_unit
= 0;
5347 map
= get_imsm_map(dev
, MAP_0
);
5348 set_pba_of_lba0(map
, super
->create_offset
);
5349 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5352 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5353 map
->failed_disk_num
= ~0;
5354 if (info
->level
> 0)
5355 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5356 : IMSM_T_STATE_UNINITIALIZED
);
5358 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5359 IMSM_T_STATE_NORMAL
;
5362 if (info
->level
== 1 && info
->raid_disks
> 2) {
5365 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5369 map
->raid_level
= info
->level
;
5370 if (info
->level
== 10) {
5371 map
->raid_level
= 1;
5372 map
->num_domains
= info
->raid_disks
/ 2;
5373 } else if (info
->level
== 1)
5374 map
->num_domains
= info
->raid_disks
;
5376 map
->num_domains
= 1;
5378 /* info->size is only int so use the 'size' parameter instead */
5379 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5380 num_data_stripes
/= map
->num_domains
;
5381 set_num_data_stripes(map
, num_data_stripes
);
5383 map
->num_members
= info
->raid_disks
;
5384 for (i
= 0; i
< map
->num_members
; i
++) {
5385 /* initialized in add_to_super */
5386 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5388 mpb
->num_raid_devs
++;
5389 mpb
->num_raid_devs_created
++;
5390 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5392 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5393 dev
->rwh_policy
= RWH_OFF
;
5394 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5395 dev
->rwh_policy
= RWH_DISTRIBUTED
;
5399 pr_err("imsm does not support consistency policy %s\n",
5400 map_num(consistency_policies
, s
->consistency_policy
));
5405 dv
->index
= super
->current_vol
;
5406 dv
->next
= super
->devlist
;
5407 super
->devlist
= dv
;
5409 imsm_update_version_info(super
);
5414 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5415 struct shape
*s
, char *name
,
5416 char *homehost
, int *uuid
,
5417 unsigned long long data_offset
)
5419 /* This is primarily called by Create when creating a new array.
5420 * We will then get add_to_super called for each component, and then
5421 * write_init_super called to write it out to each device.
5422 * For IMSM, Create can create on fresh devices or on a pre-existing
5424 * To create on a pre-existing array a different method will be called.
5425 * This one is just for fresh drives.
5427 struct intel_super
*super
;
5428 struct imsm_super
*mpb
;
5432 if (data_offset
!= INVALID_SECTORS
) {
5433 pr_err("data-offset not supported by imsm\n");
5438 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5442 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5444 mpb_size
= MAX_SECTOR_SIZE
;
5446 super
= alloc_super();
5448 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5453 pr_err("could not allocate superblock\n");
5456 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5457 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5458 pr_err("could not allocate migr_rec buffer\n");
5463 memset(super
->buf
, 0, mpb_size
);
5465 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5469 /* zeroing superblock */
5473 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5475 version
= (char *) mpb
->sig
;
5476 strcpy(version
, MPB_SIGNATURE
);
5477 version
+= strlen(MPB_SIGNATURE
);
5478 strcpy(version
, MPB_VERSION_RAID0
);
5483 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5485 unsigned int member_sector_size
;
5488 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5492 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5494 if (member_sector_size
!= super
->sector_size
)
5499 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5500 int fd
, char *devname
)
5502 struct intel_super
*super
= st
->sb
;
5503 struct imsm_super
*mpb
= super
->anchor
;
5504 struct imsm_disk
*_disk
;
5505 struct imsm_dev
*dev
;
5506 struct imsm_map
*map
;
5510 dev
= get_imsm_dev(super
, super
->current_vol
);
5511 map
= get_imsm_map(dev
, MAP_0
);
5513 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5514 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5520 /* we're doing autolayout so grab the pre-marked (in
5521 * validate_geometry) raid_disk
5523 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5524 if (dl
->raiddisk
== dk
->raid_disk
)
5527 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5528 if (dl
->major
== dk
->major
&&
5529 dl
->minor
== dk
->minor
)
5534 pr_err("%s is not a member of the same container\n", devname
);
5538 if (!drive_validate_sector_size(super
, dl
)) {
5539 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5543 /* add a pristine spare to the metadata */
5544 if (dl
->index
< 0) {
5545 dl
->index
= super
->anchor
->num_disks
;
5546 super
->anchor
->num_disks
++;
5548 /* Check the device has not already been added */
5549 slot
= get_imsm_disk_slot(map
, dl
->index
);
5551 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5552 pr_err("%s has been included in this array twice\n",
5556 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5557 dl
->disk
.status
= CONFIGURED_DISK
;
5559 /* update size of 'missing' disks to be at least as large as the
5560 * largest acitve member (we only have dummy missing disks when
5561 * creating the first volume)
5563 if (super
->current_vol
== 0) {
5564 for (df
= super
->missing
; df
; df
= df
->next
) {
5565 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5566 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5567 _disk
= __get_imsm_disk(mpb
, df
->index
);
5572 /* refresh unset/failed slots to point to valid 'missing' entries */
5573 for (df
= super
->missing
; df
; df
= df
->next
)
5574 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5575 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5577 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5579 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5580 if (is_gen_migration(dev
)) {
5581 struct imsm_map
*map2
= get_imsm_map(dev
,
5583 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5584 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5585 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5588 if ((unsigned)df
->index
==
5590 set_imsm_ord_tbl_ent(map2
,
5596 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5600 /* if we are creating the first raid device update the family number */
5601 if (super
->current_vol
== 0) {
5603 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5605 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5606 if (!_dev
|| !_disk
) {
5607 pr_err("BUG mpb setup error\n");
5613 sum
+= __gen_imsm_checksum(mpb
);
5614 mpb
->family_num
= __cpu_to_le32(sum
);
5615 mpb
->orig_family_num
= mpb
->family_num
;
5617 super
->current_disk
= dl
;
5622 * Function marks disk as spare and restores disk serial
5623 * in case it was previously marked as failed by takeover operation
5625 * -1 : critical error
5626 * 0 : disk is marked as spare but serial is not set
5629 int mark_spare(struct dl
*disk
)
5631 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5638 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5639 /* Restore disk serial number, because takeover marks disk
5640 * as failed and adds to serial ':0' before it becomes
5643 serialcpy(disk
->serial
, serial
);
5644 serialcpy(disk
->disk
.serial
, serial
);
5647 disk
->disk
.status
= SPARE_DISK
;
5653 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5654 int fd
, char *devname
,
5655 unsigned long long data_offset
)
5657 struct intel_super
*super
= st
->sb
;
5659 unsigned long long size
;
5660 unsigned int member_sector_size
;
5665 /* If we are on an RAID enabled platform check that the disk is
5666 * attached to the raid controller.
5667 * We do not need to test disks attachment for container based additions,
5668 * they shall be already tested when container was created/assembled.
5670 rv
= find_intel_hba_capability(fd
, super
, devname
);
5671 /* no orom/efi or non-intel hba of the disk */
5673 dprintf("capability: %p fd: %d ret: %d\n",
5674 super
->orom
, fd
, rv
);
5678 if (super
->current_vol
>= 0)
5679 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5682 dd
= xcalloc(sizeof(*dd
), 1);
5683 dd
->major
= major(stb
.st_rdev
);
5684 dd
->minor
= minor(stb
.st_rdev
);
5685 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5688 dd
->action
= DISK_ADD
;
5689 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5691 pr_err("failed to retrieve scsi serial, aborting\n");
5697 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5698 (super
->hba
->type
== SYS_DEV_VMD
))) {
5700 char *devpath
= diskfd_to_devpath(fd
);
5701 char controller_path
[PATH_MAX
];
5704 pr_err("failed to get devpath, aborting\n");
5711 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5714 if (devpath_to_vendor(controller_path
) == 0x8086) {
5716 * If Intel's NVMe drive has serial ended with
5717 * "-A","-B","-1" or "-2" it means that this is "x8"
5718 * device (double drive on single PCIe card).
5719 * User should be warned about potential data loss.
5721 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5722 /* Skip empty character at the end */
5723 if (dd
->serial
[i
] == 0)
5726 if (((dd
->serial
[i
] == 'A') ||
5727 (dd
->serial
[i
] == 'B') ||
5728 (dd
->serial
[i
] == '1') ||
5729 (dd
->serial
[i
] == '2')) &&
5730 (dd
->serial
[i
-1] == '-'))
5731 pr_err("\tThe action you are about to take may put your data at risk.\n"
5732 "\tPlease note that x8 devices may consist of two separate x4 devices "
5733 "located on a single PCIe port.\n"
5734 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5737 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5738 !imsm_orom_has_tpv_support(super
->orom
)) {
5739 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5740 "\tPlease refer to Intel(R) RSTe user guide.\n");
5747 get_dev_size(fd
, NULL
, &size
);
5748 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5750 if (super
->sector_size
== 0) {
5751 /* this a first device, so sector_size is not set yet */
5752 super
->sector_size
= member_sector_size
;
5755 /* clear migr_rec when adding disk to container */
5756 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5757 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5759 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5760 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5761 MIGR_REC_BUF_SECTORS
*member_sector_size
)
5762 perror("Write migr_rec failed");
5766 serialcpy(dd
->disk
.serial
, dd
->serial
);
5767 set_total_blocks(&dd
->disk
, size
);
5768 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5769 struct imsm_super
*mpb
= super
->anchor
;
5770 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5773 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5774 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5776 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5778 if (st
->update_tail
) {
5779 dd
->next
= super
->disk_mgmt_list
;
5780 super
->disk_mgmt_list
= dd
;
5782 dd
->next
= super
->disks
;
5784 super
->updates_pending
++;
5790 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5792 struct intel_super
*super
= st
->sb
;
5795 /* remove from super works only in mdmon - for communication
5796 * manager - monitor. Check if communication memory buffer
5799 if (!st
->update_tail
) {
5800 pr_err("shall be used in mdmon context only\n");
5803 dd
= xcalloc(1, sizeof(*dd
));
5804 dd
->major
= dk
->major
;
5805 dd
->minor
= dk
->minor
;
5808 dd
->action
= DISK_REMOVE
;
5810 dd
->next
= super
->disk_mgmt_list
;
5811 super
->disk_mgmt_list
= dd
;
5816 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5819 char buf
[MAX_SECTOR_SIZE
];
5820 struct imsm_super anchor
;
5821 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5823 /* spare records have their own family number and do not have any defined raid
5826 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5828 struct imsm_super
*mpb
= super
->anchor
;
5829 struct imsm_super
*spare
= &spare_record
.anchor
;
5833 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5834 spare
->generation_num
= __cpu_to_le32(1UL);
5835 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5836 spare
->num_disks
= 1;
5837 spare
->num_raid_devs
= 0;
5838 spare
->cache_size
= mpb
->cache_size
;
5839 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5841 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5842 MPB_SIGNATURE MPB_VERSION_RAID0
);
5844 for (d
= super
->disks
; d
; d
= d
->next
) {
5848 spare
->disk
[0] = d
->disk
;
5849 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5850 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5852 if (super
->sector_size
== 4096)
5853 convert_to_4k_imsm_disk(&spare
->disk
[0]);
5855 sum
= __gen_imsm_checksum(spare
);
5856 spare
->family_num
= __cpu_to_le32(sum
);
5857 spare
->orig_family_num
= 0;
5858 sum
= __gen_imsm_checksum(spare
);
5859 spare
->check_sum
= __cpu_to_le32(sum
);
5861 if (store_imsm_mpb(d
->fd
, spare
)) {
5862 pr_err("failed for device %d:%d %s\n",
5863 d
->major
, d
->minor
, strerror(errno
));
5875 static int write_super_imsm(struct supertype
*st
, int doclose
)
5877 struct intel_super
*super
= st
->sb
;
5878 unsigned int sector_size
= super
->sector_size
;
5879 struct imsm_super
*mpb
= super
->anchor
;
5885 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5887 int clear_migration_record
= 1;
5890 /* 'generation' is incremented everytime the metadata is written */
5891 generation
= __le32_to_cpu(mpb
->generation_num
);
5893 mpb
->generation_num
= __cpu_to_le32(generation
);
5895 /* fix up cases where previous mdadm releases failed to set
5898 if (mpb
->orig_family_num
== 0)
5899 mpb
->orig_family_num
= mpb
->family_num
;
5901 for (d
= super
->disks
; d
; d
= d
->next
) {
5905 mpb
->disk
[d
->index
] = d
->disk
;
5909 for (d
= super
->missing
; d
; d
= d
->next
) {
5910 mpb
->disk
[d
->index
] = d
->disk
;
5913 mpb
->num_disks
= num_disks
;
5914 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5916 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5917 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5918 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5920 imsm_copy_dev(dev
, dev2
);
5921 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5923 if (is_gen_migration(dev2
))
5924 clear_migration_record
= 0;
5927 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
5930 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
5931 mpb
->attributes
|= MPB_ATTRIB_BBM
;
5933 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
5935 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
5936 mpb_size
+= bbm_log_size
;
5937 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5940 assert(super
->len
== 0 || mpb_size
<= super
->len
);
5943 /* recalculate checksum */
5944 sum
= __gen_imsm_checksum(mpb
);
5945 mpb
->check_sum
= __cpu_to_le32(sum
);
5947 if (super
->clean_migration_record_by_mdmon
) {
5948 clear_migration_record
= 1;
5949 super
->clean_migration_record_by_mdmon
= 0;
5951 if (clear_migration_record
)
5952 memset(super
->migr_rec_buf
, 0,
5953 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5955 if (sector_size
== 4096)
5956 convert_to_4k(super
);
5958 /* write the mpb for disks that compose raid devices */
5959 for (d
= super
->disks
; d
; d
= d
->next
) {
5960 if (d
->index
< 0 || is_failed(&d
->disk
))
5963 if (clear_migration_record
) {
5964 unsigned long long dsize
;
5966 get_dev_size(d
->fd
, NULL
, &dsize
);
5967 if (lseek64(d
->fd
, dsize
- sector_size
,
5969 if ((unsigned int)write(d
->fd
,
5970 super
->migr_rec_buf
,
5971 MIGR_REC_BUF_SECTORS
*sector_size
) !=
5972 MIGR_REC_BUF_SECTORS
*sector_size
)
5973 perror("Write migr_rec failed");
5977 if (store_imsm_mpb(d
->fd
, mpb
))
5979 "failed for device %d:%d (fd: %d)%s\n",
5981 d
->fd
, strerror(errno
));
5990 return write_super_imsm_spares(super
, doclose
);
5995 static int create_array(struct supertype
*st
, int dev_idx
)
5998 struct imsm_update_create_array
*u
;
5999 struct intel_super
*super
= st
->sb
;
6000 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
6001 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6002 struct disk_info
*inf
;
6003 struct imsm_disk
*disk
;
6006 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6007 sizeof(*inf
) * map
->num_members
;
6009 u
->type
= update_create_array
;
6010 u
->dev_idx
= dev_idx
;
6011 imsm_copy_dev(&u
->dev
, dev
);
6012 inf
= get_disk_info(u
);
6013 for (i
= 0; i
< map
->num_members
; i
++) {
6014 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6016 disk
= get_imsm_disk(super
, idx
);
6018 disk
= get_imsm_missing(super
, idx
);
6019 serialcpy(inf
[i
].serial
, disk
->serial
);
6021 append_metadata_update(st
, u
, len
);
6026 static int mgmt_disk(struct supertype
*st
)
6028 struct intel_super
*super
= st
->sb
;
6030 struct imsm_update_add_remove_disk
*u
;
6032 if (!super
->disk_mgmt_list
)
6037 u
->type
= update_add_remove_disk
;
6038 append_metadata_update(st
, u
, len
);
6043 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6045 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6047 struct intel_super
*super
= st
->sb
;
6049 struct ppl_header
*ppl_hdr
;
6052 ret
= posix_memalign(&buf
, 4096, PPL_HEADER_SIZE
);
6054 pr_err("Failed to allocate PPL header buffer\n");
6058 memset(buf
, 0, PPL_HEADER_SIZE
);
6060 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6061 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6062 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6064 if (lseek64(fd
, info
->ppl_sector
* 512, SEEK_SET
) < 0) {
6066 perror("Failed to seek to PPL header location");
6069 if (!ret
&& write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6071 perror("Write PPL header failed");
6081 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6082 struct mdinfo
*disk
)
6084 struct intel_super
*super
= st
->sb
;
6088 struct ppl_header
*ppl_hdr
;
6090 struct imsm_dev
*dev
;
6091 struct imsm_map
*map
;
6094 if (disk
->disk
.raid_disk
< 0)
6097 if (posix_memalign(&buf
, 4096, PPL_HEADER_SIZE
)) {
6098 pr_err("Failed to allocate PPL header buffer\n");
6102 dev
= get_imsm_dev(super
, info
->container_member
);
6103 map
= get_imsm_map(dev
, MAP_X
);
6104 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_X
);
6105 d
= get_imsm_dl_disk(super
, idx
);
6107 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6110 if (lseek64(d
->fd
, info
->ppl_sector
* 512, SEEK_SET
) < 0) {
6111 perror("Failed to seek to PPL header location");
6116 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6117 perror("Read PPL header failed");
6124 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6125 ppl_hdr
->checksum
= 0;
6127 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6128 dprintf("Wrong PPL header checksum on %s\n",
6133 if (!ret
&& (__le32_to_cpu(ppl_hdr
->signature
) !=
6134 super
->anchor
->orig_family_num
)) {
6135 dprintf("Wrong PPL header signature on %s\n",
6143 if (ret
== 1 && map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6144 return st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6149 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6151 struct intel_super
*super
= st
->sb
;
6155 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6156 info
->array
.level
!= 5)
6159 for (d
= super
->disks
; d
; d
= d
->next
) {
6160 if (d
->index
< 0 || is_failed(&d
->disk
))
6163 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6171 static int write_init_super_imsm(struct supertype
*st
)
6173 struct intel_super
*super
= st
->sb
;
6174 int current_vol
= super
->current_vol
;
6178 getinfo_super_imsm(st
, &info
, NULL
);
6180 /* we are done with current_vol reset it to point st at the container */
6181 super
->current_vol
= -1;
6183 if (st
->update_tail
) {
6184 /* queue the recently created array / added disk
6185 * as a metadata update */
6187 /* determine if we are creating a volume or adding a disk */
6188 if (current_vol
< 0) {
6189 /* in the mgmt (add/remove) disk case we are running
6190 * in mdmon context, so don't close fd's
6194 rv
= write_init_ppl_imsm_all(st
, &info
);
6196 rv
= create_array(st
, current_vol
);
6200 for (d
= super
->disks
; d
; d
= d
->next
)
6201 Kill(d
->devname
, NULL
, 0, -1, 1);
6202 if (current_vol
>= 0)
6203 rv
= write_init_ppl_imsm_all(st
, &info
);
6205 rv
= write_super_imsm(st
, 1);
6211 static int store_super_imsm(struct supertype
*st
, int fd
)
6213 struct intel_super
*super
= st
->sb
;
6214 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6219 if (super
->sector_size
== 4096)
6220 convert_to_4k(super
);
6221 return store_imsm_mpb(fd
, mpb
);
6224 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6225 int layout
, int raiddisks
, int chunk
,
6226 unsigned long long size
,
6227 unsigned long long data_offset
,
6229 unsigned long long *freesize
,
6233 unsigned long long ldsize
;
6234 struct intel_super
*super
;
6237 if (level
!= LEVEL_CONTAINER
)
6242 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6245 pr_err("imsm: Cannot open %s: %s\n",
6246 dev
, strerror(errno
));
6249 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6254 /* capabilities retrieve could be possible
6255 * note that there is no fd for the disks in array.
6257 super
= alloc_super();
6262 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6268 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6272 fd2devname(fd
, str
);
6273 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6274 fd
, str
, super
->orom
, rv
, raiddisks
);
6276 /* no orom/efi or non-intel hba of the disk */
6283 if (raiddisks
> super
->orom
->tds
) {
6285 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6286 raiddisks
, super
->orom
->tds
);
6290 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6291 (ldsize
>> 9) >> 32 > 0) {
6293 pr_err("%s exceeds maximum platform supported size\n", dev
);
6299 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6305 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6307 const unsigned long long base_start
= e
[*idx
].start
;
6308 unsigned long long end
= base_start
+ e
[*idx
].size
;
6311 if (base_start
== end
)
6315 for (i
= *idx
; i
< num_extents
; i
++) {
6316 /* extend overlapping extents */
6317 if (e
[i
].start
>= base_start
&&
6318 e
[i
].start
<= end
) {
6321 if (e
[i
].start
+ e
[i
].size
> end
)
6322 end
= e
[i
].start
+ e
[i
].size
;
6323 } else if (e
[i
].start
> end
) {
6329 return end
- base_start
;
6332 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6334 /* build a composite disk with all known extents and generate a new
6335 * 'maxsize' given the "all disks in an array must share a common start
6336 * offset" constraint
6338 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6342 unsigned long long pos
;
6343 unsigned long long start
= 0;
6344 unsigned long long maxsize
;
6345 unsigned long reserve
;
6347 /* coalesce and sort all extents. also, check to see if we need to
6348 * reserve space between member arrays
6351 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6354 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6357 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6362 while (i
< sum_extents
) {
6363 e
[j
].start
= e
[i
].start
;
6364 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6366 if (e
[j
-1].size
== 0)
6375 unsigned long long esize
;
6377 esize
= e
[i
].start
- pos
;
6378 if (esize
>= maxsize
) {
6383 pos
= e
[i
].start
+ e
[i
].size
;
6385 } while (e
[i
-1].size
);
6391 /* FIXME assumes volume at offset 0 is the first volume in a
6394 if (start_extent
> 0)
6395 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6399 if (maxsize
< reserve
)
6402 super
->create_offset
= ~((unsigned long long) 0);
6403 if (start
+ reserve
> super
->create_offset
)
6404 return 0; /* start overflows create_offset */
6405 super
->create_offset
= start
+ reserve
;
6407 return maxsize
- reserve
;
6410 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6412 if (level
< 0 || level
== 6 || level
== 4)
6415 /* if we have an orom prevent invalid raid levels */
6418 case 0: return imsm_orom_has_raid0(orom
);
6421 return imsm_orom_has_raid1e(orom
);
6422 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6423 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6424 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6427 return 1; /* not on an Intel RAID platform so anything goes */
6433 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6434 int dpa
, int verbose
)
6436 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6437 struct mdstat_ent
*memb
;
6443 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6444 if (memb
->metadata_version
&&
6445 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6446 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6447 !is_subarray(memb
->metadata_version
+9) &&
6449 struct dev_member
*dev
= memb
->members
;
6451 while(dev
&& (fd
< 0)) {
6452 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6453 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6455 fd
= open(path
, O_RDONLY
, 0);
6456 if (num
<= 0 || fd
< 0) {
6457 pr_vrb("Cannot open %s: %s\n",
6458 dev
->name
, strerror(errno
));
6464 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6465 struct mdstat_ent
*vol
;
6466 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6467 if (vol
->active
> 0 &&
6468 vol
->metadata_version
&&
6469 is_container_member(vol
, memb
->devnm
)) {
6474 if (*devlist
&& (found
< dpa
)) {
6475 dv
= xcalloc(1, sizeof(*dv
));
6476 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6477 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6480 dv
->next
= *devlist
;
6488 free_mdstat(mdstat
);
6493 static struct md_list
*
6494 get_loop_devices(void)
6497 struct md_list
*devlist
= NULL
;
6500 for(i
= 0; i
< 12; i
++) {
6501 dv
= xcalloc(1, sizeof(*dv
));
6502 dv
->devname
= xmalloc(40);
6503 sprintf(dv
->devname
, "/dev/loop%d", i
);
6511 static struct md_list
*
6512 get_devices(const char *hba_path
)
6514 struct md_list
*devlist
= NULL
;
6521 devlist
= get_loop_devices();
6524 /* scroll through /sys/dev/block looking for devices attached to
6527 dir
= opendir("/sys/dev/block");
6528 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6533 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6535 path
= devt_to_devpath(makedev(major
, minor
));
6538 if (!path_attached_to_hba(path
, hba_path
)) {
6545 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6547 fd2devname(fd
, buf
);
6550 pr_err("cannot open device: %s\n",
6555 dv
= xcalloc(1, sizeof(*dv
));
6556 dv
->devname
= xstrdup(buf
);
6563 devlist
= devlist
->next
;
6573 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6574 int verbose
, int *found
)
6576 struct md_list
*tmpdev
;
6578 struct supertype
*st
;
6580 /* first walk the list of devices to find a consistent set
6581 * that match the criterea, if that is possible.
6582 * We flag the ones we like with 'used'.
6585 st
= match_metadata_desc_imsm("imsm");
6587 pr_vrb("cannot allocate memory for imsm supertype\n");
6591 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6592 char *devname
= tmpdev
->devname
;
6594 struct supertype
*tst
;
6596 if (tmpdev
->used
> 1)
6598 tst
= dup_super(st
);
6600 pr_vrb("cannot allocate memory for imsm supertype\n");
6603 tmpdev
->container
= 0;
6604 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6606 dprintf("cannot open device %s: %s\n",
6607 devname
, strerror(errno
));
6609 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
6611 } else if (must_be_container(dfd
)) {
6612 struct supertype
*cst
;
6613 cst
= super_by_fd(dfd
, NULL
);
6615 dprintf("cannot recognize container type %s\n",
6618 } else if (tst
->ss
!= st
->ss
) {
6619 dprintf("non-imsm container - ignore it: %s\n",
6622 } else if (!tst
->ss
->load_container
||
6623 tst
->ss
->load_container(tst
, dfd
, NULL
))
6626 tmpdev
->container
= 1;
6629 cst
->ss
->free_super(cst
);
6631 tmpdev
->st_rdev
= rdev
;
6632 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6633 dprintf("no RAID superblock on %s\n",
6636 } else if (tst
->ss
->compare_super
== NULL
) {
6637 dprintf("Cannot assemble %s metadata on %s\n",
6638 tst
->ss
->name
, devname
);
6644 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6645 /* Ignore unrecognised devices during auto-assembly */
6650 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6652 if (st
->minor_version
== -1)
6653 st
->minor_version
= tst
->minor_version
;
6655 if (memcmp(info
.uuid
, uuid_zero
,
6656 sizeof(int[4])) == 0) {
6657 /* this is a floating spare. It cannot define
6658 * an array unless there are no more arrays of
6659 * this type to be found. It can be included
6660 * in an array of this type though.
6666 if (st
->ss
!= tst
->ss
||
6667 st
->minor_version
!= tst
->minor_version
||
6668 st
->ss
->compare_super(st
, tst
) != 0) {
6669 /* Some mismatch. If exactly one array matches this host,
6670 * we can resolve on that one.
6671 * Or, if we are auto assembling, we just ignore the second
6674 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6680 dprintf("found: devname: %s\n", devname
);
6684 tst
->ss
->free_super(tst
);
6688 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6689 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6690 for (iter
= head
; iter
; iter
= iter
->next
) {
6691 dprintf("content->text_version: %s vol\n",
6692 iter
->text_version
);
6693 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6694 /* do not assemble arrays with unsupported
6696 dprintf("Cannot activate member %s.\n",
6697 iter
->text_version
);
6704 dprintf("No valid super block on device list: err: %d %p\n",
6708 dprintf("no more devices to examine\n");
6711 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6712 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6714 if (count
< tmpdev
->found
)
6717 count
-= tmpdev
->found
;
6720 if (tmpdev
->used
== 1)
6725 st
->ss
->free_super(st
);
6729 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
6732 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6734 const struct orom_entry
*entry
;
6735 struct devid_list
*dv
, *devid_list
;
6740 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
6741 if (strstr(idev
->path
, hba_path
))
6745 if (!idev
|| !idev
->dev_id
)
6748 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
6750 if (!entry
|| !entry
->devid_list
)
6753 devid_list
= entry
->devid_list
;
6754 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
6755 struct md_list
*devlist
;
6756 struct sys_dev
*device
= NULL
;
6761 device
= device_by_id_and_path(dv
->devid
, hba_path
);
6763 device
= device_by_id(dv
->devid
);
6766 hpath
= device
->path
;
6770 devlist
= get_devices(hpath
);
6771 /* if no intel devices return zero volumes */
6772 if (devlist
== NULL
)
6775 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
6777 dprintf("path: %s active arrays: %d\n", hpath
, count
);
6778 if (devlist
== NULL
)
6782 count
+= count_volumes_list(devlist
,
6786 dprintf("found %d count: %d\n", found
, count
);
6789 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
6792 struct md_list
*dv
= devlist
;
6793 devlist
= devlist
->next
;
6801 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
6805 if (hba
->type
== SYS_DEV_VMD
) {
6806 struct sys_dev
*dev
;
6809 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
6810 if (dev
->type
== SYS_DEV_VMD
)
6811 count
+= __count_volumes(dev
->path
, dpa
,
6816 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
6819 static int imsm_default_chunk(const struct imsm_orom
*orom
)
6821 /* up to 512 if the plaform supports it, otherwise the platform max.
6822 * 128 if no platform detected
6824 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
6826 return min(512, (1 << fs
));
6830 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
6831 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
6833 /* check/set platform and metadata limits/defaults */
6834 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
6835 pr_vrb("platform supports a maximum of %d disks per array\n",
6840 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6841 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
6842 pr_vrb("platform does not support raid%d with %d disk%s\n",
6843 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
6847 if (*chunk
== 0 || *chunk
== UnSet
)
6848 *chunk
= imsm_default_chunk(super
->orom
);
6850 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
6851 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
6855 if (layout
!= imsm_level_to_layout(level
)) {
6857 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
6858 else if (level
== 10)
6859 pr_vrb("imsm raid 10 only supports the n2 layout\n");
6861 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
6866 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
6867 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
6868 pr_vrb("platform does not support a volume size over 2TB\n");
6875 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
6876 * FIX ME add ahci details
6878 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
6879 int layout
, int raiddisks
, int *chunk
,
6880 unsigned long long size
,
6881 unsigned long long data_offset
,
6883 unsigned long long *freesize
,
6887 struct intel_super
*super
= st
->sb
;
6888 struct imsm_super
*mpb
;
6890 unsigned long long pos
= 0;
6891 unsigned long long maxsize
;
6895 /* We must have the container info already read in. */
6899 mpb
= super
->anchor
;
6901 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
6902 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6906 /* General test: make sure there is space for
6907 * 'raiddisks' device extents of size 'size' at a given
6910 unsigned long long minsize
= size
;
6911 unsigned long long start_offset
= MaxSector
;
6914 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
6915 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6920 e
= get_extents(super
, dl
);
6923 unsigned long long esize
;
6924 esize
= e
[i
].start
- pos
;
6925 if (esize
>= minsize
)
6927 if (found
&& start_offset
== MaxSector
) {
6930 } else if (found
&& pos
!= start_offset
) {
6934 pos
= e
[i
].start
+ e
[i
].size
;
6936 } while (e
[i
-1].size
);
6941 if (dcnt
< raiddisks
) {
6943 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6950 /* This device must be a member of the set */
6951 if (!stat_is_blkdev(dev
, &rdev
))
6953 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6954 if (dl
->major
== (int)major(rdev
) &&
6955 dl
->minor
== (int)minor(rdev
))
6960 pr_err("%s is not in the same imsm set\n", dev
);
6962 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6963 /* If a volume is present then the current creation attempt
6964 * cannot incorporate new spares because the orom may not
6965 * understand this configuration (all member disks must be
6966 * members of each array in the container).
6968 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6969 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6971 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6972 mpb
->num_disks
!= raiddisks
) {
6973 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6977 /* retrieve the largest free space block */
6978 e
= get_extents(super
, dl
);
6983 unsigned long long esize
;
6985 esize
= e
[i
].start
- pos
;
6986 if (esize
>= maxsize
)
6988 pos
= e
[i
].start
+ e
[i
].size
;
6990 } while (e
[i
-1].size
);
6995 pr_err("unable to determine free space for: %s\n",
6999 if (maxsize
< size
) {
7001 pr_err("%s not enough space (%llu < %llu)\n",
7002 dev
, maxsize
, size
);
7006 /* count total number of extents for merge */
7008 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7010 i
+= dl
->extent_cnt
;
7012 maxsize
= merge_extents(super
, i
);
7014 if (!check_env("IMSM_NO_PLATFORM") &&
7015 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7016 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7020 if (maxsize
< size
|| maxsize
== 0) {
7023 pr_err("no free space left on device. Aborting...\n");
7025 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7031 *freesize
= maxsize
;
7034 int count
= count_volumes(super
->hba
,
7035 super
->orom
->dpa
, verbose
);
7036 if (super
->orom
->vphba
<= count
) {
7037 pr_vrb("platform does not support more than %d raid volumes.\n",
7038 super
->orom
->vphba
);
7045 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7046 unsigned long long size
, int chunk
,
7047 unsigned long long *freesize
)
7049 struct intel_super
*super
= st
->sb
;
7050 struct imsm_super
*mpb
= super
->anchor
;
7055 unsigned long long maxsize
;
7056 unsigned long long minsize
;
7060 /* find the largest common start free region of the possible disks */
7064 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7070 /* don't activate new spares if we are orom constrained
7071 * and there is already a volume active in the container
7073 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7076 e
= get_extents(super
, dl
);
7079 for (i
= 1; e
[i
-1].size
; i
++)
7087 maxsize
= merge_extents(super
, extent_cnt
);
7091 minsize
= chunk
* 2;
7093 if (cnt
< raiddisks
||
7094 (super
->orom
&& used
&& used
!= raiddisks
) ||
7095 maxsize
< minsize
||
7097 pr_err("not enough devices with space to create array.\n");
7098 return 0; /* No enough free spaces large enough */
7109 if (!check_env("IMSM_NO_PLATFORM") &&
7110 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7111 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7115 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7117 dl
->raiddisk
= cnt
++;
7121 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7126 static int reserve_space(struct supertype
*st
, int raiddisks
,
7127 unsigned long long size
, int chunk
,
7128 unsigned long long *freesize
)
7130 struct intel_super
*super
= st
->sb
;
7135 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7138 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7140 dl
->raiddisk
= cnt
++;
7147 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7148 int raiddisks
, int *chunk
, unsigned long long size
,
7149 unsigned long long data_offset
,
7150 char *dev
, unsigned long long *freesize
,
7151 int consistency_policy
, int verbose
)
7158 * if given unused devices create a container
7159 * if given given devices in a container create a member volume
7161 if (level
== LEVEL_CONTAINER
) {
7162 /* Must be a fresh device to add to a container */
7163 return validate_geometry_imsm_container(st
, level
, layout
,
7173 struct intel_super
*super
= st
->sb
;
7174 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7175 raiddisks
, chunk
, size
,
7178 /* we are being asked to automatically layout a
7179 * new volume based on the current contents of
7180 * the container. If the the parameters can be
7181 * satisfied reserve_space will record the disks,
7182 * start offset, and size of the volume to be
7183 * created. add_to_super and getinfo_super
7184 * detect when autolayout is in progress.
7186 /* assuming that freesize is always given when array is
7188 if (super
->orom
&& freesize
) {
7190 count
= count_volumes(super
->hba
,
7191 super
->orom
->dpa
, verbose
);
7192 if (super
->orom
->vphba
<= count
) {
7193 pr_vrb("platform does not support more than %d raid volumes.\n",
7194 super
->orom
->vphba
);
7199 return reserve_space(st
, raiddisks
, size
,
7205 /* creating in a given container */
7206 return validate_geometry_imsm_volume(st
, level
, layout
,
7207 raiddisks
, chunk
, size
,
7209 dev
, freesize
, verbose
);
7212 /* This device needs to be a device in an 'imsm' container */
7213 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7216 pr_err("Cannot create this array on device %s\n",
7221 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7223 pr_err("Cannot open %s: %s\n",
7224 dev
, strerror(errno
));
7227 /* Well, it is in use by someone, maybe an 'imsm' container. */
7228 cfd
= open_container(fd
);
7232 pr_err("Cannot use %s: It is busy\n",
7236 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7237 if (sra
&& sra
->array
.major_version
== -1 &&
7238 strcmp(sra
->text_version
, "imsm") == 0)
7242 /* This is a member of a imsm container. Load the container
7243 * and try to create a volume
7245 struct intel_super
*super
;
7247 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7249 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7251 return validate_geometry_imsm_volume(st
, level
, layout
,
7253 size
, data_offset
, dev
,
7260 pr_err("failed container membership check\n");
7266 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7268 struct intel_super
*super
= st
->sb
;
7270 if (level
&& *level
== UnSet
)
7271 *level
= LEVEL_CONTAINER
;
7273 if (level
&& layout
&& *layout
== UnSet
)
7274 *layout
= imsm_level_to_layout(*level
);
7276 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7277 *chunk
= imsm_default_chunk(super
->orom
);
7280 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7282 static int kill_subarray_imsm(struct supertype
*st
)
7284 /* remove the subarray currently referenced by ->current_vol */
7286 struct intel_dev
**dp
;
7287 struct intel_super
*super
= st
->sb
;
7288 __u8 current_vol
= super
->current_vol
;
7289 struct imsm_super
*mpb
= super
->anchor
;
7291 if (super
->current_vol
< 0)
7293 super
->current_vol
= -1; /* invalidate subarray cursor */
7295 /* block deletions that would change the uuid of active subarrays
7297 * FIXME when immutable ids are available, but note that we'll
7298 * also need to fixup the invalidated/active subarray indexes in
7301 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7304 if (i
< current_vol
)
7306 sprintf(subarray
, "%u", i
);
7307 if (is_subarray_active(subarray
, st
->devnm
)) {
7308 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7315 if (st
->update_tail
) {
7316 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7318 u
->type
= update_kill_array
;
7319 u
->dev_idx
= current_vol
;
7320 append_metadata_update(st
, u
, sizeof(*u
));
7325 for (dp
= &super
->devlist
; *dp
;)
7326 if ((*dp
)->index
== current_vol
) {
7329 handle_missing(super
, (*dp
)->dev
);
7330 if ((*dp
)->index
> current_vol
)
7335 /* no more raid devices, all active components are now spares,
7336 * but of course failed are still failed
7338 if (--mpb
->num_raid_devs
== 0) {
7341 for (d
= super
->disks
; d
; d
= d
->next
)
7346 super
->updates_pending
++;
7351 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7352 char *update
, struct mddev_ident
*ident
)
7354 /* update the subarray currently referenced by ->current_vol */
7355 struct intel_super
*super
= st
->sb
;
7356 struct imsm_super
*mpb
= super
->anchor
;
7358 if (strcmp(update
, "name") == 0) {
7359 char *name
= ident
->name
;
7363 if (is_subarray_active(subarray
, st
->devnm
)) {
7364 pr_err("Unable to update name of active subarray\n");
7368 if (!check_name(super
, name
, 0))
7371 vol
= strtoul(subarray
, &ep
, 10);
7372 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7375 if (st
->update_tail
) {
7376 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7378 u
->type
= update_rename_array
;
7380 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7381 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7382 append_metadata_update(st
, u
, sizeof(*u
));
7384 struct imsm_dev
*dev
;
7387 dev
= get_imsm_dev(super
, vol
);
7388 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
7389 dev
->volume
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7390 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7391 dev
= get_imsm_dev(super
, i
);
7392 handle_missing(super
, dev
);
7394 super
->updates_pending
++;
7396 } else if (strcmp(update
, "ppl") == 0 ||
7397 strcmp(update
, "no-ppl") == 0) {
7400 int vol
= strtoul(subarray
, &ep
, 10);
7402 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7405 if (strcmp(update
, "ppl") == 0)
7406 new_policy
= RWH_DISTRIBUTED
;
7408 new_policy
= RWH_OFF
;
7410 if (st
->update_tail
) {
7411 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7413 u
->type
= update_rwh_policy
;
7415 u
->new_policy
= new_policy
;
7416 append_metadata_update(st
, u
, sizeof(*u
));
7418 struct imsm_dev
*dev
;
7420 dev
= get_imsm_dev(super
, vol
);
7421 dev
->rwh_policy
= new_policy
;
7422 super
->updates_pending
++;
7430 static int is_gen_migration(struct imsm_dev
*dev
)
7435 if (!dev
->vol
.migr_state
)
7438 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7444 static int is_rebuilding(struct imsm_dev
*dev
)
7446 struct imsm_map
*migr_map
;
7448 if (!dev
->vol
.migr_state
)
7451 if (migr_type(dev
) != MIGR_REBUILD
)
7454 migr_map
= get_imsm_map(dev
, MAP_1
);
7456 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7462 static int is_initializing(struct imsm_dev
*dev
)
7464 struct imsm_map
*migr_map
;
7466 if (!dev
->vol
.migr_state
)
7469 if (migr_type(dev
) != MIGR_INIT
)
7472 migr_map
= get_imsm_map(dev
, MAP_1
);
7474 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7480 static void update_recovery_start(struct intel_super
*super
,
7481 struct imsm_dev
*dev
,
7482 struct mdinfo
*array
)
7484 struct mdinfo
*rebuild
= NULL
;
7488 if (!is_rebuilding(dev
))
7491 /* Find the rebuild target, but punt on the dual rebuild case */
7492 for (d
= array
->devs
; d
; d
= d
->next
)
7493 if (d
->recovery_start
== 0) {
7500 /* (?) none of the disks are marked with
7501 * IMSM_ORD_REBUILD, so assume they are missing and the
7502 * disk_ord_tbl was not correctly updated
7504 dprintf("failed to locate out-of-sync disk\n");
7508 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7509 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7512 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7514 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7516 /* Given a container loaded by load_super_imsm_all,
7517 * extract information about all the arrays into
7519 * If 'subarray' is given, just extract info about that array.
7521 * For each imsm_dev create an mdinfo, fill it in,
7522 * then look for matching devices in super->disks
7523 * and create appropriate device mdinfo.
7525 struct intel_super
*super
= st
->sb
;
7526 struct imsm_super
*mpb
= super
->anchor
;
7527 struct mdinfo
*rest
= NULL
;
7531 int spare_disks
= 0;
7533 /* do not assemble arrays when not all attributes are supported */
7534 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7536 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7539 /* count spare devices, not used in maps
7541 for (d
= super
->disks
; d
; d
= d
->next
)
7545 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7546 struct imsm_dev
*dev
;
7547 struct imsm_map
*map
;
7548 struct imsm_map
*map2
;
7549 struct mdinfo
*this;
7555 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7558 dev
= get_imsm_dev(super
, i
);
7559 map
= get_imsm_map(dev
, MAP_0
);
7560 map2
= get_imsm_map(dev
, MAP_1
);
7562 /* do not publish arrays that are in the middle of an
7563 * unsupported migration
7565 if (dev
->vol
.migr_state
&&
7566 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7567 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7571 /* do not publish arrays that are not support by controller's
7575 this = xmalloc(sizeof(*this));
7577 super
->current_vol
= i
;
7578 getinfo_super_imsm_volume(st
, this, NULL
);
7580 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7581 /* mdadm does not support all metadata features- set the bit in all arrays state */
7582 if (!validate_geometry_imsm_orom(super
,
7583 get_imsm_raid_level(map
), /* RAID level */
7584 imsm_level_to_layout(get_imsm_raid_level(map
)),
7585 map
->num_members
, /* raid disks */
7586 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
7588 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7590 this->array
.state
|=
7591 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7592 (1<<MD_SB_BLOCK_VOLUME
);
7595 /* if array has bad blocks, set suitable bit in all arrays state */
7597 this->array
.state
|=
7598 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7599 (1<<MD_SB_BLOCK_VOLUME
);
7601 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7602 unsigned long long recovery_start
;
7603 struct mdinfo
*info_d
;
7610 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7611 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7612 for (d
= super
->disks
; d
; d
= d
->next
)
7613 if (d
->index
== idx
)
7616 recovery_start
= MaxSector
;
7619 if (d
&& is_failed(&d
->disk
))
7621 if (ord
& IMSM_ORD_REBUILD
)
7625 * if we skip some disks the array will be assmebled degraded;
7626 * reset resync start to avoid a dirty-degraded
7627 * situation when performing the intial sync
7629 * FIXME handle dirty degraded
7631 if ((skip
|| recovery_start
== 0) &&
7632 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
))
7633 this->resync_start
= MaxSector
;
7637 info_d
= xcalloc(1, sizeof(*info_d
));
7638 info_d
->next
= this->devs
;
7639 this->devs
= info_d
;
7641 info_d
->disk
.number
= d
->index
;
7642 info_d
->disk
.major
= d
->major
;
7643 info_d
->disk
.minor
= d
->minor
;
7644 info_d
->disk
.raid_disk
= slot
;
7645 info_d
->recovery_start
= recovery_start
;
7647 if (slot
< map2
->num_members
)
7648 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7650 this->array
.spare_disks
++;
7652 if (slot
< map
->num_members
)
7653 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7655 this->array
.spare_disks
++;
7657 if (info_d
->recovery_start
== MaxSector
)
7658 this->array
.working_disks
++;
7660 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7661 info_d
->data_offset
= pba_of_lba0(map
);
7663 if (map
->raid_level
== 5) {
7664 info_d
->component_size
=
7665 num_data_stripes(map
) *
7666 map
->blocks_per_strip
;
7667 info_d
->ppl_sector
= this->ppl_sector
;
7668 info_d
->ppl_size
= this->ppl_size
;
7670 info_d
->component_size
= blocks_per_member(map
);
7672 info_d
->consistency_policy
= this->consistency_policy
;
7674 info_d
->bb
.supported
= 1;
7675 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7676 info_d
->data_offset
,
7677 info_d
->component_size
,
7680 /* now that the disk list is up-to-date fixup recovery_start */
7681 update_recovery_start(super
, dev
, this);
7682 this->array
.spare_disks
+= spare_disks
;
7684 /* check for reshape */
7685 if (this->reshape_active
== 1)
7686 recover_backup_imsm(st
, this);
7693 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7694 int failed
, int look_in_map
)
7696 struct imsm_map
*map
;
7698 map
= get_imsm_map(dev
, look_in_map
);
7701 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
7702 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
7704 switch (get_imsm_raid_level(map
)) {
7706 return IMSM_T_STATE_FAILED
;
7709 if (failed
< map
->num_members
)
7710 return IMSM_T_STATE_DEGRADED
;
7712 return IMSM_T_STATE_FAILED
;
7717 * check to see if any mirrors have failed, otherwise we
7718 * are degraded. Even numbered slots are mirrored on
7722 /* gcc -Os complains that this is unused */
7723 int insync
= insync
;
7725 for (i
= 0; i
< map
->num_members
; i
++) {
7726 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
7727 int idx
= ord_to_idx(ord
);
7728 struct imsm_disk
*disk
;
7730 /* reset the potential in-sync count on even-numbered
7731 * slots. num_copies is always 2 for imsm raid10
7736 disk
= get_imsm_disk(super
, idx
);
7737 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7740 /* no in-sync disks left in this mirror the
7744 return IMSM_T_STATE_FAILED
;
7747 return IMSM_T_STATE_DEGRADED
;
7751 return IMSM_T_STATE_DEGRADED
;
7753 return IMSM_T_STATE_FAILED
;
7759 return map
->map_state
;
7762 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
7767 struct imsm_disk
*disk
;
7768 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7769 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
7770 struct imsm_map
*map_for_loop
;
7775 /* at the beginning of migration we set IMSM_ORD_REBUILD on
7776 * disks that are being rebuilt. New failures are recorded to
7777 * map[0]. So we look through all the disks we started with and
7778 * see if any failures are still present, or if any new ones
7782 if (prev
&& (map
->num_members
< prev
->num_members
))
7783 map_for_loop
= prev
;
7785 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
7787 /* when MAP_X is passed both maps failures are counted
7790 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
7791 i
< prev
->num_members
) {
7792 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
7793 idx_1
= ord_to_idx(ord
);
7795 disk
= get_imsm_disk(super
, idx_1
);
7796 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7799 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
7800 i
< map
->num_members
) {
7801 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
7802 idx
= ord_to_idx(ord
);
7805 disk
= get_imsm_disk(super
, idx
);
7806 if (!disk
|| is_failed(disk
) ||
7807 ord
& IMSM_ORD_REBUILD
)
7816 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
7819 struct intel_super
*super
= c
->sb
;
7820 struct imsm_super
*mpb
= super
->anchor
;
7821 struct imsm_update_prealloc_bb_mem u
;
7823 if (atoi(inst
) >= mpb
->num_raid_devs
) {
7824 pr_err("subarry index %d, out of range\n", atoi(inst
));
7828 dprintf("imsm: open_new %s\n", inst
);
7829 a
->info
.container_member
= atoi(inst
);
7831 u
.type
= update_prealloc_badblocks_mem
;
7832 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
7837 static int is_resyncing(struct imsm_dev
*dev
)
7839 struct imsm_map
*migr_map
;
7841 if (!dev
->vol
.migr_state
)
7844 if (migr_type(dev
) == MIGR_INIT
||
7845 migr_type(dev
) == MIGR_REPAIR
)
7848 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7851 migr_map
= get_imsm_map(dev
, MAP_1
);
7853 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
7854 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
7860 /* return true if we recorded new information */
7861 static int mark_failure(struct intel_super
*super
,
7862 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7866 struct imsm_map
*map
;
7867 char buf
[MAX_RAID_SERIAL_LEN
+3];
7868 unsigned int len
, shift
= 0;
7870 /* new failures are always set in map[0] */
7871 map
= get_imsm_map(dev
, MAP_0
);
7873 slot
= get_imsm_disk_slot(map
, idx
);
7877 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
7878 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
7881 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
7882 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
7884 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
7885 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
7886 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
7888 disk
->status
|= FAILED_DISK
;
7889 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
7890 /* mark failures in second map if second map exists and this disk
7892 * This is valid for migration, initialization and rebuild
7894 if (dev
->vol
.migr_state
) {
7895 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
7896 int slot2
= get_imsm_disk_slot(map2
, idx
);
7898 if (slot2
< map2
->num_members
&& slot2
>= 0)
7899 set_imsm_ord_tbl_ent(map2
, slot2
,
7900 idx
| IMSM_ORD_REBUILD
);
7902 if (map
->failed_disk_num
== 0xff)
7903 map
->failed_disk_num
= slot
;
7905 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
7910 static void mark_missing(struct intel_super
*super
,
7911 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7913 mark_failure(super
, dev
, disk
, idx
);
7915 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
7918 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
7919 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
7922 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7926 if (!super
->missing
)
7929 /* When orom adds replacement for missing disk it does
7930 * not remove entry of missing disk, but just updates map with
7931 * new added disk. So it is not enough just to test if there is
7932 * any missing disk, we have to look if there are any failed disks
7933 * in map to stop migration */
7935 dprintf("imsm: mark missing\n");
7936 /* end process for initialization and rebuild only
7938 if (is_gen_migration(dev
) == 0) {
7939 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7943 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7944 struct imsm_map
*map1
;
7945 int i
, ord
, ord_map1
;
7948 for (i
= 0; i
< map
->num_members
; i
++) {
7949 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
7950 if (!(ord
& IMSM_ORD_REBUILD
))
7953 map1
= get_imsm_map(dev
, MAP_1
);
7957 ord_map1
= __le32_to_cpu(map1
->disk_ord_tbl
[i
]);
7958 if (ord_map1
& IMSM_ORD_REBUILD
)
7963 map_state
= imsm_check_degraded(super
, dev
,
7965 end_migration(dev
, super
, map_state
);
7969 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7970 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
7971 super
->updates_pending
++;
7974 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7977 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7978 unsigned long long array_blocks
;
7979 struct imsm_map
*map
;
7981 if (used_disks
== 0) {
7982 /* when problems occures
7983 * return current array_blocks value
7985 array_blocks
= __le32_to_cpu(dev
->size_high
);
7986 array_blocks
= array_blocks
<< 32;
7987 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7989 return array_blocks
;
7992 /* set array size in metadata
7994 if (new_size
<= 0) {
7995 /* OLCE size change is caused by added disks
7997 map
= get_imsm_map(dev
, MAP_0
);
7998 array_blocks
= blocks_per_member(map
) * used_disks
;
8000 /* Online Volume Size Change
8001 * Using available free space
8003 array_blocks
= new_size
;
8006 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
8007 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
8008 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
8010 return array_blocks
;
8013 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
8015 static void imsm_progress_container_reshape(struct intel_super
*super
)
8017 /* if no device has a migr_state, but some device has a
8018 * different number of members than the previous device, start
8019 * changing the number of devices in this device to match
8022 struct imsm_super
*mpb
= super
->anchor
;
8023 int prev_disks
= -1;
8027 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8028 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8029 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8030 struct imsm_map
*map2
;
8031 int prev_num_members
;
8033 if (dev
->vol
.migr_state
)
8036 if (prev_disks
== -1)
8037 prev_disks
= map
->num_members
;
8038 if (prev_disks
== map
->num_members
)
8041 /* OK, this array needs to enter reshape mode.
8042 * i.e it needs a migr_state
8045 copy_map_size
= sizeof_imsm_map(map
);
8046 prev_num_members
= map
->num_members
;
8047 map
->num_members
= prev_disks
;
8048 dev
->vol
.migr_state
= 1;
8049 dev
->vol
.curr_migr_unit
= 0;
8050 set_migr_type(dev
, MIGR_GEN_MIGR
);
8051 for (i
= prev_num_members
;
8052 i
< map
->num_members
; i
++)
8053 set_imsm_ord_tbl_ent(map
, i
, i
);
8054 map2
= get_imsm_map(dev
, MAP_1
);
8055 /* Copy the current map */
8056 memcpy(map2
, map
, copy_map_size
);
8057 map2
->num_members
= prev_num_members
;
8059 imsm_set_array_size(dev
, -1);
8060 super
->clean_migration_record_by_mdmon
= 1;
8061 super
->updates_pending
++;
8065 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8066 * states are handled in imsm_set_disk() with one exception, when a
8067 * resync is stopped due to a new failure this routine will set the
8068 * 'degraded' state for the array.
8070 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8072 int inst
= a
->info
.container_member
;
8073 struct intel_super
*super
= a
->container
->sb
;
8074 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8075 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8076 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8077 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8078 __u32 blocks_per_unit
;
8080 if (dev
->vol
.migr_state
&&
8081 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8082 /* array state change is blocked due to reshape action
8084 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8085 * - finish the reshape (if last_checkpoint is big and action != reshape)
8086 * - update curr_migr_unit
8088 if (a
->curr_action
== reshape
) {
8089 /* still reshaping, maybe update curr_migr_unit */
8090 goto mark_checkpoint
;
8092 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8093 /* for some reason we aborted the reshape.
8095 * disable automatic metadata rollback
8096 * user action is required to recover process
8099 struct imsm_map
*map2
=
8100 get_imsm_map(dev
, MAP_1
);
8101 dev
->vol
.migr_state
= 0;
8102 set_migr_type(dev
, 0);
8103 dev
->vol
.curr_migr_unit
= 0;
8105 sizeof_imsm_map(map2
));
8106 super
->updates_pending
++;
8109 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8110 unsigned long long array_blocks
;
8114 used_disks
= imsm_num_data_members(dev
, MAP_0
);
8115 if (used_disks
> 0) {
8117 blocks_per_member(map
) *
8120 round_size_to_mb(array_blocks
,
8122 a
->info
.custom_array_size
= array_blocks
;
8123 /* encourage manager to update array
8127 a
->check_reshape
= 1;
8129 /* finalize online capacity expansion/reshape */
8130 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8132 mdi
->disk
.raid_disk
,
8135 imsm_progress_container_reshape(super
);
8140 /* before we activate this array handle any missing disks */
8141 if (consistent
== 2)
8142 handle_missing(super
, dev
);
8144 if (consistent
== 2 &&
8145 (!is_resync_complete(&a
->info
) ||
8146 map_state
!= IMSM_T_STATE_NORMAL
||
8147 dev
->vol
.migr_state
))
8150 if (is_resync_complete(&a
->info
)) {
8151 /* complete intialization / resync,
8152 * recovery and interrupted recovery is completed in
8155 if (is_resyncing(dev
)) {
8156 dprintf("imsm: mark resync done\n");
8157 end_migration(dev
, super
, map_state
);
8158 super
->updates_pending
++;
8159 a
->last_checkpoint
= 0;
8161 } else if ((!is_resyncing(dev
) && !failed
) &&
8162 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8163 /* mark the start of the init process if nothing is failed */
8164 dprintf("imsm: mark resync start\n");
8165 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8166 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8168 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8169 super
->updates_pending
++;
8173 /* skip checkpointing for general migration,
8174 * it is controlled in mdadm
8176 if (is_gen_migration(dev
))
8177 goto skip_mark_checkpoint
;
8179 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8180 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8181 if (blocks_per_unit
) {
8185 units
= a
->last_checkpoint
/ blocks_per_unit
;
8188 /* check that we did not overflow 32-bits, and that
8189 * curr_migr_unit needs updating
8191 if (units32
== units
&&
8193 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8194 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8195 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8196 super
->updates_pending
++;
8200 skip_mark_checkpoint
:
8201 /* mark dirty / clean */
8202 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8203 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8204 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8206 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8208 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8209 if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
8210 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8212 super
->updates_pending
++;
8218 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8220 int inst
= a
->info
.container_member
;
8221 struct intel_super
*super
= a
->container
->sb
;
8222 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8223 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8225 if (slot
> map
->num_members
) {
8226 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8227 slot
, map
->num_members
- 1);
8234 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8237 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8239 int inst
= a
->info
.container_member
;
8240 struct intel_super
*super
= a
->container
->sb
;
8241 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8242 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8243 struct imsm_disk
*disk
;
8245 int recovery_not_finished
= 0;
8249 int rebuild_done
= 0;
8252 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_X
);
8256 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8257 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8259 /* check for new failures */
8260 if (state
& DS_FAULTY
) {
8261 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8262 super
->updates_pending
++;
8265 /* check if in_sync */
8266 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8267 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8269 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8271 super
->updates_pending
++;
8274 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8275 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8277 /* check if recovery complete, newly degraded, or failed */
8278 dprintf("imsm: Detected transition to state ");
8279 switch (map_state
) {
8280 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8281 dprintf("normal: ");
8282 if (is_rebuilding(dev
)) {
8283 dprintf_cont("while rebuilding");
8284 /* check if recovery is really finished */
8285 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8286 if (mdi
->recovery_start
!= MaxSector
) {
8287 recovery_not_finished
= 1;
8290 if (recovery_not_finished
) {
8292 dprintf("Rebuild has not finished yet, state not changed");
8293 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8294 a
->last_checkpoint
= mdi
->recovery_start
;
8295 super
->updates_pending
++;
8299 end_migration(dev
, super
, map_state
);
8300 map
= get_imsm_map(dev
, MAP_0
);
8301 map
->failed_disk_num
= ~0;
8302 super
->updates_pending
++;
8303 a
->last_checkpoint
= 0;
8306 if (is_gen_migration(dev
)) {
8307 dprintf_cont("while general migration");
8308 if (a
->last_checkpoint
>= a
->info
.component_size
)
8309 end_migration(dev
, super
, map_state
);
8311 map
->map_state
= map_state
;
8312 map
= get_imsm_map(dev
, MAP_0
);
8313 map
->failed_disk_num
= ~0;
8314 super
->updates_pending
++;
8318 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8319 dprintf_cont("degraded: ");
8320 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8321 dprintf_cont("mark degraded");
8322 map
->map_state
= map_state
;
8323 super
->updates_pending
++;
8324 a
->last_checkpoint
= 0;
8327 if (is_rebuilding(dev
)) {
8328 dprintf_cont("while rebuilding.");
8329 if (map
->map_state
!= map_state
) {
8330 dprintf_cont(" Map state change");
8331 end_migration(dev
, super
, map_state
);
8332 super
->updates_pending
++;
8333 } else if (!rebuild_done
) {
8337 /* check if recovery is really finished */
8338 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8339 if (mdi
->recovery_start
!= MaxSector
) {
8340 recovery_not_finished
= 1;
8343 if (recovery_not_finished
) {
8345 dprintf("Rebuild has not finished yet, state not changed");
8346 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8347 a
->last_checkpoint
=
8348 mdi
->recovery_start
;
8349 super
->updates_pending
++;
8354 dprintf_cont(" Rebuild done, still degraded");
8355 dev
->vol
.migr_state
= 0;
8356 set_migr_type(dev
, 0);
8357 dev
->vol
.curr_migr_unit
= 0;
8359 for (i
= 0; i
< map
->num_members
; i
++) {
8360 int idx
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
8362 if (idx
& IMSM_ORD_REBUILD
)
8363 map
->failed_disk_num
= i
;
8365 super
->updates_pending
++;
8368 if (is_gen_migration(dev
)) {
8369 dprintf_cont("while general migration");
8370 if (a
->last_checkpoint
>= a
->info
.component_size
)
8371 end_migration(dev
, super
, map_state
);
8373 map
->map_state
= map_state
;
8374 manage_second_map(super
, dev
);
8376 super
->updates_pending
++;
8379 if (is_initializing(dev
)) {
8380 dprintf_cont("while initialization.");
8381 map
->map_state
= map_state
;
8382 super
->updates_pending
++;
8386 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8387 dprintf_cont("failed: ");
8388 if (is_gen_migration(dev
)) {
8389 dprintf_cont("while general migration");
8390 map
->map_state
= map_state
;
8391 super
->updates_pending
++;
8394 if (map
->map_state
!= map_state
) {
8395 dprintf_cont("mark failed");
8396 end_migration(dev
, super
, map_state
);
8397 super
->updates_pending
++;
8398 a
->last_checkpoint
= 0;
8403 dprintf_cont("state %i\n", map_state
);
8408 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8411 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8412 unsigned long long dsize
;
8413 unsigned long long sectors
;
8414 unsigned int sector_size
;
8416 get_dev_sector_size(fd
, NULL
, §or_size
);
8417 get_dev_size(fd
, NULL
, &dsize
);
8419 if (mpb_size
> sector_size
) {
8420 /* -1 to account for anchor */
8421 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8423 /* write the extended mpb to the sectors preceeding the anchor */
8424 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8428 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8429 sector_size
* sectors
) != sector_size
* sectors
)
8433 /* first block is stored on second to last sector of the disk */
8434 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8437 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8443 static void imsm_sync_metadata(struct supertype
*container
)
8445 struct intel_super
*super
= container
->sb
;
8447 dprintf("sync metadata: %d\n", super
->updates_pending
);
8448 if (!super
->updates_pending
)
8451 write_super_imsm(container
, 0);
8453 super
->updates_pending
= 0;
8456 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8458 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8459 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8462 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8466 if (dl
&& is_failed(&dl
->disk
))
8470 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8475 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8476 struct active_array
*a
, int activate_new
,
8477 struct mdinfo
*additional_test_list
)
8479 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8480 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8481 struct imsm_super
*mpb
= super
->anchor
;
8482 struct imsm_map
*map
;
8483 unsigned long long pos
;
8488 __u32 array_start
= 0;
8489 __u32 array_end
= 0;
8491 struct mdinfo
*test_list
;
8493 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8494 /* If in this array, skip */
8495 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8496 if (d
->state_fd
>= 0 &&
8497 d
->disk
.major
== dl
->major
&&
8498 d
->disk
.minor
== dl
->minor
) {
8499 dprintf("%x:%x already in array\n",
8500 dl
->major
, dl
->minor
);
8505 test_list
= additional_test_list
;
8507 if (test_list
->disk
.major
== dl
->major
&&
8508 test_list
->disk
.minor
== dl
->minor
) {
8509 dprintf("%x:%x already in additional test list\n",
8510 dl
->major
, dl
->minor
);
8513 test_list
= test_list
->next
;
8518 /* skip in use or failed drives */
8519 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8521 dprintf("%x:%x status (failed: %d index: %d)\n",
8522 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8526 /* skip pure spares when we are looking for partially
8527 * assimilated drives
8529 if (dl
->index
== -1 && !activate_new
)
8532 if (!drive_validate_sector_size(super
, dl
))
8535 /* Does this unused device have the requisite free space?
8536 * It needs to be able to cover all member volumes
8538 ex
= get_extents(super
, dl
);
8540 dprintf("cannot get extents\n");
8543 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8544 dev
= get_imsm_dev(super
, i
);
8545 map
= get_imsm_map(dev
, MAP_0
);
8547 /* check if this disk is already a member of
8550 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8556 array_start
= pba_of_lba0(map
);
8557 array_end
= array_start
+
8558 blocks_per_member(map
) - 1;
8561 /* check that we can start at pba_of_lba0 with
8562 * blocks_per_member of space
8564 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8568 pos
= ex
[j
].start
+ ex
[j
].size
;
8570 } while (ex
[j
-1].size
);
8577 if (i
< mpb
->num_raid_devs
) {
8578 dprintf("%x:%x does not have %u to %u available\n",
8579 dl
->major
, dl
->minor
, array_start
, array_end
);
8589 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8591 struct imsm_dev
*dev2
;
8592 struct imsm_map
*map
;
8598 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8600 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8601 if (state
== IMSM_T_STATE_FAILED
) {
8602 map
= get_imsm_map(dev2
, MAP_0
);
8605 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8607 * Check if failed disks are deleted from intel
8608 * disk list or are marked to be deleted
8610 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8611 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8613 * Do not rebuild the array if failed disks
8614 * from failed sub-array are not removed from
8618 is_failed(&idisk
->disk
) &&
8619 (idisk
->action
!= DISK_REMOVE
))
8627 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8628 struct metadata_update
**updates
)
8631 * Find a device with unused free space and use it to replace a
8632 * failed/vacant region in an array. We replace failed regions one a
8633 * array at a time. The result is that a new spare disk will be added
8634 * to the first failed array and after the monitor has finished
8635 * propagating failures the remainder will be consumed.
8637 * FIXME add a capability for mdmon to request spares from another
8641 struct intel_super
*super
= a
->container
->sb
;
8642 int inst
= a
->info
.container_member
;
8643 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8644 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8645 int failed
= a
->info
.array
.raid_disks
;
8646 struct mdinfo
*rv
= NULL
;
8649 struct metadata_update
*mu
;
8651 struct imsm_update_activate_spare
*u
;
8656 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8657 if ((d
->curr_state
& DS_FAULTY
) &&
8659 /* wait for Removal to happen */
8661 if (d
->state_fd
>= 0)
8665 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8666 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8668 if (imsm_reshape_blocks_arrays_changes(super
))
8671 /* Cannot activate another spare if rebuild is in progress already
8673 if (is_rebuilding(dev
)) {
8674 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8678 if (a
->info
.array
.level
== 4)
8679 /* No repair for takeovered array
8680 * imsm doesn't support raid4
8684 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8685 IMSM_T_STATE_DEGRADED
)
8688 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8689 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8694 * If there are any failed disks check state of the other volume.
8695 * Block rebuild if the another one is failed until failed disks
8696 * are removed from container.
8699 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8700 MAX_RAID_SERIAL_LEN
, dev
->volume
);
8701 /* check if states of the other volumes allow for rebuild */
8702 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
8704 allowed
= imsm_rebuild_allowed(a
->container
,
8712 /* For each slot, if it is not working, find a spare */
8713 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
8714 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8715 if (d
->disk
.raid_disk
== i
)
8717 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
8718 if (d
&& (d
->state_fd
>= 0))
8722 * OK, this device needs recovery. Try to re-add the
8723 * previous occupant of this slot, if this fails see if
8724 * we can continue the assimilation of a spare that was
8725 * partially assimilated, finally try to activate a new
8728 dl
= imsm_readd(super
, i
, a
);
8730 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
8732 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
8736 /* found a usable disk with enough space */
8737 di
= xcalloc(1, sizeof(*di
));
8739 /* dl->index will be -1 in the case we are activating a
8740 * pristine spare. imsm_process_update() will create a
8741 * new index in this case. Once a disk is found to be
8742 * failed in all member arrays it is kicked from the
8745 di
->disk
.number
= dl
->index
;
8747 /* (ab)use di->devs to store a pointer to the device
8750 di
->devs
= (struct mdinfo
*) dl
;
8752 di
->disk
.raid_disk
= i
;
8753 di
->disk
.major
= dl
->major
;
8754 di
->disk
.minor
= dl
->minor
;
8756 di
->recovery_start
= 0;
8757 di
->data_offset
= pba_of_lba0(map
);
8758 di
->component_size
= a
->info
.component_size
;
8759 di
->container_member
= inst
;
8760 di
->bb
.supported
= 1;
8761 if (dev
->rwh_policy
== RWH_DISTRIBUTED
) {
8762 di
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
8763 di
->ppl_sector
= get_ppl_sector(super
, inst
);
8764 di
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
) >> 9;
8766 super
->random
= random32();
8770 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
8771 i
, di
->data_offset
);
8775 /* No spares found */
8777 /* Now 'rv' has a list of devices to return.
8778 * Create a metadata_update record to update the
8779 * disk_ord_tbl for the array
8781 mu
= xmalloc(sizeof(*mu
));
8782 mu
->buf
= xcalloc(num_spares
,
8783 sizeof(struct imsm_update_activate_spare
));
8785 mu
->space_list
= NULL
;
8786 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
8787 mu
->next
= *updates
;
8788 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
8790 for (di
= rv
; di
; di
= di
->next
) {
8791 u
->type
= update_activate_spare
;
8792 u
->dl
= (struct dl
*) di
->devs
;
8794 u
->slot
= di
->disk
.raid_disk
;
8805 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
8807 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
8808 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8809 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8810 struct disk_info
*inf
= get_disk_info(u
);
8811 struct imsm_disk
*disk
;
8815 for (i
= 0; i
< map
->num_members
; i
++) {
8816 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
8817 for (j
= 0; j
< new_map
->num_members
; j
++)
8818 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
8825 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
8829 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8830 if (dl
->major
== major
&& dl
->minor
== minor
)
8835 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
8841 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8842 if (dl
->major
== major
&& dl
->minor
== minor
) {
8845 prev
->next
= dl
->next
;
8847 super
->disks
= dl
->next
;
8849 __free_imsm_disk(dl
);
8850 dprintf("removed %x:%x\n", major
, minor
);
8858 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
8860 static int add_remove_disk_update(struct intel_super
*super
)
8862 int check_degraded
= 0;
8865 /* add/remove some spares to/from the metadata/contrainer */
8866 while (super
->disk_mgmt_list
) {
8867 struct dl
*disk_cfg
;
8869 disk_cfg
= super
->disk_mgmt_list
;
8870 super
->disk_mgmt_list
= disk_cfg
->next
;
8871 disk_cfg
->next
= NULL
;
8873 if (disk_cfg
->action
== DISK_ADD
) {
8874 disk_cfg
->next
= super
->disks
;
8875 super
->disks
= disk_cfg
;
8877 dprintf("added %x:%x\n",
8878 disk_cfg
->major
, disk_cfg
->minor
);
8879 } else if (disk_cfg
->action
== DISK_REMOVE
) {
8880 dprintf("Disk remove action processed: %x.%x\n",
8881 disk_cfg
->major
, disk_cfg
->minor
);
8882 disk
= get_disk_super(super
,
8886 /* store action status */
8887 disk
->action
= DISK_REMOVE
;
8888 /* remove spare disks only */
8889 if (disk
->index
== -1) {
8890 remove_disk_super(super
,
8895 /* release allocate disk structure */
8896 __free_imsm_disk(disk_cfg
);
8899 return check_degraded
;
8902 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
8903 struct intel_super
*super
,
8906 struct intel_dev
*id
;
8907 void **tofree
= NULL
;
8910 dprintf("(enter)\n");
8911 if (u
->subdev
< 0 || u
->subdev
> 1) {
8912 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8915 if (space_list
== NULL
|| *space_list
== NULL
) {
8916 dprintf("imsm: Error: Memory is not allocated\n");
8920 for (id
= super
->devlist
; id
; id
= id
->next
) {
8921 if (id
->index
== (unsigned)u
->subdev
) {
8922 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8923 struct imsm_map
*map
;
8924 struct imsm_dev
*new_dev
=
8925 (struct imsm_dev
*)*space_list
;
8926 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8928 struct dl
*new_disk
;
8930 if (new_dev
== NULL
)
8932 *space_list
= **space_list
;
8933 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
8934 map
= get_imsm_map(new_dev
, MAP_0
);
8936 dprintf("imsm: Error: migration in progress");
8940 to_state
= map
->map_state
;
8941 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
8943 /* this should not happen */
8944 if (u
->new_disks
[0] < 0) {
8945 map
->failed_disk_num
=
8946 map
->num_members
- 1;
8947 to_state
= IMSM_T_STATE_DEGRADED
;
8949 to_state
= IMSM_T_STATE_NORMAL
;
8951 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
8952 if (u
->new_level
> -1)
8953 map
->raid_level
= u
->new_level
;
8954 migr_map
= get_imsm_map(new_dev
, MAP_1
);
8955 if ((u
->new_level
== 5) &&
8956 (migr_map
->raid_level
== 0)) {
8957 int ord
= map
->num_members
- 1;
8958 migr_map
->num_members
--;
8959 if (u
->new_disks
[0] < 0)
8960 ord
|= IMSM_ORD_REBUILD
;
8961 set_imsm_ord_tbl_ent(map
,
8962 map
->num_members
- 1,
8966 tofree
= (void **)dev
;
8968 /* update chunk size
8970 if (u
->new_chunksize
> 0) {
8971 unsigned long long num_data_stripes
;
8973 imsm_num_data_members(dev
, MAP_0
);
8975 if (used_disks
== 0)
8978 map
->blocks_per_strip
=
8979 __cpu_to_le16(u
->new_chunksize
* 2);
8981 (join_u32(dev
->size_low
, dev
->size_high
)
8983 num_data_stripes
/= map
->blocks_per_strip
;
8984 num_data_stripes
/= map
->num_domains
;
8985 set_num_data_stripes(map
, num_data_stripes
);
8990 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
8991 migr_map
->raid_level
== map
->raid_level
)
8994 if (u
->new_disks
[0] >= 0) {
8997 new_disk
= get_disk_super(super
,
8998 major(u
->new_disks
[0]),
8999 minor(u
->new_disks
[0]));
9000 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9001 major(u
->new_disks
[0]),
9002 minor(u
->new_disks
[0]),
9003 new_disk
, new_disk
->index
);
9004 if (new_disk
== NULL
)
9005 goto error_disk_add
;
9007 new_disk
->index
= map
->num_members
- 1;
9008 /* slot to fill in autolayout
9010 new_disk
->raiddisk
= new_disk
->index
;
9011 new_disk
->disk
.status
|= CONFIGURED_DISK
;
9012 new_disk
->disk
.status
&= ~SPARE_DISK
;
9014 goto error_disk_add
;
9017 *tofree
= *space_list
;
9018 /* calculate new size
9020 imsm_set_array_size(new_dev
, -1);
9027 *space_list
= tofree
;
9031 dprintf("Error: imsm: Cannot find disk.\n");
9035 static int apply_size_change_update(struct imsm_update_size_change
*u
,
9036 struct intel_super
*super
)
9038 struct intel_dev
*id
;
9041 dprintf("(enter)\n");
9042 if (u
->subdev
< 0 || u
->subdev
> 1) {
9043 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
9047 for (id
= super
->devlist
; id
; id
= id
->next
) {
9048 if (id
->index
== (unsigned)u
->subdev
) {
9049 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
9050 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9051 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
9052 unsigned long long blocks_per_member
;
9053 unsigned long long num_data_stripes
;
9055 /* calculate new size
9057 blocks_per_member
= u
->new_size
/ used_disks
;
9058 num_data_stripes
= blocks_per_member
/
9059 map
->blocks_per_strip
;
9060 num_data_stripes
/= map
->num_domains
;
9061 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9062 u
->new_size
, blocks_per_member
,
9064 set_blocks_per_member(map
, blocks_per_member
);
9065 set_num_data_stripes(map
, num_data_stripes
);
9066 imsm_set_array_size(dev
, u
->new_size
);
9076 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9077 struct intel_super
*super
,
9078 struct active_array
*active_array
)
9080 struct imsm_super
*mpb
= super
->anchor
;
9081 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9082 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9083 struct imsm_map
*migr_map
;
9084 struct active_array
*a
;
9085 struct imsm_disk
*disk
;
9092 int second_map_created
= 0;
9094 for (; u
; u
= u
->next
) {
9095 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9100 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9105 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9110 /* count failures (excluding rebuilds and the victim)
9111 * to determine map[0] state
9114 for (i
= 0; i
< map
->num_members
; i
++) {
9117 disk
= get_imsm_disk(super
,
9118 get_imsm_disk_idx(dev
, i
, MAP_X
));
9119 if (!disk
|| is_failed(disk
))
9123 /* adding a pristine spare, assign a new index */
9124 if (dl
->index
< 0) {
9125 dl
->index
= super
->anchor
->num_disks
;
9126 super
->anchor
->num_disks
++;
9129 disk
->status
|= CONFIGURED_DISK
;
9130 disk
->status
&= ~SPARE_DISK
;
9133 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9134 if (!second_map_created
) {
9135 second_map_created
= 1;
9136 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9137 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9139 map
->map_state
= to_state
;
9140 migr_map
= get_imsm_map(dev
, MAP_1
);
9141 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9142 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9143 dl
->index
| IMSM_ORD_REBUILD
);
9145 /* update the family_num to mark a new container
9146 * generation, being careful to record the existing
9147 * family_num in orig_family_num to clean up after
9148 * earlier mdadm versions that neglected to set it.
9150 if (mpb
->orig_family_num
== 0)
9151 mpb
->orig_family_num
= mpb
->family_num
;
9152 mpb
->family_num
+= super
->random
;
9154 /* count arrays using the victim in the metadata */
9156 for (a
= active_array
; a
; a
= a
->next
) {
9157 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9158 map
= get_imsm_map(dev
, MAP_0
);
9160 if (get_imsm_disk_slot(map
, victim
) >= 0)
9164 /* delete the victim if it is no longer being
9170 /* We know that 'manager' isn't touching anything,
9171 * so it is safe to delete
9173 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9174 if ((*dlp
)->index
== victim
)
9177 /* victim may be on the missing list */
9179 for (dlp
= &super
->missing
; *dlp
;
9180 dlp
= &(*dlp
)->next
)
9181 if ((*dlp
)->index
== victim
)
9183 imsm_delete(super
, dlp
, victim
);
9190 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9191 struct intel_super
*super
,
9194 struct dl
*new_disk
;
9195 struct intel_dev
*id
;
9197 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9198 int disk_count
= u
->old_raid_disks
;
9199 void **tofree
= NULL
;
9200 int devices_to_reshape
= 1;
9201 struct imsm_super
*mpb
= super
->anchor
;
9203 unsigned int dev_id
;
9205 dprintf("(enter)\n");
9207 /* enable spares to use in array */
9208 for (i
= 0; i
< delta_disks
; i
++) {
9209 new_disk
= get_disk_super(super
,
9210 major(u
->new_disks
[i
]),
9211 minor(u
->new_disks
[i
]));
9212 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9213 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9214 new_disk
, new_disk
->index
);
9215 if (new_disk
== NULL
||
9216 (new_disk
->index
>= 0 &&
9217 new_disk
->index
< u
->old_raid_disks
))
9218 goto update_reshape_exit
;
9219 new_disk
->index
= disk_count
++;
9220 /* slot to fill in autolayout
9222 new_disk
->raiddisk
= new_disk
->index
;
9223 new_disk
->disk
.status
|=
9225 new_disk
->disk
.status
&= ~SPARE_DISK
;
9228 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9229 mpb
->num_raid_devs
);
9230 /* manage changes in volume
9232 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9233 void **sp
= *space_list
;
9234 struct imsm_dev
*newdev
;
9235 struct imsm_map
*newmap
, *oldmap
;
9237 for (id
= super
->devlist
; id
; id
= id
->next
) {
9238 if (id
->index
== dev_id
)
9247 /* Copy the dev, but not (all of) the map */
9248 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9249 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9250 newmap
= get_imsm_map(newdev
, MAP_0
);
9251 /* Copy the current map */
9252 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9253 /* update one device only
9255 if (devices_to_reshape
) {
9256 dprintf("imsm: modifying subdev: %i\n",
9258 devices_to_reshape
--;
9259 newdev
->vol
.migr_state
= 1;
9260 newdev
->vol
.curr_migr_unit
= 0;
9261 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9262 newmap
->num_members
= u
->new_raid_disks
;
9263 for (i
= 0; i
< delta_disks
; i
++) {
9264 set_imsm_ord_tbl_ent(newmap
,
9265 u
->old_raid_disks
+ i
,
9266 u
->old_raid_disks
+ i
);
9268 /* New map is correct, now need to save old map
9270 newmap
= get_imsm_map(newdev
, MAP_1
);
9271 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9273 imsm_set_array_size(newdev
, -1);
9276 sp
= (void **)id
->dev
;
9281 /* Clear migration record */
9282 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9285 *space_list
= tofree
;
9288 update_reshape_exit
:
9293 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9294 struct intel_super
*super
,
9297 struct imsm_dev
*dev
= NULL
;
9298 struct intel_dev
*dv
;
9299 struct imsm_dev
*dev_new
;
9300 struct imsm_map
*map
;
9304 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9305 if (dv
->index
== (unsigned int)u
->subarray
) {
9313 map
= get_imsm_map(dev
, MAP_0
);
9315 if (u
->direction
== R10_TO_R0
) {
9316 unsigned long long num_data_stripes
;
9318 map
->num_domains
= 1;
9319 num_data_stripes
= blocks_per_member(map
);
9320 num_data_stripes
/= map
->blocks_per_strip
;
9321 num_data_stripes
/= map
->num_domains
;
9322 set_num_data_stripes(map
, num_data_stripes
);
9324 /* Number of failed disks must be half of initial disk number */
9325 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9326 (map
->num_members
/ 2))
9329 /* iterate through devices to mark removed disks as spare */
9330 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9331 if (dm
->disk
.status
& FAILED_DISK
) {
9332 int idx
= dm
->index
;
9333 /* update indexes on the disk list */
9334 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9335 the index values will end up being correct.... NB */
9336 for (du
= super
->disks
; du
; du
= du
->next
)
9337 if (du
->index
> idx
)
9339 /* mark as spare disk */
9344 map
->num_members
= map
->num_members
/ 2;
9345 map
->map_state
= IMSM_T_STATE_NORMAL
;
9346 map
->num_domains
= 1;
9347 map
->raid_level
= 0;
9348 map
->failed_disk_num
= -1;
9351 if (u
->direction
== R0_TO_R10
) {
9353 /* update slots in current disk list */
9354 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9358 /* create new *missing* disks */
9359 for (i
= 0; i
< map
->num_members
; i
++) {
9360 space
= *space_list
;
9363 *space_list
= *space
;
9365 memcpy(du
, super
->disks
, sizeof(*du
));
9369 du
->index
= (i
* 2) + 1;
9370 sprintf((char *)du
->disk
.serial
,
9371 " MISSING_%d", du
->index
);
9372 sprintf((char *)du
->serial
,
9373 "MISSING_%d", du
->index
);
9374 du
->next
= super
->missing
;
9375 super
->missing
= du
;
9377 /* create new dev and map */
9378 space
= *space_list
;
9381 *space_list
= *space
;
9382 dev_new
= (void *)space
;
9383 memcpy(dev_new
, dev
, sizeof(*dev
));
9384 /* update new map */
9385 map
= get_imsm_map(dev_new
, MAP_0
);
9386 map
->num_members
= map
->num_members
* 2;
9387 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9388 map
->num_domains
= 2;
9389 map
->raid_level
= 1;
9390 /* replace dev<->dev_new */
9393 /* update disk order table */
9394 for (du
= super
->disks
; du
; du
= du
->next
)
9396 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9397 for (du
= super
->missing
; du
; du
= du
->next
)
9398 if (du
->index
>= 0) {
9399 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9400 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9406 static void imsm_process_update(struct supertype
*st
,
9407 struct metadata_update
*update
)
9410 * crack open the metadata_update envelope to find the update record
9411 * update can be one of:
9412 * update_reshape_container_disks - all the arrays in the container
9413 * are being reshaped to have more devices. We need to mark
9414 * the arrays for general migration and convert selected spares
9415 * into active devices.
9416 * update_activate_spare - a spare device has replaced a failed
9417 * device in an array, update the disk_ord_tbl. If this disk is
9418 * present in all member arrays then also clear the SPARE_DISK
9420 * update_create_array
9422 * update_rename_array
9423 * update_add_remove_disk
9425 struct intel_super
*super
= st
->sb
;
9426 struct imsm_super
*mpb
;
9427 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9429 /* update requires a larger buf but the allocation failed */
9430 if (super
->next_len
&& !super
->next_buf
) {
9431 super
->next_len
= 0;
9435 if (super
->next_buf
) {
9436 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9438 super
->len
= super
->next_len
;
9439 super
->buf
= super
->next_buf
;
9441 super
->next_len
= 0;
9442 super
->next_buf
= NULL
;
9445 mpb
= super
->anchor
;
9448 case update_general_migration_checkpoint
: {
9449 struct intel_dev
*id
;
9450 struct imsm_update_general_migration_checkpoint
*u
=
9451 (void *)update
->buf
;
9453 dprintf("called for update_general_migration_checkpoint\n");
9455 /* find device under general migration */
9456 for (id
= super
->devlist
; id
; id
= id
->next
) {
9457 if (is_gen_migration(id
->dev
)) {
9458 id
->dev
->vol
.curr_migr_unit
=
9459 __cpu_to_le32(u
->curr_migr_unit
);
9460 super
->updates_pending
++;
9465 case update_takeover
: {
9466 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9467 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9468 imsm_update_version_info(super
);
9469 super
->updates_pending
++;
9474 case update_reshape_container_disks
: {
9475 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9476 if (apply_reshape_container_disks_update(
9477 u
, super
, &update
->space_list
))
9478 super
->updates_pending
++;
9481 case update_reshape_migration
: {
9482 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9483 if (apply_reshape_migration_update(
9484 u
, super
, &update
->space_list
))
9485 super
->updates_pending
++;
9488 case update_size_change
: {
9489 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9490 if (apply_size_change_update(u
, super
))
9491 super
->updates_pending
++;
9494 case update_activate_spare
: {
9495 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9496 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9497 super
->updates_pending
++;
9500 case update_create_array
: {
9501 /* someone wants to create a new array, we need to be aware of
9502 * a few races/collisions:
9503 * 1/ 'Create' called by two separate instances of mdadm
9504 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9505 * devices that have since been assimilated via
9507 * In the event this update can not be carried out mdadm will
9508 * (FIX ME) notice that its update did not take hold.
9510 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9511 struct intel_dev
*dv
;
9512 struct imsm_dev
*dev
;
9513 struct imsm_map
*map
, *new_map
;
9514 unsigned long long start
, end
;
9515 unsigned long long new_start
, new_end
;
9517 struct disk_info
*inf
;
9520 /* handle racing creates: first come first serve */
9521 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9522 dprintf("subarray %d already defined\n", u
->dev_idx
);
9526 /* check update is next in sequence */
9527 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9528 dprintf("can not create array %d expected index %d\n",
9529 u
->dev_idx
, mpb
->num_raid_devs
);
9533 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9534 new_start
= pba_of_lba0(new_map
);
9535 new_end
= new_start
+ blocks_per_member(new_map
);
9536 inf
= get_disk_info(u
);
9538 /* handle activate_spare versus create race:
9539 * check to make sure that overlapping arrays do not include
9542 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9543 dev
= get_imsm_dev(super
, i
);
9544 map
= get_imsm_map(dev
, MAP_0
);
9545 start
= pba_of_lba0(map
);
9546 end
= start
+ blocks_per_member(map
);
9547 if ((new_start
>= start
&& new_start
<= end
) ||
9548 (start
>= new_start
&& start
<= new_end
))
9553 if (disks_overlap(super
, i
, u
)) {
9554 dprintf("arrays overlap\n");
9559 /* check that prepare update was successful */
9560 if (!update
->space
) {
9561 dprintf("prepare update failed\n");
9565 /* check that all disks are still active before committing
9566 * changes. FIXME: could we instead handle this by creating a
9567 * degraded array? That's probably not what the user expects,
9568 * so better to drop this update on the floor.
9570 for (i
= 0; i
< new_map
->num_members
; i
++) {
9571 dl
= serial_to_dl(inf
[i
].serial
, super
);
9573 dprintf("disk disappeared\n");
9578 super
->updates_pending
++;
9580 /* convert spares to members and fixup ord_tbl */
9581 for (i
= 0; i
< new_map
->num_members
; i
++) {
9582 dl
= serial_to_dl(inf
[i
].serial
, super
);
9583 if (dl
->index
== -1) {
9584 dl
->index
= mpb
->num_disks
;
9586 dl
->disk
.status
|= CONFIGURED_DISK
;
9587 dl
->disk
.status
&= ~SPARE_DISK
;
9589 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9594 update
->space
= NULL
;
9595 imsm_copy_dev(dev
, &u
->dev
);
9596 dv
->index
= u
->dev_idx
;
9597 dv
->next
= super
->devlist
;
9598 super
->devlist
= dv
;
9599 mpb
->num_raid_devs
++;
9601 imsm_update_version_info(super
);
9604 /* mdmon knows how to release update->space, but not
9605 * ((struct intel_dev *) update->space)->dev
9607 if (update
->space
) {
9613 case update_kill_array
: {
9614 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9615 int victim
= u
->dev_idx
;
9616 struct active_array
*a
;
9617 struct intel_dev
**dp
;
9618 struct imsm_dev
*dev
;
9620 /* sanity check that we are not affecting the uuid of
9621 * active arrays, or deleting an active array
9623 * FIXME when immutable ids are available, but note that
9624 * we'll also need to fixup the invalidated/active
9625 * subarray indexes in mdstat
9627 for (a
= st
->arrays
; a
; a
= a
->next
)
9628 if (a
->info
.container_member
>= victim
)
9630 /* by definition if mdmon is running at least one array
9631 * is active in the container, so checking
9632 * mpb->num_raid_devs is just extra paranoia
9634 dev
= get_imsm_dev(super
, victim
);
9635 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9636 dprintf("failed to delete subarray-%d\n", victim
);
9640 for (dp
= &super
->devlist
; *dp
;)
9641 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9644 if ((*dp
)->index
> (unsigned)victim
)
9648 mpb
->num_raid_devs
--;
9649 super
->updates_pending
++;
9652 case update_rename_array
: {
9653 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9654 char name
[MAX_RAID_SERIAL_LEN
+1];
9655 int target
= u
->dev_idx
;
9656 struct active_array
*a
;
9657 struct imsm_dev
*dev
;
9659 /* sanity check that we are not affecting the uuid of
9662 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9663 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9664 for (a
= st
->arrays
; a
; a
= a
->next
)
9665 if (a
->info
.container_member
== target
)
9667 dev
= get_imsm_dev(super
, u
->dev_idx
);
9668 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9669 dprintf("failed to rename subarray-%d\n", target
);
9673 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
9674 super
->updates_pending
++;
9677 case update_add_remove_disk
: {
9678 /* we may be able to repair some arrays if disks are
9679 * being added, check the status of add_remove_disk
9680 * if discs has been added.
9682 if (add_remove_disk_update(super
)) {
9683 struct active_array
*a
;
9685 super
->updates_pending
++;
9686 for (a
= st
->arrays
; a
; a
= a
->next
)
9687 a
->check_degraded
= 1;
9691 case update_prealloc_badblocks_mem
:
9693 case update_rwh_policy
: {
9694 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
9695 int target
= u
->dev_idx
;
9696 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
9698 dprintf("could not find subarray-%d\n", target
);
9702 if (dev
->rwh_policy
!= u
->new_policy
) {
9703 dev
->rwh_policy
= u
->new_policy
;
9704 super
->updates_pending
++;
9709 pr_err("error: unsuported process update type:(type: %d)\n", type
);
9713 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
9715 static int imsm_prepare_update(struct supertype
*st
,
9716 struct metadata_update
*update
)
9719 * Allocate space to hold new disk entries, raid-device entries or a new
9720 * mpb if necessary. The manager synchronously waits for updates to
9721 * complete in the monitor, so new mpb buffers allocated here can be
9722 * integrated by the monitor thread without worrying about live pointers
9723 * in the manager thread.
9725 enum imsm_update_type type
;
9726 struct intel_super
*super
= st
->sb
;
9727 unsigned int sector_size
= super
->sector_size
;
9728 struct imsm_super
*mpb
= super
->anchor
;
9732 if (update
->len
< (int)sizeof(type
))
9735 type
= *(enum imsm_update_type
*) update
->buf
;
9738 case update_general_migration_checkpoint
:
9739 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
9741 dprintf("called for update_general_migration_checkpoint\n");
9743 case update_takeover
: {
9744 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9745 if (update
->len
< (int)sizeof(*u
))
9747 if (u
->direction
== R0_TO_R10
) {
9748 void **tail
= (void **)&update
->space_list
;
9749 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
9750 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9751 int num_members
= map
->num_members
;
9754 /* allocate memory for added disks */
9755 for (i
= 0; i
< num_members
; i
++) {
9756 size
= sizeof(struct dl
);
9757 space
= xmalloc(size
);
9762 /* allocate memory for new device */
9763 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
9764 (num_members
* sizeof(__u32
));
9765 space
= xmalloc(size
);
9769 len
= disks_to_mpb_size(num_members
* 2);
9774 case update_reshape_container_disks
: {
9775 /* Every raid device in the container is about to
9776 * gain some more devices, and we will enter a
9778 * So each 'imsm_map' will be bigger, and the imsm_vol
9779 * will now hold 2 of them.
9780 * Thus we need new 'struct imsm_dev' allocations sized
9781 * as sizeof_imsm_dev but with more devices in both maps.
9783 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9784 struct intel_dev
*dl
;
9785 void **space_tail
= (void**)&update
->space_list
;
9787 if (update
->len
< (int)sizeof(*u
))
9790 dprintf("for update_reshape\n");
9792 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
9793 int size
= sizeof_imsm_dev(dl
->dev
, 1);
9795 if (u
->new_raid_disks
> u
->old_raid_disks
)
9796 size
+= sizeof(__u32
)*2*
9797 (u
->new_raid_disks
- u
->old_raid_disks
);
9804 len
= disks_to_mpb_size(u
->new_raid_disks
);
9805 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9808 case update_reshape_migration
: {
9809 /* for migration level 0->5 we need to add disks
9810 * so the same as for container operation we will copy
9811 * device to the bigger location.
9812 * in memory prepared device and new disk area are prepared
9813 * for usage in process update
9815 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9816 struct intel_dev
*id
;
9817 void **space_tail
= (void **)&update
->space_list
;
9820 int current_level
= -1;
9822 if (update
->len
< (int)sizeof(*u
))
9825 dprintf("for update_reshape\n");
9827 /* add space for bigger array in update
9829 for (id
= super
->devlist
; id
; id
= id
->next
) {
9830 if (id
->index
== (unsigned)u
->subdev
) {
9831 size
= sizeof_imsm_dev(id
->dev
, 1);
9832 if (u
->new_raid_disks
> u
->old_raid_disks
)
9833 size
+= sizeof(__u32
)*2*
9834 (u
->new_raid_disks
- u
->old_raid_disks
);
9842 if (update
->space_list
== NULL
)
9845 /* add space for disk in update
9847 size
= sizeof(struct dl
);
9853 /* add spare device to update
9855 for (id
= super
->devlist
; id
; id
= id
->next
)
9856 if (id
->index
== (unsigned)u
->subdev
) {
9857 struct imsm_dev
*dev
;
9858 struct imsm_map
*map
;
9860 dev
= get_imsm_dev(super
, u
->subdev
);
9861 map
= get_imsm_map(dev
, MAP_0
);
9862 current_level
= map
->raid_level
;
9865 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
9866 struct mdinfo
*spares
;
9868 spares
= get_spares_for_grow(st
);
9876 makedev(dev
->disk
.major
,
9878 dl
= get_disk_super(super
,
9881 dl
->index
= u
->old_raid_disks
;
9887 len
= disks_to_mpb_size(u
->new_raid_disks
);
9888 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9891 case update_size_change
: {
9892 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
9896 case update_activate_spare
: {
9897 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
9901 case update_create_array
: {
9902 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9903 struct intel_dev
*dv
;
9904 struct imsm_dev
*dev
= &u
->dev
;
9905 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9907 struct disk_info
*inf
;
9911 if (update
->len
< (int)sizeof(*u
))
9914 inf
= get_disk_info(u
);
9915 len
= sizeof_imsm_dev(dev
, 1);
9916 /* allocate a new super->devlist entry */
9917 dv
= xmalloc(sizeof(*dv
));
9918 dv
->dev
= xmalloc(len
);
9921 /* count how many spares will be converted to members */
9922 for (i
= 0; i
< map
->num_members
; i
++) {
9923 dl
= serial_to_dl(inf
[i
].serial
, super
);
9925 /* hmm maybe it failed?, nothing we can do about
9930 if (count_memberships(dl
, super
) == 0)
9933 len
+= activate
* sizeof(struct imsm_disk
);
9936 case update_kill_array
: {
9937 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
9941 case update_rename_array
: {
9942 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
9946 case update_add_remove_disk
:
9947 /* no update->len needed */
9949 case update_prealloc_badblocks_mem
:
9950 super
->extra_space
+= sizeof(struct bbm_log
) -
9951 get_imsm_bbm_log_size(super
->bbm_log
);
9953 case update_rwh_policy
: {
9954 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
9962 /* check if we need a larger metadata buffer */
9963 if (super
->next_buf
)
9964 buf_len
= super
->next_len
;
9966 buf_len
= super
->len
;
9968 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
9969 /* ok we need a larger buf than what is currently allocated
9970 * if this allocation fails process_update will notice that
9971 * ->next_len is set and ->next_buf is NULL
9973 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
9974 super
->extra_space
+ len
, sector_size
);
9975 if (super
->next_buf
)
9976 free(super
->next_buf
);
9978 super
->next_len
= buf_len
;
9979 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
9980 memset(super
->next_buf
, 0, buf_len
);
9982 super
->next_buf
= NULL
;
9987 /* must be called while manager is quiesced */
9988 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
9990 struct imsm_super
*mpb
= super
->anchor
;
9992 struct imsm_dev
*dev
;
9993 struct imsm_map
*map
;
9994 unsigned int i
, j
, num_members
;
9995 __u32 ord
, ord_map0
;
9996 struct bbm_log
*log
= super
->bbm_log
;
9998 dprintf("deleting device[%d] from imsm_super\n", index
);
10000 /* shift all indexes down one */
10001 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
10002 if (iter
->index
> (int)index
)
10004 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
10005 if (iter
->index
> (int)index
)
10008 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
10009 dev
= get_imsm_dev(super
, i
);
10010 map
= get_imsm_map(dev
, MAP_0
);
10011 num_members
= map
->num_members
;
10012 for (j
= 0; j
< num_members
; j
++) {
10013 /* update ord entries being careful not to propagate
10014 * ord-flags to the first map
10016 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
10017 ord_map0
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
10019 if (ord_to_idx(ord
) <= index
)
10022 map
= get_imsm_map(dev
, MAP_0
);
10023 set_imsm_ord_tbl_ent(map
, j
, ord_map0
- 1);
10024 map
= get_imsm_map(dev
, MAP_1
);
10026 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
10030 for (i
= 0; i
< log
->entry_count
; i
++) {
10031 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
10033 if (entry
->disk_ordinal
<= index
)
10035 entry
->disk_ordinal
--;
10039 super
->updates_pending
++;
10041 struct dl
*dl
= *dlp
;
10043 *dlp
= (*dlp
)->next
;
10044 __free_imsm_disk(dl
);
10048 static void close_targets(int *targets
, int new_disks
)
10055 for (i
= 0; i
< new_disks
; i
++) {
10056 if (targets
[i
] >= 0) {
10063 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10064 struct intel_super
*super
,
10065 struct imsm_dev
*dev
)
10071 struct imsm_map
*map
;
10074 ret_val
= raid_disks
/2;
10075 /* check map if all disks pairs not failed
10078 map
= get_imsm_map(dev
, MAP_0
);
10079 for (i
= 0; i
< ret_val
; i
++) {
10080 int degradation
= 0;
10081 if (get_imsm_disk(super
, i
) == NULL
)
10083 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10085 if (degradation
== 2)
10088 map
= get_imsm_map(dev
, MAP_1
);
10089 /* if there is no second map
10090 * result can be returned
10094 /* check degradation in second map
10096 for (i
= 0; i
< ret_val
; i
++) {
10097 int degradation
= 0;
10098 if (get_imsm_disk(super
, i
) == NULL
)
10100 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10102 if (degradation
== 2)
10116 /*******************************************************************************
10117 * Function: open_backup_targets
10118 * Description: Function opens file descriptors for all devices given in
10121 * info : general array info
10122 * raid_disks : number of disks
10123 * raid_fds : table of device's file descriptors
10124 * super : intel super for raid10 degradation check
10125 * dev : intel device for raid10 degradation check
10129 ******************************************************************************/
10130 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10131 struct intel_super
*super
, struct imsm_dev
*dev
)
10137 for (i
= 0; i
< raid_disks
; i
++)
10140 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10143 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10144 dprintf("disk is faulty!!\n");
10148 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10151 dn
= map_dev(sd
->disk
.major
,
10152 sd
->disk
.minor
, 1);
10153 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10154 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10155 pr_err("cannot open component\n");
10160 /* check if maximum array degradation level is not exceeded
10162 if ((raid_disks
- opened
) >
10163 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10165 pr_err("Not enough disks can be opened.\n");
10166 close_targets(raid_fds
, raid_disks
);
10172 /*******************************************************************************
10173 * Function: validate_container_imsm
10174 * Description: This routine validates container after assemble,
10175 * eg. if devices in container are under the same controller.
10178 * info : linked list with info about devices used in array
10182 ******************************************************************************/
10183 int validate_container_imsm(struct mdinfo
*info
)
10185 if (check_env("IMSM_NO_PLATFORM"))
10188 struct sys_dev
*idev
;
10189 struct sys_dev
*hba
= NULL
;
10190 struct sys_dev
*intel_devices
= find_intel_devices();
10191 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10192 info
->disk
.minor
));
10194 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10195 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10204 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10205 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10209 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10210 struct mdinfo
*dev
;
10212 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10213 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10215 struct sys_dev
*hba2
= NULL
;
10216 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10217 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10225 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10226 get_orom_by_device_id(hba2
->dev_id
);
10228 if (hba2
&& hba
->type
!= hba2
->type
) {
10229 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10230 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10234 if (orom
!= orom2
) {
10235 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10236 " This operation is not supported and can lead to data loss.\n");
10241 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10242 " This operation is not supported and can lead to data loss.\n");
10250 /*******************************************************************************
10251 * Function: imsm_record_badblock
10252 * Description: This routine stores new bad block record in BBM log
10255 * a : array containing a bad block
10256 * slot : disk number containing a bad block
10257 * sector : bad block sector
10258 * length : bad block sectors range
10262 ******************************************************************************/
10263 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10264 unsigned long long sector
, int length
)
10266 struct intel_super
*super
= a
->container
->sb
;
10270 ord
= imsm_disk_slot_to_ord(a
, slot
);
10274 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10277 super
->updates_pending
++;
10281 /*******************************************************************************
10282 * Function: imsm_clear_badblock
10283 * Description: This routine clears bad block record from BBM log
10286 * a : array containing a bad block
10287 * slot : disk number containing a bad block
10288 * sector : bad block sector
10289 * length : bad block sectors range
10293 ******************************************************************************/
10294 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10295 unsigned long long sector
, int length
)
10297 struct intel_super
*super
= a
->container
->sb
;
10301 ord
= imsm_disk_slot_to_ord(a
, slot
);
10305 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10307 super
->updates_pending
++;
10311 /*******************************************************************************
10312 * Function: imsm_get_badblocks
10313 * Description: This routine get list of bad blocks for an array
10317 * slot : disk number
10319 * bb : structure containing bad blocks
10321 ******************************************************************************/
10322 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10324 int inst
= a
->info
.container_member
;
10325 struct intel_super
*super
= a
->container
->sb
;
10326 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10327 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10330 ord
= imsm_disk_slot_to_ord(a
, slot
);
10334 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10335 blocks_per_member(map
), &super
->bb
);
10339 /*******************************************************************************
10340 * Function: examine_badblocks_imsm
10341 * Description: Prints list of bad blocks on a disk to the standard output
10344 * st : metadata handler
10345 * fd : open file descriptor for device
10346 * devname : device name
10350 ******************************************************************************/
10351 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10353 struct intel_super
*super
= st
->sb
;
10354 struct bbm_log
*log
= super
->bbm_log
;
10355 struct dl
*d
= NULL
;
10358 for (d
= super
->disks
; d
; d
= d
->next
) {
10359 if (strcmp(d
->devname
, devname
) == 0)
10363 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10364 pr_err("%s doesn't appear to be part of a raid array\n",
10371 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10373 for (i
= 0; i
< log
->entry_count
; i
++) {
10374 if (entry
[i
].disk_ordinal
== d
->index
) {
10375 unsigned long long sector
= __le48_to_cpu(
10376 &entry
[i
].defective_block_start
);
10377 int cnt
= entry
[i
].marked_count
+ 1;
10380 printf("Bad-blocks on %s:\n", devname
);
10384 printf("%20llu for %d sectors\n", sector
, cnt
);
10390 printf("No bad-blocks list configured on %s\n", devname
);
10394 /*******************************************************************************
10395 * Function: init_migr_record_imsm
10396 * Description: Function inits imsm migration record
10398 * super : imsm internal array info
10399 * dev : device under migration
10400 * info : general array info to find the smallest device
10403 ******************************************************************************/
10404 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10405 struct mdinfo
*info
)
10407 struct intel_super
*super
= st
->sb
;
10408 struct migr_record
*migr_rec
= super
->migr_rec
;
10409 int new_data_disks
;
10410 unsigned long long dsize
, dev_sectors
;
10411 long long unsigned min_dev_sectors
= -1LLU;
10415 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10416 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10417 unsigned long long num_migr_units
;
10418 unsigned long long array_blocks
;
10420 memset(migr_rec
, 0, sizeof(struct migr_record
));
10421 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10423 /* only ascending reshape supported now */
10424 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10426 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10427 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10428 migr_rec
->dest_depth_per_unit
*=
10429 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10430 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
10431 migr_rec
->blocks_per_unit
=
10432 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10433 migr_rec
->dest_depth_per_unit
=
10434 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10435 array_blocks
= info
->component_size
* new_data_disks
;
10437 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10439 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10441 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
10443 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10444 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10446 /* Find the smallest dev */
10447 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10448 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10449 fd
= dev_open(nm
, O_RDONLY
);
10452 get_dev_size(fd
, NULL
, &dsize
);
10453 dev_sectors
= dsize
/ 512;
10454 if (dev_sectors
< min_dev_sectors
)
10455 min_dev_sectors
= dev_sectors
;
10458 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
10459 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10461 write_imsm_migr_rec(st
);
10466 /*******************************************************************************
10467 * Function: save_backup_imsm
10468 * Description: Function saves critical data stripes to Migration Copy Area
10469 * and updates the current migration unit status.
10470 * Use restore_stripes() to form a destination stripe,
10471 * and to write it to the Copy Area.
10473 * st : supertype information
10474 * dev : imsm device that backup is saved for
10475 * info : general array info
10476 * buf : input buffer
10477 * length : length of data to backup (blocks_per_unit)
10481 ******************************************************************************/
10482 int save_backup_imsm(struct supertype
*st
,
10483 struct imsm_dev
*dev
,
10484 struct mdinfo
*info
,
10489 struct intel_super
*super
= st
->sb
;
10490 unsigned long long *target_offsets
;
10493 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10494 int new_disks
= map_dest
->num_members
;
10495 int dest_layout
= 0;
10497 unsigned long long start
;
10498 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
10500 targets
= xmalloc(new_disks
* sizeof(int));
10502 for (i
= 0; i
< new_disks
; i
++)
10505 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10507 start
= info
->reshape_progress
* 512;
10508 for (i
= 0; i
< new_disks
; i
++) {
10509 target_offsets
[i
] = (unsigned long long)
10510 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
10511 /* move back copy area adderss, it will be moved forward
10512 * in restore_stripes() using start input variable
10514 target_offsets
[i
] -= start
/data_disks
;
10517 if (open_backup_targets(info
, new_disks
, targets
,
10521 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10522 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10524 if (restore_stripes(targets
, /* list of dest devices */
10525 target_offsets
, /* migration record offsets */
10528 map_dest
->raid_level
,
10530 -1, /* source backup file descriptor */
10531 0, /* input buf offset
10532 * always 0 buf is already offseted */
10536 pr_err("Error restoring stripes\n");
10544 close_targets(targets
, new_disks
);
10547 free(target_offsets
);
10552 /*******************************************************************************
10553 * Function: save_checkpoint_imsm
10554 * Description: Function called for current unit status update
10555 * in the migration record. It writes it to disk.
10557 * super : imsm internal array info
10558 * info : general array info
10562 * 2: failure, means no valid migration record
10563 * / no general migration in progress /
10564 ******************************************************************************/
10565 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10567 struct intel_super
*super
= st
->sb
;
10568 unsigned long long blocks_per_unit
;
10569 unsigned long long curr_migr_unit
;
10571 if (load_imsm_migr_rec(super
, info
) != 0) {
10572 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10576 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10577 if (blocks_per_unit
== 0) {
10578 dprintf("imsm: no migration in progress.\n");
10581 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10582 /* check if array is alligned to copy area
10583 * if it is not alligned, add one to current migration unit value
10584 * this can happend on array reshape finish only
10586 if (info
->reshape_progress
% blocks_per_unit
)
10589 super
->migr_rec
->curr_migr_unit
=
10590 __cpu_to_le32(curr_migr_unit
);
10591 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10592 super
->migr_rec
->dest_1st_member_lba
=
10593 __cpu_to_le32(curr_migr_unit
*
10594 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
10595 if (write_imsm_migr_rec(st
) < 0) {
10596 dprintf("imsm: Cannot write migration record outside backup area\n");
10603 /*******************************************************************************
10604 * Function: recover_backup_imsm
10605 * Description: Function recovers critical data from the Migration Copy Area
10606 * while assembling an array.
10608 * super : imsm internal array info
10609 * info : general array info
10611 * 0 : success (or there is no data to recover)
10613 ******************************************************************************/
10614 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10616 struct intel_super
*super
= st
->sb
;
10617 struct migr_record
*migr_rec
= super
->migr_rec
;
10618 struct imsm_map
*map_dest
;
10619 struct intel_dev
*id
= NULL
;
10620 unsigned long long read_offset
;
10621 unsigned long long write_offset
;
10623 int *targets
= NULL
;
10624 int new_disks
, i
, err
;
10627 unsigned int sector_size
= super
->sector_size
;
10628 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
10629 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
10631 int skipped_disks
= 0;
10633 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10637 /* recover data only during assemblation */
10638 if (strncmp(buffer
, "inactive", 8) != 0)
10640 /* no data to recover */
10641 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10643 if (curr_migr_unit
>= num_migr_units
)
10646 /* find device during reshape */
10647 for (id
= super
->devlist
; id
; id
= id
->next
)
10648 if (is_gen_migration(id
->dev
))
10653 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10654 new_disks
= map_dest
->num_members
;
10656 read_offset
= (unsigned long long)
10657 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
10659 write_offset
= ((unsigned long long)
10660 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
10661 pba_of_lba0(map_dest
)) * 512;
10663 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10664 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10666 targets
= xcalloc(new_disks
, sizeof(int));
10668 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10669 pr_err("Cannot open some devices belonging to array.\n");
10673 for (i
= 0; i
< new_disks
; i
++) {
10674 if (targets
[i
] < 0) {
10678 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
10679 pr_err("Cannot seek to block: %s\n",
10684 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
10685 pr_err("Cannot read copy area block: %s\n",
10690 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
10691 pr_err("Cannot seek to block: %s\n",
10696 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
10697 pr_err("Cannot restore block: %s\n",
10704 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
10708 pr_err("Cannot restore data from backup. Too many failed disks\n");
10712 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
10713 /* ignore error == 2, this can mean end of reshape here
10715 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
10721 for (i
= 0; i
< new_disks
; i
++)
10730 static char disk_by_path
[] = "/dev/disk/by-path/";
10732 static const char *imsm_get_disk_controller_domain(const char *path
)
10734 char disk_path
[PATH_MAX
];
10738 strcpy(disk_path
, disk_by_path
);
10739 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
10740 if (stat(disk_path
, &st
) == 0) {
10741 struct sys_dev
* hba
;
10744 path
= devt_to_devpath(st
.st_rdev
);
10747 hba
= find_disk_attached_hba(-1, path
);
10748 if (hba
&& hba
->type
== SYS_DEV_SAS
)
10750 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
10752 else if (hba
&& hba
->type
== SYS_DEV_VMD
)
10754 else if (hba
&& hba
->type
== SYS_DEV_NVME
)
10758 dprintf("path: %s hba: %s attached: %s\n",
10759 path
, (hba
) ? hba
->path
: "NULL", drv
);
10765 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
10767 static char devnm
[32];
10768 char subdev_name
[20];
10769 struct mdstat_ent
*mdstat
;
10771 sprintf(subdev_name
, "%d", subdev
);
10772 mdstat
= mdstat_by_subdev(subdev_name
, container
);
10776 strcpy(devnm
, mdstat
->devnm
);
10777 free_mdstat(mdstat
);
10781 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
10782 struct geo_params
*geo
,
10783 int *old_raid_disks
,
10786 /* currently we only support increasing the number of devices
10787 * for a container. This increases the number of device for each
10788 * member array. They must all be RAID0 or RAID5.
10791 struct mdinfo
*info
, *member
;
10792 int devices_that_can_grow
= 0;
10794 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
10796 if (geo
->size
> 0 ||
10797 geo
->level
!= UnSet
||
10798 geo
->layout
!= UnSet
||
10799 geo
->chunksize
!= 0 ||
10800 geo
->raid_disks
== UnSet
) {
10801 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
10805 if (direction
== ROLLBACK_METADATA_CHANGES
) {
10806 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
10810 info
= container_content_imsm(st
, NULL
);
10811 for (member
= info
; member
; member
= member
->next
) {
10814 dprintf("imsm: checking device_num: %i\n",
10815 member
->container_member
);
10817 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
10818 /* we work on container for Online Capacity Expansion
10819 * only so raid_disks has to grow
10821 dprintf("imsm: for container operation raid disks increase is required\n");
10825 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
10826 /* we cannot use this container with other raid level
10828 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
10829 info
->array
.level
);
10832 /* check for platform support
10833 * for this raid level configuration
10835 struct intel_super
*super
= st
->sb
;
10836 if (!is_raid_level_supported(super
->orom
,
10837 member
->array
.level
,
10838 geo
->raid_disks
)) {
10839 dprintf("platform does not support raid%d with %d disk%s\n",
10842 geo
->raid_disks
> 1 ? "s" : "");
10845 /* check if component size is aligned to chunk size
10847 if (info
->component_size
%
10848 (info
->array
.chunk_size
/512)) {
10849 dprintf("Component size is not aligned to chunk size\n");
10854 if (*old_raid_disks
&&
10855 info
->array
.raid_disks
!= *old_raid_disks
)
10857 *old_raid_disks
= info
->array
.raid_disks
;
10859 /* All raid5 and raid0 volumes in container
10860 * have to be ready for Online Capacity Expansion
10861 * so they need to be assembled. We have already
10862 * checked that no recovery etc is happening.
10864 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
10865 st
->container_devnm
);
10866 if (result
== NULL
) {
10867 dprintf("imsm: cannot find array\n");
10870 devices_that_can_grow
++;
10873 if (!member
&& devices_that_can_grow
)
10877 dprintf("Container operation allowed\n");
10879 dprintf("Error: %i\n", ret_val
);
10884 /* Function: get_spares_for_grow
10885 * Description: Allocates memory and creates list of spare devices
10886 * avaliable in container. Checks if spare drive size is acceptable.
10887 * Parameters: Pointer to the supertype structure
10888 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
10891 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
10893 struct spare_criteria sc
;
10895 get_spare_criteria_imsm(st
, &sc
);
10896 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
10899 /******************************************************************************
10900 * function: imsm_create_metadata_update_for_reshape
10901 * Function creates update for whole IMSM container.
10903 ******************************************************************************/
10904 static int imsm_create_metadata_update_for_reshape(
10905 struct supertype
*st
,
10906 struct geo_params
*geo
,
10907 int old_raid_disks
,
10908 struct imsm_update_reshape
**updatep
)
10910 struct intel_super
*super
= st
->sb
;
10911 struct imsm_super
*mpb
= super
->anchor
;
10912 int update_memory_size
;
10913 struct imsm_update_reshape
*u
;
10914 struct mdinfo
*spares
;
10917 struct mdinfo
*dev
;
10919 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
10921 delta_disks
= geo
->raid_disks
- old_raid_disks
;
10923 /* size of all update data without anchor */
10924 update_memory_size
= sizeof(struct imsm_update_reshape
);
10926 /* now add space for spare disks that we need to add. */
10927 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
10929 u
= xcalloc(1, update_memory_size
);
10930 u
->type
= update_reshape_container_disks
;
10931 u
->old_raid_disks
= old_raid_disks
;
10932 u
->new_raid_disks
= geo
->raid_disks
;
10934 /* now get spare disks list
10936 spares
= get_spares_for_grow(st
);
10938 if (spares
== NULL
|| delta_disks
> spares
->array
.spare_disks
) {
10939 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
10944 /* we have got spares
10945 * update disk list in imsm_disk list table in anchor
10947 dprintf("imsm: %i spares are available.\n\n",
10948 spares
->array
.spare_disks
);
10950 dev
= spares
->devs
;
10951 for (i
= 0; i
< delta_disks
; i
++) {
10956 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
10958 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
10959 dl
->index
= mpb
->num_disks
;
10967 sysfs_free(spares
);
10969 dprintf("imsm: reshape update preparation :");
10970 if (i
== delta_disks
) {
10971 dprintf_cont(" OK\n");
10973 return update_memory_size
;
10976 dprintf_cont(" Error\n");
10981 /******************************************************************************
10982 * function: imsm_create_metadata_update_for_size_change()
10983 * Creates update for IMSM array for array size change.
10985 ******************************************************************************/
10986 static int imsm_create_metadata_update_for_size_change(
10987 struct supertype
*st
,
10988 struct geo_params
*geo
,
10989 struct imsm_update_size_change
**updatep
)
10991 struct intel_super
*super
= st
->sb
;
10992 int update_memory_size
;
10993 struct imsm_update_size_change
*u
;
10995 dprintf("(enter) New size = %llu\n", geo
->size
);
10997 /* size of all update data without anchor */
10998 update_memory_size
= sizeof(struct imsm_update_size_change
);
11000 u
= xcalloc(1, update_memory_size
);
11001 u
->type
= update_size_change
;
11002 u
->subdev
= super
->current_vol
;
11003 u
->new_size
= geo
->size
;
11005 dprintf("imsm: reshape update preparation : OK\n");
11008 return update_memory_size
;
11011 /******************************************************************************
11012 * function: imsm_create_metadata_update_for_migration()
11013 * Creates update for IMSM array.
11015 ******************************************************************************/
11016 static int imsm_create_metadata_update_for_migration(
11017 struct supertype
*st
,
11018 struct geo_params
*geo
,
11019 struct imsm_update_reshape_migration
**updatep
)
11021 struct intel_super
*super
= st
->sb
;
11022 int update_memory_size
;
11023 struct imsm_update_reshape_migration
*u
;
11024 struct imsm_dev
*dev
;
11025 int previous_level
= -1;
11027 dprintf("(enter) New Level = %i\n", geo
->level
);
11029 /* size of all update data without anchor */
11030 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
11032 u
= xcalloc(1, update_memory_size
);
11033 u
->type
= update_reshape_migration
;
11034 u
->subdev
= super
->current_vol
;
11035 u
->new_level
= geo
->level
;
11036 u
->new_layout
= geo
->layout
;
11037 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
11038 u
->new_disks
[0] = -1;
11039 u
->new_chunksize
= -1;
11041 dev
= get_imsm_dev(super
, u
->subdev
);
11043 struct imsm_map
*map
;
11045 map
= get_imsm_map(dev
, MAP_0
);
11047 int current_chunk_size
=
11048 __le16_to_cpu(map
->blocks_per_strip
) / 2;
11050 if (geo
->chunksize
!= current_chunk_size
) {
11051 u
->new_chunksize
= geo
->chunksize
/ 1024;
11052 dprintf("imsm: chunk size change from %i to %i\n",
11053 current_chunk_size
, u
->new_chunksize
);
11055 previous_level
= map
->raid_level
;
11058 if (geo
->level
== 5 && previous_level
== 0) {
11059 struct mdinfo
*spares
= NULL
;
11061 u
->new_raid_disks
++;
11062 spares
= get_spares_for_grow(st
);
11063 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11065 sysfs_free(spares
);
11066 update_memory_size
= 0;
11067 pr_err("cannot get spare device for requested migration\n");
11070 sysfs_free(spares
);
11072 dprintf("imsm: reshape update preparation : OK\n");
11075 return update_memory_size
;
11078 static void imsm_update_metadata_locally(struct supertype
*st
,
11079 void *buf
, int len
)
11081 struct metadata_update mu
;
11086 mu
.space_list
= NULL
;
11088 if (imsm_prepare_update(st
, &mu
))
11089 imsm_process_update(st
, &mu
);
11091 while (mu
.space_list
) {
11092 void **space
= mu
.space_list
;
11093 mu
.space_list
= *space
;
11098 /***************************************************************************
11099 * Function: imsm_analyze_change
11100 * Description: Function analyze change for single volume
11101 * and validate if transition is supported
11102 * Parameters: Geometry parameters, supertype structure,
11103 * metadata change direction (apply/rollback)
11104 * Returns: Operation type code on success, -1 if fail
11105 ****************************************************************************/
11106 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11107 struct geo_params
*geo
,
11110 struct mdinfo info
;
11112 int check_devs
= 0;
11114 /* number of added/removed disks in operation result */
11115 int devNumChange
= 0;
11116 /* imsm compatible layout value for array geometry verification */
11117 int imsm_layout
= -1;
11119 struct imsm_dev
*dev
;
11120 struct intel_super
*super
;
11121 unsigned long long current_size
;
11122 unsigned long long free_size
;
11123 unsigned long long max_size
;
11126 getinfo_super_imsm_volume(st
, &info
, NULL
);
11127 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11128 geo
->level
!= UnSet
) {
11129 switch (info
.array
.level
) {
11131 if (geo
->level
== 5) {
11132 change
= CH_MIGRATION
;
11133 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11134 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11136 goto analyse_change_exit
;
11138 imsm_layout
= geo
->layout
;
11140 devNumChange
= 1; /* parity disk added */
11141 } else if (geo
->level
== 10) {
11142 change
= CH_TAKEOVER
;
11144 devNumChange
= 2; /* two mirrors added */
11145 imsm_layout
= 0x102; /* imsm supported layout */
11150 if (geo
->level
== 0) {
11151 change
= CH_TAKEOVER
;
11153 devNumChange
= -(geo
->raid_disks
/2);
11154 imsm_layout
= 0; /* imsm raid0 layout */
11158 if (change
== -1) {
11159 pr_err("Error. Level Migration from %d to %d not supported!\n",
11160 info
.array
.level
, geo
->level
);
11161 goto analyse_change_exit
;
11164 geo
->level
= info
.array
.level
;
11166 if (geo
->layout
!= info
.array
.layout
&&
11167 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11168 change
= CH_MIGRATION
;
11169 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11170 geo
->layout
== 5) {
11171 /* reshape 5 -> 4 */
11172 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11173 geo
->layout
== 0) {
11174 /* reshape 4 -> 5 */
11178 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11179 info
.array
.layout
, geo
->layout
);
11181 goto analyse_change_exit
;
11184 geo
->layout
= info
.array
.layout
;
11185 if (imsm_layout
== -1)
11186 imsm_layout
= info
.array
.layout
;
11189 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11190 geo
->chunksize
!= info
.array
.chunk_size
) {
11191 if (info
.array
.level
== 10) {
11192 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11194 goto analyse_change_exit
;
11195 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11196 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11197 geo
->chunksize
/1024, info
.component_size
/2);
11199 goto analyse_change_exit
;
11201 change
= CH_MIGRATION
;
11203 geo
->chunksize
= info
.array
.chunk_size
;
11206 chunk
= geo
->chunksize
/ 1024;
11209 dev
= get_imsm_dev(super
, super
->current_vol
);
11210 data_disks
= imsm_num_data_members(dev
, MAP_0
);
11211 /* compute current size per disk member
11213 current_size
= info
.custom_array_size
/ data_disks
;
11215 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11216 /* align component size
11218 geo
->size
= imsm_component_size_aligment_check(
11219 get_imsm_raid_level(dev
->vol
.map
),
11220 chunk
* 1024, super
->sector_size
,
11222 if (geo
->size
== 0) {
11223 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11225 goto analyse_change_exit
;
11229 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11230 if (change
!= -1) {
11231 pr_err("Error. Size change should be the only one at a time.\n");
11233 goto analyse_change_exit
;
11235 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11236 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11237 super
->current_vol
, st
->devnm
);
11238 goto analyse_change_exit
;
11240 /* check the maximum available size
11242 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11243 0, chunk
, &free_size
);
11245 /* Cannot find maximum available space
11249 max_size
= free_size
+ current_size
;
11250 /* align component size
11252 max_size
= imsm_component_size_aligment_check(
11253 get_imsm_raid_level(dev
->vol
.map
),
11254 chunk
* 1024, super
->sector_size
,
11257 if (geo
->size
== MAX_SIZE
) {
11258 /* requested size change to the maximum available size
11260 if (max_size
== 0) {
11261 pr_err("Error. Cannot find maximum available space.\n");
11263 goto analyse_change_exit
;
11265 geo
->size
= max_size
;
11268 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11269 /* accept size for rollback only
11272 /* round size due to metadata compatibility
11274 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11275 << SECT_PER_MB_SHIFT
;
11276 dprintf("Prepare update for size change to %llu\n",
11278 if (current_size
>= geo
->size
) {
11279 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11280 current_size
, geo
->size
);
11281 goto analyse_change_exit
;
11283 if (max_size
&& geo
->size
> max_size
) {
11284 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11285 max_size
, geo
->size
);
11286 goto analyse_change_exit
;
11289 geo
->size
*= data_disks
;
11290 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11291 change
= CH_ARRAY_SIZE
;
11293 if (!validate_geometry_imsm(st
,
11296 geo
->raid_disks
+ devNumChange
,
11298 geo
->size
, INVALID_SECTORS
,
11299 0, 0, info
.consistency_policy
, 1))
11303 struct intel_super
*super
= st
->sb
;
11304 struct imsm_super
*mpb
= super
->anchor
;
11306 if (mpb
->num_raid_devs
> 1) {
11307 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11313 analyse_change_exit
:
11314 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11315 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11316 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11322 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11324 struct intel_super
*super
= st
->sb
;
11325 struct imsm_update_takeover
*u
;
11327 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11329 u
->type
= update_takeover
;
11330 u
->subarray
= super
->current_vol
;
11332 /* 10->0 transition */
11333 if (geo
->level
== 0)
11334 u
->direction
= R10_TO_R0
;
11336 /* 0->10 transition */
11337 if (geo
->level
== 10)
11338 u
->direction
= R0_TO_R10
;
11340 /* update metadata locally */
11341 imsm_update_metadata_locally(st
, u
,
11342 sizeof(struct imsm_update_takeover
));
11343 /* and possibly remotely */
11344 if (st
->update_tail
)
11345 append_metadata_update(st
, u
,
11346 sizeof(struct imsm_update_takeover
));
11353 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11355 int layout
, int chunksize
, int raid_disks
,
11356 int delta_disks
, char *backup
, char *dev
,
11357 int direction
, int verbose
)
11360 struct geo_params geo
;
11362 dprintf("(enter)\n");
11364 memset(&geo
, 0, sizeof(struct geo_params
));
11366 geo
.dev_name
= dev
;
11367 strcpy(geo
.devnm
, st
->devnm
);
11370 geo
.layout
= layout
;
11371 geo
.chunksize
= chunksize
;
11372 geo
.raid_disks
= raid_disks
;
11373 if (delta_disks
!= UnSet
)
11374 geo
.raid_disks
+= delta_disks
;
11376 dprintf("for level : %i\n", geo
.level
);
11377 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11379 if (experimental() == 0)
11382 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11383 /* On container level we can only increase number of devices. */
11384 dprintf("imsm: info: Container operation\n");
11385 int old_raid_disks
= 0;
11387 if (imsm_reshape_is_allowed_on_container(
11388 st
, &geo
, &old_raid_disks
, direction
)) {
11389 struct imsm_update_reshape
*u
= NULL
;
11392 len
= imsm_create_metadata_update_for_reshape(
11393 st
, &geo
, old_raid_disks
, &u
);
11396 dprintf("imsm: Cannot prepare update\n");
11397 goto exit_imsm_reshape_super
;
11401 /* update metadata locally */
11402 imsm_update_metadata_locally(st
, u
, len
);
11403 /* and possibly remotely */
11404 if (st
->update_tail
)
11405 append_metadata_update(st
, u
, len
);
11410 pr_err("(imsm) Operation is not allowed on this container\n");
11413 /* On volume level we support following operations
11414 * - takeover: raid10 -> raid0; raid0 -> raid10
11415 * - chunk size migration
11416 * - migration: raid5 -> raid0; raid0 -> raid5
11418 struct intel_super
*super
= st
->sb
;
11419 struct intel_dev
*dev
= super
->devlist
;
11421 dprintf("imsm: info: Volume operation\n");
11422 /* find requested device */
11425 imsm_find_array_devnm_by_subdev(
11426 dev
->index
, st
->container_devnm
);
11427 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11432 pr_err("Cannot find %s (%s) subarray\n",
11433 geo
.dev_name
, geo
.devnm
);
11434 goto exit_imsm_reshape_super
;
11436 super
->current_vol
= dev
->index
;
11437 change
= imsm_analyze_change(st
, &geo
, direction
);
11440 ret_val
= imsm_takeover(st
, &geo
);
11442 case CH_MIGRATION
: {
11443 struct imsm_update_reshape_migration
*u
= NULL
;
11445 imsm_create_metadata_update_for_migration(
11448 dprintf("imsm: Cannot prepare update\n");
11452 /* update metadata locally */
11453 imsm_update_metadata_locally(st
, u
, len
);
11454 /* and possibly remotely */
11455 if (st
->update_tail
)
11456 append_metadata_update(st
, u
, len
);
11461 case CH_ARRAY_SIZE
: {
11462 struct imsm_update_size_change
*u
= NULL
;
11464 imsm_create_metadata_update_for_size_change(
11467 dprintf("imsm: Cannot prepare update\n");
11471 /* update metadata locally */
11472 imsm_update_metadata_locally(st
, u
, len
);
11473 /* and possibly remotely */
11474 if (st
->update_tail
)
11475 append_metadata_update(st
, u
, len
);
11485 exit_imsm_reshape_super
:
11486 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11490 #define COMPLETED_OK 0
11491 #define COMPLETED_NONE 1
11492 #define COMPLETED_DELAYED 2
11494 static int read_completed(int fd
, unsigned long long *val
)
11499 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11503 ret
= COMPLETED_OK
;
11504 if (strncmp(buf
, "none", 4) == 0) {
11505 ret
= COMPLETED_NONE
;
11506 } else if (strncmp(buf
, "delayed", 7) == 0) {
11507 ret
= COMPLETED_DELAYED
;
11510 *val
= strtoull(buf
, &ep
, 0);
11511 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11517 /*******************************************************************************
11518 * Function: wait_for_reshape_imsm
11519 * Description: Function writes new sync_max value and waits until
11520 * reshape process reach new position
11522 * sra : general array info
11523 * ndata : number of disks in new array's layout
11526 * 1 : there is no reshape in progress,
11528 ******************************************************************************/
11529 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11531 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11533 unsigned long long completed
;
11534 /* to_complete : new sync_max position */
11535 unsigned long long to_complete
= sra
->reshape_progress
;
11536 unsigned long long position_to_set
= to_complete
/ ndata
;
11539 dprintf("cannot open reshape_position\n");
11544 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11546 dprintf("cannot read reshape_position (no reshape in progres)\n");
11555 if (completed
> position_to_set
) {
11556 dprintf("wrong next position to set %llu (%llu)\n",
11557 to_complete
, position_to_set
);
11561 dprintf("Position set: %llu\n", position_to_set
);
11562 if (sysfs_set_num(sra
, NULL
, "sync_max",
11563 position_to_set
) != 0) {
11564 dprintf("cannot set reshape position to %llu\n",
11573 int timeout
= 3000;
11575 sysfs_wait(fd
, &timeout
);
11576 if (sysfs_get_str(sra
, NULL
, "sync_action",
11578 strncmp(action
, "reshape", 7) != 0) {
11579 if (strncmp(action
, "idle", 4) == 0)
11585 rc
= read_completed(fd
, &completed
);
11587 dprintf("cannot read reshape_position (in loop)\n");
11590 } else if (rc
== COMPLETED_NONE
)
11592 } while (completed
< position_to_set
);
11598 /*******************************************************************************
11599 * Function: check_degradation_change
11600 * Description: Check that array hasn't become failed.
11602 * info : for sysfs access
11603 * sources : source disks descriptors
11604 * degraded: previous degradation level
11606 * degradation level
11607 ******************************************************************************/
11608 int check_degradation_change(struct mdinfo
*info
,
11612 unsigned long long new_degraded
;
11615 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11616 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11617 /* check each device to ensure it is still working */
11620 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11621 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11623 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11626 if (sysfs_get_str(info
,
11627 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11628 strstr(sbuf
, "faulty") ||
11629 strstr(sbuf
, "in_sync") == NULL
) {
11630 /* this device is dead */
11631 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11632 if (sd
->disk
.raid_disk
>= 0 &&
11633 sources
[sd
->disk
.raid_disk
] >= 0) {
11635 sd
->disk
.raid_disk
]);
11636 sources
[sd
->disk
.raid_disk
] =
11645 return new_degraded
;
11648 /*******************************************************************************
11649 * Function: imsm_manage_reshape
11650 * Description: Function finds array under reshape and it manages reshape
11651 * process. It creates stripes backups (if required) and sets
11654 * afd : Backup handle (nattive) - not used
11655 * sra : general array info
11656 * reshape : reshape parameters - not used
11657 * st : supertype structure
11658 * blocks : size of critical section [blocks]
11659 * fds : table of source device descriptor
11660 * offsets : start of array (offest per devices)
11662 * destfd : table of destination device descriptor
11663 * destoffsets : table of destination offsets (per device)
11665 * 1 : success, reshape is done
11667 ******************************************************************************/
11668 static int imsm_manage_reshape(
11669 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
11670 struct supertype
*st
, unsigned long backup_blocks
,
11671 int *fds
, unsigned long long *offsets
,
11672 int dests
, int *destfd
, unsigned long long *destoffsets
)
11675 struct intel_super
*super
= st
->sb
;
11676 struct intel_dev
*dv
;
11677 unsigned int sector_size
= super
->sector_size
;
11678 struct imsm_dev
*dev
= NULL
;
11679 struct imsm_map
*map_src
;
11680 int migr_vol_qan
= 0;
11681 int ndata
, odata
; /* [bytes] */
11682 int chunk
; /* [bytes] */
11683 struct migr_record
*migr_rec
;
11685 unsigned int buf_size
; /* [bytes] */
11686 unsigned long long max_position
; /* array size [bytes] */
11687 unsigned long long next_step
; /* [blocks]/[bytes] */
11688 unsigned long long old_data_stripe_length
;
11689 unsigned long long start_src
; /* [bytes] */
11690 unsigned long long start
; /* [bytes] */
11691 unsigned long long start_buf_shift
; /* [bytes] */
11693 int source_layout
= 0;
11698 if (!fds
|| !offsets
)
11701 /* Find volume during the reshape */
11702 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
11703 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
&&
11704 dv
->dev
->vol
.migr_state
== 1) {
11709 /* Only one volume can migrate at the same time */
11710 if (migr_vol_qan
!= 1) {
11711 pr_err("%s", migr_vol_qan
?
11712 "Number of migrating volumes greater than 1\n" :
11713 "There is no volume during migrationg\n");
11717 map_src
= get_imsm_map(dev
, MAP_1
);
11718 if (map_src
== NULL
)
11721 ndata
= imsm_num_data_members(dev
, MAP_0
);
11722 odata
= imsm_num_data_members(dev
, MAP_1
);
11724 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
11725 old_data_stripe_length
= odata
* chunk
;
11727 migr_rec
= super
->migr_rec
;
11729 /* initialize migration record for start condition */
11730 if (sra
->reshape_progress
== 0)
11731 init_migr_record_imsm(st
, dev
, sra
);
11733 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
11734 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
11737 /* Save checkpoint to update migration record for current
11738 * reshape position (in md). It can be farther than current
11739 * reshape position in metadata.
11741 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11742 /* ignore error == 2, this can mean end of reshape here
11744 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
11749 /* size for data */
11750 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
11751 /* extend buffer size for parity disk */
11752 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11753 /* add space for stripe aligment */
11754 buf_size
+= old_data_stripe_length
;
11755 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
11756 dprintf("imsm: Cannot allocate checkpoint buffer\n");
11760 max_position
= sra
->component_size
* ndata
;
11761 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
11763 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
11764 __le32_to_cpu(migr_rec
->num_migr_units
)) {
11765 /* current reshape position [blocks] */
11766 unsigned long long current_position
=
11767 __le32_to_cpu(migr_rec
->blocks_per_unit
)
11768 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
11769 unsigned long long border
;
11771 /* Check that array hasn't become failed.
11773 degraded
= check_degradation_change(sra
, fds
, degraded
);
11774 if (degraded
> 1) {
11775 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
11779 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
11781 if ((current_position
+ next_step
) > max_position
)
11782 next_step
= max_position
- current_position
;
11784 start
= current_position
* 512;
11786 /* align reading start to old geometry */
11787 start_buf_shift
= start
% old_data_stripe_length
;
11788 start_src
= start
- start_buf_shift
;
11790 border
= (start_src
/ odata
) - (start
/ ndata
);
11792 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
11793 /* save critical stripes to buf
11794 * start - start address of current unit
11795 * to backup [bytes]
11796 * start_src - start address of current unit
11797 * to backup alligned to source array
11800 unsigned long long next_step_filler
;
11801 unsigned long long copy_length
= next_step
* 512;
11803 /* allign copy area length to stripe in old geometry */
11804 next_step_filler
= ((copy_length
+ start_buf_shift
)
11805 % old_data_stripe_length
);
11806 if (next_step_filler
)
11807 next_step_filler
= (old_data_stripe_length
11808 - next_step_filler
);
11809 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
11810 start
, start_src
, copy_length
,
11811 start_buf_shift
, next_step_filler
);
11813 if (save_stripes(fds
, offsets
, map_src
->num_members
,
11814 chunk
, map_src
->raid_level
,
11815 source_layout
, 0, NULL
, start_src
,
11817 next_step_filler
+ start_buf_shift
,
11819 dprintf("imsm: Cannot save stripes to buffer\n");
11822 /* Convert data to destination format and store it
11823 * in backup general migration area
11825 if (save_backup_imsm(st
, dev
, sra
,
11826 buf
+ start_buf_shift
, copy_length
)) {
11827 dprintf("imsm: Cannot save stripes to target devices\n");
11830 if (save_checkpoint_imsm(st
, sra
,
11831 UNIT_SRC_IN_CP_AREA
)) {
11832 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
11836 /* set next step to use whole border area */
11837 border
/= next_step
;
11839 next_step
*= border
;
11841 /* When data backed up, checkpoint stored,
11842 * kick the kernel to reshape unit of data
11844 next_step
= next_step
+ sra
->reshape_progress
;
11845 /* limit next step to array max position */
11846 if (next_step
> max_position
)
11847 next_step
= max_position
;
11848 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
11849 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
11850 sra
->reshape_progress
= next_step
;
11852 /* wait until reshape finish */
11853 if (wait_for_reshape_imsm(sra
, ndata
)) {
11854 dprintf("wait_for_reshape_imsm returned error!\n");
11860 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11861 /* ignore error == 2, this can mean end of reshape here
11863 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
11869 /* clear migr_rec on disks after successful migration */
11872 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
11873 for (d
= super
->disks
; d
; d
= d
->next
) {
11874 if (d
->index
< 0 || is_failed(&d
->disk
))
11876 unsigned long long dsize
;
11878 get_dev_size(d
->fd
, NULL
, &dsize
);
11879 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
11881 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
11882 MIGR_REC_BUF_SECTORS
*sector_size
) !=
11883 MIGR_REC_BUF_SECTORS
*sector_size
)
11884 perror("Write migr_rec failed");
11888 /* return '1' if done */
11892 /* See Grow.c: abort_reshape() for further explanation */
11893 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
11894 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
11895 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
11900 struct superswitch super_imsm
= {
11901 .examine_super
= examine_super_imsm
,
11902 .brief_examine_super
= brief_examine_super_imsm
,
11903 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
11904 .export_examine_super
= export_examine_super_imsm
,
11905 .detail_super
= detail_super_imsm
,
11906 .brief_detail_super
= brief_detail_super_imsm
,
11907 .write_init_super
= write_init_super_imsm
,
11908 .validate_geometry
= validate_geometry_imsm
,
11909 .add_to_super
= add_to_super_imsm
,
11910 .remove_from_super
= remove_from_super_imsm
,
11911 .detail_platform
= detail_platform_imsm
,
11912 .export_detail_platform
= export_detail_platform_imsm
,
11913 .kill_subarray
= kill_subarray_imsm
,
11914 .update_subarray
= update_subarray_imsm
,
11915 .load_container
= load_container_imsm
,
11916 .default_geometry
= default_geometry_imsm
,
11917 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
11918 .reshape_super
= imsm_reshape_super
,
11919 .manage_reshape
= imsm_manage_reshape
,
11920 .recover_backup
= recover_backup_imsm
,
11921 .copy_metadata
= copy_metadata_imsm
,
11922 .examine_badblocks
= examine_badblocks_imsm
,
11923 .match_home
= match_home_imsm
,
11924 .uuid_from_super
= uuid_from_super_imsm
,
11925 .getinfo_super
= getinfo_super_imsm
,
11926 .getinfo_super_disks
= getinfo_super_disks_imsm
,
11927 .update_super
= update_super_imsm
,
11929 .avail_size
= avail_size_imsm
,
11930 .get_spare_criteria
= get_spare_criteria_imsm
,
11932 .compare_super
= compare_super_imsm
,
11934 .load_super
= load_super_imsm
,
11935 .init_super
= init_super_imsm
,
11936 .store_super
= store_super_imsm
,
11937 .free_super
= free_super_imsm
,
11938 .match_metadata_desc
= match_metadata_desc_imsm
,
11939 .container_content
= container_content_imsm
,
11940 .validate_container
= validate_container_imsm
,
11942 .write_init_ppl
= write_init_ppl_imsm
,
11943 .validate_ppl
= validate_ppl_imsm
,
11949 .open_new
= imsm_open_new
,
11950 .set_array_state
= imsm_set_array_state
,
11951 .set_disk
= imsm_set_disk
,
11952 .sync_metadata
= imsm_sync_metadata
,
11953 .activate_spare
= imsm_activate_spare
,
11954 .process_update
= imsm_process_update
,
11955 .prepare_update
= imsm_prepare_update
,
11956 .record_bad_block
= imsm_record_badblock
,
11957 .clear_bad_block
= imsm_clear_badblock
,
11958 .get_bad_blocks
= imsm_get_badblocks
,